Capital Inflows, Sovereign Debt and Bank Lending: Micro-Evidence from an Emerging Market

Capital Inflows, Sovereign Debt and Bank Lending: Micro-Evidence from an Emerging Market Abstract This paper uses a natural experiment to show that government access to foreign credit increases private access to credit. I identify a sudden, and unanticipated increase in capital inflows to the sovereign debt market in Colombia, due to a rebalancing in a government bond index by J.P. Morgan. I find that market makers banks in the treasury market reduced their sovereign debt by 7.8 percentage points of assets and increased their credit availability by 4.2 percentage points of assets. Using industry level data, I show that a higher exposure to market makers led to a higher growth in economic activity. Received August 17, 2017; editorial decision January 25, 2018 by Editor Philip Strahan. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online. How do public and private access to credit interact? Economic theory suggests that when governments borrow from domestic institutions this may lead to a crowding out of private credit.1 This crowding out depends crucially on public access to foreign credit. When there is little access to foreign investors, governments rely heavily on local financial institutions to absorb the issuance of debt. If foreign investors become willing to purchase sovereign debt, the reliance on local financial institutions may be reduced, freeing resources for the private sector. As a result, local firms may be able to finance investment projects and boost economic activity. Although intuitive, there is no clear evidence on this topic because of the identification challenges. Notice that sovereign risk, sovereign bond holdings of banks and foreigners, and loans to the private sector are all jointly determined. For instance, an improvement in local economic conditions may increase both foreign investor demand for sovereign debt and credit demand from local firms. In such a case, one would observe both an increase in the share of sovereign debt held by foreign investors and an increase in private credit, but this correlation would not imply causation. This common problem illustrates the difficulty of finding causal evidence on this issue. The main contribution of this paper is using a novel episode to overcome the identification problems previously encountered in the literature. By doing so, I provide clear evidence that government access to foreign credit increases private access to credit and boosts economic activity. I use a sudden, unanticipated shock that triggered the entrance of foreign investors to the local currency sovereign debt market in Colombia.2 In March 2014, J.P. Morgan announced the inclusion of several Colombian treasury bonds into its emerging markets local currency government debt index. Since many international mutual funds track their performance against this index, they changed their portfolio suddenly, directing capital flows to the Colombian local currency sovereign debt market. This shock had a number of appealing features for identification. First, the timing of the decision by J.P. Morgan appears, at least in the short window of time I study, unrelated to changes in the future economic prospects of Colombia. Second, it was sizable: the share of debt held by foreigners in this market went from 8.5 to 19 percent in only 7 months (Figure 1). Third, the event was unanticipated. Fourth, provided that the index is specific to government debt, it does not directly affect flows to the private sector. Figure 1 View largeDownload slide Foreign share of domestic government debt securities This figure presents the evolution of the share of domestic government debt securities (TES) held by foreigners. The dashed line shows a linear trend using the average growth during the 12 months prior to the announcement of the change in the index by J.P. Morgan. The gray bar represents the announcement of the rebalancing by J. P. Morgan. Figure 1 View largeDownload slide Foreign share of domestic government debt securities This figure presents the evolution of the share of domestic government debt securities (TES) held by foreigners. The dashed line shows a linear trend using the average growth during the 12 months prior to the announcement of the change in the index by J.P. Morgan. The gray bar represents the announcement of the rebalancing by J. P. Morgan. I find that the entrance of foreign investors had sizable effects on commercial banks. Moreover, it had heterogenous effects on banks according to their participation in the local currency sovereign debt market. In Colombia, the Ministry of Finance selects financial institutions to act as market makers or official intermediaries in the treasury market. Each of the intermediaries participating in the program is obliged to absorb 4.5 percent of the total debt issued by the government in the primary market. I find that market maker banks reduced their domestic sovereign debt holdings by 7.8 percentage points of assets, compared to the rest of the banks. Using data either at the city-zone or industry level, I also show that they increased differentially their commercial credit availability by 4.2 percentage points of their assets. Results show a considerable substitution between sovereign debt and commercial credit. This effect is also economically significant, around 2 percent of Colombia’s GDP. I analyze whether the shock had real effects by looking at evidence at the industry level. I obtain data on monthly employment, production and sales from the Monthly Manufacturing Polls conducted by the Departamento Administrativo Nacional de Estadistica (DANE), and data on GDP by industry from the Quarterly National Accounts database. I construct a proxy for the exposure to market maker banks at the industry level, and find that industries more exposed to market makers had higher growth of employment, production, sales and GDP during this period. I conduct several robustness checks to confirm the results. I estimate the cross-sectional coefficients of a regression of credit growth on a dummy variable indicating whether a bank is a market maker or not. I find that the coefficients are only statistically significant during the rebalancing, showing an important support for the identification strategy. I also conduct an analyisis on the potential omitted variable bias using tests initially developed by Altonji et al. (2005) and later formalized by Oster (2014). I find that omitted variables are not likely to be biasing the results. Moreover, I show that the effect on credit growth is not driven by valuation effects on the balance sheet of banks and is likely coming from a change in the access to credit by banks. Finally, the results are robust when I use a matching estimator using the geographical areas and bank observable characteristics such as assets and capital ratios to match market makers and non market makers. The evidence is consistent with the following narrative. Before the entrance of foreign investors, there was a crowding out of private credit. The domestic sovereign debt market was dominated by local participants and the investor base was undiversified. Therefore, the government used market makers to absorb debt issued in the primary market. These financial institutions were paid a premium to absorb this debt, since the secondary market was less liquid, and thus market makers kept part of the issued debt in their balance sheet. As foreign institutional investors entered the domestic sovereign debt market, the government reduced the premium it was paying to market makers, and these financial institutions sold the excess of debt that could not offload before and used the proceeds to extend credit. As defined by Reinhart and Sbrancia (2011) this could be a form of financial repression, as the government pays banks to aborb (and ultimately) hold government debt and this restricts the extension of credit.3 This paper contributes to two broad strands of literature. From a macroeconomic perspective, it is related to the literature on the interrelationship between the supply of credit to public and private sectors. This line of research has received a lot of attention during the recent European Sovereign Debt Crisis, emphasizing two different mechanisms. On the one hand, there is a line of research highlighting that an increase in the home bias of sovereign debt holdings crowds out private credit. For instance, Broner et al. (2014) propose a model with creditor discrimination and crowding out effects that accounts for the reallocation of credit from the private to the public sector observed in the euro zone periphery during the European Sovereign Debt crisis. Becker and Ivashina (2014) and Altavilla et al. (2015) also provide empirical evidence consistent with this reallocation channel. My paper paper is closely related to these two studies since it explores how financial repression affects the extension of credit to the private sector. On the other hand, there is a part of the literature that emphasizes how shifts in sovereign risk affect the balance sheets of banks. For example, Bolton and Jeanne (2011), Gennaioli et al. (2014) and Perez (2015) propose models in which sovereign defaults hurt the balance sheet of banks and reduce private credit. From an empirical point of view, Bofondi et al. (2013), Acharya et al. (2014), Gennaioli et al. (2014), and Baskaya and Kalemli-Ozcan (2016) present evidence consistent with this channel.4 Most of these papers have problems identifying a shock that exclusively affects foreign demand for sovereign debt. I provide such a shock and to the best of my knowledge this paper is the first to use this type of event and separate between the two channels highlighted by the literature. Second, from a finance perspective, this paper contributes to a growing literature on the aggregate effects of institutional investors. Since the Global Financial Crisis, there has been an increased interest in the activities of financial intermediaries other than traditional banks. Investment activities by mutual funds have been at the core of the discussion and index-tracking funds have received special attention because of their exponential growth in size.5 There have been several studies analyzing the consequences on financial markets of the presence of these funds. For instance, Chang et al. (2014) and Raddatz et al. (2015) document the price effects generated by these funds during index rebalancing periods. Sullivan and Xiong (2012), Bhattacharya et al. (2013), Ben-David et al. (2014), and Israeli et al. (2015) show that index-tracking investors increase market vulnerability and volatility.6 However, the evidence on the possible economic consequences of index-tracking investors is slim as highlighted by Wurgler (2010).7 To the best of my knowledge, this paper is the first attempt to document that international capital flows by index-tracking investors have effects on the real economy. More broadly, this paper contributes to the literature on the relationship between international capital flows, credit booms and economic activity. On the one hand, there are several studies analyzing whether capital inflows lead to higher credit growth and an increase in economic activity.8 On the other hand, there are several studies analyzing the relationship between large capital inflows or capital account liberalizations and the consequences for the economy.9 Most of these studies have problems addressing endogeneity issues, since capital flows are almost always related to local economic conditions. I contribute to this literature by using a sudden and unanticipated increase in capital inflows for identification. Thus, I provide evidence that capital inflows to the sovereign debt market cause an increase in credit growth and an expansion in economic activity. The rest of the paper is organized as follows. Section 1 provides the empirical setting, with a detailed account of the index rebalancing. Section 2 provides evidence on the consequences of the event for the sovereign debt market in Colombia. Section 3 presents the identification strategy and empirical analysis for the results on bank lending. Section 4 shows the results for the real economic activity. Section 5 discusses the external validity of the results. Section 6 concludes. 1. Empirical Setting 1.1 Indexing in International Markets International indexes (or international benchmarks) are broad market indexes of different assets that involve several countries. They are constructed by different companies (index providers) such as Morgan Stanley Capital International (MSCI) for international equities or J.P. Morgan for international debt securities. The former constructs, for instance, the MSCI Emerging Markets Index and, the latter, the J.P. Morgan Emerging Markets Bond Index (EMBI), two of the most recognized indexes in the world for emerging countries. The construction process for these indexes involves different broad steps that are used by almost all index providers. They first define the main scope of a benchmark index (such as, geography, industry, and type of firms) and in which category each country is classified at each point in time (developed, emerging, or frontier). Then, they select a number of securities that fall within the scope and meet the size, market capitalization, liquidity, and other requirements. Each of these securities gets a loading (or inclusion factor) in the index portfolio assigned by the index producer according to how much it meets the index-construction criteria and how accessible it is to investors (given by the free-float market capitalization, restrictions to foreign investors, and so forth). The return of the index consists of the returns of its constituent securities, using various approaches to aggregate fluctuations in individual instruments (e.g., Laspeyres, chain-weighting). Namely, each index captures the market capitalization weighted returns of all constituents included in the index.10 The indexes are periodically rebalanced to ensure their continuity and representativeness. Countries’ weights in a specific index are assembled with the portfolio weights of individual securities included in a benchmark index, aggregated at the country level according to the market where the security was issued. That is, international benchmark indexes are typically constructed using a bottom-up approach and consist of composite stock or bond market indexes that include securities from many countries as constituents. The market for local currency sovereign debt indexes is mainly dominated by the World Government Bond Index (WGBI) by Citigroup and the Government Bond Index Emerging Markets (GBI-EM) by J.P. Morgan. The former is a local currency government bond index that includes securities mainly from developed markets. The latter only includes emerging market government debt in local currency.11 While many more funds track the WGBI (approximately 1.5 trillions U.S. dollars) than the GBI-EM (200 billions U.S. dollars), the weights of emerging countries significantly differ in both indexes. For instance, Mexico (one of the few emerging countries included in both indexes) has a weight of around 0.7 percent in the WGBI and of 10 percent in the GBI-EM. Thus, the exposure of emerging markets is generally lower in the WGBI. These indexes have become popular and are frequently used as benchmarks by international mutual funds, which manage a significant part of international assets. By helping alleviate agency problems, benchmarks allow the underlying investors and supervisors to evaluate and discipline fund managers on a short-run basis using, for example, the tracking error of the fund (the deviation of its returns from the benchmark returns). To the extent that the investment strategy of these funds is pinned down by the composition of their benchmark indexes, changes in the weights that a popular benchmark gives to different countries can trigger a similar rebalancing among the funds that track it and result in sizeable movements in international portfolio allocations, capital flows and asset prices. 1.2 Benchmark Change in Colombia On March 19th 2014, J.P. Morgan announced the inclusion of five Colombian bonds into its benchmark indexes. J.P. Morgan constructs three type of major international indexes: (i) foreign currency denominated sovereign debt; (ii) local currency denominated sovereign debt and; (iii) corporate debt. The addition of these bonds involved only local currency sovereign debt indexes, namely the Government Bond Emerging Markets Indexes (GBI-EM). The securities introduced were treasury bonds (named TES) issued by the Colombian government with maturities in 2016, 2018, 2022, 2024 and 2028. The process was done in a phased approach starting at the end of May 2014 and finishing at the end of September 2014. The most popular index, the GBI-EM Global Diversified saw a large rebalancing of Colombia’s benchmark weight.12 It went from nearly 3 to almost 8 percent, representing the largest restructuring by J.P. Morgan in one of its indexes. At the time of the announcement there were estimations of 10 billions U.S. dollars in inflows into the Colombian government debt market with an estimated outstanding debt of 90 billions U.S. dollars.13 This event has several appealing features from an identification perspective. First, the timing of the decision by J.P. Morgan appears to be arbitrary. The reason for Colombia’s inclusion in the index revolves around an improvement in market accessibility and transparency. The note provided to investors by J.P. Morgan stated: “As a result of improved transparency and accesibility for international investors in the local TES market, Colombia sufficiently meets inclusion requirements for complete GBI-EM inclusion.” This note did not mention a specific policy change as the trigger for this inclusion. At the time of the announcement, many newspapers highlighted that this decision could have been motivated by Law 1607 of January 2013, which reduced taxes on foreign investors’ earnings from domestic securities from 33 to 14 percent. Of course, it is impossible to say that this law and more broadly improving economic and institutional conditions in Colombia did not play a role in the decision by J.P. Morgan. However, J.P. Morgan only announced the inclusion of these bonds in March 2014, despite having monthly periodical rebalancings. The gap of more than 1 year between the change in the law and the benchmark change by J.P. Morgan with the use of quarterly data allows me to exploit the timing of the decision to separate the two events. From an identification perspective, the decision by J.P. Morgan should not be related to other changes affecting the foreign demand for sovereign debt in the quarters from March 2014 until September 2014 (the whole process of the rebalancing by J.P. Morgan).14 Second, the event was unanticipated by the markets. Around the years 2013-2014 there were many events affecting Colombia as a country. Then, it is useful to observe the evolution of the price of domestic sovereign bonds in Colombia. (Figure 2, Panel A). The tax reform seems to have a positive impact on the price of these bonds (while part of this upward trend was due to a global factor driving up bond prices of emerging markets during large part of 2012). Not long after the tax reform, the Federal Reserve of the United States started considering unwinding quantitative easing (Taper Tantrum). The beginning of this event was marked by the Federal Reserve Chairman Ben Bernanke’s suggestion of this unwinding in his testimony before Congress on May 22, 2013. Bond prices of emerging markets dropped sharply around that period, and Colombia’s bonds were not an exception as they experienced a drop of nearly 12 percent during that summer. Prices remained low during the rest of 2013, mostly due to the uncertainty generated by Taper Tantrum talks. Upon the announcement by J.P. Morgan in March, Colombian bonds increased in price by almost 5 percent in two weeks. Most of this gain was reversed by the end of the rebalancing in October. The pattern in the price of these bonds suggests that the event was unexpected as there was a sharp and sudden increase upon its announcement.15 Figure 2 View largeDownload slide Domestic sovereign bonds index and capital inflows This figure presents the evolution of the price of the domestic sovereign bonds in Colombia and the capital inflows to the Balance of Payments in Colombia. Panel A shows the bond prices. The bond price index weights the price return of each bond by its outstanding amount. Only the bonds included in the index rebalancing by J. P. Morgan are included and the index equals 100 at the day before the announcement of the rebalancing (March 18, 2014). Panel B depicts the gross inflows from balance of payments data. The dark bars depicts the average inflows during the period 2011–2013, and the gray bars show the inflows during 2014, the year of J. P. Morgan’s index rebalancing. All values are in percentage of nominal GDP. FDI is foreign direct investment; private portfolio flows are liability flows in private portfolio debt and equity; and public debt inflows are liabiity flows to government debt securities. Figure 2 View largeDownload slide Domestic sovereign bonds index and capital inflows This figure presents the evolution of the price of the domestic sovereign bonds in Colombia and the capital inflows to the Balance of Payments in Colombia. Panel A shows the bond prices. The bond price index weights the price return of each bond by its outstanding amount. Only the bonds included in the index rebalancing by J. P. Morgan are included and the index equals 100 at the day before the announcement of the rebalancing (March 18, 2014). Panel B depicts the gross inflows from balance of payments data. The dark bars depicts the average inflows during the period 2011–2013, and the gray bars show the inflows during 2014, the year of J. P. Morgan’s index rebalancing. All values are in percentage of nominal GDP. FDI is foreign direct investment; private portfolio flows are liability flows in private portfolio debt and equity; and public debt inflows are liabiity flows to government debt securities. Third, as the affected index contains only government debt, the event only affected capital flows to the sovereign debt market, and not private capital flows. This is further confirmed by looking at capital flows from the Balance of Payments in Colombia (Figure 2, Panel B). Foreign direct investment and private portfolio flows only experienced a slight increase in inflows relative to GDP. Instead, public debt securities had gross inflows of 0.7 percent of GDP on average before the rebalancing. This number increased to 2.9 percent during 2014, indicating an increase of almost 400 percent. Fourth, the rebalancing by J.P. Morgan did not bring a renewed appetite to issue more debt by the government. In Colombia, the government has a tight fiscal rule and debt issuance programming that could not be affected by the time J.P. Morgan announced this change. Therefore, during the period under analysis, the Colombian government maintained a relatively constant growth of its local currency debt securities, and was unable to change its expenditure pattern significantly.16 The evidence suggests that the event generated a sudden, unanticipated and sizable increase in capital inflows exclusively to domestic sovereign debt markets in Colombia. Moreover, the timing of the event, around the short window of time I study, does not seem to be contemporaneous to sizable changes that would affect the foreign demand for sovereign debt and at the same time credit to the private sector. Thus, one can use the timing of this event to understand how the entrance of foreign investors in the sovereign debt market affects the holdings of government debt by domestic agents, and the credit availability to the private sector. 2. Sovereign Debt Institutional investors that track these indexes closely were forced to rebalance their portfolio. These portfolio changes had aggregate consequences for Colombia in terms of capital flows to local currency sovereign debt as shown in Figure 3, Panel A. This figure presents the rolling 6-month net purchases of TES securities by foreigners and commercial banks.17 Prior to the announcement by J.P. Morgan there were some capital inflows to domestic sovereign debt. However, after the announcement, foreigners started massively buying local currency sovereign debt in Colombia. Purchases made between the end and the beginning of the rebalancing were 8 percent of the total outstanding local currency sovereign debt securities.18 During the same period, foreigners more than doubled their participation in the affected local currency sovereign debt market (Figure 3, Panel B). Furthermore, by the end of 2014 they were the largest holders of the affected bonds. Figure 3 View largeDownload slide Holdings of domestic sovereign bonds This figure presents the net purchases of domestic sovereign bonds in Colombia around the index rebalancing by J. P. Morgan. Panel A depicts 6-month rolling purchases by foreigners and commercial banks. The gray bars indicate the events described in the picture. Panel B shows the percentage of TES bonds held by the different economic agents in the economy before and after the rebalancing. PF, IC, and MF are pension funds, insurance companies, and domestic mutual funds, respectively. Figure 3 View largeDownload slide Holdings of domestic sovereign bonds This figure presents the net purchases of domestic sovereign bonds in Colombia around the index rebalancing by J. P. Morgan. Panel A depicts 6-month rolling purchases by foreigners and commercial banks. The gray bars indicate the events described in the picture. Panel B shows the percentage of TES bonds held by the different economic agents in the economy before and after the rebalancing. PF, IC, and MF are pension funds, insurance companies, and domestic mutual funds, respectively. Another interesting feature of Figure 3 is the different agents that were on the other side of the purchases of domestic sovereign debt by foreigners. Commercial banks, with relatively stable purchases before the announcement, started selling treasury securities in an image that mirrors the one by foreigners (Panel A). Compared to the rest of the agents in the economy, commercial banks were the main providers of liquidity during the rebalancing (Panel B). Out of the 10 percentage point increase in the participation of foreigners in this market, 7 were accomodated by banks (almost a 30 percent decline in their participation in the TES market). Alternatively, pension funds, insurance companies and domestic mutual funds only reduced their participation by 0.3 percentage points (1 percent decline), while public institutions reduced their share in this market by 2.5 percentage points (7 percent decline). Both of these figures suggest that commercial banks in Colombia reduced their holdings of affected bonds by much more than the rest of the agents in the economy. 2.1 Market Makers in TES bonds In Colombia, commercial banks are important participants in the sovereign debt market. Before the benchmark change, banks were the largest participant in the TES market that was affected by the rebalancing. They held collectively around 25 percent of government issued bonds. Moreover, around 11 percent of their assets were local currency sovereign bonds as of December 2013. As noted above, commercial banks were on the other side of the transactions from foreigners in the market for TES bonds during the rebalancing. This suggests that there might be a special feature about commercial banks that may have generated this large response. One noticeable feature of the market for TES in Colombia is that the Finance Ministry designates official market makers in this market. By law, these market makers are within the set of commercial banks, financial corporations and brokerage firms. The objective of this program is to promote adequate conditions for the financing of the government in capital markets by developing the domestic sovereign debt market. Every year, each institution decides whether it wants to participate in the program or not. In order to participate, they need to fulfil a minimum net worth and corporate rating criteria. Then, the Finance Ministry ranks institutions according to their activity in primary and secondary debt markets and designates the official market makers. At the end of every year, a maximum of 20 entities are designated as such. There are two main obligations for these institutions. First, they need to absorb at least 4.5 percent of all primary market debt issuances during the year. Second, they need to quote permanently and simultaneously bid and ask prices in secondary markets subject to a maximum bid-ask price determined by the government.19 On the other hand, designated market makers benefit from having special access to debt issuances from the government, constant access to officials from the Ministry of Finance, and access to a liquidity window in case of problems.20 Under this program, at the end of 2013, 9 commercial banks in Colombia were designated as market makers among 14 participants in the program. 2.2 Conceptual Framework The difference between commercial banks that are designated official market makers and those that are not could be insightful to understand the channel at work during the benchmark rebalancing. These financial institutions are obliged to absorb certain amount of debt issued by the government in the primary market. Every institution by law has to be awarded at least 4.5 percent of total debt issued by the government during the year. For this service, they get a benefit in the form of participating in non-competitive auctions (at lower prices than secondary markets), and potential liquidity assistance if needed. In the case of both frequent auctions and the absence of a diversified investor base in the secondary market, these banks have a lower probability of offloading this debt to other investors. As a result, they keep a considerable amount of it in their balance sheet. After the entrance of foreign investors, these banks can offload debt absorbed in the primary auction more easily and can use the proceeds for other purposes. The type of foreign institutional investors that enter the market is also important. In this episode, these were index-tracking investors. Moreover, the indexes affected (GBI-EM) are market capitalization weighted indexes. As such, every time the government issues this type of debt, it gets included into these indexes, and index-tracking investors have to inmediately buy it, creating an almost certain demand in the secondary market for this debt. This reduces the need for the government to rely on market makers since it increases the probability that market maker banks offload the debt from the primary market. This would predict both a decrease in the holdings of sovereign debt by market makers, and a reduction in the discount the government gives market maker in primary auctions. Figure 4 presents evidence in this direction. Before the rebalancing, the average holdings of market makers were considerably higher than those of the rest of the banks. Between the announcement and the end of J.P. Morgan’s index rebalancing, the ratio of local currency sovereign debt to total assets decreases considerably for commercial banks that are market makers, while it is almost the same for the rest of the banks. This effect is driven entirely by reductions in sovereign debt positions.21 This illustration is consistent with regression results in Table 1. Market makers reduced differentially their exposure to government debt by 7.8 percentage points of assets during the rebalancing by J.P. Morgan.22 Figure 4 View largeDownload slide Domestic sovereign debt exposure in commercial banks This figure shows the evolution of sovereign debt over assets dividing by market maker and non-market-maker banks at the end of 2013. The index is constructed by averaging the growth of domestic debt over total assets at each point in time. The index is normalized to the average holdings of sovereign debt over assets for the two groups in February 2014. The gray bars indicate the events described in the picture. Figure 4 View largeDownload slide Domestic sovereign debt exposure in commercial banks This figure shows the evolution of sovereign debt over assets dividing by market maker and non-market-maker banks at the end of 2013. The index is constructed by averaging the growth of domestic debt over total assets at each point in time. The index is normalized to the average holdings of sovereign debt over assets for the two groups in February 2014. The gray bars indicate the events described in the picture. Table 1 Sovereign debt exposure and market makers Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 This table presents OLS estimations of the growth of local public debt to total assets against a treatment variable using balance sheet quarterly data. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013-2014. The dependent variable is winsorized at the 5th and 95th percent level. Errors are bootstrapped clustered at the bank level. *, **, and *** denote 10, 5 and 1 percent level of significance respectively. Table 1 Sovereign debt exposure and market makers Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 This table presents OLS estimations of the growth of local public debt to total assets against a treatment variable using balance sheet quarterly data. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013-2014. The dependent variable is winsorized at the 5th and 95th percent level. Errors are bootstrapped clustered at the bank level. *, **, and *** denote 10, 5 and 1 percent level of significance respectively. Figure 5 shows additional evidence consistent with this channel. After the rebalancing by J.P. Morgan there is an important reduction in the discount paid by the government to market makers in non-competitive auctions for the affected government debt securities. Both the figure and table results show a permanent decrease in this discount. This is consistent with the following narrative. As the sovereign debt market is populated by domestic investors, the government uses these financial institutions as an insurance against desserted auctions and pays them a premium to absorb the issuance of debt in primary markets. Since this debt is not easy to offload, and they receive a discount in the price they keep it in their balance sheet. When foreign investors massively enter into this market, they offload this debt, and there is a reduction in the profit of being market makers as the government does not need these financial institutions as much.23 This channel is consistent with a large literature on financial repression in emerging markets.24 Figure 5 View largeDownload slide Discount in the primary market This figure shows the evolution of the discount in noncompetitive primary market issuances of sovereign bonds in Colombia. It shows the time evolution of this discount. The discount is calculated for each bond in each issuance as the secondary market price divided by primary market price and substracting 1 from that ratio. The solid line is the average across bonds for each month with at least some issuance activity. The dashed line is the average before March 2014 (pre) and the average after the end of May 2014 (post). The gray bars indicate the events described in the picture. Figure 5 View largeDownload slide Discount in the primary market This figure shows the evolution of the discount in noncompetitive primary market issuances of sovereign bonds in Colombia. It shows the time evolution of this discount. The discount is calculated for each bond in each issuance as the secondary market price divided by primary market price and substracting 1 from that ratio. The solid line is the average across bonds for each month with at least some issuance activity. The dashed line is the average before March 2014 (pre) and the average after the end of May 2014 (post). The gray bars indicate the events described in the picture. 3. Bank Lending 3.1 Data and Identification In the previous section, I established that commercial banks that were market makers sold a sizable amount of their positions in local currency government bonds. This section presents the data and methodology to understand whether the proceeds from these sales resulted in an increase in credit by these banks. I use the following baseline specification: $$\Delta L_{ibt}=\theta_{it}+\theta_{b}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{2}X_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\varepsilon_{ibt}$$ (1) where |$L_{ijt}$| is the log of credit to total assets for a city-zone or an industry |$i$|⁠, bank |$b$| at time |$t$|⁠. |$\theta_{it}$| are fixed effects at the city-zone-time or at the industry-time level. |$X_{b}$| is a set of observable variables at the bank-level. |${\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}$| is a dummy variable that indicates whether a commercial bank was a market maker at the end of 2013 or not. |${\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}$| is a dummy variable that takes the value 1 from March until the end of September 2014.25 The identification comes from the difference-in-difference estimation of credit growth for market maker and non market maker banks. During the rebalancing, market makers were more affected than the rest of the banks, since they could sell domestic sovereign debt more easily. The rest of the banks should not be affected by this channel. Therefore, as long as the evolution of credit for both groups before the rebalancing was similar, we can use |${\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}$| as a treatment variable and analyze whether the entrance of index-tracking investors in the sovereign debt market had an effect on bank lending through this channel. |$\beta_{1}$| identifies this by comparing the differential average growth in credit between market maker and non market maker banks during the rebalancing within a city-zone or industry. By comparing observations within a city-zone or an industry I can partially control for any concerns that credit demand may be affecting this estimation.26 For all the estimations, I use standard errors bootstrapped clustered at the bank level.27 I use data from Colombia’s banking system. I gather data from Superintendencia Financiera de Colombia on credit by banks. The main database consists on different types of credit to a specific city-zone (for instance Bogota-Centro Internacional) at the bank level. I match this data on credit with balance sheet data for each bank to use different bank-level variables. I complement data with the official designation of market makers by the Finance Ministry. Data is on a quarterly basis for the 2012-2014 period and contains data for 24 commercial banks on 86 city-zones (with 10 zones). While most of the results use the city-zone credit database, I rely on an alternative database at the industry level for robustnes. This database contains information for 94 industries.28Table 2 presents a list of all the commercial banks with their classification into market makers at the end of 2013, and whether they are domestic or foreign banks.29 Table 2 Commercial banks in Colombia Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No This table shows the commercial banks in Colombia during 2013–2014 and their classification into market makers in 2013 and into foreign or domestic banks. Table 2 Commercial banks in Colombia Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No This table shows the commercial banks in Colombia during 2013–2014 and their classification into market makers in 2013 and into foreign or domestic banks. Table 3 shows the descriptive statistics of the balance sheet structure divided by whether a bank was an official market maker at the end of 2013 or not. There are substantial observable differences between the two type of banks. Market makers are larger both in assets and liabilities than non market maker banks. Market makers hold more investments and more local public debt, and thus less total credit in their asset side. On average, market makers hold 15.4 percent of their assets in local public debt, while non market makers hold 8.9 percent of their assets in local currency sovereign debt. Within credit, they seem more exposed to commercial credit, while non market makers lean more to consumer credit. Regarding balance sheet health, all Colombian banks are above the minimum solvency ratio (9 percent for the total solvency ratio) and non market maker banks have a larger solvency ratio than market-maker banks. Table 3 Balance sheet structure of commercial banks (December 2013) Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 This table presents the structure of the balance sheet for commercial banks before Colombia’s benchmark rebalancing by J.P. Morgan. Panel A depicts the asset structure, with all variables in percentage of total assets unless indicated. Panel B shows the liability structure with all variables in percentage of total liabilities unless indicated. Panel C presents other relevant variables. The exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. The solvency ratio is the tier 1 capital divided by risk-weighted assets and market risk. Table 3 Balance sheet structure of commercial banks (December 2013) Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 This table presents the structure of the balance sheet for commercial banks before Colombia’s benchmark rebalancing by J.P. Morgan. Panel A depicts the asset structure, with all variables in percentage of total assets unless indicated. Panel B shows the liability structure with all variables in percentage of total liabilities unless indicated. Panel C presents other relevant variables. The exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. The solvency ratio is the tier 1 capital divided by risk-weighted assets and market risk. 3.2 Empirical Analysis Before going in-depth into the full formal analysis, I start by estimating Equation (1) in Table 4 for the period 2013-2014 for different types of credit. Market maker banks significantly increased their total credit growth during J.P. Morgan’s rebalancing. This increase is exclusively driven by commercial credit, rather than by consumer credit. A possible explanation is that banks usually have a relationship established with firms, and this type of credit provides the next best substitute to sovereign debt in Colombia. In the rest of the paper, I will only look at commercial credit growth, since it is the driver large differential in credit between market maker and non market makers. Table 4 Credit and market makers City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 This table presents OLS estimations of the growth of credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. Other credit is the sum of consumer, housing, and micro credit. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 4 Credit and market makers City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 This table presents OLS estimations of the growth of credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. Other credit is the sum of consumer, housing, and micro credit. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 5 presents the results from the main empirical specification. When I control only for time fixed effects, there is a positive and significant differential credit growth of 4.1 percent during the rebalancing (Column 1). The coefficient and standard errors are almost the same when I include city-zone-time fixed effects (Column 2). Moreover, the R-squared goes from 1.9 percent to 13.2 percent indicating that these fixed effects are capturing an important amount of the credit demand varying at the city-zone-time level. When I include bank fixed effects, which control for any unobserved fixed bank characteristics, there is a differential average growth of 3.8 percent for market makers (Column 3). Effects estimated at the city-zone level are very similar when I use the industry database (Columns 4 to 6). The coefficients are statistically similar, and the analysis mirrors the one at the geographical level. One thing to notice is that by using the industry database I partially control for any credit demand shock that may affect tradable and non-tradable industries differently due to the exchange rate changes during the rebalancing. Table 5 Commercial credit and market makers City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5% and 1% level of significance, respectively. Table 5 Commercial credit and market makers City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5% and 1% level of significance, respectively. The economic size of these estimations are also meaningful. The results in Column 2 suggest that the differential average growth for market makers versus non market makers during the period was 12.8 percent. This implies an increase in commercial credit of 4.2 percentage points of assets. Using the total assets of market maker banks, this amounts to a differential growth in commercial credit of 13.7 trillions of Colombian Pesos.30 These numbers suggest that the shock induced an increase in commercial credit of 2.41 percent of GDP. Moreover, the estimations suggest a considerable substitution between local public debt and commercial credit. The difference between the two changes could be explained by regulatory purposes. If banks do not want to change their risk profile, they would need to increase credit by less than the decrease in sovereign debt since government debt is zero-risk weighted for regulatory purposes. 3.3 Threats to Identification 3.3.1 Parallel Trends There are important identification threats that I address in this section. The key assumption behind the difference-in-difference estimation in Equation 1 is that of parallel trends. I test for the presence of differential credit growth between market maker and non market maker banks before the rebalancing with the following specification: $$\Delta L_{ibt}=\theta_{it}+\theta_{b}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{2} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Pre Rebalancing}}_{t}\varepsilon_{ibt}$$ (2) which is very similar to Equation 1 but with the inclusion of an interaction between the market maker dummy and a dummy that indicates the period previous to the rebalancing. More specifically, this dummy is equal to 1 from June 2013-December 2013, to have the same time length of the period of the rebalancing (3 quarters). The results in Table 6 show that this interaction is not statistically significant, and quantitatively the coefficient is very close to zero. Moreover, the treatment variable remains unchanged with respect to Table 5. All in all, this provides evidence consistent with no differential prior trends in credit growth before the rebalancing period. Table 6 Commercial credit and market makers: Parallel trends test City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. Pre-rebalancing is a dummy that indicates the pre-rebalancing period of the same quarter length (June 2013–December 2013). All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 6 Commercial credit and market makers: Parallel trends test City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. Pre-rebalancing is a dummy that indicates the pre-rebalancing period of the same quarter length (June 2013–December 2013). All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. To further illustrate this result, I estimate a cross-sectional version of Equation (1). More specifically, I run the following regression: $$\Delta L_{ibt}=\theta_{i}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{Market Maker 2013}}_{b}+\varepsilon_{ib}$$ (3) where I use bootstrapped errors clustered at the bank level. I do this for every quarter from 2013-2014 and plot this time series in Figure 6. The evidence shows that the market maker dummy is only positive and significant during the rebalancing, giving further support to the identification strategy. Figure 6 View largeDownload slide Cross-sectional estimation betas This figure presents the coefficient from an estimation of the growth of commercial credit over assets to a market maker dummy with city zone fixed effects. Errors are constructed with bootstrapping and are clustered at the bank level. The dashed lines indicate the 5%–95% confidence interval. The gray bars indicate the events described in the picture. Figure 6 View largeDownload slide Cross-sectional estimation betas This figure presents the coefficient from an estimation of the growth of commercial credit over assets to a market maker dummy with city zone fixed effects. Errors are constructed with bootstrapping and are clustered at the bank level. The dashed lines indicate the 5%–95% confidence interval. The gray bars indicate the events described in the picture. Another possible threat to identification is the existence of a differential prior evolution of the outcome variable. While the placebo test and plot are partially showing this, I present a figure with the actual evolution of the growth of commercial credit over assets during the period of study. Figure 7, Panel A presents these trends, and shows that the evolution of credit between market makers and the rest of the banks before the rebalancing is quite similar, and it differs considerably afterwards, with both a large increase in growth for market makers and a an important decrease in the growth for non market makers. Figure 7 View largeDownload slide Commercial credit growth evolution This figure shows the evolution of the commercial credit over total assets in time. Panel A shows the growth of commercial credit over total assets dividing by market maker and non market maker banks at the end of 2013. Panel B depicts the aggregate commercial credit over total assets. The gray bars indicate the events described in the picture. Figure 7 View largeDownload slide Commercial credit growth evolution This figure shows the evolution of the commercial credit over total assets in time. Panel A shows the growth of commercial credit over total assets dividing by market maker and non market maker banks at the end of 2013. Panel B depicts the aggregate commercial credit over total assets. The gray bars indicate the events described in the picture. The credit growth decrease for non market makers brings an additional concern for the difference-in-difference analysis, since it might be possible that market makers ended up taking business away from non-market maker banks. To address these concerns, I complement this evidence with the aggregate commercial credit over assets on a monthly frequency in Figure 7, Panel B. Commercial credit seems to be decreasing constantly during 2013 just until the start of the rebalancing by J.P. Morgan, when it starts to grow considerably until June 2016. This bottom panel shows that the banking sector as a whole was experiencing an important decline in the growth of credit of almost 2.5 percent from January 2013 until March 2014. Thus, in the difference-in-difference analysis non-market makers potentially act as a control group for the economic conditions in Colombia and are accounting for the fact that absent the shock commercial banks should have experienced a further decrease in credit growth during that period. This explanation is also sustained by the evidence in Section 4 which looks at the real effects in the economy. In the case of a pure redistribution of credit between market and non market makers, one should not observe any effects on the real economy. As shown in that section, industries with a higher exposure to market makers experienced increase in the growth of employment, production, sales and GDP, which suggests that the effect documented in Table 5 is not driven by a pure redistribution in credit between the two types of banks. 3.3.2 Omitted Variable Bias The positive differential effect on credit growth between market makers and non market makers estimated in Table 5 could also be affected by omitted variables. As acknowledge above, there could be institutional or economic changes during the period of the rebalancing that could be affecting the estimations. The fixed effects at the city-zone-time level partially controls for this. But there could still be other unobserved factors that could be biasing the estimates. In this section, I analyze the robustness of these estimates to omitted variable bias using a method developed by Oster (2014) that builds on Altonji et al. (2005) that uses both changes in coefficients and the R-squared after introducing observable controls.31 The basic idea both works is to provide a lower bound of the coefficients estimated, and a degree of selection that would make the estimated coefficient zero. Altonji et al. (2005) use this degree of selection to argue whether an estimate is likely to be causal or not. Oster (2014) builds on this and provides a lower bound on the estimates using both the movements in estimated coefficients and changes in the R-squared after including controls in the regression. More formally, she proposes to report a bias-adjusted coefficient: $$\beta^{*}=\tilde{\beta}-\delta \left[\mathring{\beta}-\tilde{\beta}\right]\frac{R_{\textrm{max}}-\tilde{R}}{\tilde{R}-\mathring{R}}$$ (4) where |$\tilde{\beta}$| (⁠|$\tilde{R}$|⁠) is the coefficient (R-squared) from a regression including control variables and |$\mathring{\beta}$| (⁠|$\mathring{R}$|⁠) is the coefficient from a regression with no controls. |$\delta$| is a parameter that indicates how strongly unobservables are correlated with the treatment variable relative to observables and |$R_{\textrm{max}}$| is the R-squared from a hypothetical regression of the outcome variable on both obersvables and unobservables. Oster (2014) suggests using |$\delta=1$| such that unobservables are equally important as observables. |$R_{\textrm{max}}$| has a natural upper bound of 1, but Oster (2014) acknowledges that it is common for the outcome not to be explained in its entirety by observables and unobservables and thus suggests using |$R_{\textrm{max}}=\min\{2.2\tilde{R},1\}$|⁠. Moreover, Oster (2014) also provides a |$\hat{\delta}$| such that |$\tilde{\beta}=0$|⁠, that indicates the degree of selection on unobservables relative to observables that would be sufficient to explain away the result. More specifically: $$\hat{\delta}=\frac{(\tilde{\beta})(\tilde{R}-\mathring{R})}{(\mathring{\beta}-\tilde{\beta})(R_{\textrm{max}}-\tilde{R})}$$ (5) Consistent with this framework, I use an uncontrolled regression and the regression with city-zone-time fixed effects in Table 7 and different values for |$\delta$| and |$R_{\textrm{max}}$| to provide a lower bound for the estimated treatment effect. Furthermore, the table and presents both the case of |$\delta=1$| (Panel A) and |$\delta=2$| (Panel B). Consider first the most conservative scenario where observables and unobservables explain all the variation in commercial credit growth (⁠|$R_{\textrm{max}}=1$|⁠) and the unobservables have double the importance of the observables |$\delta=2$|⁠, the bias-adjusted coefficient is different from zero. This scenario, however is highly unlikely as it is difficult to explain fully quarter to quarter growth in credit in general. In a more realistic scenario, where |$R_{\textrm{max}}=\min\{2.2\tilde{R},1\}=0.29$| and observables and unobservables are equally important, the bias-adjusted coefficient is 0.039, which is very close to the 0.041 coefficient estimated in Table 5. More importantly, under this scenario, |$\hat{\delta}=30.475$|⁠. This signals that unobservables would have to be 30 times more important than observables to attribute the entire treatment effect to these unobservable variables. Even in the most conservative scenario, where |$R_{\textrm{max}}=1$| the unobservables would have to be 5 times as important as the observables to explain to result on the differential credit growth away. All in all, these results provide evidence that the estimated treatment effect is robust to omitted variable bias. Table 7 Commercial credit and market makers: Coefficient stability |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 This table presents the coefficient from the main estimations in Table 5 with city zone time fixed effects and the bias-adjusted coefficients with the method developed by Oster (2013). Panel A presents results when the unobservable variables have equal importance as the unobservables, and panel B shows results when the unobservables have double importance than the observables. Bias adjusted |$\beta$| is computed like in Oster (2013). R|$_{\rm max}$| denotes the unknown overall R-squared in a model that controls for both observables and unobservables. |$\delta$| for |$\beta=0$| is the proportionality coefficient that would imply a bias-adjusted coefficient of 0. Table 7 Commercial credit and market makers: Coefficient stability |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 This table presents the coefficient from the main estimations in Table 5 with city zone time fixed effects and the bias-adjusted coefficients with the method developed by Oster (2013). Panel A presents results when the unobservable variables have equal importance as the unobservables, and panel B shows results when the unobservables have double importance than the observables. Bias adjusted |$\beta$| is computed like in Oster (2013). R|$_{\rm max}$| denotes the unknown overall R-squared in a model that controls for both observables and unobservables. |$\delta$| for |$\beta=0$| is the proportionality coefficient that would imply a bias-adjusted coefficient of 0. 3.3.3 Credit Supply or Demand? Another potential concern with the estimated treatment effect is whether it is actually capturing an increase in the credit availability by banks, or an increase in credit demand by local firms. The analysis regarding the potential omitted variable bias already suggests that the effect is mainly driven by an increase in credit supply by banks. Notice that the fixed effects at the city-zone-time level act as a proxy for credit demand at that level under the assumption that a single city-zone changes its demand equally for the two types of banks. By the analysis above, any other unobservable related to credit demand that are correlated with these fixed effects should not be affecting considerably the estimated coefficients. This lends support to the hypothesis that this effect is coming from an increase in credit supply rather than by changes in credit demand. I complement these results with evidence from interest rates. If the effect is mainly driven by credit supply we should observe a higher decrease in interest rates by market maker banks during the period of the rebalancing. I use the average commercial credit interest rates for commercial banks in Colombia and estimate a specification analogous to the one for credit growth: $$\Delta r_{bt}=\theta_{t}+\theta_{b}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\varepsilon_{bt}$$ (6) where |$r_{bt}$| is the log of interest rates on commercial credit for bank |$b$| at time |$t$|⁠. Results are presented in Table 8. During the period of the rebalancing market makers had a significantly lower growth of interest rates on commercial credit, even when I include different type of fixed effects. Quantitatively, the average commercial interest rate before the rebalancing for market makers was 18.2 percent, and thus the estimates suggest a decrease of 5.37 percentage points in the interest rates on commercial credit relative to non market maker banks. Thus, the evidence points towards a considerable decrease in interest rates from market maker banks, which is consistent with a change in the credit availability from market makers.32 Table 8 Commercial credit and market makers: Interest rates Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 This table presents OLS estimations of the growth of the interest rate on commercial credit against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% levels of significance, respectively. Table 8 Commercial credit and market makers: Interest rates Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 This table presents OLS estimations of the growth of the interest rate on commercial credit against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% levels of significance, respectively. 3.3.4 Treatment and Control Groups A possible concern with the treatment effects estimated in Table 5 is the fact that market makers are not randomly selected. Their participation in the program is endogenous and might depend among others on their trading expertise, their assets and in the end the two type of banks might be considerably different which might affect the main estimations. In this section I use the matching estimator developed by Abadie and Imbens (2002) and Abadie et al. (2004) to more closely match market and non market maker banks.33 I use an exact matching by the city-zone a bank is present, and the nearest matching through both assets and the bank solvency ratio. Furthermore, I allow for 1, 2 and 5 matches in order to use several non market makers as controls.34 I use the matching estimator for the period of the rebalancing and consider whether a bank is a market maker or not as the treatment variable (Table 9). Thus, the estimator matches any observation from the treatment group (i.e. credit growth in a city-zone by a market maker bank), with one or more observations from the control group (i.e. credit growth in the same city-zone by a non-market maker bank). Then, it selects from the control group the closest observations, in distance, using the total assets and solvency ratio to complete the matching. The table also presents the observations used in each estimation, and the percentage of exact matches.35 The results show coefficients very similars to the ones in my main specification, showing a positive and significant differential credit growth for market makers. As the number of exact matches decay, the coefficient is somewhat lower, but still statisitcally significant. These findings suggest that the potential differences between market maker and non market maker banks are not entirely driving the estimated effects. Table 9 Commercial credit and market makers: Matching estimators m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 This table presents the average treatment effect through the corrected matching estimators developed by Abadiee Imbens (2002). The dependent variable is the the growth of commercial credit to total assets. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy. All estimations are for the period March 2014–Sep 2014. Observations are matched, as possible, exactly through their city zone and with their nearest match for assets and solvency ratio in the different panels. Percentage of exact matches denotes the number of observations that could be exactly matched through their city zone. If not possible, the observations are matched through the closest observation in the same city. The dependent variable is winsorized at the 5th and the 95th percentile. m(.) indicates the number of matches in each estimation. Robust standard errors are in parentheses. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 9 Commercial credit and market makers: Matching estimators m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 This table presents the average treatment effect through the corrected matching estimators developed by Abadiee Imbens (2002). The dependent variable is the the growth of commercial credit to total assets. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy. All estimations are for the period March 2014–Sep 2014. Observations are matched, as possible, exactly through their city zone and with their nearest match for assets and solvency ratio in the different panels. Percentage of exact matches denotes the number of observations that could be exactly matched through their city zone. If not possible, the observations are matched through the closest observation in the same city. The dependent variable is winsorized at the 5th and the 95th percentile. m(.) indicates the number of matches in each estimation. Robust standard errors are in parentheses. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. 3.4 Robustness Tests In this section, I control for the different alternative hypothesis that could be affecting the main market making channel. In Table 10, Columns 1-3 I present three different tests.36 First, I include several bank level controls interacted with the rebalancing dummy to control for other potential channels (Column 1). I include the sovereign debt and exchange rate exposure, the size of banks, their return on assets, their liquid assets over total assets, their solvency ratio, their corporate debt issuance divided by total liabilities and whether a bank is foreign or not. While some of these controls are significant, the treatment variable is of similar size as the one estimated in Table 5, and still significant. Second, I remove the only state-owned bank (Banagrario) from the regression. The results show that there is no difference in the results from excluding this bank (Column 2). Third, I remove Bancolombia, the largest and most important bank in Colombia, and this does not influence the main results (Column 3).37 Table 10 Commercial credit and market makers: Robustness tests City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Corporate debt is the debt securities issued by a bank divided by total liabilities. Foreign is a dummy indicating whether a bank is foreign. Domestic sovereign debt exposure is the initial local public debt divided by assets. Exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. Assets is the initial log of assets. Liquid assets is the initial liquid assets to total assets. ROA is initial return on assets. Solvency ratio is the initial tier 1 capital divided by risk-weighted assets and market risk. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 10 Commercial credit and market makers: Robustness tests City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Corporate debt is the debt securities issued by a bank divided by total liabilities. Foreign is a dummy indicating whether a bank is foreign. Domestic sovereign debt exposure is the initial local public debt divided by assets. Exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. Assets is the initial log of assets. Liquid assets is the initial liquid assets to total assets. ROA is initial return on assets. Solvency ratio is the initial tier 1 capital divided by risk-weighted assets and market risk. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. There is also the possibility that market maker banks have trading expertise and the results capture a similar channel to that in Abbassi et al. (2016). As the affected bonds increase in price, banks with expertise reduce their holdings since these securities are overvalued, and thus increase the credit supply. For the evidence to be consistent with this channel, I should observe that after an initial overvaluation of the bonds’ price, the banks return to a similar level of ex-ante holdings of debt. However, the evidence is not consistent with a reversal of sovereign debt holdings. The price of sovereign bonds have a maximum peak the first week of April 2014. After that, these bonds suffer a 3 percent decrease in their price until the end of July. However, the average holdings of sovereign debt by market makers have a constanst decrease with a minimum at the end of July. This suggests that the evidence is not consistent with the channel presented by Abbassi et al. (2016) in the case of Germany. 4. Real Effects 4.1 Manufacturing Industries Data Until now I have shown that the entrance of foreign investors had consequences for the extension of private credit in Colombia. However, it remains to be seen whether this credit shock had consequences for the real economy. A problem towards this end is that I do not have access to the information at the bank-firm level such as the ones provided by the credit registry of each country. Therefore I need to rely on alternative data to gain understanding of the impact of the shock on the real economy. I gather data for a subsample of industries from the Monthly Manufacturing Polls conducted by the Departamento Administrativo Nacional de Estadistica (DANE) in Colombia. This database contains the yearly growth at monthly level of employment, production and sales for each manufacturing industry. Additionally, I obtain quarterly data from national accounts on the real annual GDP growth for each of these manufacturing industries. Then, I can construct a proxy of the exposure to market maker banks at the industry level. More specifically, $$Exp_{i,2013}^{MM}=\frac{\sum_{b\in MM}C_{ib,2013}}{\sum_{b\in B}C_{ib,2013}}$$ (7) where |$Exp_{i,2013}^{MM}$| is the exposure of industry |$i$| to market makers at the end of 2013. The numerator indicates the total credit extended by market makers to industry |$i$| at the end of 2013. The denominator contains the total credit extended by all banks to industry |$i$|⁠. This information is helpful to understand whether a given industry was more likely to be exposed to these banks before the benchmark change. Internet Appendix Table 9 shows the 20 industries that I was able to match between the two databases, along with the specific exposure at the end of 2013. This exposure is also presented graphically in Figure 8. The exposure to market makers from most of the industries is high, probably a consequence of these banks having a sizable part of the total assets in the banking system in Colombia. While this is a subsample of 20 industries, there is still significant variation in the exposure to market makers. The average exposure is 82.65, with a standard deviation of 8.4. Figure 8 View largeDownload slide Exposure to market makers across industries This figure presents the exposure of each industry to market maker banks at the end of 2013. The exposure is constructed by summing the commercial credit of market maker banks to each industry and dividing it by the total credit to the same industry by all commercial banks. Figure 8 View largeDownload slide Exposure to market makers across industries This figure presents the exposure of each industry to market maker banks at the end of 2013. The exposure is constructed by summing the commercial credit of market maker banks to each industry and dividing it by the total credit to the same industry by all commercial banks. To analyze whether the shock to credit had an impact on the real variables for these industries, I estimate the following specification: $$\Delta y_{it}=\theta_{t}+\theta_{i}+\gamma_{1}Exp_{i,2013}^{MM}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\varepsilon_{it}$$ (8) where |$y_{it}$| is either employment, production, sales or GDP. |$\theta_{t}$| and |$\theta_{i}$| are time and industry fixed effects, respectively. |$Exp_{i,2013}^{MM}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}$| is the treatment variable, with the exposure to market makers interacted with a time dummy for the period after the rebalancing to take into account any potential lags. The estimation is a classical differences-in-differences approach, and |$\gamma_{1}$| captures the differential effect on real variables of being more exposed to the credit shock. Errors are bootstrapped clustered at the industry level.38 Results show that being more exposed to a market maker bank led to a statistically significant increase in employment, production, sales and GDP during the rebalancing period (Table 11). The economic size of the effect is important. For example, consider an interquartile movement for an industry in the exposure to market makers, which implies an almost 8.7 percentage point increase in the exposure to market makers. This would have implied an increase in the average yearly growth of 1.2, 2.9, 2.9, and 2.8 percentage points for employment, production, sales, and GDP respectively. Overall, there were important effects for economic activity from the credit shock. Moreover, most of the effect for employment is driven by increases in employment directed towards production activities. Table 11 Real effects Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 This table presents OLS estimations of the yearly growth of real variables against different set of fixed effects and a treatment variable for manufacturing industries. The growth of the dependent variable is constructed as the difference in logs of a month/quarter versus the month/quarter of the year before. The treatment variable is the exposure of an industry to market makers in 2013 multiplied by a dummy indicating the period after the index rebalancing by J. P. Morgan. All estimations are for the period 2012m3–2015m2/2012q2–2015q1. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the industry level. *, **, and *** denote 10%, 5%, and 1% level of significance, respectively. Table 11 Real effects Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 This table presents OLS estimations of the yearly growth of real variables against different set of fixed effects and a treatment variable for manufacturing industries. The growth of the dependent variable is constructed as the difference in logs of a month/quarter versus the month/quarter of the year before. The treatment variable is the exposure of an industry to market makers in 2013 multiplied by a dummy indicating the period after the index rebalancing by J. P. Morgan. All estimations are for the period 2012m3–2015m2/2012q2–2015q1. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the industry level. *, **, and *** denote 10%, 5%, and 1% level of significance, respectively. 5. External Validity The results showing that governemnt acess to foreign credit increases the private access to credit are based in the rebalancing event that happened in Colombia. However, these results could well be applicable to other countries. First, the rebalancing events in government debt index of this style has become more and more common. Countries such as Mexico and South Africa were added to the World Government Bond Index (constructed by Citigroup) and Argentina, Czech Republic, Romania and Nigeria were included in the same index studied in this paper. China and India have yet to be included in one of these indexes, which could trigger similar effects to the ones shown for Colombia for the private access to credit and economic activity. In the end, the size of the rebalancing for Colombia, and the access to very detailed data on credit and economic activity, provided a natural laboratory to study this question. Second, the market maker program used in Colombia is not specific to that country. Other emerging countries in Latin America such as Mexico and Peru use this kind of program, with similar obligations and benefits for the financial corporations that participate in it. The same happens for emerging countries in Asia and Europe such as Indonesia, Poland and Romania among others. Third, Colombia is an average country in terms of the size of its local currency debt market. Du et al. (2016) report that Colombia has a ratio of local currency debt to central government debt close to 60 percent. This number is similar to the numbers of countries such as Czech Republic, Hungary, Indonesia, Philippines, Poland, Russia, South Africa, and only somewhat higher than countries like Brazil, Israel, Mexico, and Peru (closer to 40 percent). Together, these three facts suggest that the findings in this paper could be applicable to at least a considerable number of emerging countries. Fourth, this episode is also important from the point of view of financial liberalization. There has been several works on how capital account liberalization affects economic growth and investment.39 Colombia is a country that has an open capital account, but previous to this episode with a low access to foreign investors in the domestic sovereign debt market. Thus, even in a country open to international capital flows, working on policies to increase the access to foreign investors in the sovereign debt market could lead to an increase in domestic credit and economic activity. This sheds light on the possible consequences on improving the access to the domestic sovereign debt markets in countries such as China. 6. Conclusions In this paper, I exploit a sudden and unanticipated shock that triggered the entrance of foreign investors to the local currency sovereign debt market in an emerging market. I use an episode in which J.P. Morgan introduced several Colombian bonds in its local currency governemnt debt indexes in emerging markets. Since foreign institutional investors often use benchmark-tracking strategies, they rebalanced their portfolio towards Colombia increasing capital inflows to the domestic sovereign debt market. As foreign investors purchased this debt, banks officially designated as market makers in the sovereign debt market decreased their exposure to these securities. On average, they reduced their local public debt over total assets by 7.8 percentage points, compared to the rest of the banks. This shock, that was originated on the government debt market, spillovered to the credit market. Market maker banks increased their commercial credit to total assets ratio by 4.2 percentage points on average, relative to the rest of the banks. This transmission channel is not found during other periods and is not driven by other observable differences between market and non market maker banks. The evidence suggests that the shock had an impact on the real economy. I construct a proxy for the exposure to market maker banks at the industry level and find that industries with more exposure to market makers had a higher growth of employment, production, sales and GDP during this period. The evidence is consistent with a crowding out of private credit before the entrance of foreign investors and with a financial repression channel. Because of the illiquidity of the sovereign debt market, the government used market makers to absorb debt issued in the primary market. As foreign institutional investors entered the domestic sovereign debt market, these domestic financial institutions were able to sell the excess of debt that they could not offload before and used the proceeds to extend credit. The findings in this paper has implications from a policy-making perspective for a number of reasons. For instance, large countries such as China and India still have less than 2 percent of their local currency debt in the hands of foreign investors. This study sheds light on the possible consequences of using policies to increase the share of foreign investors in domestic sovereign debt markets. Moreover, on March 2016, J.P. Morgan included China on a watchlist to enter the GBI-EM. The evidence suggests that the confirmation of this process could lead to a boost in private credit for the Chinese economy. Another consideration is that China would have the largest weight in the index (10 percent). As a consequence, its introduction to the index could lead to a decrease in the weights of the rest of the countries. This might ultimately lead to negative spillovers to the other constituents of the GBI-EM. Finally, results also suggest that sovereign debt index rebalancing can have effects on the economy beyond the usual price effects found in the literature. Thus, one policy concern is related to the regulation of activities of both institutional investors with index-tracking strategies and index providers. Ultimately, the effects documented in this paper were started by a decision made by a single index provider. Is this desirable? Should there be regulation on the construction of benchmark indexes and their reconstitutions? Moreover, a decision to remove Colombia from the index may produce the opposite effect. The evidence suggests that activities by index-tracking investors and index providers should be followed more closely by policy makers. I am indebted to Fernando Broner, Alberto Martin and José Luis Peydró for their continuing guidance and support throughout this project. This paper benefited from numerous discussions with Lorenzo Pandolfi. I would like to thank Philip Strahan and two anonymous referees for their insightful suggestions that improved the paper significantly. I would also like to thank Gene Ambrocio, Aitor Erce, Juanita Gonzalez Uribe, Graciela Kaminsky, Mauricio Larrain, Eduardo Levy Yeyati, Maria Lombardi, Roberto Pancrazi, Daniel Paravisini, Carolin Pflueger, Francisco Queiros, Veronica Rappoport, Diego Restrepo Tobon, David Reeb, Francesc Rodriguez Tous, Tom Schmitz, Sergio Schmukler, Jesse Schreger, Jay Shambaugh and Jagdish Tripathy for insightful comments. This work has also benefited from comments by participants at the Barcelona GSE PhD Jamboree, Barcelona GSE Summer Forum, the Belgrade Young Economists Meeting, the CREI International Lunch Seminar, the European Doctoral Program Jamboree, the International Conference on Sovereign Bond Markets, the LSE Finance Ongoing Work Seminar, NBER IFM Meetings, the SAE Bilbao, the SFS Cavalcade, the UPF Finance Lunch Seminar, the World Finance Conference. I am extremely grateful to Carla Maenza for her invaluable and constant feedback during the writing of this paper. All errors are my own. Supplementary data can be found on The Review of Financial Studies web site. Footnotes 1 See for instance Diamond (1965). 2 As shown by Du et al. (2016) and Du and Schreger (2016) most of the local (foreign) currency sovereign bonds in emerging markets are issued under domestic (foreign) law, and traded in domestic (foreign) markets. Thus, the terms domestic sovereign debt and local currency sovereign debt will be used interchangeably throughout the paper. 3 Ongena et al. (2016) document a similar channel for banks during the European Sovereign Debt Crisis. 4 This empirical literature is closely related to the growing literature on the real effects from credit supply changes. See among others Gan (2007), Ivashina and Scharfstein (2010), Iyer and Peydro (2011), Jimenez et al. (2012), Jimenez et al. (2014), Iyer et al. (2014). Within this literature my paper is more related to several papers studying the effect of international shocks to emerging markets and lending by banks. See Khwaja and Mian (2008), Paravisini (2008), and Schnabl (2012) for episodes in Pakistan, Argentina and Peru, respectively. 5 This phenomenon is the consequence of a large switch of investor funds from active to passive funds and a documented movement of active funds into more passive investment strategies. See among others Cremers and Petajisto (2009). 6 More broadly, there is a large literature on the aggregate effects of international mutual funds on financial markets. See among others Broner et al. (2006), Jotikasthira et al. (2012), Levy-Yeyati and Williams (2012) and Raddatz and Schmukler (2012). 7 There has been a recent literature focusing on the real effects of institutional investors flows in general. Chernenko and Sunderam (2012) analyze how flows into high-yield mutual funds have effects on the issuance of firms and their investments. Adelino et al. (2014) document how changes in credit ratings by Municipalities in the United States have consequences for public financing and for economic activity. Almeida et al. (2015) and Adelino and Ferreira (2015) document how credit ratings upgrades and downgrades affect firms’ real investment decisions and banks’ credit supply. 8 For instance, Mendoza and Terrones (2012) find that credit booms are positively correlated with net capital inflows. Calderon and Kubota (2012) suggest that private capital inflows are good predictors of credit booms. In a more granular approach, Lane and McQuade (2014) argue that only net debt inflows generate domestic credit growth in European countries. In a related theoretical and empirical work, Blanchard et al. (2015) find that only equity inflows are correlated to credit expansions. 9 Reinhart and Reinhart (2009) study how capital flow bonanzas affect the likelihood of economic crises. Caballero (2016) shows that capital inflows bonanzas increase the probability of banking crises. Kalantzis (2015) and Benigno et al. (2015) study the changes in the sectorial allocation of resources due to large capital inflows. There is a large literature on how capital account liberalizations and how they affect investment and economic growth. For a detailed review see Henry (2007). More recently, there has been studies on how capital account liberalization affects inequality. For instance, Larrain (2015) shows that opening the capital account leads to a greater wage inequality. 10 More recently, index providers have focused on constructing alternative indexes not based on market capitalization (such as GDP-weighted indexes or fundamentals based indexes). 11 J.P. Morgan also constructs the Emerging Markets Bond Index (EMBI) which is a foreign currency sovereign debt index. Since emerging markets governments shifted their preference towards local currency debt, this index has been steadily declining in popularity. 12 The benchmark weight of a country is defined as the sum of the market capitalization of all securities issued in a country divided by the total market capitalization of all the securities included in the benchmark index. 13 Reuters (2014). 14 Even during those months it is impossible to assert that there were no changes at all in the economy of a country. Section 3.3 presents different tests to alleviate these concerns. 15 The Internet Appendix provides an account of the exchange rate evolution during this period. Since there were large capital inflows, the exchange rate appreciated considerably during the first part of the rebalancing. 16 See Internet Appendix Figure 1, Panel A. 17 I use the 6-month rolling sum to compare the aggregate activity by foreigners and commercial banks by the end of the period of the rebalancing. Internet Appendix Figure 2 presents the same figure with monthly data which shows a similar albeit noisier pattern. 18 This number increases to 10.3 percent if we consider the March-December 2014 period, accounting for the fact that some funds slowly change their positions. 19 The other two obligations are to construct monthly reports about the state of the domestic sovereign debt market and to inform of any mergers/sales to the treasury department. 20 (Ministerio Hacienda y Credito Publico, 2010) 21 More strikingly, all market maker banks were net sellers of sovereign debt during the event, and half of the banks among non market makers were net sellers of government bonds. 22 This number is obtained by using the average local currency sovereign debt to total assets ratio in December 2013 multiplied by 1 minus the coefficient estimated for the three quarters of the rebalancing. 23 Anecdotal evidence from several people working in banks around the period suggest that the reduction in sovereign debt from market makers was due to the entrance of foreign investors, and a reduction in profitability in the market making business for these banks. 24 See Reinhart and Sbrancia (2011) for a detailed account of this literature. 25 While this is the full fledge specification, I also estimate models with only time fixed effects. More specifically, \begin{equation*} \Delta L_{ibt}=\theta_{t}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{2}X_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{3}{\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}+\beta_{4}X_{b}+\varepsilon_{ibt} \end{equation*} which includes the levels of the market maker dummy, and the levels of any bank observable characteristic used without the interaction with the rebalancing dummy. 26 As explained by Adelino et al. (2015) this might not be a perfect control for credit demand. However, I will try to present suggestive evidence on the unlikely possibility that this shock is coming through credit demand. While a more sound strategy would be controlling at the firm level, I do not have data on the credit register of Colombia. In some cases, even data at the bank-firm level might not capture perfectly credit demand as suggested by Paravisini et al. (2015). 27 Additionally, credit growth is winsorized at the 5 and 95 percent level since this variable has very large values, with its standard deviation being 0.469 (or 46.9%) for non market makers and 0.841 (84.1%) for market makers (see Internet Appendix Table 2 for the summary statistics on credit growth). Internet Appendix Table 3 presents results using different winsorizations, that do not affect the main conclusions. 28 Throughout the paper I mainly rely on the city-zone database due to its balance among different banks and for brevity. Results are qualitatively similar at the industry level. 29 For all the estimations I exclude Banco Corpbanca after 2014q1 and GNB Sudameris after 2014q2 since both were part of two different mergers. 30 These numbers are obtained by using the estimated differential credit growth during the period multiplied by the commercial credit for market makers in December 2013. After this, I multiply it by the total assets of market makers in December 2013 to obtain the differential growth in commercial credit. 31 See Hener et al. (2015) for a simple exposition of these tests. 32 This is further illustrated in Internet Appendix Figure 3 where I show the average commercial interest rates for both types of banks in time. 33 This can also partially be controlled by bank observable characteristics, which I explore in the next sections. 34 I also consider more than 5 exact matches, but after 5 the percentage of exact matches start to decay very fast. 35 The percentage of exact matches specifies the percentage of observations from the treatment group that were exactly matched to an observation from the control group through their city-zone. The observations that are not exactly matched, are matched to observations in another zone but in the same city. 36 The Internet Appendix contains a more in depth analysis of the different channels that could be affecting the baseline specification. 37 I also perform a robustness test using the commercial credit growth (not normalized by assets) in Internet Appendix Table 3. The results are quantitatively similar to when I do normalize credit by assets. 38 Internet Appendix Table 10 presents summary statistics of the growth of these real variables. 39 See for instance Henry (2000) and Henry (2003) among others. References Abadie, A., Drukker, D. Herr, J. L. and Imbens. G. W. 2004 . Implementing matching estimators for average treatment effects in Stata. Stata Journal 4 : 290 – 311 . Abadie, A., and Imbens. G. W. 2002 . Simple and bias-corrected matching estimators for average treatment effects. Working Paper, NBER . Abbassi, P., Iyer, R. Peydro, J. L. and Tous. F. R. 2016 . Securities trading by banks and credit supply: Micro evidence from the crisis. Journal of Financial Economics 121 : 569 – 94 . Google Scholar Crossref Search ADS Acharya, V. V., Eisert, T. Eufinger, C. and Hirsch. C. 2014 . Real effects of the sovereign debt crisis in Europe: Evidence from syndicated loans. Discussion Paper, CEPR . Adelino, M., Cunha, I. and Ferreira. M. 2014 . The economic effects of public financing: Evidence from municipal bond ratings recalibration. Technical Report . Adelino, M., and Ferreira. M. 2015 . Bank ratings and lending supply: Evidence from sovereign downgrades. Technical Report . Adelino, M., Schoar, A. and Severino. F. 2016 . Loan originations and defaults in the mortgage crisis: The role of the middle class. Review of Financial Studies 29 : 1635 – 70 . Google Scholar Crossref Search ADS Almeida, H., Cunha, I. Ferreira, M. and Restrepo. F. 2015 . The real effects of credit ratings: The sovereign ceiling channel. Technical Report . Altavilla, C., Pagano, M. and Simonelli. S. 2017 . Bank exposures and sovereign stress transmission. Review of Finance 21 : 2103 – 39 . Google Scholar Crossref Search ADS Altonji, J. G., Elder, T. E. and Taber. C. R. 2005 . Selection on observed and unobserved variables: Assessing the effectiveness of Catholic schools. Journal of Political Economy 113 : 151 – 84 . Google Scholar Crossref Search ADS Baskaya, Y. S., and Kalemli-Ozcan. S. 2016 . Sovereign risk and bank lending: Evidence from 1999 Turkish earthquake. Working Paper, NBER . Becker, B., and Ivashina. V. 2018 . Financial repression in the European sovereign debt crisis. Review of Finance 22 : 83 – 115 . Google Scholar Crossref Search ADS Ben-David, I., Franzoni, F. and Moussawi. R. 2014 . Do ETFs increase volatility? Working Paper, NBER . Benigno, G., Converse, N. and Fornaro. L. 2015 . Large capital inflows, sectoral allocation, and economic performance. Journal of International Money and Finance 55 : 60 – 87 . Google Scholar Crossref Search ADS Bhattacharya, U., Loos, B. Meyer, S. Hackethal, A. and Kaesler. S. 2013 . The dark side of ETFs. Technical Report . Blanchard, O., Ostry, J. D. Ghosh, A. R. and Chamon. M. 2015 . Are capital inflows expansionary or contractionary? Theory, policy implications, and some evidence. Working Paper, NBER . Bofondi, M., Carpinelli, L. and Sette. E. 2013 . Credit supply during a sovereign debt crisis. Temi di discussione , Bank of Italy . Bolton, P., and Jeanne. O. 2011 . Sovereign default risk and bank fragility in financially integrated economies. IMF Economic Review 59 : 162 – 94 . Google Scholar Crossref Search ADS Broner, F., Erce, A. Martin, A. and Ventura. J. 2014 . Sovereign debt markets in turbulent times: Creditor discrimination and crowding-out effects. Journal of Monetary Economics 61 : 114 – 42 . Google Scholar Crossref Search ADS Broner, F., Gelos, G. and Reinhart. C. 2006 . When in peril, retrench: Testing the portfolio channel of contagion. Journal of International Economics 69 : 203 – 30 . Google Scholar Crossref Search ADS Caballero, J. A. 2016 . Do surges in international capital inflows influence the likelihood of banking crises? Economic Journal 126 : 281 – 316 . Google Scholar Crossref Search ADS Calderon, C., and Kubota. M. 2012 . Gross inflows gone wild: Gross capital inflows, credit booms and crises. Working Paper , the World Bank . Chang, Y.-C., Hong, H. and Liskovich. I. 2014 . Regression discontinuity and the price effects of stock market indexing. Review of Financial Studies 28 : 212 – 46 . Google Scholar Crossref Search ADS Chernenko, S., and Sunderam. A. 2012 . The real consequences of market segmentation. Review of Financial Studies 25 : 2041 – 69 . Google Scholar Crossref Search ADS Cremers, M., and Petajisto. A. 2009 . How active is your fund manager? A new measure that predicts performance. Review of Financial Studies 22 : 3329 – 65 . Google Scholar Crossref Search ADS Diamond, P. A. 1965 . National debt in a neoclassical growth model. American Economic Review 55 : 1126 – 50 . Du, W., Pueger, C. E. and Schreger. J. 2016 . Sovereign debt portfolios, bond risks, and the credibility of monetary policy. Working Paper, NBER . Du, W., and Schreger. J. 2016 . Local currency sovereign risk. Journal of Finance 71 : 1027 – 70 . Google Scholar Crossref Search ADS Gan, J. 2007 . The real effects of asset market bubbles: Loan and firm-level evidence of a lending channel. Review of Financial Studies 20 : 1941 – 73 . Google Scholar Crossref Search ADS Gennaioli, N., Martin, A. and Rossi. S. 2014 . Banks, government bonds, and default: What do the data say? Technical Report . Hener, T., Rainer, H. and Siedler. T. 2015 . Political socialization in flux? Linking family non-Intactness during childhood to adult civic engagement. Discussion Paper , Institute for the Study of Labor (IZA). Henry, P. B. 2000 . Stock market liberalization, economic reform, and emerging market equity prices. Journal of Finance 55 : 529 – 64 . Google Scholar Crossref Search ADS Henry, P. B. 2003 . Capital-account liberalization, the cost of capital, and economic growth. American Economic Review 93 : 91 – 6 . Google Scholar Crossref Search ADS Henry, P. B. 2007 . Capital account liberalization: Theory, evidence, and speculation. Journal of Economic Literature 45 : 887 – 935 . Google Scholar Crossref Search ADS Israeli, D., Lee, C. and Sridharan S. 2015 . Is there a dark side to exchange traded funds (ETFs)? An information perspective. Technical Report . Ivashina, V., and Scharfstein. D. 2010 . Bank lending during the financial crisis of 2008. Journal of Financial Economics 97 : 319 – 338 . Google Scholar Crossref Search ADS Iyer, R., and Peydro. J. L. 2011 . Interbank contagion at work: Evidence from a natural experiment. Review of Financial Studies 24 : 1337 – 77 . Google Scholar Crossref Search ADS Iyer, R., Peydro, J. L. da Rocha-Lopes, S. and Schoar. A. 2014 . Interbank liquidity crunch and the firm credit crunch: Evidence from the 2007–2009 crisis. Review of Financial Studies 27 : 347 – 72 . Google Scholar Crossref Search ADS Jimenez, G., Ongena, S. Peydro, J. L. and Saurina. J. 2012 . Credit supply and monetary policy: Identifying the bank balance-sheet channel with loan applications. American Economic Review 102 : 2301 – 26 . Google Scholar Crossref Search ADS Jimenez, G., Ongena, S. Peydro, J. L. and Saurina. J. 2014 . Hazardous times for monetary policy: What do twenty-three million bank loans say about the effects of monetary policy on credit risk-taking? Econometrica 82 : 463 – 505 . Google Scholar Crossref Search ADS Jotikasthira, C., Lundblad, C. and Ramadorai. T. 2012 . Asset fire sales and purchases and the international transmission of funding shocks. Journal of Finance 67 : 2015 – 50 . Google Scholar Crossref Search ADS Kalantzis, Y. 2015 . Financial fragility in small open economies: Firm balance sheets and the sectoral structure. Review of Economic Studies 82 : 1194 – 222 . Google Scholar Crossref Search ADS Khwaja, A. I., and Mian. A. 2008 . Tracing the impact of bank liquidity shocks: Evidence from an emerging market. American Economic Review 98 : 1413 – 42 . Google Scholar Crossref Search ADS Lane, P. R., and McQuade. P. 2014 . Domestic credit growth and international capital flows. Scandinavian Journal of Economics 116 : 218 – 52 . Google Scholar Crossref Search ADS Larrain, M. 2015 . Capital account opening and wage inequality. Review of Financial Studies 28 : 1555 – 87 . Google Scholar Crossref Search ADS Levy-Yeyati, E., and Williams. T. 2012 . Emerging economies in the 2000s: Real decoupling and financial recoupling. Journal of International Money and Finance 31 : 2102 – 26 . Google Scholar Crossref Search ADS Mendoza, E., and Terrones. M. 2012 . An anatomy of credit booms and their demise. Working Paper , Central Bank of Chile . Ministerio Hacienda y Credito Publico . 2010 . Programa de creadores de mercado para titulos de deuda publica. Technical Report , Colombia . Ongena, S., Popov, A. and Horen. N. V. 2016 . The invisible hand of the government: ‘Moral suasion’ during the European sovereign debt crisis. Working Paper , Netherlands Central Bank . Oster, E. 2017 . Unobservable Selection and Coefficient Stability: Theory and Validation. Journal of Business & Economic Statistics . Advance access published online June 1, 2017, https://doi.org/10.1080/07350015.2016.1227711 . Paravisini, D. 2008 . Local bank financial constraints and firm access to external finance. Journal of Finance 63 : 2161 – 93 . Google Scholar Crossref Search ADS Paravisini, D., Rappoport, V. and Schnabl P. 2015 . Comparative Advantage and Specialization in Bank Lending. Technical Report . Perez, D. J. 2015 . Sovereign debt, domestic Banks and the provision of public liquidity. Discussion Papers , Stanford Institute for Economic Policy Research. Raddatz, C., and Schmukler. S. 2012 . On the international transmission of shocks: Micro-evidence from mutual fund portfolios. Journal of International Economics 88 : 357 – 74 . Google Scholar Crossref Search ADS Raddatz, C., Schmukler, S. and Williams. T. 2017 . International asset allocations and capital flows: The benchmark effect. Journal of International Economics 108 : 413 – 30 . Google Scholar Crossref Search ADS Reinhart, C., and Reinhart. V. 2009 . Capital flow bonanzas: An encompassing view of the past and present. In NBER international seminar on macroeconomics 2008 , pp. 9 – 62 . Cambridge, MA : National Bureau of Economic Research. Reinhart, C. M., and Sbrancia. M. B. 2011 . The liquidation of government debt. Working Paper , Bank for International Settlements. Reuters. 2014 . Colombia bonds to see 10 billion inflows, Turkey, Russia lose in index changes-banks. Reuters , March 20. https://af.reuters.com/article/commoditiesNews/idAFL6N0MH46O20140320. Schnabl, P. 2012 . The international transmission of bank liquidity shocks: Evidence from an emerging market. Journal of Finance 67 : 897 – 932 . Google Scholar Crossref Search ADS Sullivan, R., and Xiong. J. 2012 . How index tracking increases market vulnerability. Financial Analysts Journal 68 : 70 – 84 . Google Scholar Crossref Search ADS Wurgler, J. 2010 . On the economic consequences of index-linked investing. Working Paper, NBER. © The Author(s) 2018. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Review of Financial Studies Oxford University Press

Capital Inflows, Sovereign Debt and Bank Lending: Micro-Evidence from an Emerging Market

, Volume 31 (12) – Dec 1, 2018
37 pages

/lp/ou_press/capital-inflows-sovereign-debt-and-bank-lending-micro-evidence-from-an-QUw69AHPZe
Publisher
Oxford University Press
ISSN
0893-9454
eISSN
1465-7368
D.O.I.
10.1093/rfs/hhy026
Publisher site
See Article on Publisher Site

Abstract

Abstract This paper uses a natural experiment to show that government access to foreign credit increases private access to credit. I identify a sudden, and unanticipated increase in capital inflows to the sovereign debt market in Colombia, due to a rebalancing in a government bond index by J.P. Morgan. I find that market makers banks in the treasury market reduced their sovereign debt by 7.8 percentage points of assets and increased their credit availability by 4.2 percentage points of assets. Using industry level data, I show that a higher exposure to market makers led to a higher growth in economic activity. Received August 17, 2017; editorial decision January 25, 2018 by Editor Philip Strahan. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online. How do public and private access to credit interact? Economic theory suggests that when governments borrow from domestic institutions this may lead to a crowding out of private credit.1 This crowding out depends crucially on public access to foreign credit. When there is little access to foreign investors, governments rely heavily on local financial institutions to absorb the issuance of debt. If foreign investors become willing to purchase sovereign debt, the reliance on local financial institutions may be reduced, freeing resources for the private sector. As a result, local firms may be able to finance investment projects and boost economic activity. Although intuitive, there is no clear evidence on this topic because of the identification challenges. Notice that sovereign risk, sovereign bond holdings of banks and foreigners, and loans to the private sector are all jointly determined. For instance, an improvement in local economic conditions may increase both foreign investor demand for sovereign debt and credit demand from local firms. In such a case, one would observe both an increase in the share of sovereign debt held by foreign investors and an increase in private credit, but this correlation would not imply causation. This common problem illustrates the difficulty of finding causal evidence on this issue. The main contribution of this paper is using a novel episode to overcome the identification problems previously encountered in the literature. By doing so, I provide clear evidence that government access to foreign credit increases private access to credit and boosts economic activity. I use a sudden, unanticipated shock that triggered the entrance of foreign investors to the local currency sovereign debt market in Colombia.2 In March 2014, J.P. Morgan announced the inclusion of several Colombian treasury bonds into its emerging markets local currency government debt index. Since many international mutual funds track their performance against this index, they changed their portfolio suddenly, directing capital flows to the Colombian local currency sovereign debt market. This shock had a number of appealing features for identification. First, the timing of the decision by J.P. Morgan appears, at least in the short window of time I study, unrelated to changes in the future economic prospects of Colombia. Second, it was sizable: the share of debt held by foreigners in this market went from 8.5 to 19 percent in only 7 months (Figure 1). Third, the event was unanticipated. Fourth, provided that the index is specific to government debt, it does not directly affect flows to the private sector. Figure 1 View largeDownload slide Foreign share of domestic government debt securities This figure presents the evolution of the share of domestic government debt securities (TES) held by foreigners. The dashed line shows a linear trend using the average growth during the 12 months prior to the announcement of the change in the index by J.P. Morgan. The gray bar represents the announcement of the rebalancing by J. P. Morgan. Figure 1 View largeDownload slide Foreign share of domestic government debt securities This figure presents the evolution of the share of domestic government debt securities (TES) held by foreigners. The dashed line shows a linear trend using the average growth during the 12 months prior to the announcement of the change in the index by J.P. Morgan. The gray bar represents the announcement of the rebalancing by J. P. Morgan. I find that the entrance of foreign investors had sizable effects on commercial banks. Moreover, it had heterogenous effects on banks according to their participation in the local currency sovereign debt market. In Colombia, the Ministry of Finance selects financial institutions to act as market makers or official intermediaries in the treasury market. Each of the intermediaries participating in the program is obliged to absorb 4.5 percent of the total debt issued by the government in the primary market. I find that market maker banks reduced their domestic sovereign debt holdings by 7.8 percentage points of assets, compared to the rest of the banks. Using data either at the city-zone or industry level, I also show that they increased differentially their commercial credit availability by 4.2 percentage points of their assets. Results show a considerable substitution between sovereign debt and commercial credit. This effect is also economically significant, around 2 percent of Colombia’s GDP. I analyze whether the shock had real effects by looking at evidence at the industry level. I obtain data on monthly employment, production and sales from the Monthly Manufacturing Polls conducted by the Departamento Administrativo Nacional de Estadistica (DANE), and data on GDP by industry from the Quarterly National Accounts database. I construct a proxy for the exposure to market maker banks at the industry level, and find that industries more exposed to market makers had higher growth of employment, production, sales and GDP during this period. I conduct several robustness checks to confirm the results. I estimate the cross-sectional coefficients of a regression of credit growth on a dummy variable indicating whether a bank is a market maker or not. I find that the coefficients are only statistically significant during the rebalancing, showing an important support for the identification strategy. I also conduct an analyisis on the potential omitted variable bias using tests initially developed by Altonji et al. (2005) and later formalized by Oster (2014). I find that omitted variables are not likely to be biasing the results. Moreover, I show that the effect on credit growth is not driven by valuation effects on the balance sheet of banks and is likely coming from a change in the access to credit by banks. Finally, the results are robust when I use a matching estimator using the geographical areas and bank observable characteristics such as assets and capital ratios to match market makers and non market makers. The evidence is consistent with the following narrative. Before the entrance of foreign investors, there was a crowding out of private credit. The domestic sovereign debt market was dominated by local participants and the investor base was undiversified. Therefore, the government used market makers to absorb debt issued in the primary market. These financial institutions were paid a premium to absorb this debt, since the secondary market was less liquid, and thus market makers kept part of the issued debt in their balance sheet. As foreign institutional investors entered the domestic sovereign debt market, the government reduced the premium it was paying to market makers, and these financial institutions sold the excess of debt that could not offload before and used the proceeds to extend credit. As defined by Reinhart and Sbrancia (2011) this could be a form of financial repression, as the government pays banks to aborb (and ultimately) hold government debt and this restricts the extension of credit.3 This paper contributes to two broad strands of literature. From a macroeconomic perspective, it is related to the literature on the interrelationship between the supply of credit to public and private sectors. This line of research has received a lot of attention during the recent European Sovereign Debt Crisis, emphasizing two different mechanisms. On the one hand, there is a line of research highlighting that an increase in the home bias of sovereign debt holdings crowds out private credit. For instance, Broner et al. (2014) propose a model with creditor discrimination and crowding out effects that accounts for the reallocation of credit from the private to the public sector observed in the euro zone periphery during the European Sovereign Debt crisis. Becker and Ivashina (2014) and Altavilla et al. (2015) also provide empirical evidence consistent with this reallocation channel. My paper paper is closely related to these two studies since it explores how financial repression affects the extension of credit to the private sector. On the other hand, there is a part of the literature that emphasizes how shifts in sovereign risk affect the balance sheets of banks. For example, Bolton and Jeanne (2011), Gennaioli et al. (2014) and Perez (2015) propose models in which sovereign defaults hurt the balance sheet of banks and reduce private credit. From an empirical point of view, Bofondi et al. (2013), Acharya et al. (2014), Gennaioli et al. (2014), and Baskaya and Kalemli-Ozcan (2016) present evidence consistent with this channel.4 Most of these papers have problems identifying a shock that exclusively affects foreign demand for sovereign debt. I provide such a shock and to the best of my knowledge this paper is the first to use this type of event and separate between the two channels highlighted by the literature. Second, from a finance perspective, this paper contributes to a growing literature on the aggregate effects of institutional investors. Since the Global Financial Crisis, there has been an increased interest in the activities of financial intermediaries other than traditional banks. Investment activities by mutual funds have been at the core of the discussion and index-tracking funds have received special attention because of their exponential growth in size.5 There have been several studies analyzing the consequences on financial markets of the presence of these funds. For instance, Chang et al. (2014) and Raddatz et al. (2015) document the price effects generated by these funds during index rebalancing periods. Sullivan and Xiong (2012), Bhattacharya et al. (2013), Ben-David et al. (2014), and Israeli et al. (2015) show that index-tracking investors increase market vulnerability and volatility.6 However, the evidence on the possible economic consequences of index-tracking investors is slim as highlighted by Wurgler (2010).7 To the best of my knowledge, this paper is the first attempt to document that international capital flows by index-tracking investors have effects on the real economy. More broadly, this paper contributes to the literature on the relationship between international capital flows, credit booms and economic activity. On the one hand, there are several studies analyzing whether capital inflows lead to higher credit growth and an increase in economic activity.8 On the other hand, there are several studies analyzing the relationship between large capital inflows or capital account liberalizations and the consequences for the economy.9 Most of these studies have problems addressing endogeneity issues, since capital flows are almost always related to local economic conditions. I contribute to this literature by using a sudden and unanticipated increase in capital inflows for identification. Thus, I provide evidence that capital inflows to the sovereign debt market cause an increase in credit growth and an expansion in economic activity. The rest of the paper is organized as follows. Section 1 provides the empirical setting, with a detailed account of the index rebalancing. Section 2 provides evidence on the consequences of the event for the sovereign debt market in Colombia. Section 3 presents the identification strategy and empirical analysis for the results on bank lending. Section 4 shows the results for the real economic activity. Section 5 discusses the external validity of the results. Section 6 concludes. 1. Empirical Setting 1.1 Indexing in International Markets International indexes (or international benchmarks) are broad market indexes of different assets that involve several countries. They are constructed by different companies (index providers) such as Morgan Stanley Capital International (MSCI) for international equities or J.P. Morgan for international debt securities. The former constructs, for instance, the MSCI Emerging Markets Index and, the latter, the J.P. Morgan Emerging Markets Bond Index (EMBI), two of the most recognized indexes in the world for emerging countries. The construction process for these indexes involves different broad steps that are used by almost all index providers. They first define the main scope of a benchmark index (such as, geography, industry, and type of firms) and in which category each country is classified at each point in time (developed, emerging, or frontier). Then, they select a number of securities that fall within the scope and meet the size, market capitalization, liquidity, and other requirements. Each of these securities gets a loading (or inclusion factor) in the index portfolio assigned by the index producer according to how much it meets the index-construction criteria and how accessible it is to investors (given by the free-float market capitalization, restrictions to foreign investors, and so forth). The return of the index consists of the returns of its constituent securities, using various approaches to aggregate fluctuations in individual instruments (e.g., Laspeyres, chain-weighting). Namely, each index captures the market capitalization weighted returns of all constituents included in the index.10 The indexes are periodically rebalanced to ensure their continuity and representativeness. Countries’ weights in a specific index are assembled with the portfolio weights of individual securities included in a benchmark index, aggregated at the country level according to the market where the security was issued. That is, international benchmark indexes are typically constructed using a bottom-up approach and consist of composite stock or bond market indexes that include securities from many countries as constituents. The market for local currency sovereign debt indexes is mainly dominated by the World Government Bond Index (WGBI) by Citigroup and the Government Bond Index Emerging Markets (GBI-EM) by J.P. Morgan. The former is a local currency government bond index that includes securities mainly from developed markets. The latter only includes emerging market government debt in local currency.11 While many more funds track the WGBI (approximately 1.5 trillions U.S. dollars) than the GBI-EM (200 billions U.S. dollars), the weights of emerging countries significantly differ in both indexes. For instance, Mexico (one of the few emerging countries included in both indexes) has a weight of around 0.7 percent in the WGBI and of 10 percent in the GBI-EM. Thus, the exposure of emerging markets is generally lower in the WGBI. These indexes have become popular and are frequently used as benchmarks by international mutual funds, which manage a significant part of international assets. By helping alleviate agency problems, benchmarks allow the underlying investors and supervisors to evaluate and discipline fund managers on a short-run basis using, for example, the tracking error of the fund (the deviation of its returns from the benchmark returns). To the extent that the investment strategy of these funds is pinned down by the composition of their benchmark indexes, changes in the weights that a popular benchmark gives to different countries can trigger a similar rebalancing among the funds that track it and result in sizeable movements in international portfolio allocations, capital flows and asset prices. 1.2 Benchmark Change in Colombia On March 19th 2014, J.P. Morgan announced the inclusion of five Colombian bonds into its benchmark indexes. J.P. Morgan constructs three type of major international indexes: (i) foreign currency denominated sovereign debt; (ii) local currency denominated sovereign debt and; (iii) corporate debt. The addition of these bonds involved only local currency sovereign debt indexes, namely the Government Bond Emerging Markets Indexes (GBI-EM). The securities introduced were treasury bonds (named TES) issued by the Colombian government with maturities in 2016, 2018, 2022, 2024 and 2028. The process was done in a phased approach starting at the end of May 2014 and finishing at the end of September 2014. The most popular index, the GBI-EM Global Diversified saw a large rebalancing of Colombia’s benchmark weight.12 It went from nearly 3 to almost 8 percent, representing the largest restructuring by J.P. Morgan in one of its indexes. At the time of the announcement there were estimations of 10 billions U.S. dollars in inflows into the Colombian government debt market with an estimated outstanding debt of 90 billions U.S. dollars.13 This event has several appealing features from an identification perspective. First, the timing of the decision by J.P. Morgan appears to be arbitrary. The reason for Colombia’s inclusion in the index revolves around an improvement in market accessibility and transparency. The note provided to investors by J.P. Morgan stated: “As a result of improved transparency and accesibility for international investors in the local TES market, Colombia sufficiently meets inclusion requirements for complete GBI-EM inclusion.” This note did not mention a specific policy change as the trigger for this inclusion. At the time of the announcement, many newspapers highlighted that this decision could have been motivated by Law 1607 of January 2013, which reduced taxes on foreign investors’ earnings from domestic securities from 33 to 14 percent. Of course, it is impossible to say that this law and more broadly improving economic and institutional conditions in Colombia did not play a role in the decision by J.P. Morgan. However, J.P. Morgan only announced the inclusion of these bonds in March 2014, despite having monthly periodical rebalancings. The gap of more than 1 year between the change in the law and the benchmark change by J.P. Morgan with the use of quarterly data allows me to exploit the timing of the decision to separate the two events. From an identification perspective, the decision by J.P. Morgan should not be related to other changes affecting the foreign demand for sovereign debt in the quarters from March 2014 until September 2014 (the whole process of the rebalancing by J.P. Morgan).14 Second, the event was unanticipated by the markets. Around the years 2013-2014 there were many events affecting Colombia as a country. Then, it is useful to observe the evolution of the price of domestic sovereign bonds in Colombia. (Figure 2, Panel A). The tax reform seems to have a positive impact on the price of these bonds (while part of this upward trend was due to a global factor driving up bond prices of emerging markets during large part of 2012). Not long after the tax reform, the Federal Reserve of the United States started considering unwinding quantitative easing (Taper Tantrum). The beginning of this event was marked by the Federal Reserve Chairman Ben Bernanke’s suggestion of this unwinding in his testimony before Congress on May 22, 2013. Bond prices of emerging markets dropped sharply around that period, and Colombia’s bonds were not an exception as they experienced a drop of nearly 12 percent during that summer. Prices remained low during the rest of 2013, mostly due to the uncertainty generated by Taper Tantrum talks. Upon the announcement by J.P. Morgan in March, Colombian bonds increased in price by almost 5 percent in two weeks. Most of this gain was reversed by the end of the rebalancing in October. The pattern in the price of these bonds suggests that the event was unexpected as there was a sharp and sudden increase upon its announcement.15 Figure 2 View largeDownload slide Domestic sovereign bonds index and capital inflows This figure presents the evolution of the price of the domestic sovereign bonds in Colombia and the capital inflows to the Balance of Payments in Colombia. Panel A shows the bond prices. The bond price index weights the price return of each bond by its outstanding amount. Only the bonds included in the index rebalancing by J. P. Morgan are included and the index equals 100 at the day before the announcement of the rebalancing (March 18, 2014). Panel B depicts the gross inflows from balance of payments data. The dark bars depicts the average inflows during the period 2011–2013, and the gray bars show the inflows during 2014, the year of J. P. Morgan’s index rebalancing. All values are in percentage of nominal GDP. FDI is foreign direct investment; private portfolio flows are liability flows in private portfolio debt and equity; and public debt inflows are liabiity flows to government debt securities. Figure 2 View largeDownload slide Domestic sovereign bonds index and capital inflows This figure presents the evolution of the price of the domestic sovereign bonds in Colombia and the capital inflows to the Balance of Payments in Colombia. Panel A shows the bond prices. The bond price index weights the price return of each bond by its outstanding amount. Only the bonds included in the index rebalancing by J. P. Morgan are included and the index equals 100 at the day before the announcement of the rebalancing (March 18, 2014). Panel B depicts the gross inflows from balance of payments data. The dark bars depicts the average inflows during the period 2011–2013, and the gray bars show the inflows during 2014, the year of J. P. Morgan’s index rebalancing. All values are in percentage of nominal GDP. FDI is foreign direct investment; private portfolio flows are liability flows in private portfolio debt and equity; and public debt inflows are liabiity flows to government debt securities. Third, as the affected index contains only government debt, the event only affected capital flows to the sovereign debt market, and not private capital flows. This is further confirmed by looking at capital flows from the Balance of Payments in Colombia (Figure 2, Panel B). Foreign direct investment and private portfolio flows only experienced a slight increase in inflows relative to GDP. Instead, public debt securities had gross inflows of 0.7 percent of GDP on average before the rebalancing. This number increased to 2.9 percent during 2014, indicating an increase of almost 400 percent. Fourth, the rebalancing by J.P. Morgan did not bring a renewed appetite to issue more debt by the government. In Colombia, the government has a tight fiscal rule and debt issuance programming that could not be affected by the time J.P. Morgan announced this change. Therefore, during the period under analysis, the Colombian government maintained a relatively constant growth of its local currency debt securities, and was unable to change its expenditure pattern significantly.16 The evidence suggests that the event generated a sudden, unanticipated and sizable increase in capital inflows exclusively to domestic sovereign debt markets in Colombia. Moreover, the timing of the event, around the short window of time I study, does not seem to be contemporaneous to sizable changes that would affect the foreign demand for sovereign debt and at the same time credit to the private sector. Thus, one can use the timing of this event to understand how the entrance of foreign investors in the sovereign debt market affects the holdings of government debt by domestic agents, and the credit availability to the private sector. 2. Sovereign Debt Institutional investors that track these indexes closely were forced to rebalance their portfolio. These portfolio changes had aggregate consequences for Colombia in terms of capital flows to local currency sovereign debt as shown in Figure 3, Panel A. This figure presents the rolling 6-month net purchases of TES securities by foreigners and commercial banks.17 Prior to the announcement by J.P. Morgan there were some capital inflows to domestic sovereign debt. However, after the announcement, foreigners started massively buying local currency sovereign debt in Colombia. Purchases made between the end and the beginning of the rebalancing were 8 percent of the total outstanding local currency sovereign debt securities.18 During the same period, foreigners more than doubled their participation in the affected local currency sovereign debt market (Figure 3, Panel B). Furthermore, by the end of 2014 they were the largest holders of the affected bonds. Figure 3 View largeDownload slide Holdings of domestic sovereign bonds This figure presents the net purchases of domestic sovereign bonds in Colombia around the index rebalancing by J. P. Morgan. Panel A depicts 6-month rolling purchases by foreigners and commercial banks. The gray bars indicate the events described in the picture. Panel B shows the percentage of TES bonds held by the different economic agents in the economy before and after the rebalancing. PF, IC, and MF are pension funds, insurance companies, and domestic mutual funds, respectively. Figure 3 View largeDownload slide Holdings of domestic sovereign bonds This figure presents the net purchases of domestic sovereign bonds in Colombia around the index rebalancing by J. P. Morgan. Panel A depicts 6-month rolling purchases by foreigners and commercial banks. The gray bars indicate the events described in the picture. Panel B shows the percentage of TES bonds held by the different economic agents in the economy before and after the rebalancing. PF, IC, and MF are pension funds, insurance companies, and domestic mutual funds, respectively. Another interesting feature of Figure 3 is the different agents that were on the other side of the purchases of domestic sovereign debt by foreigners. Commercial banks, with relatively stable purchases before the announcement, started selling treasury securities in an image that mirrors the one by foreigners (Panel A). Compared to the rest of the agents in the economy, commercial banks were the main providers of liquidity during the rebalancing (Panel B). Out of the 10 percentage point increase in the participation of foreigners in this market, 7 were accomodated by banks (almost a 30 percent decline in their participation in the TES market). Alternatively, pension funds, insurance companies and domestic mutual funds only reduced their participation by 0.3 percentage points (1 percent decline), while public institutions reduced their share in this market by 2.5 percentage points (7 percent decline). Both of these figures suggest that commercial banks in Colombia reduced their holdings of affected bonds by much more than the rest of the agents in the economy. 2.1 Market Makers in TES bonds In Colombia, commercial banks are important participants in the sovereign debt market. Before the benchmark change, banks were the largest participant in the TES market that was affected by the rebalancing. They held collectively around 25 percent of government issued bonds. Moreover, around 11 percent of their assets were local currency sovereign bonds as of December 2013. As noted above, commercial banks were on the other side of the transactions from foreigners in the market for TES bonds during the rebalancing. This suggests that there might be a special feature about commercial banks that may have generated this large response. One noticeable feature of the market for TES in Colombia is that the Finance Ministry designates official market makers in this market. By law, these market makers are within the set of commercial banks, financial corporations and brokerage firms. The objective of this program is to promote adequate conditions for the financing of the government in capital markets by developing the domestic sovereign debt market. Every year, each institution decides whether it wants to participate in the program or not. In order to participate, they need to fulfil a minimum net worth and corporate rating criteria. Then, the Finance Ministry ranks institutions according to their activity in primary and secondary debt markets and designates the official market makers. At the end of every year, a maximum of 20 entities are designated as such. There are two main obligations for these institutions. First, they need to absorb at least 4.5 percent of all primary market debt issuances during the year. Second, they need to quote permanently and simultaneously bid and ask prices in secondary markets subject to a maximum bid-ask price determined by the government.19 On the other hand, designated market makers benefit from having special access to debt issuances from the government, constant access to officials from the Ministry of Finance, and access to a liquidity window in case of problems.20 Under this program, at the end of 2013, 9 commercial banks in Colombia were designated as market makers among 14 participants in the program. 2.2 Conceptual Framework The difference between commercial banks that are designated official market makers and those that are not could be insightful to understand the channel at work during the benchmark rebalancing. These financial institutions are obliged to absorb certain amount of debt issued by the government in the primary market. Every institution by law has to be awarded at least 4.5 percent of total debt issued by the government during the year. For this service, they get a benefit in the form of participating in non-competitive auctions (at lower prices than secondary markets), and potential liquidity assistance if needed. In the case of both frequent auctions and the absence of a diversified investor base in the secondary market, these banks have a lower probability of offloading this debt to other investors. As a result, they keep a considerable amount of it in their balance sheet. After the entrance of foreign investors, these banks can offload debt absorbed in the primary auction more easily and can use the proceeds for other purposes. The type of foreign institutional investors that enter the market is also important. In this episode, these were index-tracking investors. Moreover, the indexes affected (GBI-EM) are market capitalization weighted indexes. As such, every time the government issues this type of debt, it gets included into these indexes, and index-tracking investors have to inmediately buy it, creating an almost certain demand in the secondary market for this debt. This reduces the need for the government to rely on market makers since it increases the probability that market maker banks offload the debt from the primary market. This would predict both a decrease in the holdings of sovereign debt by market makers, and a reduction in the discount the government gives market maker in primary auctions. Figure 4 presents evidence in this direction. Before the rebalancing, the average holdings of market makers were considerably higher than those of the rest of the banks. Between the announcement and the end of J.P. Morgan’s index rebalancing, the ratio of local currency sovereign debt to total assets decreases considerably for commercial banks that are market makers, while it is almost the same for the rest of the banks. This effect is driven entirely by reductions in sovereign debt positions.21 This illustration is consistent with regression results in Table 1. Market makers reduced differentially their exposure to government debt by 7.8 percentage points of assets during the rebalancing by J.P. Morgan.22 Figure 4 View largeDownload slide Domestic sovereign debt exposure in commercial banks This figure shows the evolution of sovereign debt over assets dividing by market maker and non-market-maker banks at the end of 2013. The index is constructed by averaging the growth of domestic debt over total assets at each point in time. The index is normalized to the average holdings of sovereign debt over assets for the two groups in February 2014. The gray bars indicate the events described in the picture. Figure 4 View largeDownload slide Domestic sovereign debt exposure in commercial banks This figure shows the evolution of sovereign debt over assets dividing by market maker and non-market-maker banks at the end of 2013. The index is constructed by averaging the growth of domestic debt over total assets at each point in time. The index is normalized to the average holdings of sovereign debt over assets for the two groups in February 2014. The gray bars indicate the events described in the picture. Table 1 Sovereign debt exposure and market makers Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 This table presents OLS estimations of the growth of local public debt to total assets against a treatment variable using balance sheet quarterly data. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013-2014. The dependent variable is winsorized at the 5th and 95th percent level. Errors are bootstrapped clustered at the bank level. *, **, and *** denote 10, 5 and 1 percent level of significance respectively. Table 1 Sovereign debt exposure and market makers Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 Explanatory variables Dependent variable: Growth local public debt/assets Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.208** –0.210** (0.087) (0.086) Market maker 0.092* (0.055) Time fixed effects Yes Yes Bank fixed effects No Yes Observations 151 151 R-squared 0.093 0.199 This table presents OLS estimations of the growth of local public debt to total assets against a treatment variable using balance sheet quarterly data. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013-2014. The dependent variable is winsorized at the 5th and 95th percent level. Errors are bootstrapped clustered at the bank level. *, **, and *** denote 10, 5 and 1 percent level of significance respectively. Figure 5 shows additional evidence consistent with this channel. After the rebalancing by J.P. Morgan there is an important reduction in the discount paid by the government to market makers in non-competitive auctions for the affected government debt securities. Both the figure and table results show a permanent decrease in this discount. This is consistent with the following narrative. As the sovereign debt market is populated by domestic investors, the government uses these financial institutions as an insurance against desserted auctions and pays them a premium to absorb the issuance of debt in primary markets. Since this debt is not easy to offload, and they receive a discount in the price they keep it in their balance sheet. When foreign investors massively enter into this market, they offload this debt, and there is a reduction in the profit of being market makers as the government does not need these financial institutions as much.23 This channel is consistent with a large literature on financial repression in emerging markets.24 Figure 5 View largeDownload slide Discount in the primary market This figure shows the evolution of the discount in noncompetitive primary market issuances of sovereign bonds in Colombia. It shows the time evolution of this discount. The discount is calculated for each bond in each issuance as the secondary market price divided by primary market price and substracting 1 from that ratio. The solid line is the average across bonds for each month with at least some issuance activity. The dashed line is the average before March 2014 (pre) and the average after the end of May 2014 (post). The gray bars indicate the events described in the picture. Figure 5 View largeDownload slide Discount in the primary market This figure shows the evolution of the discount in noncompetitive primary market issuances of sovereign bonds in Colombia. It shows the time evolution of this discount. The discount is calculated for each bond in each issuance as the secondary market price divided by primary market price and substracting 1 from that ratio. The solid line is the average across bonds for each month with at least some issuance activity. The dashed line is the average before March 2014 (pre) and the average after the end of May 2014 (post). The gray bars indicate the events described in the picture. 3. Bank Lending 3.1 Data and Identification In the previous section, I established that commercial banks that were market makers sold a sizable amount of their positions in local currency government bonds. This section presents the data and methodology to understand whether the proceeds from these sales resulted in an increase in credit by these banks. I use the following baseline specification: $$\Delta L_{ibt}=\theta_{it}+\theta_{b}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{2}X_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\varepsilon_{ibt}$$ (1) where |$L_{ijt}$| is the log of credit to total assets for a city-zone or an industry |$i$|⁠, bank |$b$| at time |$t$|⁠. |$\theta_{it}$| are fixed effects at the city-zone-time or at the industry-time level. |$X_{b}$| is a set of observable variables at the bank-level. |${\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}$| is a dummy variable that indicates whether a commercial bank was a market maker at the end of 2013 or not. |${\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}$| is a dummy variable that takes the value 1 from March until the end of September 2014.25 The identification comes from the difference-in-difference estimation of credit growth for market maker and non market maker banks. During the rebalancing, market makers were more affected than the rest of the banks, since they could sell domestic sovereign debt more easily. The rest of the banks should not be affected by this channel. Therefore, as long as the evolution of credit for both groups before the rebalancing was similar, we can use |${\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}$| as a treatment variable and analyze whether the entrance of index-tracking investors in the sovereign debt market had an effect on bank lending through this channel. |$\beta_{1}$| identifies this by comparing the differential average growth in credit between market maker and non market maker banks during the rebalancing within a city-zone or industry. By comparing observations within a city-zone or an industry I can partially control for any concerns that credit demand may be affecting this estimation.26 For all the estimations, I use standard errors bootstrapped clustered at the bank level.27 I use data from Colombia’s banking system. I gather data from Superintendencia Financiera de Colombia on credit by banks. The main database consists on different types of credit to a specific city-zone (for instance Bogota-Centro Internacional) at the bank level. I match this data on credit with balance sheet data for each bank to use different bank-level variables. I complement data with the official designation of market makers by the Finance Ministry. Data is on a quarterly basis for the 2012-2014 period and contains data for 24 commercial banks on 86 city-zones (with 10 zones). While most of the results use the city-zone credit database, I rely on an alternative database at the industry level for robustnes. This database contains information for 94 industries.28Table 2 presents a list of all the commercial banks with their classification into market makers at the end of 2013, and whether they are domestic or foreign banks.29 Table 2 Commercial banks in Colombia Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No This table shows the commercial banks in Colombia during 2013–2014 and their classification into market makers in 2013 and into foreign or domestic banks. Table 2 Commercial banks in Colombia Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No Bank name Market maker Foreign Banagrario Yes No Banco de Bogota Yes No Bancolombia Yes No BBVA Yes Yes Citibank Yes Yes Corpbanca Yes Yes Davivienda Yes No GNB Sudameris Yes Yes Popular Yes No Red Multibanca Colpatria Yes Yes AV Villas No No Bancamia No No Bancoomeva No No BCSC No No Coopcentral No No Falabella No Yes Finandina No No GNB Colombia No Yes Helm Bank No No Occidente No No Pichincha No Yes Procredit No Yes Santander de Negocios No Yes WWB No No This table shows the commercial banks in Colombia during 2013–2014 and their classification into market makers in 2013 and into foreign or domestic banks. Table 3 shows the descriptive statistics of the balance sheet structure divided by whether a bank was an official market maker at the end of 2013 or not. There are substantial observable differences between the two type of banks. Market makers are larger both in assets and liabilities than non market maker banks. Market makers hold more investments and more local public debt, and thus less total credit in their asset side. On average, market makers hold 15.4 percent of their assets in local public debt, while non market makers hold 8.9 percent of their assets in local currency sovereign debt. Within credit, they seem more exposed to commercial credit, while non market makers lean more to consumer credit. Regarding balance sheet health, all Colombian banks are above the minimum solvency ratio (9 percent for the total solvency ratio) and non market maker banks have a larger solvency ratio than market-maker banks. Table 3 Balance sheet structure of commercial banks (December 2013) Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 This table presents the structure of the balance sheet for commercial banks before Colombia’s benchmark rebalancing by J.P. Morgan. Panel A depicts the asset structure, with all variables in percentage of total assets unless indicated. Panel B shows the liability structure with all variables in percentage of total liabilities unless indicated. Panel C presents other relevant variables. The exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. The solvency ratio is the tier 1 capital divided by risk-weighted assets and market risk. Table 3 Balance sheet structure of commercial banks (December 2013) Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 Total sum Average Median Variable Non-market-markers Market markers Non-market-markers Market markers Non-market-markers Market markers A. Assets Total assets (in trillions COP) 72.3 328.8 5.2 29.9 1.3 16.6 Liquid assets 8.7 8.8 9.9 8.2 8.8 7.2 Investments 14.6 20.3 12.5 23.3 6.2 19.2 Local public debt 8.2 11.5 8.9 15.4 2.3 10.8 Total credit 63.5 63.8 70.3 62.8 75.4 64.8 Commercial credit 31.5 37.7 25.5 32.5 19.0 35.1 Consumer credit 22.6 18.2 27.8 22.8 19.4 20.4 Microcredit 3.0 1.7 12.4 2.4 0.0 0.0 Mortgages 6.4 6.2 4.5 5.1 0.0 4.6 Other assets 6.3 7.3 6.8 6.5 5.7 5.8 ROA 1.0 1.4 0.4 1.2 0.9 1.2 B. Liabilities Total liabilities (in trillions COP) 62.3 281.5 4.4 25.6 1.1 14.9 Total deposits 79.5 76.6 65.5 77.0 75.6 73.1 Credit other institutions 9.3 10.7 20.4 11.5 11.7 11.8 External credit 2.5 4.3 0.7 3.4 0.0 3.0 Debt 7.6 8.8 8.2 7.3 2.7 5.6 C. Other Variables Exchange rate exposure 0.2 1.0 0.3 0.9 0.1 0.9 Solvency ratio 15.1 14.7 28.5 15.6 15.5 13.4 This table presents the structure of the balance sheet for commercial banks before Colombia’s benchmark rebalancing by J.P. Morgan. Panel A depicts the asset structure, with all variables in percentage of total assets unless indicated. Panel B shows the liability structure with all variables in percentage of total liabilities unless indicated. Panel C presents other relevant variables. The exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. The solvency ratio is the tier 1 capital divided by risk-weighted assets and market risk. 3.2 Empirical Analysis Before going in-depth into the full formal analysis, I start by estimating Equation (1) in Table 4 for the period 2013-2014 for different types of credit. Market maker banks significantly increased their total credit growth during J.P. Morgan’s rebalancing. This increase is exclusively driven by commercial credit, rather than by consumer credit. A possible explanation is that banks usually have a relationship established with firms, and this type of credit provides the next best substitute to sovereign debt in Colombia. In the rest of the paper, I will only look at commercial credit growth, since it is the driver large differential in credit between market maker and non market makers. Table 4 Credit and market makers City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 This table presents OLS estimations of the growth of credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. Other credit is the sum of consumer, housing, and micro credit. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 4 Credit and market makers City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 City zone database Total credit Commercial credit Consumer credit Explanatory variables Dependent variable: Growth credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.021*** 0.041** 0.010 (0.008) (0.017) (0.009) Market maker –0.015* –0.004 –0.012 (0.008) (0.018) (0.010) Time fixed effects Yes Yes Yes City zone time fixed effects No No No Industry time fixed effects No No No Bank fixed effects No No No Observations 5,677 5,357 5,654 R-squared 0.033 0.019 0.061 This table presents OLS estimations of the growth of credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. Other credit is the sum of consumer, housing, and micro credit. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 5 presents the results from the main empirical specification. When I control only for time fixed effects, there is a positive and significant differential credit growth of 4.1 percent during the rebalancing (Column 1). The coefficient and standard errors are almost the same when I include city-zone-time fixed effects (Column 2). Moreover, the R-squared goes from 1.9 percent to 13.2 percent indicating that these fixed effects are capturing an important amount of the credit demand varying at the city-zone-time level. When I include bank fixed effects, which control for any unobserved fixed bank characteristics, there is a differential average growth of 3.8 percent for market makers (Column 3). Effects estimated at the city-zone level are very similar when I use the industry database (Columns 4 to 6). The coefficients are statistically similar, and the analysis mirrors the one at the geographical level. One thing to notice is that by using the industry database I partially control for any credit demand shock that may affect tradable and non-tradable industries differently due to the exchange rate changes during the rebalancing. Table 5 Commercial credit and market makers City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5% and 1% level of significance, respectively. Table 5 Commercial credit and market makers City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 City zone database Industry database Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker* 0.041** 0.041** 0.038*** 0.045** 0.045** 0.037*** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.017) (0.016) (0.014) (0.021) (0.022) (0.012) Market maker –0.004 –0.006 0.001 0.004 (0.018) (0.017) (0.014) (0.015) Time fixed effects Yes No No Yes No No City zone time fixed effects No Yes Yes No No No Industry time fixed effects No No No No Yes Yes Bank fixed effects No No Yes No No Yes Observations 5,357 5,357 5,357 7,677 7,677 7,677 R-squared 0.019 0.132 0.189 0.007 0.125 0.139 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5% and 1% level of significance, respectively. The economic size of these estimations are also meaningful. The results in Column 2 suggest that the differential average growth for market makers versus non market makers during the period was 12.8 percent. This implies an increase in commercial credit of 4.2 percentage points of assets. Using the total assets of market maker banks, this amounts to a differential growth in commercial credit of 13.7 trillions of Colombian Pesos.30 These numbers suggest that the shock induced an increase in commercial credit of 2.41 percent of GDP. Moreover, the estimations suggest a considerable substitution between local public debt and commercial credit. The difference between the two changes could be explained by regulatory purposes. If banks do not want to change their risk profile, they would need to increase credit by less than the decrease in sovereign debt since government debt is zero-risk weighted for regulatory purposes. 3.3 Threats to Identification 3.3.1 Parallel Trends There are important identification threats that I address in this section. The key assumption behind the difference-in-difference estimation in Equation 1 is that of parallel trends. I test for the presence of differential credit growth between market maker and non market maker banks before the rebalancing with the following specification: $$\Delta L_{ibt}=\theta_{it}+\theta_{b}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{2} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Pre Rebalancing}}_{t}\varepsilon_{ibt}$$ (2) which is very similar to Equation 1 but with the inclusion of an interaction between the market maker dummy and a dummy that indicates the period previous to the rebalancing. More specifically, this dummy is equal to 1 from June 2013-December 2013, to have the same time length of the period of the rebalancing (3 quarters). The results in Table 6 show that this interaction is not statistically significant, and quantitatively the coefficient is very close to zero. Moreover, the treatment variable remains unchanged with respect to Table 5. All in all, this provides evidence consistent with no differential prior trends in credit growth before the rebalancing period. Table 6 Commercial credit and market makers: Parallel trends test City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. Pre-rebalancing is a dummy that indicates the pre-rebalancing period of the same quarter length (June 2013–December 2013). All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 6 Commercial credit and market makers: Parallel trends test City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 City zone database Explanatory Variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.040** 0.038** 0.038*** (0.019) (0.019) (0.015) Market maker*Pre-rebalancing|$_{\textrm{Jun 2013-Dec 2013}}$| –0.002 –0.004 0.000 (0.018) (0.018) (0.010) Market maker –0.003 –0.003 (0.019) (0.017) Time fixed effects Yes No No City zone time fixed effects No Yes Yes Industry time fixed effects No No No Bank fixed effects No No Yes Observations 5,357 5,357 5,357 R-squared 0.019 0.132 0.189 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. Pre-rebalancing is a dummy that indicates the pre-rebalancing period of the same quarter length (June 2013–December 2013). All estimations are for the period 2013–2014. Estimations without bank fixed effects include levels of all interactions. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. To further illustrate this result, I estimate a cross-sectional version of Equation (1). More specifically, I run the following regression: $$\Delta L_{ibt}=\theta_{i}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{Market Maker 2013}}_{b}+\varepsilon_{ib}$$ (3) where I use bootstrapped errors clustered at the bank level. I do this for every quarter from 2013-2014 and plot this time series in Figure 6. The evidence shows that the market maker dummy is only positive and significant during the rebalancing, giving further support to the identification strategy. Figure 6 View largeDownload slide Cross-sectional estimation betas This figure presents the coefficient from an estimation of the growth of commercial credit over assets to a market maker dummy with city zone fixed effects. Errors are constructed with bootstrapping and are clustered at the bank level. The dashed lines indicate the 5%–95% confidence interval. The gray bars indicate the events described in the picture. Figure 6 View largeDownload slide Cross-sectional estimation betas This figure presents the coefficient from an estimation of the growth of commercial credit over assets to a market maker dummy with city zone fixed effects. Errors are constructed with bootstrapping and are clustered at the bank level. The dashed lines indicate the 5%–95% confidence interval. The gray bars indicate the events described in the picture. Another possible threat to identification is the existence of a differential prior evolution of the outcome variable. While the placebo test and plot are partially showing this, I present a figure with the actual evolution of the growth of commercial credit over assets during the period of study. Figure 7, Panel A presents these trends, and shows that the evolution of credit between market makers and the rest of the banks before the rebalancing is quite similar, and it differs considerably afterwards, with both a large increase in growth for market makers and a an important decrease in the growth for non market makers. Figure 7 View largeDownload slide Commercial credit growth evolution This figure shows the evolution of the commercial credit over total assets in time. Panel A shows the growth of commercial credit over total assets dividing by market maker and non market maker banks at the end of 2013. Panel B depicts the aggregate commercial credit over total assets. The gray bars indicate the events described in the picture. Figure 7 View largeDownload slide Commercial credit growth evolution This figure shows the evolution of the commercial credit over total assets in time. Panel A shows the growth of commercial credit over total assets dividing by market maker and non market maker banks at the end of 2013. Panel B depicts the aggregate commercial credit over total assets. The gray bars indicate the events described in the picture. The credit growth decrease for non market makers brings an additional concern for the difference-in-difference analysis, since it might be possible that market makers ended up taking business away from non-market maker banks. To address these concerns, I complement this evidence with the aggregate commercial credit over assets on a monthly frequency in Figure 7, Panel B. Commercial credit seems to be decreasing constantly during 2013 just until the start of the rebalancing by J.P. Morgan, when it starts to grow considerably until June 2016. This bottom panel shows that the banking sector as a whole was experiencing an important decline in the growth of credit of almost 2.5 percent from January 2013 until March 2014. Thus, in the difference-in-difference analysis non-market makers potentially act as a control group for the economic conditions in Colombia and are accounting for the fact that absent the shock commercial banks should have experienced a further decrease in credit growth during that period. This explanation is also sustained by the evidence in Section 4 which looks at the real effects in the economy. In the case of a pure redistribution of credit between market and non market makers, one should not observe any effects on the real economy. As shown in that section, industries with a higher exposure to market makers experienced increase in the growth of employment, production, sales and GDP, which suggests that the effect documented in Table 5 is not driven by a pure redistribution in credit between the two types of banks. 3.3.2 Omitted Variable Bias The positive differential effect on credit growth between market makers and non market makers estimated in Table 5 could also be affected by omitted variables. As acknowledge above, there could be institutional or economic changes during the period of the rebalancing that could be affecting the estimations. The fixed effects at the city-zone-time level partially controls for this. But there could still be other unobserved factors that could be biasing the estimates. In this section, I analyze the robustness of these estimates to omitted variable bias using a method developed by Oster (2014) that builds on Altonji et al. (2005) that uses both changes in coefficients and the R-squared after introducing observable controls.31 The basic idea both works is to provide a lower bound of the coefficients estimated, and a degree of selection that would make the estimated coefficient zero. Altonji et al. (2005) use this degree of selection to argue whether an estimate is likely to be causal or not. Oster (2014) builds on this and provides a lower bound on the estimates using both the movements in estimated coefficients and changes in the R-squared after including controls in the regression. More formally, she proposes to report a bias-adjusted coefficient: $$\beta^{*}=\tilde{\beta}-\delta \left[\mathring{\beta}-\tilde{\beta}\right]\frac{R_{\textrm{max}}-\tilde{R}}{\tilde{R}-\mathring{R}}$$ (4) where |$\tilde{\beta}$| (⁠|$\tilde{R}$|⁠) is the coefficient (R-squared) from a regression including control variables and |$\mathring{\beta}$| (⁠|$\mathring{R}$|⁠) is the coefficient from a regression with no controls. |$\delta$| is a parameter that indicates how strongly unobservables are correlated with the treatment variable relative to observables and |$R_{\textrm{max}}$| is the R-squared from a hypothetical regression of the outcome variable on both obersvables and unobservables. Oster (2014) suggests using |$\delta=1$| such that unobservables are equally important as observables. |$R_{\textrm{max}}$| has a natural upper bound of 1, but Oster (2014) acknowledges that it is common for the outcome not to be explained in its entirety by observables and unobservables and thus suggests using |$R_{\textrm{max}}=\min\{2.2\tilde{R},1\}$|⁠. Moreover, Oster (2014) also provides a |$\hat{\delta}$| such that |$\tilde{\beta}=0$|⁠, that indicates the degree of selection on unobservables relative to observables that would be sufficient to explain away the result. More specifically: $$\hat{\delta}=\frac{(\tilde{\beta})(\tilde{R}-\mathring{R})}{(\mathring{\beta}-\tilde{\beta})(R_{\textrm{max}}-\tilde{R})}$$ (5) Consistent with this framework, I use an uncontrolled regression and the regression with city-zone-time fixed effects in Table 7 and different values for |$\delta$| and |$R_{\textrm{max}}$| to provide a lower bound for the estimated treatment effect. Furthermore, the table and presents both the case of |$\delta=1$| (Panel A) and |$\delta=2$| (Panel B). Consider first the most conservative scenario where observables and unobservables explain all the variation in commercial credit growth (⁠|$R_{\textrm{max}}=1$|⁠) and the unobservables have double the importance of the observables |$\delta=2$|⁠, the bias-adjusted coefficient is different from zero. This scenario, however is highly unlikely as it is difficult to explain fully quarter to quarter growth in credit in general. In a more realistic scenario, where |$R_{\textrm{max}}=\min\{2.2\tilde{R},1\}=0.29$| and observables and unobservables are equally important, the bias-adjusted coefficient is 0.039, which is very close to the 0.041 coefficient estimated in Table 5. More importantly, under this scenario, |$\hat{\delta}=30.475$|⁠. This signals that unobservables would have to be 30 times more important than observables to attribute the entire treatment effect to these unobservable variables. Even in the most conservative scenario, where |$R_{\textrm{max}}=1$| the unobservables would have to be 5 times as important as the observables to explain to result on the differential credit growth away. All in all, these results provide evidence that the estimated treatment effect is robust to omitted variable bias. Table 7 Commercial credit and market makers: Coefficient stability |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 This table presents the coefficient from the main estimations in Table 5 with city zone time fixed effects and the bias-adjusted coefficients with the method developed by Oster (2013). Panel A presents results when the unobservable variables have equal importance as the unobservables, and panel B shows results when the unobservables have double importance than the observables. Bias adjusted |$\beta$| is computed like in Oster (2013). R|$_{\rm max}$| denotes the unknown overall R-squared in a model that controls for both observables and unobservables. |$\delta$| for |$\beta=0$| is the proportionality coefficient that would imply a bias-adjusted coefficient of 0. Table 7 Commercial credit and market makers: Coefficient stability |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 |$\beta$| controlled Bias adjusted |$\beta$| R|$_{\text{max}}$| |$\sigma$| for |$\beta=0$| A. |$\sigma=1$| 0.041 0.040 0.198 73.140 0.041 0.039 0.290 30.475 0.041 0.036 0.660 9.143 0.041 0.033 1.000 5.561 B. |$\sigma=2$| 0.041 0.039 0.198 73.140 0.041 0.038 0.290 30.475 0.041 0.032 0.660 9.143 0.041 0.026 1.000 5.561 This table presents the coefficient from the main estimations in Table 5 with city zone time fixed effects and the bias-adjusted coefficients with the method developed by Oster (2013). Panel A presents results when the unobservable variables have equal importance as the unobservables, and panel B shows results when the unobservables have double importance than the observables. Bias adjusted |$\beta$| is computed like in Oster (2013). R|$_{\rm max}$| denotes the unknown overall R-squared in a model that controls for both observables and unobservables. |$\delta$| for |$\beta=0$| is the proportionality coefficient that would imply a bias-adjusted coefficient of 0. 3.3.3 Credit Supply or Demand? Another potential concern with the estimated treatment effect is whether it is actually capturing an increase in the credit availability by banks, or an increase in credit demand by local firms. The analysis regarding the potential omitted variable bias already suggests that the effect is mainly driven by an increase in credit supply by banks. Notice that the fixed effects at the city-zone-time level act as a proxy for credit demand at that level under the assumption that a single city-zone changes its demand equally for the two types of banks. By the analysis above, any other unobservable related to credit demand that are correlated with these fixed effects should not be affecting considerably the estimated coefficients. This lends support to the hypothesis that this effect is coming from an increase in credit supply rather than by changes in credit demand. I complement these results with evidence from interest rates. If the effect is mainly driven by credit supply we should observe a higher decrease in interest rates by market maker banks during the period of the rebalancing. I use the average commercial credit interest rates for commercial banks in Colombia and estimate a specification analogous to the one for credit growth: $$\Delta r_{bt}=\theta_{t}+\theta_{b}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\varepsilon_{bt}$$ (6) where |$r_{bt}$| is the log of interest rates on commercial credit for bank |$b$| at time |$t$|⁠. Results are presented in Table 8. During the period of the rebalancing market makers had a significantly lower growth of interest rates on commercial credit, even when I include different type of fixed effects. Quantitatively, the average commercial interest rate before the rebalancing for market makers was 18.2 percent, and thus the estimates suggest a decrease of 5.37 percentage points in the interest rates on commercial credit relative to non market maker banks. Thus, the evidence points towards a considerable decrease in interest rates from market maker banks, which is consistent with a change in the credit availability from market makers.32 Table 8 Commercial credit and market makers: Interest rates Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 This table presents OLS estimations of the growth of the interest rate on commercial credit against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% levels of significance, respectively. Table 8 Commercial credit and market makers: Interest rates Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 Explanatory variables Dependent variable: Growth interest rate commercial credit (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.115** –0.111** –0.119** (0.058) (0.054) (0.060) Market maker 0.043** 0.041** (0.022) (0.019) Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.12** 0.124** (0.058) (0.059) Time fixed effects No Yes No Bank fixed effects No No Yes Observations 159 159 159 R-squared 0.086 0.166 0.158 This table presents OLS estimations of the growth of the interest rate on commercial credit against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in logs. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% levels of significance, respectively. 3.3.4 Treatment and Control Groups A possible concern with the treatment effects estimated in Table 5 is the fact that market makers are not randomly selected. Their participation in the program is endogenous and might depend among others on their trading expertise, their assets and in the end the two type of banks might be considerably different which might affect the main estimations. In this section I use the matching estimator developed by Abadie and Imbens (2002) and Abadie et al. (2004) to more closely match market and non market maker banks.33 I use an exact matching by the city-zone a bank is present, and the nearest matching through both assets and the bank solvency ratio. Furthermore, I allow for 1, 2 and 5 matches in order to use several non market makers as controls.34 I use the matching estimator for the period of the rebalancing and consider whether a bank is a market maker or not as the treatment variable (Table 9). Thus, the estimator matches any observation from the treatment group (i.e. credit growth in a city-zone by a market maker bank), with one or more observations from the control group (i.e. credit growth in the same city-zone by a non-market maker bank). Then, it selects from the control group the closest observations, in distance, using the total assets and solvency ratio to complete the matching. The table also presents the observations used in each estimation, and the percentage of exact matches.35 The results show coefficients very similars to the ones in my main specification, showing a positive and significant differential credit growth for market makers. As the number of exact matches decay, the coefficient is somewhat lower, but still statisitcally significant. These findings suggest that the potential differences between market maker and non market maker banks are not entirely driving the estimated effects. Table 9 Commercial credit and market makers: Matching estimators m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 This table presents the average treatment effect through the corrected matching estimators developed by Abadiee Imbens (2002). The dependent variable is the the growth of commercial credit to total assets. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy. All estimations are for the period March 2014–Sep 2014. Observations are matched, as possible, exactly through their city zone and with their nearest match for assets and solvency ratio in the different panels. Percentage of exact matches denotes the number of observations that could be exactly matched through their city zone. If not possible, the observations are matched through the closest observation in the same city. The dependent variable is winsorized at the 5th and the 95th percentile. m(.) indicates the number of matches in each estimation. Robust standard errors are in parentheses. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 9 Commercial credit and market makers: Matching estimators m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 m(1) m(2) m(5) Explanatory variables Dependent variable: Growth commercial credit/assets (Rebalancing|$_{\text{Mar 2014-Sep 2014)}}$| A. Exact matching city zone Market maker 0.035*** 0.035*** 0.036*** (0.007) (0.007) (0.007) Observations 1,944 1,944 1,944 Percentage of exact matches 97.53 97.31 93.15 B. Exact matching city zone, nearest matching assets Market maker 0.035*** 0.032*** 0.031*** (0.011) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 90.72 92.51 84.11 C. Exact matching city zone, nearest matching assets, and solvency ratio Market maker 0.033*** 0.028*** 0.028*** (0.009) (0.008) (0.008) Observations 1,944 1,944 1,944 Percentage of exact matches 87.87 88.39 79.47 This table presents the average treatment effect through the corrected matching estimators developed by Abadiee Imbens (2002). The dependent variable is the the growth of commercial credit to total assets. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy. All estimations are for the period March 2014–Sep 2014. Observations are matched, as possible, exactly through their city zone and with their nearest match for assets and solvency ratio in the different panels. Percentage of exact matches denotes the number of observations that could be exactly matched through their city zone. If not possible, the observations are matched through the closest observation in the same city. The dependent variable is winsorized at the 5th and the 95th percentile. m(.) indicates the number of matches in each estimation. Robust standard errors are in parentheses. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. 3.4 Robustness Tests In this section, I control for the different alternative hypothesis that could be affecting the main market making channel. In Table 10, Columns 1-3 I present three different tests.36 First, I include several bank level controls interacted with the rebalancing dummy to control for other potential channels (Column 1). I include the sovereign debt and exchange rate exposure, the size of banks, their return on assets, their liquid assets over total assets, their solvency ratio, their corporate debt issuance divided by total liabilities and whether a bank is foreign or not. While some of these controls are significant, the treatment variable is of similar size as the one estimated in Table 5, and still significant. Second, I remove the only state-owned bank (Banagrario) from the regression. The results show that there is no difference in the results from excluding this bank (Column 2). Third, I remove Bancolombia, the largest and most important bank in Colombia, and this does not influence the main results (Column 3).37 Table 10 Commercial credit and market makers: Robustness tests City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Corporate debt is the debt securities issued by a bank divided by total liabilities. Foreign is a dummy indicating whether a bank is foreign. Domestic sovereign debt exposure is the initial local public debt divided by assets. Exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. Assets is the initial log of assets. Liquid assets is the initial liquid assets to total assets. ROA is initial return on assets. Solvency ratio is the initial tier 1 capital divided by risk-weighted assets and market risk. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. Table 10 Commercial credit and market makers: Robustness tests City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 City zone database Full sample Without Banagrario Without Bancolombia Explanatory variables Dependent variable: Growth commercial credit/assets (2013–2014) Market maker*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.070*** 0.073*** 0.069** (0.006) (0.021) (0.028) Assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.005 –0.006 –0.006 (0.004) (0.005) (0.009) Liquid assets*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.004** 0.004* 0.004 (0.002) (0.002) (0.006) ROA*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.019*** 0.019*** 0.019 (0.004) (0.006) (0.015) Solvency ratio*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.001 –0.001 –0.001 (0.001) (0.001) (0.001) Domestic sovereign debt exposure* –0.197*** –0.136 –0.187 |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.071) (0.239) (0.162) Exchange rate exposure* –0.023*** –0.025*** –0.023** |$\quad$| Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| (0.004) (0.008) (0.009) Corporate debt*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| –0.000 –0.000 –0.000 (0.001) (0.001) (0.001) Foreign*Rebalancing|$_{\textrm{Mar 2014-Sep 2014}}$| 0.028*** 0.021 0.029** (0.009) (0.015) (0.013) Time fixed effects No No No City zone time fixed effects Yes Yes Yes Bank fixed effects Yes Yes Yes Observations 5,184 5,025 4,710 R-squared 0.202 0.206 0.212 This table presents OLS estimations of the growth of commercial credit to total assets against different explanatory variables for commercial banks and a treatment variable. The growth of the dependent variable is constructed as the difference in log. The treatment variable is a market maker dummy multiplied by a dummy indicating the period of the index rebalancing by J. P. Morgan. All estimations are for the period 2013–2014. Corporate debt is the debt securities issued by a bank divided by total liabilities. Foreign is a dummy indicating whether a bank is foreign. Domestic sovereign debt exposure is the initial local public debt divided by assets. Exchange rate exposure is the total assets minus total liabilities denominated in foreign currency divided by the net worth. Assets is the initial log of assets. Liquid assets is the initial liquid assets to total assets. ROA is initial return on assets. Solvency ratio is the initial tier 1 capital divided by risk-weighted assets and market risk. The dependent variable is winsorized at the 5th and the 95th percentile. Errors are bootstrapped clustered at the bank level. *, **, and *** denote the 10%, 5%, and 1% level of significance, respectively. There is also the possibility that market maker banks have trading expertise and the results capture a similar channel to that in Abbassi et al. (2016). As the affected bonds increase in price, banks with expertise reduce their holdings since these securities are overvalued, and thus increase the credit supply. For the evidence to be consistent with this channel, I should observe that after an initial overvaluation of the bonds’ price, the banks return to a similar level of ex-ante holdings of debt. However, the evidence is not consistent with a reversal of sovereign debt holdings. The price of sovereign bonds have a maximum peak the first week of April 2014. After that, these bonds suffer a 3 percent decrease in their price until the end of July. However, the average holdings of sovereign debt by market makers have a constanst decrease with a minimum at the end of July. This suggests that the evidence is not consistent with the channel presented by Abbassi et al. (2016) in the case of Germany. 4. Real Effects 4.1 Manufacturing Industries Data Until now I have shown that the entrance of foreign investors had consequences for the extension of private credit in Colombia. However, it remains to be seen whether this credit shock had consequences for the real economy. A problem towards this end is that I do not have access to the information at the bank-firm level such as the ones provided by the credit registry of each country. Therefore I need to rely on alternative data to gain understanding of the impact of the shock on the real economy. I gather data for a subsample of industries from the Monthly Manufacturing Polls conducted by the Departamento Administrativo Nacional de Estadistica (DANE) in Colombia. This database contains the yearly growth at monthly level of employment, production and sales for each manufacturing industry. Additionally, I obtain quarterly data from national accounts on the real annual GDP growth for each of these manufacturing industries. Then, I can construct a proxy of the exposure to market maker banks at the industry level. More specifically, $$Exp_{i,2013}^{MM}=\frac{\sum_{b\in MM}C_{ib,2013}}{\sum_{b\in B}C_{ib,2013}}$$ (7) where |$Exp_{i,2013}^{MM}$| is the exposure of industry |$i$| to market makers at the end of 2013. The numerator indicates the total credit extended by market makers to industry |$i$| at the end of 2013. The denominator contains the total credit extended by all banks to industry |$i$|⁠. This information is helpful to understand whether a given industry was more likely to be exposed to these banks before the benchmark change. Internet Appendix Table 9 shows the 20 industries that I was able to match between the two databases, along with the specific exposure at the end of 2013. This exposure is also presented graphically in Figure 8. The exposure to market makers from most of the industries is high, probably a consequence of these banks having a sizable part of the total assets in the banking system in Colombia. While this is a subsample of 20 industries, there is still significant variation in the exposure to market makers. The average exposure is 82.65, with a standard deviation of 8.4. Figure 8 View largeDownload slide Exposure to market makers across industries This figure presents the exposure of each industry to market maker banks at the end of 2013. The exposure is constructed by summing the commercial credit of market maker banks to each industry and dividing it by the total credit to the same industry by all commercial banks. Figure 8 View largeDownload slide Exposure to market makers across industries This figure presents the exposure of each industry to market maker banks at the end of 2013. The exposure is constructed by summing the commercial credit of market maker banks to each industry and dividing it by the total credit to the same industry by all commercial banks. To analyze whether the shock to credit had an impact on the real variables for these industries, I estimate the following specification: $$\Delta y_{it}=\theta_{t}+\theta_{i}+\gamma_{1}Exp_{i,2013}^{MM}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\varepsilon_{it}$$ (8) where |$y_{it}$| is either employment, production, sales or GDP. |$\theta_{t}$| and |$\theta_{i}$| are time and industry fixed effects, respectively. |$Exp_{i,2013}^{MM}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}$| is the treatment variable, with the exposure to market makers interacted with a time dummy for the period after the rebalancing to take into account any potential lags. The estimation is a classical differences-in-differences approach, and |$\gamma_{1}$| captures the differential effect on real variables of being more exposed to the credit shock. Errors are bootstrapped clustered at the industry level.38 Results show that being more exposed to a market maker bank led to a statistically significant increase in employment, production, sales and GDP during the rebalancing period (Table 11). The economic size of the effect is important. For example, consider an interquartile movement for an industry in the exposure to market makers, which implies an almost 8.7 percentage point increase in the exposure to market makers. This would have implied an increase in the average yearly growth of 1.2, 2.9, 2.9, and 2.8 percentage points for employment, production, sales, and GDP respectively. Overall, there were important effects for economic activity from the credit shock. Moreover, most of the effect for employment is driven by increases in employment directed towards production activities. Table 11 Real effects Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 This table presents OLS estimations of the yearly growth of real variables against different set of fixed effects and a treatment variable for manufacturing industries. The growth of the dependent variable is constructed as the difference in logs of a month/quarter versus the month/quarter of the year before. The treatment variable is the exposure of an industry to market makers in 2013 multiplied by a dummy indicating the period after the index rebalancing by J. P. Morgan. All estimations are for the period 2012m3–2015m2/2012q2–2015q1. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the industry level. *, **, and *** denote 10%, 5%, and 1% level of significance, respectively. Table 11 Real effects Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 Monthly Manufacturing database Quarterly National Accounts database Employment Administrative employment Production employment Production Sales GDP Explanatory variables Dependent variable: Growth real variable (2012–2015) Exposure market maker* 0.150** 0.121 0.168*** 0.339** 0.336*** 0.325** |$\quad$| Rebalancing|$_{\text{Mar 2014-Feb 2015}}$| (0.068) (0.111) (0.065) (0.148) (0.117) (0.145) Time fixed effects Yes Yes Yes Yes Yes Yes Industry fixed effects Yes Yes Yes Yes Yes Yes Observations 680 680 680 680 680 196 R-squared 0.344 0.309 0.380 0.353 0.363 0.311 This table presents OLS estimations of the yearly growth of real variables against different set of fixed effects and a treatment variable for manufacturing industries. The growth of the dependent variable is constructed as the difference in logs of a month/quarter versus the month/quarter of the year before. The treatment variable is the exposure of an industry to market makers in 2013 multiplied by a dummy indicating the period after the index rebalancing by J. P. Morgan. All estimations are for the period 2012m3–2015m2/2012q2–2015q1. The dependent variable is winsorized at the 1st and the 99th percentile. Errors are bootstrapped clustered at the industry level. *, **, and *** denote 10%, 5%, and 1% level of significance, respectively. 5. External Validity The results showing that governemnt acess to foreign credit increases the private access to credit are based in the rebalancing event that happened in Colombia. However, these results could well be applicable to other countries. First, the rebalancing events in government debt index of this style has become more and more common. Countries such as Mexico and South Africa were added to the World Government Bond Index (constructed by Citigroup) and Argentina, Czech Republic, Romania and Nigeria were included in the same index studied in this paper. China and India have yet to be included in one of these indexes, which could trigger similar effects to the ones shown for Colombia for the private access to credit and economic activity. In the end, the size of the rebalancing for Colombia, and the access to very detailed data on credit and economic activity, provided a natural laboratory to study this question. Second, the market maker program used in Colombia is not specific to that country. Other emerging countries in Latin America such as Mexico and Peru use this kind of program, with similar obligations and benefits for the financial corporations that participate in it. The same happens for emerging countries in Asia and Europe such as Indonesia, Poland and Romania among others. Third, Colombia is an average country in terms of the size of its local currency debt market. Du et al. (2016) report that Colombia has a ratio of local currency debt to central government debt close to 60 percent. This number is similar to the numbers of countries such as Czech Republic, Hungary, Indonesia, Philippines, Poland, Russia, South Africa, and only somewhat higher than countries like Brazil, Israel, Mexico, and Peru (closer to 40 percent). Together, these three facts suggest that the findings in this paper could be applicable to at least a considerable number of emerging countries. Fourth, this episode is also important from the point of view of financial liberalization. There has been several works on how capital account liberalization affects economic growth and investment.39 Colombia is a country that has an open capital account, but previous to this episode with a low access to foreign investors in the domestic sovereign debt market. Thus, even in a country open to international capital flows, working on policies to increase the access to foreign investors in the sovereign debt market could lead to an increase in domestic credit and economic activity. This sheds light on the possible consequences on improving the access to the domestic sovereign debt markets in countries such as China. 6. Conclusions In this paper, I exploit a sudden and unanticipated shock that triggered the entrance of foreign investors to the local currency sovereign debt market in an emerging market. I use an episode in which J.P. Morgan introduced several Colombian bonds in its local currency governemnt debt indexes in emerging markets. Since foreign institutional investors often use benchmark-tracking strategies, they rebalanced their portfolio towards Colombia increasing capital inflows to the domestic sovereign debt market. As foreign investors purchased this debt, banks officially designated as market makers in the sovereign debt market decreased their exposure to these securities. On average, they reduced their local public debt over total assets by 7.8 percentage points, compared to the rest of the banks. This shock, that was originated on the government debt market, spillovered to the credit market. Market maker banks increased their commercial credit to total assets ratio by 4.2 percentage points on average, relative to the rest of the banks. This transmission channel is not found during other periods and is not driven by other observable differences between market and non market maker banks. The evidence suggests that the shock had an impact on the real economy. I construct a proxy for the exposure to market maker banks at the industry level and find that industries with more exposure to market makers had a higher growth of employment, production, sales and GDP during this period. The evidence is consistent with a crowding out of private credit before the entrance of foreign investors and with a financial repression channel. Because of the illiquidity of the sovereign debt market, the government used market makers to absorb debt issued in the primary market. As foreign institutional investors entered the domestic sovereign debt market, these domestic financial institutions were able to sell the excess of debt that they could not offload before and used the proceeds to extend credit. The findings in this paper has implications from a policy-making perspective for a number of reasons. For instance, large countries such as China and India still have less than 2 percent of their local currency debt in the hands of foreign investors. This study sheds light on the possible consequences of using policies to increase the share of foreign investors in domestic sovereign debt markets. Moreover, on March 2016, J.P. Morgan included China on a watchlist to enter the GBI-EM. The evidence suggests that the confirmation of this process could lead to a boost in private credit for the Chinese economy. Another consideration is that China would have the largest weight in the index (10 percent). As a consequence, its introduction to the index could lead to a decrease in the weights of the rest of the countries. This might ultimately lead to negative spillovers to the other constituents of the GBI-EM. Finally, results also suggest that sovereign debt index rebalancing can have effects on the economy beyond the usual price effects found in the literature. Thus, one policy concern is related to the regulation of activities of both institutional investors with index-tracking strategies and index providers. Ultimately, the effects documented in this paper were started by a decision made by a single index provider. Is this desirable? Should there be regulation on the construction of benchmark indexes and their reconstitutions? Moreover, a decision to remove Colombia from the index may produce the opposite effect. The evidence suggests that activities by index-tracking investors and index providers should be followed more closely by policy makers. I am indebted to Fernando Broner, Alberto Martin and José Luis Peydró for their continuing guidance and support throughout this project. This paper benefited from numerous discussions with Lorenzo Pandolfi. I would like to thank Philip Strahan and two anonymous referees for their insightful suggestions that improved the paper significantly. I would also like to thank Gene Ambrocio, Aitor Erce, Juanita Gonzalez Uribe, Graciela Kaminsky, Mauricio Larrain, Eduardo Levy Yeyati, Maria Lombardi, Roberto Pancrazi, Daniel Paravisini, Carolin Pflueger, Francisco Queiros, Veronica Rappoport, Diego Restrepo Tobon, David Reeb, Francesc Rodriguez Tous, Tom Schmitz, Sergio Schmukler, Jesse Schreger, Jay Shambaugh and Jagdish Tripathy for insightful comments. This work has also benefited from comments by participants at the Barcelona GSE PhD Jamboree, Barcelona GSE Summer Forum, the Belgrade Young Economists Meeting, the CREI International Lunch Seminar, the European Doctoral Program Jamboree, the International Conference on Sovereign Bond Markets, the LSE Finance Ongoing Work Seminar, NBER IFM Meetings, the SAE Bilbao, the SFS Cavalcade, the UPF Finance Lunch Seminar, the World Finance Conference. I am extremely grateful to Carla Maenza for her invaluable and constant feedback during the writing of this paper. All errors are my own. Supplementary data can be found on The Review of Financial Studies web site. Footnotes 1 See for instance Diamond (1965). 2 As shown by Du et al. (2016) and Du and Schreger (2016) most of the local (foreign) currency sovereign bonds in emerging markets are issued under domestic (foreign) law, and traded in domestic (foreign) markets. Thus, the terms domestic sovereign debt and local currency sovereign debt will be used interchangeably throughout the paper. 3 Ongena et al. (2016) document a similar channel for banks during the European Sovereign Debt Crisis. 4 This empirical literature is closely related to the growing literature on the real effects from credit supply changes. See among others Gan (2007), Ivashina and Scharfstein (2010), Iyer and Peydro (2011), Jimenez et al. (2012), Jimenez et al. (2014), Iyer et al. (2014). Within this literature my paper is more related to several papers studying the effect of international shocks to emerging markets and lending by banks. See Khwaja and Mian (2008), Paravisini (2008), and Schnabl (2012) for episodes in Pakistan, Argentina and Peru, respectively. 5 This phenomenon is the consequence of a large switch of investor funds from active to passive funds and a documented movement of active funds into more passive investment strategies. See among others Cremers and Petajisto (2009). 6 More broadly, there is a large literature on the aggregate effects of international mutual funds on financial markets. See among others Broner et al. (2006), Jotikasthira et al. (2012), Levy-Yeyati and Williams (2012) and Raddatz and Schmukler (2012). 7 There has been a recent literature focusing on the real effects of institutional investors flows in general. Chernenko and Sunderam (2012) analyze how flows into high-yield mutual funds have effects on the issuance of firms and their investments. Adelino et al. (2014) document how changes in credit ratings by Municipalities in the United States have consequences for public financing and for economic activity. Almeida et al. (2015) and Adelino and Ferreira (2015) document how credit ratings upgrades and downgrades affect firms’ real investment decisions and banks’ credit supply. 8 For instance, Mendoza and Terrones (2012) find that credit booms are positively correlated with net capital inflows. Calderon and Kubota (2012) suggest that private capital inflows are good predictors of credit booms. In a more granular approach, Lane and McQuade (2014) argue that only net debt inflows generate domestic credit growth in European countries. In a related theoretical and empirical work, Blanchard et al. (2015) find that only equity inflows are correlated to credit expansions. 9 Reinhart and Reinhart (2009) study how capital flow bonanzas affect the likelihood of economic crises. Caballero (2016) shows that capital inflows bonanzas increase the probability of banking crises. Kalantzis (2015) and Benigno et al. (2015) study the changes in the sectorial allocation of resources due to large capital inflows. There is a large literature on how capital account liberalizations and how they affect investment and economic growth. For a detailed review see Henry (2007). More recently, there has been studies on how capital account liberalization affects inequality. For instance, Larrain (2015) shows that opening the capital account leads to a greater wage inequality. 10 More recently, index providers have focused on constructing alternative indexes not based on market capitalization (such as GDP-weighted indexes or fundamentals based indexes). 11 J.P. Morgan also constructs the Emerging Markets Bond Index (EMBI) which is a foreign currency sovereign debt index. Since emerging markets governments shifted their preference towards local currency debt, this index has been steadily declining in popularity. 12 The benchmark weight of a country is defined as the sum of the market capitalization of all securities issued in a country divided by the total market capitalization of all the securities included in the benchmark index. 13 Reuters (2014). 14 Even during those months it is impossible to assert that there were no changes at all in the economy of a country. Section 3.3 presents different tests to alleviate these concerns. 15 The Internet Appendix provides an account of the exchange rate evolution during this period. Since there were large capital inflows, the exchange rate appreciated considerably during the first part of the rebalancing. 16 See Internet Appendix Figure 1, Panel A. 17 I use the 6-month rolling sum to compare the aggregate activity by foreigners and commercial banks by the end of the period of the rebalancing. Internet Appendix Figure 2 presents the same figure with monthly data which shows a similar albeit noisier pattern. 18 This number increases to 10.3 percent if we consider the March-December 2014 period, accounting for the fact that some funds slowly change their positions. 19 The other two obligations are to construct monthly reports about the state of the domestic sovereign debt market and to inform of any mergers/sales to the treasury department. 20 (Ministerio Hacienda y Credito Publico, 2010) 21 More strikingly, all market maker banks were net sellers of sovereign debt during the event, and half of the banks among non market makers were net sellers of government bonds. 22 This number is obtained by using the average local currency sovereign debt to total assets ratio in December 2013 multiplied by 1 minus the coefficient estimated for the three quarters of the rebalancing. 23 Anecdotal evidence from several people working in banks around the period suggest that the reduction in sovereign debt from market makers was due to the entrance of foreign investors, and a reduction in profitability in the market making business for these banks. 24 See Reinhart and Sbrancia (2011) for a detailed account of this literature. 25 While this is the full fledge specification, I also estimate models with only time fixed effects. More specifically, \begin{equation*} \Delta L_{ibt}=\theta_{t}+\beta_{1} {\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{2}X_{b}{\rm 1}\kern-0.24em{\rm I}^{\textrm{Rebalancing}}_{t}+\beta_{3}{\rm 1}\kern-0.24em{\rm I}^{\textrm{MM 2013}}_{b}+\beta_{4}X_{b}+\varepsilon_{ibt} \end{equation*} which includes the levels of the market maker dummy, and the levels of any bank observable characteristic used without the interaction with the rebalancing dummy. 26 As explained by Adelino et al. (2015) this might not be a perfect control for credit demand. However, I will try to present suggestive evidence on the unlikely possibility that this shock is coming through credit demand. While a more sound strategy would be controlling at the firm level, I do not have data on the credit register of Colombia. In some cases, even data at the bank-firm level might not capture perfectly credit demand as suggested by Paravisini et al. (2015). 27 Additionally, credit growth is winsorized at the 5 and 95 percent level since this variable has very large values, with its standard deviation being 0.469 (or 46.9%) for non market makers and 0.841 (84.1%) for market makers (see Internet Appendix Table 2 for the summary statistics on credit growth). Internet Appendix Table 3 presents results using different winsorizations, that do not affect the main conclusions. 28 Throughout the paper I mainly rely on the city-zone database due to its balance among different banks and for brevity. Results are qualitatively similar at the industry level. 29 For all the estimations I exclude Banco Corpbanca after 2014q1 and GNB Sudameris after 2014q2 since both were part of two different mergers. 30 These numbers are obtained by using the estimated differential credit growth during the period multiplied by the commercial credit for market makers in December 2013. After this, I multiply it by the total assets of market makers in December 2013 to obtain the differential growth in commercial credit. 31 See Hener et al. (2015) for a simple exposition of these tests. 32 This is further illustrated in Internet Appendix Figure 3 where I show the average commercial interest rates for both types of banks in time. 33 This can also partially be controlled by bank observable characteristics, which I explore in the next sections. 34 I also consider more than 5 exact matches, but after 5 the percentage of exact matches start to decay very fast. 35 The percentage of exact matches specifies the percentage of observations from the treatment group that were exactly matched to an observation from the control group through their city-zone. The observations that are not exactly matched, are matched to observations in another zone but in the same city. 36 The Internet Appendix contains a more in depth analysis of the different channels that could be affecting the baseline specification. 37 I also perform a robustness test using the commercial credit growth (not normalized by assets) in Internet Appendix Table 3. The results are quantitatively similar to when I do normalize credit by assets. 38 Internet Appendix Table 10 presents summary statistics of the growth of these real variables. 39 See for instance Henry (2000) and Henry (2003) among others. References Abadie, A., Drukker, D. Herr, J. L. and Imbens. G. W. 2004 . Implementing matching estimators for average treatment effects in Stata. Stata Journal 4 : 290 – 311 . Abadie, A., and Imbens. G. W. 2002 . Simple and bias-corrected matching estimators for average treatment effects. Working Paper, NBER . Abbassi, P., Iyer, R. Peydro, J. L. and Tous. F. R. 2016 . Securities trading by banks and credit supply: Micro evidence from the crisis. Journal of Financial Economics 121 : 569 – 94 . Google Scholar Crossref Search ADS Acharya, V. V., Eisert, T. Eufinger, C. and Hirsch. C. 2014 . Real effects of the sovereign debt crisis in Europe: Evidence from syndicated loans. Discussion Paper, CEPR . Adelino, M., Cunha, I. and Ferreira. M. 2014 . The economic effects of public financing: Evidence from municipal bond ratings recalibration. Technical Report . Adelino, M., and Ferreira. M. 2015 . Bank ratings and lending supply: Evidence from sovereign downgrades. Technical Report . Adelino, M., Schoar, A. and Severino. F. 2016 . Loan originations and defaults in the mortgage crisis: The role of the middle class. Review of Financial Studies 29 : 1635 – 70 . Google Scholar Crossref Search ADS Almeida, H., Cunha, I. Ferreira, M. and Restrepo. F. 2015 . The real effects of credit ratings: The sovereign ceiling channel. Technical Report . Altavilla, C., Pagano, M. and Simonelli. S. 2017 . Bank exposures and sovereign stress transmission. Review of Finance 21 : 2103 – 39 . Google Scholar Crossref Search ADS Altonji, J. G., Elder, T. E. and Taber. C. R. 2005 . Selection on observed and unobserved variables: Assessing the effectiveness of Catholic schools. Journal of Political Economy 113 : 151 – 84 . Google Scholar Crossref Search ADS Baskaya, Y. S., and Kalemli-Ozcan. S. 2016 . Sovereign risk and bank lending: Evidence from 1999 Turkish earthquake. Working Paper, NBER . Becker, B., and Ivashina. V. 2018 . Financial repression in the European sovereign debt crisis. Review of Finance 22 : 83 – 115 . Google Scholar Crossref Search ADS Ben-David, I., Franzoni, F. and Moussawi. R. 2014 . Do ETFs increase volatility? Working Paper, NBER . Benigno, G., Converse, N. and Fornaro. L. 2015 . Large capital inflows, sectoral allocation, and economic performance. Journal of International Money and Finance 55 : 60 – 87 . Google Scholar Crossref Search ADS Bhattacharya, U., Loos, B. Meyer, S. Hackethal, A. and Kaesler. S. 2013 . The dark side of ETFs. Technical Report . Blanchard, O., Ostry, J. D. Ghosh, A. R. and Chamon. M. 2015 . Are capital inflows expansionary or contractionary? Theory, policy implications, and some evidence. Working Paper, NBER . Bofondi, M., Carpinelli, L. and Sette. E. 2013 . Credit supply during a sovereign debt crisis. Temi di discussione , Bank of Italy . Bolton, P., and Jeanne. O. 2011 . Sovereign default risk and bank fragility in financially integrated economies. IMF Economic Review 59 : 162 – 94 . Google Scholar Crossref Search ADS Broner, F., Erce, A. Martin, A. and Ventura. J. 2014 . Sovereign debt markets in turbulent times: Creditor discrimination and crowding-out effects. Journal of Monetary Economics 61 : 114 – 42 . Google Scholar Crossref Search ADS Broner, F., Gelos, G. and Reinhart. C. 2006 . When in peril, retrench: Testing the portfolio channel of contagion. Journal of International Economics 69 : 203 – 30 . Google Scholar Crossref Search ADS Caballero, J. A. 2016 . Do surges in international capital inflows influence the likelihood of banking crises? Economic Journal 126 : 281 – 316 . Google Scholar Crossref Search ADS Calderon, C., and Kubota. M. 2012 . Gross inflows gone wild: Gross capital inflows, credit booms and crises. Working Paper , the World Bank . Chang, Y.-C., Hong, H. and Liskovich. I. 2014 . Regression discontinuity and the price effects of stock market indexing. Review of Financial Studies 28 : 212 – 46 . Google Scholar Crossref Search ADS Chernenko, S., and Sunderam. A. 2012 . The real consequences of market segmentation. Review of Financial Studies 25 : 2041 – 69 . Google Scholar Crossref Search ADS Cremers, M., and Petajisto. A. 2009 . How active is your fund manager? A new measure that predicts performance. Review of Financial Studies 22 : 3329 – 65 . Google Scholar Crossref Search ADS Diamond, P. A. 1965 . National debt in a neoclassical growth model. American Economic Review 55 : 1126 – 50 . Du, W., Pueger, C. E. and Schreger. J. 2016 . Sovereign debt portfolios, bond risks, and the credibility of monetary policy. Working Paper, NBER . Du, W., and Schreger. J. 2016 . Local currency sovereign risk. Journal of Finance 71 : 1027 – 70 . Google Scholar Crossref Search ADS Gan, J. 2007 . The real effects of asset market bubbles: Loan and firm-level evidence of a lending channel. Review of Financial Studies 20 : 1941 – 73 . Google Scholar Crossref Search ADS Gennaioli, N., Martin, A. and Rossi. S. 2014 . Banks, government bonds, and default: What do the data say? Technical Report . Hener, T., Rainer, H. and Siedler. T. 2015 . Political socialization in flux? Linking family non-Intactness during childhood to adult civic engagement. Discussion Paper , Institute for the Study of Labor (IZA). Henry, P. B. 2000 . Stock market liberalization, economic reform, and emerging market equity prices. Journal of Finance 55 : 529 – 64 . Google Scholar Crossref Search ADS Henry, P. B. 2003 . Capital-account liberalization, the cost of capital, and economic growth. American Economic Review 93 : 91 – 6 . Google Scholar Crossref Search ADS Henry, P. B. 2007 . Capital account liberalization: Theory, evidence, and speculation. Journal of Economic Literature 45 : 887 – 935 . Google Scholar Crossref Search ADS Israeli, D., Lee, C. and Sridharan S. 2015 . Is there a dark side to exchange traded funds (ETFs)? An information perspective. Technical Report . Ivashina, V., and Scharfstein. D. 2010 . Bank lending during the financial crisis of 2008. Journal of Financial Economics 97 : 319 – 338 . Google Scholar Crossref Search ADS Iyer, R., and Peydro. J. L. 2011 . Interbank contagion at work: Evidence from a natural experiment. Review of Financial Studies 24 : 1337 – 77 . Google Scholar Crossref Search ADS Iyer, R., Peydro, J. L. da Rocha-Lopes, S. and Schoar. A. 2014 . Interbank liquidity crunch and the firm credit crunch: Evidence from the 2007–2009 crisis. Review of Financial Studies 27 : 347 – 72 . Google Scholar Crossref Search ADS Jimenez, G., Ongena, S. Peydro, J. L. and Saurina. J. 2012 . Credit supply and monetary policy: Identifying the bank balance-sheet channel with loan applications. American Economic Review 102 : 2301 – 26 . Google Scholar Crossref Search ADS Jimenez, G., Ongena, S. Peydro, J. L. and Saurina. J. 2014 . Hazardous times for monetary policy: What do twenty-three million bank loans say about the effects of monetary policy on credit risk-taking? Econometrica 82 : 463 – 505 . Google Scholar Crossref Search ADS Jotikasthira, C., Lundblad, C. and Ramadorai. T. 2012 . Asset fire sales and purchases and the international transmission of funding shocks. Journal of Finance 67 : 2015 – 50 . Google Scholar Crossref Search ADS Kalantzis, Y. 2015 . Financial fragility in small open economies: Firm balance sheets and the sectoral structure. Review of Economic Studies 82 : 1194 – 222 . Google Scholar Crossref Search ADS Khwaja, A. I., and Mian. A. 2008 . Tracing the impact of bank liquidity shocks: Evidence from an emerging market. American Economic Review 98 : 1413 – 42 . Google Scholar Crossref Search ADS Lane, P. R., and McQuade. P. 2014 . Domestic credit growth and international capital flows. Scandinavian Journal of Economics 116 : 218 – 52 . Google Scholar Crossref Search ADS Larrain, M. 2015 . Capital account opening and wage inequality. Review of Financial Studies 28 : 1555 – 87 . Google Scholar Crossref Search ADS Levy-Yeyati, E., and Williams. T. 2012 . Emerging economies in the 2000s: Real decoupling and financial recoupling. Journal of International Money and Finance 31 : 2102 – 26 . Google Scholar Crossref Search ADS Mendoza, E., and Terrones. M. 2012 . An anatomy of credit booms and their demise. Working Paper , Central Bank of Chile . Ministerio Hacienda y Credito Publico . 2010 . Programa de creadores de mercado para titulos de deuda publica. Technical Report , Colombia . Ongena, S., Popov, A. and Horen. N. V. 2016 . The invisible hand of the government: ‘Moral suasion’ during the European sovereign debt crisis. Working Paper , Netherlands Central Bank . Oster, E. 2017 . Unobservable Selection and Coefficient Stability: Theory and Validation. Journal of Business & Economic Statistics . Advance access published online June 1, 2017, https://doi.org/10.1080/07350015.2016.1227711 . Paravisini, D. 2008 . Local bank financial constraints and firm access to external finance. Journal of Finance 63 : 2161 – 93 . Google Scholar Crossref Search ADS Paravisini, D., Rappoport, V. and Schnabl P. 2015 . Comparative Advantage and Specialization in Bank Lending. Technical Report . Perez, D. J. 2015 . Sovereign debt, domestic Banks and the provision of public liquidity. Discussion Papers , Stanford Institute for Economic Policy Research. Raddatz, C., and Schmukler. S. 2012 . On the international transmission of shocks: Micro-evidence from mutual fund portfolios. Journal of International Economics 88 : 357 – 74 . Google Scholar Crossref Search ADS Raddatz, C., Schmukler, S. and Williams. T. 2017 . International asset allocations and capital flows: The benchmark effect. Journal of International Economics 108 : 413 – 30 . Google Scholar Crossref Search ADS Reinhart, C., and Reinhart. V. 2009 . Capital flow bonanzas: An encompassing view of the past and present. In NBER international seminar on macroeconomics 2008 , pp. 9 – 62 . Cambridge, MA : National Bureau of Economic Research. Reinhart, C. M., and Sbrancia. M. B. 2011 . The liquidation of government debt. Working Paper , Bank for International Settlements. Reuters. 2014 . Colombia bonds to see 10 billion inflows, Turkey, Russia lose in index changes-banks. Reuters , March 20. https://af.reuters.com/article/commoditiesNews/idAFL6N0MH46O20140320. Schnabl, P. 2012 . The international transmission of bank liquidity shocks: Evidence from an emerging market. Journal of Finance 67 : 897 – 932 . Google Scholar Crossref Search ADS Sullivan, R., and Xiong. J. 2012 . How index tracking increases market vulnerability. Financial Analysts Journal 68 : 70 – 84 . Google Scholar Crossref Search ADS Wurgler, J. 2010 . On the economic consequences of index-linked investing. Working Paper, NBER. © The Author(s) 2018. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

The Review of Financial StudiesOxford University Press

Published: Dec 1, 2018

DeepDyve is your personal research library

It’s your single place to instantly
that matters to you.

over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month Explore the DeepDyve Library Search Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly Organize Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place. Access Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals. Your journals are on DeepDyve Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more. All the latest content is available, no embargo periods. DeepDyve Freelancer DeepDyve Pro Price FREE$49/month
\$360/year

Save searches from
PubMed

Create lists to

Export lists, citations