TY - JOUR
AU - Ramcharan,, Rodney
AB - Abstract This paper finds that declining bank equity or liquidity reduces liquidation values of bank-owned real estate and accelerates the pace of asset sales. Buyers of these assets earn significant returns for providing liquidity to banks, as prices tend to rebound sharply after sales by illiquid banks. Lower liquidation values also depress the prices of nearby real estate transactions. Policy interventions, such as equity injections and central bank asset purchases, increase liquidation values by providing institutions with the balance sheet capacity to slow asset sales. This evidence suggests that balance sheet adjustments at financial institutions can explain real asset price dynamics. 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. This paper investigates the impact of bank equity constraints and illiquidity on the liquidation value of residential real estate collateral. It also studies the spillover effects of depressed liquidation values onto the broader housing market. Prominent theories of intermediation predict that the solvency and liquidity of intermediary balance sheets might directly shape asset price dynamics through the sale of distressed assets.1 These arguments begin with the fact that dislocations in labor and financial markets bring troubled assets onto the balance sheet of financial institutions. In 2011, nearly 25% of the vacant homes for sale in the United States were foreclosed or became real-estate-owned (REO) properties held by banks and securitization trusts. Balance sheet pressures, such as binding equity constraints, can cause a bank to deleverage, primarily through the sale of assets. Difficulties in meeting the liquidity demands of depositors can also induce a bank to sell quickly illiquid assets. Either because of equity constraints or illiquidity, the rapid dispossession of these assets when potential buyers might themselves have limited financing capacity can depress liquidation values relative to fundamentals, possibly spilling over onto the broader housing market (Shleifer and Vishny 1992; the survey in Shleifer and Vishny 2011).2 This idea that balance sheet adjustments at financial institutions can impact asset prices and the broader economy is not only at the center of models that connect intermediaries to economic fluctuations but is the main rationale for new approaches to liquidity and capital regulations.3 Yet although there has been much progress in understanding how household balance sheets affect real estate markets, causal evidence connecting the balance sheet of financial institutions to asset price declines, especially for major asset classes like real estate—the chief source of collateral in the U.S. economy—remains limited.4 To be sure, compelling microeconomic evidence suggests that aggregate credit availability, such as changes in banking competition within a geographic area, might inflate local real asset prices (Favara and Imbs 2015; Rajan and Ramcharan 2015).5 But little is known about the underlying mechanism behind this relationship and whether in particular balance sheet illiquidity can directly affect real asset price movements (Diamond and Kashyap 2015). Three principal factors make it difficult to identify the effects of balance sheets on liquidation values. First, endogenous matching driven by persistent unobserved bank characteristics can both determine the quality of collateral retained on balance sheet and subsequent balance sheet observables, such as solvency. This persistence can, in turn, lead to spurious associations between balance sheet observables and realized liquidation values. Second, current unobserved local economic conditions can simultaneously affect both liquidation values and current balance sheet outcomes. For example, weak economic conditions can both depress liquidation values and cause depositors to run or create losses that deplete bank equity. Finally, in the case of selection bias, loan delinquency is often precipitated by a borrower-specific shock. But a bank’s current balance sheet health could shape its incentives to renegotiate loan terms, and “select” a property into foreclosure, again leading to possibly biased estimates of the impact of balance sheet observables on liquidation values. The research design uses detailed transaction-level information on collateral matched to the selling bank and various identification schemes to help address these endogeneity concerns. Information on the precise location of the collateral as well as the exact date of the auction allows the basic specifications to absorb most plausible controls for local economic conditions, including census tract by quarter fixed effects, as well as collateral characteristics, such as the year built, the size of the property, and, even, balance sheet health at the time the intermediary first originated the loan. In the case of selection bias, I collect information on the population of 1.6 million delinquent properties for the 680 banks in the sample—the population from which foreclosed properties are selected—to model selection into foreclosure directly. The geographic diversification in the data set—300,000 liquidations spanning some 5,000 ZIP codes, across 12 states and 680 banks—also aid in identification. Banks that either anticipate or directly face funding pressures often increase their deposit rates to stem deposit outflows. But many large banks set their deposit rates both at the headquarters and at regional “rate-setting” branches. Branch rates might reflect local economic conditions and the relative supply of deposits. But among the sample of banks operating across all twelve states, economic conditions in the ZIP code or census tract in which the asset is liquidated are unlikely to determine changes in the headquarters’ rate (Figure 1). Figure 1 Open in new tabDownload slide Multistate banks and the local endogeneity concern This figure illustrates the research design used in Table 4 to address the concern that unobserved local shocks simultaneously affect balance sheet outcomes and liquidation values. Large banks, such as Bank of America, set their deposit rates both at the headquarters (Charlotte) and at regional “rate-setting” branches. At such banks, asset liquidations are centralized. The research design estimates the impact of changes in Bank of America’s deposit rate set in Charlotte, North Carolina, on properties liquidated in ZIP codes across California, Arizona, and Florida. The headquarters’ rate likely reflects balance sheet pressures at the bank level and provides the impetus for asset liquidations, whereas the absorptive capacity in the ZIP code helps determine liquidation values. Figure 1 Open in new tabDownload slide Multistate banks and the local endogeneity concern This figure illustrates the research design used in Table 4 to address the concern that unobserved local shocks simultaneously affect balance sheet outcomes and liquidation values. Large banks, such as Bank of America, set their deposit rates both at the headquarters (Charlotte) and at regional “rate-setting” branches. At such banks, asset liquidations are centralized. The research design estimates the impact of changes in Bank of America’s deposit rate set in Charlotte, North Carolina, on properties liquidated in ZIP codes across California, Arizona, and Florida. The headquarters’ rate likely reflects balance sheet pressures at the bank level and provides the impetus for asset liquidations, whereas the absorptive capacity in the ZIP code helps determine liquidation values. Thus, some specifications use this headquarters’ rate to study the impact of funding pressures on liquidation values among multistate banks. At such banks, asset liquidations are centralized. And the headquarters’ rate likely reflect funding pressures at the bank-level which provides the impetus for asset liquidations, while the absorptive capacity in the ZIP code help determine liquidation values. Bank of America created, for example, a legacy asset division to handle Countrywide asset dispositions. The exercise is analogous then to estimating the impact of changes in Bank of America’s deposit rate set in Charlotte, North Carolina, on properties liquidated in ZIP codes across California, Arizona, and Florida. Either using the change in deposits scaled by assets or changes in the deposit rate for term deposits as measures of a liquidity shock, significant evidence suggests that increased funding pressures and lower regulatory capital ratios are associated with lower liquidation values. A 1-standard-deviation decrease (increase) in deposits (interest rates) is associated with about a 1.5% drop in the liquidation value of real estate collateral in the next quarter or about $\$$ 2,200 for the average property. The magnitudes on the tier 1 ratio coefficient is similar in the baseline specification. Illiquidity is also associated with faster asset sales, as banks likely discount properties to raise cash more quickly. A 1-standard-deviation decrease in the deposit change variable is associated with about a 2-week decrease in the time remaining until a property sells. The results on the returns to arbitrage or price rebound also suggest that liquidation values are discounted relative to their fundamental value—the fire-sale hypothesis. Under the fire-sale hypothesis, the size of the liquidation discount or the buyer’s returns to arbitrage is related to the magnitude of the seller’s distress. For the sample of properties that resold during the period under observation, I computed the daily internal rate of return. The returns to providing liquidity to banks appear large. Buying the asset from a bank after it experiences a negative 1-standard-deviation liquidity shock and reselling it 160 days later—the median resale time—augments returns by about 3.1 percentage points. This result is difficult to reconcile with endogeneity narratives, as local economic fundamentals are highly persistent in neighborhoods with foreclosures. The variation in balance sheets induced by policy interventions, such as TARP—government equity injections into banks—and quantitative easing (QE)—central bank asset purchases—help further identify the underlying mechanism behind REO discounts and balance sheet observables. In the quarter before a bank receives TARP, a 1-standard-deviation decrease in the tier 1 ratio is associated with about a 5% drop in liquidation values. But after a government equity injection—which both increases the distance to insolvency and likely helps the recipient bank more easily raise outside equity—a similar loss of equity suggests only a 1.1% drop in liquidation values. The pace of asset sales also slows sharply after TARP, as recipient banks have less incentive to deleverage though discounted asset sales. In addition, QE1, in which the Federal Reserve purchased mortgage-backed securities (MBS) and U.S. Treasuries is also associated with increased liquidation values and slowed asset sales among banks more exposed to the capital gains induced by QE1 asset purchases, as measured by banks’ preexisting holdings of MBS and Treasuries. Finally, using detailed data on about 800,000 nonforeclosure market transactions, I provide evidence that REO liquidation values impact nearby house prices. A key challenge to causal inference stems from the fact that latent local economic shocks can simultaneously determine liquidation values and the price of non-REO sales. The evidence linking balance sheets to liquidation values offers a new way to address this identification challenge, and I instrument the REO liquidation value with the balance sheet of the selling bank. Simple back-of-the-envelope calculations suggest that the average sale price for arm’s length transactions in the sample would have been about 69 basis points (bps), or about $\$$ 1,994 higher, had banks been operating with 2017 equity ratios and deposit flows outcomes. This paper provides the first detailed evidence that asset sales by financial institutions can significantly influence the liquidation values of real estate collateral and affect the broader residential housing market. These results are thus supportive of theories that emphasize the importance of financial intermediary balance sheets in shaping asset prices and economic fluctuations (see the survey in Gertler and Kiyotaki 2010). The evidence in this paper also help us understand better the effects of some of the extraordinary policy interventions during the 2008–2009 crisis, like TARP and QE, and the results bear on the ongoing debate over bank capital and liquidity regulation. 1. Hypothesis Development and Data 1.1 Hypothesis development A financial intermediary’s balance sheet might affect the liquidation value of distressed or troubled assets through at least two major channels. First, models of financial intermediation emanating from Diamond and Dybvig (1983) observe that difficulties in rolling over short-term liabilities or an increase in the demand for liquidity among depositors can force a financial institution to sell quickly illiquid assets to meet the liquidity demands of creditors. When cash-in-the-market is limited, this rapid selling of illiquid assets can in turn depress liquidation values and the prices of similar assets (Allen and Gale 1994). Illiquidity on both the asset and liability sides of the balance sheet can also interact to shape the liquidation values of collateral: when confronted with an increase in the demand for liquidity from depositors, banks with more liquid assets might face less pressure to sell quickly illiquid assets and depress liquidation values.6 A second key channel centers on the high cost and slow pace of raising equity during times of crisis in conjunction with the incentives provided by risk-based capital regulation. Unable to raise outside equity easily when the financial sector is in distress, banks have powerful incentives to deleverage through asset sales, primarily of assets with greater risk weights. Figure 2 shows the extent of deleveraging in the banking system during the 2006–2015 sample period. The ratio of tier 1 capital to risk-weighted assets—a key indicator of regulatory solvency—in the total banking system rises sharply beginning in 2008. But much of this increase stems from the shedding of capital-intensive assets, as the ratio of risk weighted to total assets declines equally sharply over this period. Note well that this deleveraging continues well after the 2008–2009 crisis, extending through the entire sample period. Figure 2 Open in new tabDownload slide Risk-weighted assets and tier 1 capital in the banking system, 2001–2015 The solid line is the sum of risk-weighted assets in the U.S. banking system divided by the sum of total assets. The dotted line is the total of tier 1 capital divided by the sum of total risk-weighted assets. The data are from the Call Report. Figure 2 Open in new tabDownload slide Risk-weighted assets and tier 1 capital in the banking system, 2001–2015 The solid line is the sum of risk-weighted assets in the U.S. banking system divided by the sum of total assets. The dotted line is the total of tier 1 capital divided by the sum of total risk-weighted assets. The data are from the Call Report. REO properties are one such capital-intensive asset class. These properties migrated onto the balance sheets of banks en masse over this period, averaging about $\$$ 20 billion in the banking system from the first quarter of 2008 through the end of 2015 (Figure 3). Given the sizeable risk weight of these assets, an intermediary facing binding equity constraints would likely prefer to liquidate rapidly on-balance-sheet REO assets.7 The risk weight on foreclosed real estate assets is 100% or twice as large as real estate loans in good standing.8 Using the typical 8% minimum equity constraint before regulators mandate “prompt and corrective action,” the capital requirement for bank owned real estate with a fair value of $\$$100,000 would be $\$$8,000; a loan of similar value would only have a capital charge of $\$$4,000. In some circumstances, realized losses from the sale of REO assets can offset the deleveraging benefits of these asset sales. But accounting practices generally limit this possible offsetting effect.9 Figure 3 Open in new tabDownload slide REO assets in the banking system This figure plots the total REO assets held on the balance sheet of commercial banks from 2001 Q2 through 2015 Q4. REO, real estate owned. Figure 3 Open in new tabDownload slide REO assets in the banking system This figure plots the total REO assets held on the balance sheet of commercial banks from 2001 Q2 through 2015 Q4. REO, real estate owned. Both economic theory and the institutional details surrounding REO assets suggest that intermediary balance sheet pressures can affect liquidation values. But establishing a causal relationship between balance sheet illiquidity and solvency and liquidation values is difficult. The endogenous matching between collateral quality and bank balance sheets can make it difficult to differentiate the effects of balance sheet pressures from intrinsic collateral quality on observed liquidation values. Contemporaneous unobserved economic conditions can also simultaneously affect the demand for liquidity or a bank’s equity constraint as well as liquidation values. And selection bias can also theoretically hamper causal inference. Foreclosed properties are drawn from the population of delinquent loans. And once a loan becomes delinquent, potentially unobserved balance sheet factors that affect liquidation values also could be correlated with lender-borrower negotiations and selection into the foreclosure subsample. Apart from these conceptual challenges to identification, data on liquidation values, especially for real assets, are generally unavailable. Regulatory financial statements—the call report—records coarse quarterly information on charge-offs and recoveries, containing no data on the prices obtained from the sale of underlying assets and the characteristics of the collateral sold. The coarseness of public regulatory information makes it impossible to address these challenges to identification when using typical data sets. 1.2 Data: REO assets To address these identification challenges, the analysis uses data from Zillow’s ZTRAX database on the liquidation of foreclosed properties collected in twelve states, including Arizona, California, and Florida—the three states with the highest number of foreclosures in the United States. The ZTRAX database contains information on the near universe of housing transactions drawn from county records across the country. Importantly for the analysis, the database lists the price and date of liquidation; the property address and key collateral characteristics, including price and leverage at origination. ZTRAX also lists in text form the name of the bank that liquidated the foreclosed property, allowing a manual match with financial institutions’ regulatory balance sheet and income data. The sample period runs from the first quarter of 2006 through the final quarter of 2015 (the appendix provides more detail). The number of bank-owned properties in the twelve states, about 500,000, reflects the raw count and about one-third of these bank-owned properties do not have recorded prices, or lack information to match the commercial bank to the liquidation; observations with missing or nonmatched data are excluded from the subsequent analysis. The remaining 301,757 properties that can be matched to a bank, have a recorded liquidation value, and was not a unique sale within a ZIP code-year-quarter observation constitute the baseline sample. A possible concern here is that this matching might systematically “miss” the smaller banks or the requirement of a nonunique sale within a ZIP code-year-quarter observation might also exclude sales by smaller banks because they have fewer properties to sell. However, the large sample size still should be representative of the U.S. banking system, and the main results are also robust to relaxing the “nonunique sale within a ZIP code-year-quarter” criteria. Table 1 tabulates by state the number of bank-owned foreclosures for this sample. California, Arizona and Florida have the highest number of bank-liquidated properties. Table 2 summarizes key collateral characteristics for the full sample of REO assets. Of note is that across the entire distribution, foreclosed properties sold at huge discounts relative to their nominal origination price. Table 1 Bank-owned properties in the sample, by state State . Freq. . Percent . Arizona 53,033 17.57 California 78,654 26.07 Colorado 12,199 4.04 Florida 70,500 23.36 Illinois 12,309 4.08 Michigan 6,836 2.27 New Jersey 995 0.33 Nevada 9,856 3.27 Ohio 34,035 11.28 Pennsylvania 5,047 1.67 Texas 8,685 2.88 Washington 9,608 3.18 Total 301,757 100 State . Freq. . Percent . Arizona 53,033 17.57 California 78,654 26.07 Colorado 12,199 4.04 Florida 70,500 23.36 Illinois 12,309 4.08 Michigan 6,836 2.27 New Jersey 995 0.33 Nevada 9,856 3.27 Ohio 34,035 11.28 Pennsylvania 5,047 1.67 Texas 8,685 2.88 Washington 9,608 3.18 Total 301,757 100 This table lists the number of liquidated bank-owned properties by state in the sample. Open in new tab Table 1 Bank-owned properties in the sample, by state State . Freq. . Percent . Arizona 53,033 17.57 California 78,654 26.07 Colorado 12,199 4.04 Florida 70,500 23.36 Illinois 12,309 4.08 Michigan 6,836 2.27 New Jersey 995 0.33 Nevada 9,856 3.27 Ohio 34,035 11.28 Pennsylvania 5,047 1.67 Texas 8,685 2.88 Washington 9,608 3.18 Total 301,757 100 State . Freq. . Percent . Arizona 53,033 17.57 California 78,654 26.07 Colorado 12,199 4.04 Florida 70,500 23.36 Illinois 12,309 4.08 Michigan 6,836 2.27 New Jersey 995 0.33 Nevada 9,856 3.27 Ohio 34,035 11.28 Pennsylvania 5,047 1.67 Texas 8,685 2.88 Washington 9,608 3.18 Total 301,757 100 This table lists the number of liquidated bank-owned properties by state in the sample. Open in new tab Table 2 Collateral Characteristics A. Collateral characteristics, REO assets, and the full sample . . Price per square feet at origination (⁠ $\$$ ⁠) . Price per square feet at foreclosure (⁠ $\$$ ⁠) . Foreclosure price (⁠ $\$$ ⁠) . Lot size (square feet) . Total bedrooms . Total bath . Year built . mean 34 17 151,542 8,316 2 2 1977 25th percentile 13 7 64,000 5,929 0 1 1958 50th percentile 24 14 116,000 7,405 3 2 1981 75th percentile 42 26 193,800 10,018 3 2 2001 90th percentile 68 43 300,000 17,860 4 3 2005 standard deviation 358 372 386,049 26,795 2 1 26 B. Below-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 35 12 144,812 8,151 2 2 1975 25th percentile 13 6 58,500 6,000 1 1 1957 50th percentile 24 13 107,000 7,475 3 2 1979 75th percentile 42 24 180,000 10,019 3 2 1999 90th percentile 70 41 294,900 17,414 4 3 2005 standard deviation 334 400 599,383 17,075 2 1 27 C. Above-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 36 12 153,367 8,178 2 2 1977 25th percentile 13 7 65,600 5,896 0 1 1959 50th percentile 24 15 119,000 7,405 3 2 1983 75th percentile 42 27 197,089 10,018 3 2 2001 90th percentile 68 44 301,745 18,000 4 3 2005 standard deviation 371 365 303,298 20,338 2 1 26 D. Below-median deposits to assets ratio, 2001–2006 mean 32 17 151,122 8,778 2 2 1976 25th percentile 14 7 63,900 6,000 0 1 1958 50th percentile 25 14 115,000 7,405 3 2 1980 75th percentile 44 26 190,191 10,019 3 2 2000 90th percentile 73 44 305,000 17,685 4 3 2005 standard deviation 352 346 408,011 20,705 2 1 26 E. Above-median deposits to assets ratio, 2001–2006 mean 28 18 152,029 8,609 2 2 1977 25th percentile 13 7 64,800 5,800 0 1 1959 50th percentile 23 15 118,746 7,405 3 2 1983 75th percentile 40 27 196,378 10,018 3 2 2001 90th percentile 62 43 296,063 18,000 4 3 2006 standard deviation 307 311 358,887 20,322 2 1 27 A. Collateral characteristics, REO assets, and the full sample . . Price per square feet at origination (⁠ $\$$ ⁠) . Price per square feet at foreclosure (⁠ $\$$ ⁠) . Foreclosure price (⁠ $\$$ ⁠) . Lot size (square feet) . Total bedrooms . Total bath . Year built . mean 34 17 151,542 8,316 2 2 1977 25th percentile 13 7 64,000 5,929 0 1 1958 50th percentile 24 14 116,000 7,405 3 2 1981 75th percentile 42 26 193,800 10,018 3 2 2001 90th percentile 68 43 300,000 17,860 4 3 2005 standard deviation 358 372 386,049 26,795 2 1 26 B. Below-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 35 12 144,812 8,151 2 2 1975 25th percentile 13 6 58,500 6,000 1 1 1957 50th percentile 24 13 107,000 7,475 3 2 1979 75th percentile 42 24 180,000 10,019 3 2 1999 90th percentile 70 41 294,900 17,414 4 3 2005 standard deviation 334 400 599,383 17,075 2 1 27 C. Above-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 36 12 153,367 8,178 2 2 1977 25th percentile 13 7 65,600 5,896 0 1 1959 50th percentile 24 15 119,000 7,405 3 2 1983 75th percentile 42 27 197,089 10,018 3 2 2001 90th percentile 68 44 301,745 18,000 4 3 2005 standard deviation 371 365 303,298 20,338 2 1 26 D. Below-median deposits to assets ratio, 2001–2006 mean 32 17 151,122 8,778 2 2 1976 25th percentile 14 7 63,900 6,000 0 1 1958 50th percentile 25 14 115,000 7,405 3 2 1980 75th percentile 44 26 190,191 10,019 3 2 2000 90th percentile 73 44 305,000 17,685 4 3 2005 standard deviation 352 346 408,011 20,705 2 1 26 E. Above-median deposits to assets ratio, 2001–2006 mean 28 18 152,029 8,609 2 2 1977 25th percentile 13 7 64,800 5,800 0 1 1959 50th percentile 23 15 118,746 7,405 3 2 1983 75th percentile 40 27 196,378 10,018 3 2 2001 90th percentile 62 43 296,063 18,000 4 3 2006 standard deviation 307 311 358,887 20,322 2 1 27 These tables report summary statistics for the collateral for the full sample of banks and by various subsamples based on balance sheet averages from 2001 to 2006. REO, real estate owned. Open in new tab Table 2 Collateral Characteristics A. Collateral characteristics, REO assets, and the full sample . . Price per square feet at origination (⁠ $\$$ ⁠) . Price per square feet at foreclosure (⁠ $\$$ ⁠) . Foreclosure price (⁠ $\$$ ⁠) . Lot size (square feet) . Total bedrooms . Total bath . Year built . mean 34 17 151,542 8,316 2 2 1977 25th percentile 13 7 64,000 5,929 0 1 1958 50th percentile 24 14 116,000 7,405 3 2 1981 75th percentile 42 26 193,800 10,018 3 2 2001 90th percentile 68 43 300,000 17,860 4 3 2005 standard deviation 358 372 386,049 26,795 2 1 26 B. Below-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 35 12 144,812 8,151 2 2 1975 25th percentile 13 6 58,500 6,000 1 1 1957 50th percentile 24 13 107,000 7,475 3 2 1979 75th percentile 42 24 180,000 10,019 3 2 1999 90th percentile 70 41 294,900 17,414 4 3 2005 standard deviation 334 400 599,383 17,075 2 1 27 C. Above-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 36 12 153,367 8,178 2 2 1977 25th percentile 13 7 65,600 5,896 0 1 1959 50th percentile 24 15 119,000 7,405 3 2 1983 75th percentile 42 27 197,089 10,018 3 2 2001 90th percentile 68 44 301,745 18,000 4 3 2005 standard deviation 371 365 303,298 20,338 2 1 26 D. Below-median deposits to assets ratio, 2001–2006 mean 32 17 151,122 8,778 2 2 1976 25th percentile 14 7 63,900 6,000 0 1 1958 50th percentile 25 14 115,000 7,405 3 2 1980 75th percentile 44 26 190,191 10,019 3 2 2000 90th percentile 73 44 305,000 17,685 4 3 2005 standard deviation 352 346 408,011 20,705 2 1 26 E. Above-median deposits to assets ratio, 2001–2006 mean 28 18 152,029 8,609 2 2 1977 25th percentile 13 7 64,800 5,800 0 1 1959 50th percentile 23 15 118,746 7,405 3 2 1983 75th percentile 40 27 196,378 10,018 3 2 2001 90th percentile 62 43 296,063 18,000 4 3 2006 standard deviation 307 311 358,887 20,322 2 1 27 A. Collateral characteristics, REO assets, and the full sample . . Price per square feet at origination (⁠ $\$$ ⁠) . Price per square feet at foreclosure (⁠ $\$$ ⁠) . Foreclosure price (⁠ $\$$ ⁠) . Lot size (square feet) . Total bedrooms . Total bath . Year built . mean 34 17 151,542 8,316 2 2 1977 25th percentile 13 7 64,000 5,929 0 1 1958 50th percentile 24 14 116,000 7,405 3 2 1981 75th percentile 42 26 193,800 10,018 3 2 2001 90th percentile 68 43 300,000 17,860 4 3 2005 standard deviation 358 372 386,049 26,795 2 1 26 B. Below-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 35 12 144,812 8,151 2 2 1975 25th percentile 13 6 58,500 6,000 1 1 1957 50th percentile 24 13 107,000 7,475 3 2 1979 75th percentile 42 24 180,000 10,019 3 2 1999 90th percentile 70 41 294,900 17,414 4 3 2005 standard deviation 334 400 599,383 17,075 2 1 27 C. Above-median tier 1 capital to risk-weighted assets ratio, 2001–2006 mean 36 12 153,367 8,178 2 2 1977 25th percentile 13 7 65,600 5,896 0 1 1959 50th percentile 24 15 119,000 7,405 3 2 1983 75th percentile 42 27 197,089 10,018 3 2 2001 90th percentile 68 44 301,745 18,000 4 3 2005 standard deviation 371 365 303,298 20,338 2 1 26 D. Below-median deposits to assets ratio, 2001–2006 mean 32 17 151,122 8,778 2 2 1976 25th percentile 14 7 63,900 6,000 0 1 1958 50th percentile 25 14 115,000 7,405 3 2 1980 75th percentile 44 26 190,191 10,019 3 2 2000 90th percentile 73 44 305,000 17,685 4 3 2005 standard deviation 352 346 408,011 20,705 2 1 26 E. Above-median deposits to assets ratio, 2001–2006 mean 28 18 152,029 8,609 2 2 1977 25th percentile 13 7 64,800 5,800 0 1 1959 50th percentile 23 15 118,746 7,405 3 2 1983 75th percentile 40 27 196,378 10,018 3 2 2001 90th percentile 62 43 296,063 18,000 4 3 2006 standard deviation 307 311 358,887 20,322 2 1 27 These tables report summary statistics for the collateral for the full sample of banks and by various subsamples based on balance sheet averages from 2001 to 2006. REO, real estate owned. Open in new tab Panels B–F of Table 2 show that while banks might vary in the quality of collateral originated, this variation appears substantially smaller among the set of assets that they actually retained on balance sheet. This implies that the potential for endogenous matching in the analysis might be limited. These tables report collateral characteristics by key bank observables, such as tier 1 capital to risk-weighted assets—a standard measure of regulatory solvency—as well as by deposits to assets—a measure of a bank’s dependence on deposit financing. To avoid endogeneity, we average these bank variables between 2001 and 2006. Table 2C provides some evidence that banks with above-median equity ratios tended to originate and retain newer homes. But across the panels of the table, the differences in REO asset quality across bank types is tiny. 1.3 Data: Bank data To measure balance sheet liquidity, in much of the analysis, I use a bank’s change in deposits relative to the same quarter in the previous year and scaled by assets. This approach builds on the evidence that the traditional banking system faced significant liquidity pressures during much of the sample period, as aggregate deposit inflows weakened and funding shortfalls increased (Acharya and Mora 2015).10 New liquidity regulations, such as the NSF and LCR, also significantly raised the demand for stable funding sources, like deposits, and the cost of illiquid assets like REOs. Table 3 summarizes some of the other balance sheet variables both in 2006 and again at the end of the sample period in 2015. Consistent with the significant changes in financial regulation over the sample period, median tier 1 capital to risk-weighted asset ratios are about 2 percentage points higher in 2015 compared to 2006. Similarly, over this period, balance sheet illiquidity, as measured by both the ratios of loans to deposits, and cash to assets appear to have decreased. Surviving banks also appear to be much larger. Table 3 Summary statistics: Bank balance sheet variables, 2006 and 2015 . Tier 1 capital / risk-weighted assets . Loans/deposits . Deposits/assets . Cash/assets . Return on assets . Assets (log) . 2006 Mean 0.166 0.63 0.8 0.046 0.007 11.789 Median 0.134 0.665 0.836 0.033 0.006 11.663 SD 0.103 0.182 0.145 0.062 0.023 1.383 2015 Mean 0.179 0.604 0.823 0.098 0.007 12.271 Median 0.153 0.635 0.85 0.066 0.005 12.112 SD 0.092 0.18 0.121 0.104 0.026 1.403 . Tier 1 capital / risk-weighted assets . Loans/deposits . Deposits/assets . Cash/assets . Return on assets . Assets (log) . 2006 Mean 0.166 0.63 0.8 0.046 0.007 11.789 Median 0.134 0.665 0.836 0.033 0.006 11.663 SD 0.103 0.182 0.145 0.062 0.023 1.383 2015 Mean 0.179 0.604 0.823 0.098 0.007 12.271 Median 0.153 0.635 0.85 0.066 0.005 12.112 SD 0.092 0.18 0.121 0.104 0.026 1.403 This table presents summary statistics for selected bank balance sheet variables in 2006 (beginning of sample) and in 2015 (end of sample) Open in new tab Table 3 Summary statistics: Bank balance sheet variables, 2006 and 2015 . Tier 1 capital / risk-weighted assets . Loans/deposits . Deposits/assets . Cash/assets . Return on assets . Assets (log) . 2006 Mean 0.166 0.63 0.8 0.046 0.007 11.789 Median 0.134 0.665 0.836 0.033 0.006 11.663 SD 0.103 0.182 0.145 0.062 0.023 1.383 2015 Mean 0.179 0.604 0.823 0.098 0.007 12.271 Median 0.153 0.635 0.85 0.066 0.005 12.112 SD 0.092 0.18 0.121 0.104 0.026 1.403 . Tier 1 capital / risk-weighted assets . Loans/deposits . Deposits/assets . Cash/assets . Return on assets . Assets (log) . 2006 Mean 0.166 0.63 0.8 0.046 0.007 11.789 Median 0.134 0.665 0.836 0.033 0.006 11.663 SD 0.103 0.182 0.145 0.062 0.023 1.383 2015 Mean 0.179 0.604 0.823 0.098 0.007 12.271 Median 0.153 0.635 0.85 0.066 0.005 12.112 SD 0.092 0.18 0.121 0.104 0.026 1.403 This table presents summary statistics for selected bank balance sheet variables in 2006 (beginning of sample) and in 2015 (end of sample) Open in new tab In some specifications, I also use the change in the interest rate on the 6-month CD, obtained from RateWatch, to proxy for liquidity pressures. This choice is motivated by the fact that a quantity-based measure of liquidity pressures is only partially informative of underlying balance sheet funding pressures. Banks that either anticipate or directly face funding pressures might increase their deposit rates to stem deposit outflows or even induce additional flows. Section A of the Internet Appendix further discusses the relationship between the price and quantity-based proxies for liquidity pressures. 2. Balance Sheets and Liquidation Values 2.1 Main results This subsection studies the impact of bank balance sheets on the liquidation values of REO bank assets. Let |$P_{\mathit{ijkt}}$| denote the liquidation value of property |$i$| located in neighborhood |$k$|—ZIP code or census tract—that is liquidated by bank |$j$| on date |$t$|⁠. To establish simply the relationship between balance sheet liquidity and liquidation values, the baseline specification uses the change in deposits relative to the same quarter in the previous year and scaled by assets as the main measure of balance sheet liquidity pressures. Illiquidity and insolvency are closely related, and the baseline specification also uses the ratio of tier 1 equity to risk-weighted assets, a key regulatory measure of solvency; all bank variables are observed in the quarter before liquidation. The estimating equation is thus $$\begin{equation}P_{\mathit{ijkt}}=\delta _j{+}\delta _{\mathit{kt}}{+}\beta _1\frac{(\mathit{deposit}_{\mathit{jt}-1}{-}\mathit{deposit}_{\mathit{jt}{-}5})}{\mathit{Asset}_{\mathit{jt}{-}1}}{+}\beta _2\frac{\mathit{tier}1\mathit{equity}_{\mathit{jt}-1}}{\mathit{Asset}_{\mathit{jt}-1}}{+}X_{\mathit{jt}{-}1}\Theta _1{+}C_i\Theta _2.\end{equation}$$ (1) The parameter |$\delta _j$| absorbs bank fixed effects, while |$\delta _{\mathit{kt}}$| absorb neighborhood by year-quarter fixed effects. The latter parameter absorbs nonparametrically time-varying shocks at the neighborhood level, such as changes in income or unemployment, that simultaneously affect liquidation values in the neighborhood and balance sheet outcomes at the selling bank. The baseline specification uses ZIP code by year-quarter fixed effects to absorb these time-varying neighborhood shocks. Table IA1 in the Internet Appendix shows that the main results are robust to defining neighborhood at the much finer census tract level, as well as when including parametric controls like ZIP code-level changes in house prices. The vector |$X_{\mathit{jt}-1}$| contains other bank observables and |$C_i$| is a vector of collateral observables that proxy for the quality of the property. The date of liquidation is the date of auction for the 62% of cases in which liquidation occurs in an auction. Note that auctions are more common in “power-of-sale” states, where creditors can generally seize the collateral without lengthy recourse to the courts. In “power-of-sale” states, some 68% of properties are sold by auction, whereas only 53% are sold by auction in “judicial” states. In the remaining cases the property is sold to an arm’s length buyer, and the date is the recording date in the county deed’s office.11 The dependent variable in Table 4 is the log price of the liquidated property. From Column 1, the deposit change point estimate is statistically significant at the 1% level and positive: a 1-standard-deviation decrease in deposits in the previous quarter is associated with about a 0.8% decline in the liquidation value of the property. Column 2 adds the ratio of tier 1 capital to risk-weighted assets—book equity. A 1-standard-deviation decrease in the tier 1 ratio is associated with a 1.1% drop in the liquidation value of the collateral; although illiquidity and solvency are closely related, the point estimate on balance sheet liquidity remains unchanged. Table 4 Impact of banks’ balance sheets on liquidation values and price rebound . (1) . (2) . (3) . (4) . (5) . (6) . (7) . . . . baseline . Deposit Rate Change . Price Rebound: Unlevered IRR . VARIABLES . change in deposits . solvency . balance sheet controls . All Banks . Headquarters . Deposits Flows . Deposit Rate Change . Year on year change in deposits, scaled by assets 0.0590*** 0.0594*** 0.114*** –0.182*** (0.0228) (0.0229) (0.0322) (0.0468) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.202* 0.205** 0.667 0.952 0.469 0.564 (0.118) (0.104) (0.496) (0.590) (0.313) (0.979) Change in 6 month CD rate, lagged one quarter –0.0642*** –0.0639** 0.118** (0.0246) (0.0276) (0.0542) zipcode*year-quarter*fixed effects yes yes yes yes yes yes yes bank fixed effects yes yes yes yes yes yes yes Observations 301,757 301,757 301,757 229,578 216,407 105,493 81,509 R-squared 0.693 0.693 0.693 0.695 0.694 0.280 0.293 . (1) . (2) . (3) . (4) . (5) . (6) . (7) . . . . baseline . Deposit Rate Change . Price Rebound: Unlevered IRR . VARIABLES . change in deposits . solvency . balance sheet controls . All Banks . Headquarters . Deposits Flows . Deposit Rate Change . Year on year change in deposits, scaled by assets 0.0590*** 0.0594*** 0.114*** –0.182*** (0.0228) (0.0229) (0.0322) (0.0468) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.202* 0.205** 0.667 0.952 0.469 0.564 (0.118) (0.104) (0.496) (0.590) (0.313) (0.979) Change in 6 month CD rate, lagged one quarter –0.0642*** –0.0639** 0.118** (0.0246) (0.0276) (0.0542) zipcode*year-quarter*fixed effects yes yes yes yes yes yes yes bank fixed effects yes yes yes yes yes yes yes Observations 301,757 301,757 301,757 229,578 216,407 105,493 81,509 R-squared 0.693 0.693 0.693 0.695 0.694 0.280 0.293 Columns 1–5 investigate the impact of bank balance sheet variables, observed the quarter before liquidation, on the liquidation value of bank-owned real estate. In Columns 1–5, the dependent variable is the liquidation value, the log price of the property obtained in liquidation. All specifications include ZIP-code-by-year-quarter fixed effects and bank fixed effects; standard errors, in parentheses, are clustered by ZIP code and bank. Columns 3–7 include the loan to deposits, deposits to total assets, cash to total assets, the log of total assets, and the return to assets, all lagged one quarter as additional controls. The dependent variable in Columns 6 and 7 is the unlevered internal rate of return based on the liquidation price and the next resale price. This is defined as |$(\frac{\mathit{resale}\ \mathit{price}_{t+n}}{\mathit{liquidation}\ \mathit{value}_t})^{\frac 1 n}-1$|⁠, where |$n$| is the number of days elapsed from liquidation by the bank to subsequent resale by the buyer. Open in new tab Table 4 Impact of banks’ balance sheets on liquidation values and price rebound . (1) . (2) . (3) . (4) . (5) . (6) . (7) . . . . baseline . Deposit Rate Change . Price Rebound: Unlevered IRR . VARIABLES . change in deposits . solvency . balance sheet controls . All Banks . Headquarters . Deposits Flows . Deposit Rate Change . Year on year change in deposits, scaled by assets 0.0590*** 0.0594*** 0.114*** –0.182*** (0.0228) (0.0229) (0.0322) (0.0468) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.202* 0.205** 0.667 0.952 0.469 0.564 (0.118) (0.104) (0.496) (0.590) (0.313) (0.979) Change in 6 month CD rate, lagged one quarter –0.0642*** –0.0639** 0.118** (0.0246) (0.0276) (0.0542) zipcode*year-quarter*fixed effects yes yes yes yes yes yes yes bank fixed effects yes yes yes yes yes yes yes Observations 301,757 301,757 301,757 229,578 216,407 105,493 81,509 R-squared 0.693 0.693 0.693 0.695 0.694 0.280 0.293 . (1) . (2) . (3) . (4) . (5) . (6) . (7) . . . . baseline . Deposit Rate Change . Price Rebound: Unlevered IRR . VARIABLES . change in deposits . solvency . balance sheet controls . All Banks . Headquarters . Deposits Flows . Deposit Rate Change . Year on year change in deposits, scaled by assets 0.0590*** 0.0594*** 0.114*** –0.182*** (0.0228) (0.0229) (0.0322) (0.0468) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.202* 0.205** 0.667 0.952 0.469 0.564 (0.118) (0.104) (0.496) (0.590) (0.313) (0.979) Change in 6 month CD rate, lagged one quarter –0.0642*** –0.0639** 0.118** (0.0246) (0.0276) (0.0542) zipcode*year-quarter*fixed effects yes yes yes yes yes yes yes bank fixed effects yes yes yes yes yes yes yes Observations 301,757 301,757 301,757 229,578 216,407 105,493 81,509 R-squared 0.693 0.693 0.693 0.695 0.694 0.280 0.293 Columns 1–5 investigate the impact of bank balance sheet variables, observed the quarter before liquidation, on the liquidation value of bank-owned real estate. In Columns 1–5, the dependent variable is the liquidation value, the log price of the property obtained in liquidation. All specifications include ZIP-code-by-year-quarter fixed effects and bank fixed effects; standard errors, in parentheses, are clustered by ZIP code and bank. Columns 3–7 include the loan to deposits, deposits to total assets, cash to total assets, the log of total assets, and the return to assets, all lagged one quarter as additional controls. The dependent variable in Columns 6 and 7 is the unlevered internal rate of return based on the liquidation price and the next resale price. This is defined as |$(\frac{\mathit{resale}\ \mathit{price}_{t+n}}{\mathit{liquidation}\ \mathit{value}_t})^{\frac 1 n}-1$|⁠, where |$n$| is the number of days elapsed from liquidation by the bank to subsequent resale by the buyer. Open in new tab Because the effects of book equity and deposit flows could proxy for other balance sheet observables, Column 3 adds other standard income and balance sheet controls. These variables all enter with a 1-quarter lag. The point estimates on the liquidity and solvency variables are slightly changed in this specification. A 1-standard-deviation decrease in deposit flows is associated with a 1.5% drop in liquidation values in the subsequent quarter. A similar decrease in the tier 1 ratio implies a 1.2% decline the value of the distressed collateral. To put these magnitudes in context, the average property in the sample sold for around $\$$ 151,542; evaluated at this average, a 1-standard-deviation increase in the deposit change coefficient suggests an additional $\$$2,219 increase in the liquidation value. Similarly, a 1-standard-deviation increase in the tier 1 ratio suggests a $\$$1,775 increase in the liquidation value for the average property. A simple “back-of-the-envelope” calculation can further quantify the aggregate economic effects implied by the baseline coefficients. The average Tier 1 capital ratio among banks in 2017 was around 13.5%, whereas the typical large bank in 2017 averaged about a 5.3% increase in deposits. To speculate on what liquidation values might have been if all banks had operated with these 2017 outcomes, I use the coefficients from Column 3 and compute the implied difference in liquidation values between the 2017 outcomes and the sample period. This entails first computing the predicted liquidation values based on the realized balance sheet data over the sample period. The second step predicts liquidation values using the point estimates in Column 3 under the assumption that all banks have the same mean 2017 capitalization ratio and deposit flows but otherwise identical balance outcomes to that observed during the sample period. This difference between the predicted values from the latter step relative to those obtained in the first step suggests that the median liquidation value would have been about 2.3% higher than otherwise if banks had 2017 deposit and tier 1 outcomes. This yields an additional $\$$ 1.5 billion in total revenues or about $\$$3,485 in additional revenue per property. Obviously, the “all else equal” assumption is fraught, but these numbers are a useful heuristic. Liquidity pressures still can be present even absent deposit outflows. Banks that either anticipate or directly face funding pressures can increase deposit rates to stem deposit outflows. Column 4 of Table 4 replaces the change in deposits with the quarter on quarter change in the bank’s 6-month certificate of deposit rate. RateWatch reports the headquarters’ interest rate, as well as the branch rate for those banks that allow local branches to set rates. I use the rate offered by the branch nearest to the property for those banks with interest-rate-setting branches. Otherwise, I use the headquarters’ rate. These data are available for a smaller number of banks, shrinking the sample size. Consistent with illiquidity pressures leading to lower liquidation values, the coefficient implies that a 1-standard-deviation increase in the deposit rate is associated with a 1.5% decline in liquidation values (p-value = .01). This magnitude is identical to that obtained using the quantity derived deposit flows variable in the baseline specification (Column 3). Because depositors often withdraw funds from undercapitalized banks, illiquidity and insolvency measures can be collinear. And in Column 4, the tier 1 variable is more noisily estimated when using the deposit rate measure of illiquidity, but the implied effect on liquidation values is unchanged relative to the baseline. In results available on request, I also use the rate on the longer duration 48-month CD as a robustness exercise; this point estimate is also significant at the 1% level, but the implied effect is about 50% smaller. The deposit rate specification can also help address the endogenous run concern: Poor local fundamentals simultaneously affect liquidation values and bank funding pressures. This approach builds on the fact that some banks set their deposit rates at the headquarters, while other banks allow for “rate-setting” branches. In the latter case, local branches set deposit rates based on local economic conditions and the relative supply of local deposits. Thus, deposit rates set at local branches are more susceptible to the endogenous run concern. But when restricting the sample to large banks, the headquarters’ rate is likely exogenous. For nearly all large banks, asset liquidations are centralized. For this sample of banks then, the headquarters’ rate is clearly not driven by local economic conditions. Concretely, Bank of America created a legacy asset division to handle Countrywide asset dispositions. Likewise, JP Morgan Chase created a centralized division to dispose of REO assets; separate management companies like REOExperts often handled the marketing and management of these properties. Therefore, using the headquarters’ rate is analogous to estimating the impact of Bank of America’s deposit rate set in Charlotte, North Carolina, on properties liquidated in ZIP codes across California, Florida, and Arizona and the other disparate states in the sample. In the context of ZIP code-by-year-quarter fixed effects, unobserved poor local economic conditions in California or Arizona are unlikely to simultaneously cause depositors to withdraw from the bank, pushing up deposit rates at the headquarters and depressing liquidation values. Using the sample of multistate banks—banks operating across all thirteen states—Column 5 shows that the negative impact of headquarters’ deposit rates on liquidation values is unchanged, suggesting that the latent fundamentals concern may not be material. Table IA2 of the Internet Appendix considers several additional robustness exercises using these multimarket banks. Data on post-liquidation resales offer perhaps the most powerful test of the firesale hypothesis. If these results reflect the causal effect of bank balance sheet distress on liquidation values—the firesale hypothesis—then the price bounce or capital gain from resale should be largest for those assets sold by the most illiquid or under-capitalized banks. Liquidations by distress sellers create an arbitrage opportunity for unconstrained buyers, allowing these buyers to purchase the liquidated property at a discount. Unconstrained buyers can then resell the property at its higher fundamental value (Shleifer and Vishny 1997). Put differently, under the fire-sale hypothesis, the size of the discount is related to the magnitude of the seller’s distress; this means that the buyer’s return to arbitrage is proportional to the seller’s distress. If, however, these results are driven by latent economic fundamentals that jointly drive liquidation values and balance sheets, then any subsequent capital gains should be unrelated to the balance sheet of the seller at the time of liquidation. If anything, given that fundamentals are highly persistent in neighborhoods with foreclosures—the autoregressive coefficient for house price changes at the quarterly frequency in these areas is 0.96—under the latent fundamentals hypothesis, observed capital gains should be lower for properties sold by illiquid banks: The unobserved negative fundamentals that weakened balance sheets and liquidation values in the first place should also keep subsequent resale prices low. To test whether balance sheet variables might explain subsequent capital gains, I collected data on the date of the first resale and the price obtained for the subset of liquidated properties in the sample that subsequently resold by end 2015. About 105,000 such cases exist. I do not have data on any improvements made after purchase or any leverage used by buyers, and for each of these properties I calculate the daily unlevered internal rate of return (IRR): $$(\frac{\mathit{resale}\ \mathit{price}_{t+n}}{\mathit{liquidation}\ \mathit{value}_t})^{\frac1 n}-1,$$ where n is the number of days elapsed from liquidation by the bank to subsequent resale by the buyer. Column 6 of Table 4 regresses this internal rate of return on the baseline set of balance sheet variables observed in the quarter before liquidation. The evidence is striking: properties sold by more illiquid banks are associated with an economically and statistically large price bounce. A 1-standard-deviation decrease in deposit growth in the quarter before a bank sells an asset increases the buyer’s subsequent daily IRR by 0.02 percentage points. Note that the median property is resold in about 160 days, and the median daily IRR is 0.08%. Equally striking again is the fact that using the change in the deposit rate as the measure of illiquidity gives identical results (Column 7). A 1-standard-deviation increase in the deposit rate in the quarter before liquidation increases a buyer’s daily IRR by 0.02 percentage points. These estimates suggest that the returns to “dry powder” are large. Buying from an illiquid bank—defined as one that in the previous quarter experienced a negative 1-standard-deviation liquidity shock—and reselling the asset 160 days later augments returns by about 3.1 percentage points. This result is difficult to reconcile with narratives about latent fundamentals that drive liquidations and balance sheets. Note also this result is robust to including measures of collateral quality, such as the year built, an indicator for whether the property has been remodeled in the last 10 years, square footage, the number of bedrooms, and the number of bathrooms, all of which are available on request. All this suggest that balance sheet pressures depress liquidation values below fundamentals. But a possible concern is that the sample of properties that are resold could be selected on unobservables that also correlate with the balance sheet observables of interest. For example, positive local fundamentals could increase the probability that a liquidated property is resold, and also increase the IRR on the transaction. At the same time, positive latent fundamentals could drive deposit flows, imparting an upward on the coefficient in Column 6. Although I cannot completely exclude this possibility, fundamentals at the ZIP code level are slowly changing over the sample period, and mostly absorbed nonparametrically in the fixed effects. And in results available on request, these balance sheet observables in the quarter before a property is liquidated do not predict whether the property is subsequently resold, suggesting that endogenous selection might not be significant source of bias. However, the endogenous selection of delinquent properties into foreclosure is a potential source of bias. Once a loan becomes delinquent, the bank and borrower can agree to revise the loan terms and return the loan to current status. Otherwise, failing agreement, the bank can foreclose upon the property; the property then enters the sample of liquidated bank collateral. This sequence of decisions implies that unobserved bank-level characteristics that drive selection into foreclosure also could be correlated with the balance sheet variables, leading to biased estimates in the pricing equation. The narrative evidence around mortgage servicing and the statistical evidence in Table IA3, which are both available in the Internet Appendix, suggest that any bias from nonrandom selection into the pool of foreclosed properties is likely to be small. 3. Mechanism 3.1 Policy, local absorptive capacity, and balance-sheet heterogeneity During the sample period, the government experimented with a range of policies to stabilize the financial system, including directly injecting capital into some banks—the Troubled Asset Relief Program (TARP)—and directly purchasing assets—QE1, QE2, and QE3.12 This subsection uses the variation in some of these programs across time and banks, along with other tests, to understand better the underlying mechanism behind REO discounts and balance sheet observables. In the case of TARP, eligible institutions sold equity interests to the U.S. Treasury in amounts equal to 1% to 3% of the institution’s risk-weighted assets. Eligibility was in part determined by the regulator’s assessment of a bank’s solvency. To limit taxpayer losses, banks in poor health, such as those with plentiful bad assets or those close to insolvency, were ineligible for TARP funds. Also, to remove any stigma associated with these injections and encourage other banks to apply, the Treasury first injected capital into the largest banks. TARP likely had two key effects. First, these capital injections mechanically increased a bank’s distance to insolvency. Second, because weak banks were ostensibly ineligible for TARP funds, Treasury equity purchases likely conveyed positive information about a bank’s solvency, helping recipients more easily raise outside equity. These effects suggest that for a given distance to insolvency—the tier 1 ratio—a bank that received TARP funds, and now viewed as solvent, might have less incentive to deleverage by selling REO assets at a discount, as it would now have the balance sheet capacity to potentially wait for a rebound in prices. Of course, recipient banks could use the capital injection to realize losses and clean up their balance sheets. To evaluate the effects of TARP then, I use a simple difference-in-difference model, creating an indicator variable that equals 1 in the quarters after a bank receives TARP funds, and 0 otherwise. I also interact this variable with a bank’s tier 1 to risk-weighted assets ratio. The timing of entry into TARP is nonrandom, as troubled banks probably sought funds earlier, and the specification in Column 1 of Table 5 also includes an indicator variable that equals 1 in the quarter before a bank receives TARP funds, and 0 otherwise. Earlier lags are insignificant. This “pretreatment” variable is also interacted with the tier 1 ratio. As before, ZIP code by year-quarter fixed effects absorb local economic conditions that might simultaneously affect liquidation values and balance sheet outcomes, including the decision to enter into TARP. Table 5 Mechanism 1: Heterogeneity . (1) . (2) . (3) . (4) . (5) . . . Policy . . Cash-in-the-local-market . Cash-on-balance sheet . VARIABLES . TARP . QE . State-Laws . non-bank foreclosures . cash . Year on year change in deposits, scaled by assets 0.113*** 0.123*** 0.145*** 0.0147 0.107*** (0.0319) (0.0333) (0.0389) (0.0401) (0.0343) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.217 0.197 0.184* 0.378* 0.503 (0.135) (0.120) (0.109) (0.215) (0.312) Tier 1 Capital/Risk Weighted Assets* TARP –0.0119 Tier 1 Capital/Risk Weighted Assets* TARP, quarter before (0.312) 0.719 (0.491) TARP –0.0228 (0.0298) TARP, quarter before –0.0751** (0.0369) QE 1*MBS and Treasury Holdings 1.136** (0.487) QE 2*MBS and Treasury Holdings –0.401 (0.385) QE 3*MBS and Treasury Holdings –0.125 (0.255) QE 1 –0.0281 (0.0259) QE 2 0.00264 (0.0167) QE 3 0.0455*** (0.0104) change in deposits*judicial foreclosure state –0.0828** (0.0348) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.0162 (0.142) change in deposits*log of non-bank foreclosures in zipcode 0.0309** (0.0147) Tier 1 Capital/Risk Weighted Assets*log number of non-bank foreclosures in zip c –0.0537 (0.0770) change in deposits*bottom quartile of cash 0.119* (0.0648) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash –0.353 (0.375) zipcode*year-quarter*fixed effects yes yes yes yes yes bank fixed effects yes yes yes yes yes Observations 301,757 301,757 301,757 297,825 301,757 R-squared 0.693 0.693 0.693 0.690 0.693 H0: Tier 1 Ratio+Tier 1Ratio*TARP=0 0.05 H0: Tier 1 Ratio+Tier 1Ratio*TARP, quarter before=0 0.01 H0:change in deposits+change in deposits*interaction term=0 0.00 0.00 0.00 H0:tier 1 ratio+tier 1 ratio*interaction term=0 0.00 0.08 0.15 . (1) . (2) . (3) . (4) . (5) . . . Policy . . Cash-in-the-local-market . Cash-on-balance sheet . VARIABLES . TARP . QE . State-Laws . non-bank foreclosures . cash . Year on year change in deposits, scaled by assets 0.113*** 0.123*** 0.145*** 0.0147 0.107*** (0.0319) (0.0333) (0.0389) (0.0401) (0.0343) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.217 0.197 0.184* 0.378* 0.503 (0.135) (0.120) (0.109) (0.215) (0.312) Tier 1 Capital/Risk Weighted Assets* TARP –0.0119 Tier 1 Capital/Risk Weighted Assets* TARP, quarter before (0.312) 0.719 (0.491) TARP –0.0228 (0.0298) TARP, quarter before –0.0751** (0.0369) QE 1*MBS and Treasury Holdings 1.136** (0.487) QE 2*MBS and Treasury Holdings –0.401 (0.385) QE 3*MBS and Treasury Holdings –0.125 (0.255) QE 1 –0.0281 (0.0259) QE 2 0.00264 (0.0167) QE 3 0.0455*** (0.0104) change in deposits*judicial foreclosure state –0.0828** (0.0348) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.0162 (0.142) change in deposits*log of non-bank foreclosures in zipcode 0.0309** (0.0147) Tier 1 Capital/Risk Weighted Assets*log number of non-bank foreclosures in zip c –0.0537 (0.0770) change in deposits*bottom quartile of cash 0.119* (0.0648) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash –0.353 (0.375) zipcode*year-quarter*fixed effects yes yes yes yes yes bank fixed effects yes yes yes yes yes Observations 301,757 301,757 301,757 297,825 301,757 R-squared 0.693 0.693 0.693 0.690 0.693 H0: Tier 1 Ratio+Tier 1Ratio*TARP=0 0.05 H0: Tier 1 Ratio+Tier 1Ratio*TARP, quarter before=0 0.01 H0:change in deposits+change in deposits*interaction term=0 0.00 0.00 0.00 H0:tier 1 ratio+tier 1 ratio*interaction term=0 0.00 0.08 0.15 The dependent variable is the log liquidation price. In Column 1, the variable “TARP” equals 1 after a bank receives TARP funds and 0 otherwise. “TARP, quarter before” equals 1 in the quarter before a bank receives TARP and 0 otherwise. In Column 2, QE “i” are indicator variables that equal 1 in the quarters in which the specific program was in place. “MBS and Treasury” holdings is the ratio of these assets to total assets in 2008. In Column 3, “Judicial Foreclosure State” equals 1 in those states that require judicial foreclosures and 0 otherwise. In Column 4, “Nonbank foreclosures” is the log of the number of nonbank foreclosures in the ZIP code in that quarter. In Column 5, “bottom quartile of cash” is an indicator variable that equals 1 if a bank’s averaged cash/assets ratio over 2001–2006 is in the bottom quartile and 0 otherwise. All specifications include the baseline controls from Column 3 of Table 4: loan to deposits, deposits to total assets, cash to total assets, the log of total assets, and the return to assets, all lagged one quarter. All specifications include ZIP code-by-year-quarter fixed effects and bank fixed effects; standard errors, in parentheses, are clustered by ZIP code and bank. Open in new tab Table 5 Mechanism 1: Heterogeneity . (1) . (2) . (3) . (4) . (5) . . . Policy . . Cash-in-the-local-market . Cash-on-balance sheet . VARIABLES . TARP . QE . State-Laws . non-bank foreclosures . cash . Year on year change in deposits, scaled by assets 0.113*** 0.123*** 0.145*** 0.0147 0.107*** (0.0319) (0.0333) (0.0389) (0.0401) (0.0343) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.217 0.197 0.184* 0.378* 0.503 (0.135) (0.120) (0.109) (0.215) (0.312) Tier 1 Capital/Risk Weighted Assets* TARP –0.0119 Tier 1 Capital/Risk Weighted Assets* TARP, quarter before (0.312) 0.719 (0.491) TARP –0.0228 (0.0298) TARP, quarter before –0.0751** (0.0369) QE 1*MBS and Treasury Holdings 1.136** (0.487) QE 2*MBS and Treasury Holdings –0.401 (0.385) QE 3*MBS and Treasury Holdings –0.125 (0.255) QE 1 –0.0281 (0.0259) QE 2 0.00264 (0.0167) QE 3 0.0455*** (0.0104) change in deposits*judicial foreclosure state –0.0828** (0.0348) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.0162 (0.142) change in deposits*log of non-bank foreclosures in zipcode 0.0309** (0.0147) Tier 1 Capital/Risk Weighted Assets*log number of non-bank foreclosures in zip c –0.0537 (0.0770) change in deposits*bottom quartile of cash 0.119* (0.0648) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash –0.353 (0.375) zipcode*year-quarter*fixed effects yes yes yes yes yes bank fixed effects yes yes yes yes yes Observations 301,757 301,757 301,757 297,825 301,757 R-squared 0.693 0.693 0.693 0.690 0.693 H0: Tier 1 Ratio+Tier 1Ratio*TARP=0 0.05 H0: Tier 1 Ratio+Tier 1Ratio*TARP, quarter before=0 0.01 H0:change in deposits+change in deposits*interaction term=0 0.00 0.00 0.00 H0:tier 1 ratio+tier 1 ratio*interaction term=0 0.00 0.08 0.15 . (1) . (2) . (3) . (4) . (5) . . . Policy . . Cash-in-the-local-market . Cash-on-balance sheet . VARIABLES . TARP . QE . State-Laws . non-bank foreclosures . cash . Year on year change in deposits, scaled by assets 0.113*** 0.123*** 0.145*** 0.0147 0.107*** (0.0319) (0.0333) (0.0389) (0.0401) (0.0343) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.217 0.197 0.184* 0.378* 0.503 (0.135) (0.120) (0.109) (0.215) (0.312) Tier 1 Capital/Risk Weighted Assets* TARP –0.0119 Tier 1 Capital/Risk Weighted Assets* TARP, quarter before (0.312) 0.719 (0.491) TARP –0.0228 (0.0298) TARP, quarter before –0.0751** (0.0369) QE 1*MBS and Treasury Holdings 1.136** (0.487) QE 2*MBS and Treasury Holdings –0.401 (0.385) QE 3*MBS and Treasury Holdings –0.125 (0.255) QE 1 –0.0281 (0.0259) QE 2 0.00264 (0.0167) QE 3 0.0455*** (0.0104) change in deposits*judicial foreclosure state –0.0828** (0.0348) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.0162 (0.142) change in deposits*log of non-bank foreclosures in zipcode 0.0309** (0.0147) Tier 1 Capital/Risk Weighted Assets*log number of non-bank foreclosures in zip c –0.0537 (0.0770) change in deposits*bottom quartile of cash 0.119* (0.0648) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash –0.353 (0.375) zipcode*year-quarter*fixed effects yes yes yes yes yes bank fixed effects yes yes yes yes yes Observations 301,757 301,757 301,757 297,825 301,757 R-squared 0.693 0.693 0.693 0.690 0.693 H0: Tier 1 Ratio+Tier 1Ratio*TARP=0 0.05 H0: Tier 1 Ratio+Tier 1Ratio*TARP, quarter before=0 0.01 H0:change in deposits+change in deposits*interaction term=0 0.00 0.00 0.00 H0:tier 1 ratio+tier 1 ratio*interaction term=0 0.00 0.08 0.15 The dependent variable is the log liquidation price. In Column 1, the variable “TARP” equals 1 after a bank receives TARP funds and 0 otherwise. “TARP, quarter before” equals 1 in the quarter before a bank receives TARP and 0 otherwise. In Column 2, QE “i” are indicator variables that equal 1 in the quarters in which the specific program was in place. “MBS and Treasury” holdings is the ratio of these assets to total assets in 2008. In Column 3, “Judicial Foreclosure State” equals 1 in those states that require judicial foreclosures and 0 otherwise. In Column 4, “Nonbank foreclosures” is the log of the number of nonbank foreclosures in the ZIP code in that quarter. In Column 5, “bottom quartile of cash” is an indicator variable that equals 1 if a bank’s averaged cash/assets ratio over 2001–2006 is in the bottom quartile and 0 otherwise. All specifications include the baseline controls from Column 3 of Table 4: loan to deposits, deposits to total assets, cash to total assets, the log of total assets, and the return to assets, all lagged one quarter. All specifications include ZIP code-by-year-quarter fixed effects and bank fixed effects; standard errors, in parentheses, are clustered by ZIP code and bank. Open in new tab The evidence in Column 1 of Table 5 suggests that TARP capital injections shaped liquidation values. After a bank receives TARP funds, the relationship between the equity constraint and liquidation values weakens: The interaction term is negative, and both variables are significant at the 1% level (p-value = .01). But in the quarter before a bank receives TARP funds, equity losses have a bigger effect on liquidation values. These effects are large. A 1-standard-deviation decrease in the tier 1 ratio is associated with a 1.1% drop in liquidation values after a bank receives TARP. But in the quarter just before the capital injection, probably when the bank is at greatest risk, a similar decrease in equity suggests a 5.3% drop in liquidation values.13 While this evidence is suggestive that equity injections might help contain destabilizing asset sales, this evidence should be interpreted cautiously as unobserved heterogeneity that determine selection into TARP could bias inference.14 Quantitative easing (QE)—the purchases of assets, such as MBS and Treasuries of various maturities by central banks—can also affect banks’ incentives to sell assets and equilibrium liquidation values. In the most direct channel, central bank purchases of MBS and Treasuries can inflate the prices of these assets and create capital gains at banks that hold these assets, thereby improving the solvency of these institutions. Improved solvency could then reduce a bank’s incentive to sell REOs at a discount. Central bank purchases of MBS can also improve the liquidity of these assets, again reducing the need for banks to sell REOs at a discount. QE can affect liquidation values through other channels, including through demand and expectations (see Foley-Fisher, Ramcharan, and Yu 2015; see also the surveys in Williamson 2017; Krishnamurthy and Vissig-Jorgensen 2013).15 To gauge the impact of these asset purchases on liquidation values, I follow Rodnyanski and Darmouni (2016) and compute each bank’s exposure to QE asset purchases based on the bank’s holdings of MBS and Treasury securities in 2007, expressed as a share of assets. This ratio is then interacted with indicators that equal 1 for the various QE policy intervention quarters; these indicator variables also appear linearly in the regressions; the cross-sectional exposure to QE—the ratio of MBS and Treasuries to assets in 2007—is absorbed in the bank fixed effects. Column 2 of Table 5 reports these results. During QE1, there is significant evidence that liquidation values are higher when a bank has more MBS and Treasury holdings. For a bank at the tenth percentile of this ratio, liquidation values during QE1 are about 0.04 bp higher than otherwise. But for a bank at the 75th percentile, and thus heavily exposed to QE1’s attempt to stabilize the MBS market, liquidation values are 1.7 percentage points higher on average. While QE1 was unexpected, QE2 and QE3 were largely anticipated by markets, and those coefficients are insignificant. Again, while this evidence is suggestive of the balance sheet channel in determining liquidation values, they cannot completely exclude other channels like unobserved demand or changes in expectations. State-level differences in foreclosure policies provide another source of plausibly exogenous variation that can also help interpret the relationship between balance sheet outcomes and liquidation values. Foreclosures and asset sales are much slower in states that require lenders to use the courts in order to foreclose upon real estate collateral (judicial foreclosure states). While in “power of sale” states, after a notice of default lenders can in many cases seize and liquidate collateral quickly (Pence 2006). Some evidence already indicates that foreclosure rates in “power of sale” states are higher and that the impact of these laws can affect local prices (Mian, Sufi, and Trebbi 2015; Rajan and Ramcharan 2016). We would expect then that balance sheet pressures should have a bigger effect on liquidation values in “power of sale” states. Column 3 shows this. The deposit change coefficient is about 57% smaller in judicial foreclosure states. Column 4 of Table 5 uses the variation in the local capacity to absorb asset sales. In ZIP codes with a larger number of ongoing nonbank asset sales that absorb the local cash available in the market, the capacity to absorb additional bank asset sales without dislocating prices is more limited (Allen and Gale 1994). Sales of bank collateral in these areas should then be associated with even lower liquidation values: the impact of balance sheet pressures on liquidation values should be larger when the local capacity to absorb more asset sales is limited. To measure the capacity to absorb asset sales in the ZIP code, I compute the number of nonbank foreclosures sales within the ZIP code in the current quarter. These are mainly sales of properties held in securitization trusts. Column 4 interacts this variable with both the deposit change and tier 1 variables; the latter two variables are included linearly, while the number of nonbank foreclosures is absorbed in the ZIP code-by-year-quarter fixed effect. The deposit change variable and its interaction term with the number of nonbank foreclosures sales within the ZIP code are positive and jointly significant at the 1% level: nonbank foreclosure sales amplify the impact of bank balance sheet outcomes on liquidation values. For a ZIP code-quarter observation at the tenth percentile of nonbank foreclosure sales, a 1-standard-deviation decrease in deposit growth is associated with a 0.8% drop in liquidation values. But at the 90th percentile of nonbank foreclosure sales, a similar loss of funding is associated with a 1.8% drop in the liquidation value of the distressed collateral, suggesting that forced sales in already depressed markets lead to greater price dislocations. Note that the sign of the interaction term between the tier 1 ratio and the number of nonbank foreclosures sales is unexpected, but this magnitude is economically small and only marginally significant. Finally, illiquidity on both sides of the balance sheet can interact to shape liquidation values. This implies that the impact of a loss of deposit financing on liquidation values should be larger among banks with less liquid assets. Unable to meet easily the liquidity demands of depositors, banks with less cash face greater pressures to liquidate quickly troubled assets. To test this prediction, Column 5 of Table 5 interacts both the deposit change and tier 1 capital ratio with an indicator variable that equals 1 if a bank’s ratio of cash to assets, averaged between 2001–2006, is in the bottom quartile and zero otherwise. Evidence shows that a loss of deposits largely affects liquidation values among banks that entered the crisis period with less cash. A 1-standard-deviation decrease in the deposit change variable is associated with a 1.4% drop in liquidation values among banks in the full sample. But for a bank in the bottom quartile of the cash to assets ratio as measured in the precrisis period, a similar decrease in deposits suggests a 2.9% drop in liquidation values. 3.2 Hazard model The speed of asset sales might help explain how balance sheet outcomes affect liquidation values. Banks with plentiful liquidity—those experiencing deposit inflows or those with sizeable cash on the balance sheet—might delay REO assets sales in order to obtain higher liquidation values. Conversely, a loss of deposits could accelerate the pace of asset of sales and depress liquidation values. If, however, deposit growth proxies for good local economic fundamentals and plentiful cash in the local market, then positive deposit growth would imply both higher liquidation values and faster asset sales on account of plentiful cash in the local market. This section thus studies the speed of asset sales. Table IA4 provides suggestive evidence that illiquid banks tend to sell faster. It reports summary statistics for the number of quarters that a property takes to sell. This duration is computed from the date the property is foreclosed upon and available sale, to the final sale date. Of the 300,000 or so properties in the sample, I can compute this duration for about 172,210 properties, as not all counties identify reliably the date of first foreclosure. The simple summary statistics suggests that liquidity might shape the pace of asset sales. During the quarters in which each property is available for sale, I compute the average deposit growth of the selling bank. Table IA4 then reports distributional statistics for this duration separately for properties sold by banks with below-median deposit growth—the illiquid banks—and for the more liquid banks—those with mean deposit growth rates over the property’s for-sale period that is above sample median. The median number of quarters until sale across the two groups is the same, but at the 75th percentile and beyond, properties sold by more liquid banks generally take an additional quarter to sell. These differences in duration between below- and above-median banks are significant at the 1% level. Recall that the maximum duration in the sample is 36 months. Table 6 considers more carefully the impact of balance sheet outcomes on the duration of an asset sale. The duration is the number of quarters elapsed from foreclosure to sale. The hazard rate, h(t), is the probability that a property sells at time t given that it has not yet sold. The table uses a simple parametric survival-time model with a Weibull distribution. The proportional hazards model takes the form: Table 6 Mechanism 2: Duration . (1) . (2) . (3) . (4) . (5) . VARIABLES . baseline . TARP . QE . state-laws . cash . Year on year change in deposits, scaled by assets –0.802*** –0.816*** –0.609*** –0.662*** –0.796*** (0.0268) (0.0270) (0.0260) (0.0282) (0.0268) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.271 –1.7046*** –1.074*** –0.676*** –2.242*** (0.165) (0.217) (0.180) (0.171) (0.257) TARP –0.06066* (0.0369) TARP, quarter before –0.138 (0.140) Tier 1 Capital/Risk Weighted Assets* TARP, quarter before –0.903*** (0.260) Tier 1 Capital/Risk Weighted Assets* TARP 0.718** (0.332) QE 1*MBS and Treasury Holdings –5.899*** (0.371) QE 2*MBS and Treasury Holdings –6.778*** (0.463) QE 3*MBS and Treasury Holdings –0.376 (0.363) QE 1 0.325*** (0.0102) QE 2 –0.111*** (0.00914) QE 3 0.210*** (0.0105) change in deposits*judicial foreclosure state 0.389*** (0.0442) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.986*** (0.158) change in deposits*bottom quartile of cash –0.421*** (0.0891) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash 5.068*** (0.336) Observations 740,129 740,129 740,129 740,129 740,129 p 1.851 1.848 1.860 1.854 1.852 . (1) . (2) . (3) . (4) . (5) . VARIABLES . baseline . TARP . QE . state-laws . cash . Year on year change in deposits, scaled by assets –0.802*** –0.816*** –0.609*** –0.662*** –0.796*** (0.0268) (0.0270) (0.0260) (0.0282) (0.0268) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.271 –1.7046*** –1.074*** –0.676*** –2.242*** (0.165) (0.217) (0.180) (0.171) (0.257) TARP –0.06066* (0.0369) TARP, quarter before –0.138 (0.140) Tier 1 Capital/Risk Weighted Assets* TARP, quarter before –0.903*** (0.260) Tier 1 Capital/Risk Weighted Assets* TARP 0.718** (0.332) QE 1*MBS and Treasury Holdings –5.899*** (0.371) QE 2*MBS and Treasury Holdings –6.778*** (0.463) QE 3*MBS and Treasury Holdings –0.376 (0.363) QE 1 0.325*** (0.0102) QE 2 –0.111*** (0.00914) QE 3 0.210*** (0.0105) change in deposits*judicial foreclosure state 0.389*** (0.0442) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.986*** (0.158) change in deposits*bottom quartile of cash –0.421*** (0.0891) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash 5.068*** (0.336) Observations 740,129 740,129 740,129 740,129 740,129 p 1.851 1.848 1.860 1.854 1.852 This table models the duration of asset sales. Duration is computed from the time a property becomes foreclosed to when it is finally liquidated. The point estimates below are the coefficients (nonexponentiated) from a Weibull hazard model. The parameter “|$p$|” is the shape parameter in the Weibull function |$pt^{p-1}$|⁠. Each regression allows the baseline hazard to vary by bank and ZIP code (banks and ZIP code fixed effects) and standard errors are clustered by ZIP code. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Open in new tab Table 6 Mechanism 2: Duration . (1) . (2) . (3) . (4) . (5) . VARIABLES . baseline . TARP . QE . state-laws . cash . Year on year change in deposits, scaled by assets –0.802*** –0.816*** –0.609*** –0.662*** –0.796*** (0.0268) (0.0270) (0.0260) (0.0282) (0.0268) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.271 –1.7046*** –1.074*** –0.676*** –2.242*** (0.165) (0.217) (0.180) (0.171) (0.257) TARP –0.06066* (0.0369) TARP, quarter before –0.138 (0.140) Tier 1 Capital/Risk Weighted Assets* TARP, quarter before –0.903*** (0.260) Tier 1 Capital/Risk Weighted Assets* TARP 0.718** (0.332) QE 1*MBS and Treasury Holdings –5.899*** (0.371) QE 2*MBS and Treasury Holdings –6.778*** (0.463) QE 3*MBS and Treasury Holdings –0.376 (0.363) QE 1 0.325*** (0.0102) QE 2 –0.111*** (0.00914) QE 3 0.210*** (0.0105) change in deposits*judicial foreclosure state 0.389*** (0.0442) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.986*** (0.158) change in deposits*bottom quartile of cash –0.421*** (0.0891) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash 5.068*** (0.336) Observations 740,129 740,129 740,129 740,129 740,129 p 1.851 1.848 1.860 1.854 1.852 . (1) . (2) . (3) . (4) . (5) . VARIABLES . baseline . TARP . QE . state-laws . cash . Year on year change in deposits, scaled by assets –0.802*** –0.816*** –0.609*** –0.662*** –0.796*** (0.0268) (0.0270) (0.0260) (0.0282) (0.0268) Tier 1 Capital/Risk Weighted Assets, lagged one quarter 0.271 –1.7046*** –1.074*** –0.676*** –2.242*** (0.165) (0.217) (0.180) (0.171) (0.257) TARP –0.06066* (0.0369) TARP, quarter before –0.138 (0.140) Tier 1 Capital/Risk Weighted Assets* TARP, quarter before –0.903*** (0.260) Tier 1 Capital/Risk Weighted Assets* TARP 0.718** (0.332) QE 1*MBS and Treasury Holdings –5.899*** (0.371) QE 2*MBS and Treasury Holdings –6.778*** (0.463) QE 3*MBS and Treasury Holdings –0.376 (0.363) QE 1 0.325*** (0.0102) QE 2 –0.111*** (0.00914) QE 3 0.210*** (0.0105) change in deposits*judicial foreclosure state 0.389*** (0.0442) Tier 1 Capital/Risk Weighted Assets*judicial foreclosure state 0.986*** (0.158) change in deposits*bottom quartile of cash –0.421*** (0.0891) Tier 1 Capital/Risk Weighted Assets*bottom quartile of cash 5.068*** (0.336) Observations 740,129 740,129 740,129 740,129 740,129 p 1.851 1.848 1.860 1.854 1.852 This table models the duration of asset sales. Duration is computed from the time a property becomes foreclosed to when it is finally liquidated. The point estimates below are the coefficients (nonexponentiated) from a Weibull hazard model. The parameter “|$p$|” is the shape parameter in the Weibull function |$pt^{p-1}$|⁠. Each regression allows the baseline hazard to vary by bank and ZIP code (banks and ZIP code fixed effects) and standard errors are clustered by ZIP code. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Open in new tab $$h\left[t,X\left(t\right),b\right]=pt^{p-1}\exp\left[X(t)'b\right],$$ where |$X(t)$| is the baseline set of bank time-varying balance sheet outcomes; |$b$| is a vector of parameters to be estimated; |$p$| is the shape parameter that is also estimated from the data. When |$p>1$|⁠, this model displays a monotonically increasing hazard rate. Standard errors are clustered at the ZIP code level, and the basic model uses bank-fixed effect and ZIP code fixed effects. These fixed effects allow the baseline hazard to vary by bank and ZIP code. The evidence in Table 6 suggests that a loss of deposits is associated with quicker asset sales. From Column 1, using the baseline controls, a 1-standard-deviation decrease in the deposit change variable is associated with a 7% increase in the hazard or the probability that the property sells next quarter given that it has not yet sold. An advantage of the Weibull model is that it can be inverted and expressed in the time domain, with the new coefficients, |$b^{\ast} $|⁠, scaled by |$p$|⁠: |$b^{\ast} =-b/p$|⁠. Using this time representation, a 1-standard-deviation decrease in deposits implies a 3.8% decrease in the time remaining until the property sells; this is about 2 additional weeks for the average property. The coefficient on the tier 1 variable is not significant. Note that because the dynamics of auction sales could differ, in results available on request, I have also included a dummy for whether the property was sold at auction—the results are broadly similar. Column 2 builds on the previous results and interacts the TARP indicator variables with the tier 1 ratio; all variables are also linearly included. For those bank-quarter observations without TARP equity injections, a decrease in the tier 1 ratio is associated with an increase in the hazard or a greater conditional likelihood of a sale. A 1-standard-deviation decline in the tier 1 ratio in the previous quarter shifts up the hazard by 9.3% or decreases the remaining time to sell by about 5.1% or 19 days—banks deleverage by selling assets more quickly. In contrast, in the quarters in which a bank receives TARP, a similar decrease in the tier 1 ratio increases the hazard by only 5.4% or increases the time to sale by about 3%. But in the quarter before TARP, these magnitudes are about twice as large relative to the TARP treatment period. This evidence is consistent with those obtained earlier, implying that equity injections slow the pace of deleveraging through asset sales, thereby reducing price discounts. Column 3 adapts the basic QE specification from Column 2 of Table 5. As with TARP, evidence suggests that asset sales significantly slowed during QE1 and QE2 among banks with greater exposure to these policies. It shows, for example, that, during QE1, a 1-standard-deviation increase in a bank’s ratio of MBS and Treasuries assets to total assets is associated with about a 12% decrease in the hazard rate. For the average property, this suggests about 22 additional days until sale. Column 4 focuses on the role of state foreclosure laws. Legal frictions delay sales generally, and at any point in time, the hazard is about 28.4% smaller in judicial foreclosure states relative to power of sale states. And as with the evidence on liquidation values, the impact of a decrease in deposits on the time-to-sell depends on state law. A 1-standard-deviation decrease in the deposit change variable is associated with about a 5.8% increase in the hazard in a power of sale state; but in a judicial foreclosure state, the conditional probability of sale increases by only 2.4%. Finally, Column 5 shows that the impact of a decrease in deposits on the conditional probability of a sale next period is about 52% larger among banks in the bottom quartile of cash. In results, which are available on request, some evidence indicates that banks limit geographic exposure by selling more quickly properties in counties that account for a bigger share of the bank’s origination portfolio during the boom period. Some properties are never sold by end 2015 and attrite out of the sample. This censoring could potentially provide only a partial picture of how balance sheet pressures might affect sales at the extensive margin—whether a property is ever sold. To model the extensive margin then, I collect data on about 105,000 additional properties that were “available for sale” properties held by banks but did not sell by the end of sample period. Adding these to the remaining 300,000 or so properties that sold, creates an unbalanced quarterly panel of 421,746 properties with 1.6 million property-quarter observations. Using this panel, I model the probability that a property sells over the sample period—the extensive margin—using a simple linear probability model. The dependent variable equal 1 in the quarter a property is sold and 0 in the quarters before it is sold. A property exits the sample once sold. If it remains unsold by the end of the sample, the dependent variable remains 0 throughout. As before, a property enters the panel when it is foreclosed upon (available for sale) and the sample period ends Q4 2015. Consistent with the hazard model results on the time-to-sell, a decrease in deposits the previous quarter increases the probability that a property is sold in the next quarter (these results are available on request). 4. Spillovers This subsection now examines whether liquidation values among bank-owned properties affect the prices of other nearby properties. Pricing in real estate is based in part on the price of previous sales of comparable assets. Low liquidation values among bank owned properties could thus depress the subsequent price of otherwise similar non-bank-owned properties (Annenberg and Kung 2014; Murfin and Pratt 2016). In this way, bank balance sheet pressures could negatively spill over onto the prices of nearby nonforeclosure or arm’s length home sales, depressing collateral values more widely and impeding economic activity (Rajan and Ramcharan 2016). To understand the spillover effects of REO sales, the analysis turns to transaction level data on all residential arm’s length real estate sales for the five states in the sample the provide the latitude and longitude of the house.16 Using the date of sale and location of each of these arm’s length transactions, I match each arm’s length transaction to an REO sale using a simple nearest neighbor approach: I identify the geographically closest REO sale within the previous 18 months that is located no farther than 1 km away from the arm’s length transaction. This criterion yields about 805,000 unique real estate transactions that match to one of the 167,000 REO transactions in these five states. The median distance is about 140 m—the same block—whereas the median REO sale precedes its nearest arm’s length neighbor by about 240 days. Given this relative geographic and temporal proximity, the nearest REO sale is likely to be a relevant comparable for the pricing of the subsequent nonbank sale. Using this simple nearest neighbor approach, Column 1 of Table 7 regresses the price obtained in the arm’s length transaction on the liquidation value of the nearest REO sale within the 18-month and 1-km window. Not surprisingly given the geographic and temporal proximity, the coefficient is positive and significant with a price elasticity of 0.4. The regression includes ZIP code and year-by-quarter fixed effects. Table 7 Spillovers . (1) . (2) . (3) . (4) . (5) . (6) . . OLS . IV . VARIABLES . no controls . hedonic controls . monthly house prices . zipcode*year-quarter . fixed effects . zipcode*bank fixed effects . liquidation value of REO asset, logs 0.402*** 0.250*** 0.253*** 0.253*** 0.298* 0.381* (0.00689) (0.00628) (0.00606) (0.00515) (0.179) (0.239) Lot size, square feet, logs 0.216*** 0.217*** 0.223*** 0.217*** 0.201*** (0.00438) (0.00461) (0.00474) (0.0249) (0.0348) Total number of bedrooms, logs 0.0299*** 0.0262*** 0.0351*** 0.0378*** 0.0350*** (0.00967) (0.00962) (0.00982) (0.0130) (0.0125) Total number of baths, logs 0.499*** 0.501*** 0.476*** 0.463*** 0.453*** (0.0130) (0.0131) (0.0117) (0.0466) (0.0635) Year built, logs 14.79*** 14.68*** 14.41*** 13.89*** 12.97*** (0.489) (0.498) (0.556) (2.220) (2.981) zipcode price index change 0.127 0.217 (0.252) (0.332) zipcode price index change, one month lag 1.886*** 1.828*** (0.272) (0.295) Observations 804,801 737,896 711,822 734,078 734,064 709,828 R-squared 0.571 0.667 0.666 0.714 0.714 0.689 . (1) . (2) . (3) . (4) . (5) . (6) . . OLS . IV . VARIABLES . no controls . hedonic controls . monthly house prices . zipcode*year-quarter . fixed effects . zipcode*bank fixed effects . liquidation value of REO asset, logs 0.402*** 0.250*** 0.253*** 0.253*** 0.298* 0.381* (0.00689) (0.00628) (0.00606) (0.00515) (0.179) (0.239) Lot size, square feet, logs 0.216*** 0.217*** 0.223*** 0.217*** 0.201*** (0.00438) (0.00461) (0.00474) (0.0249) (0.0348) Total number of bedrooms, logs 0.0299*** 0.0262*** 0.0351*** 0.0378*** 0.0350*** (0.00967) (0.00962) (0.00982) (0.0130) (0.0125) Total number of baths, logs 0.499*** 0.501*** 0.476*** 0.463*** 0.453*** (0.0130) (0.0131) (0.0117) (0.0466) (0.0635) Year built, logs 14.79*** 14.68*** 14.41*** 13.89*** 12.97*** (0.489) (0.498) (0.556) (2.220) (2.981) zipcode price index change 0.127 0.217 (0.252) (0.332) zipcode price index change, one month lag 1.886*** 1.828*** (0.272) (0.295) Observations 804,801 737,896 711,822 734,078 734,064 709,828 R-squared 0.571 0.667 0.666 0.714 0.714 0.689 This table examines the impact of bank liquidations on the prices of nearby sales. The dependent variable is the log price of a nonforeclosure transaction. “Nearby” is defined as those properties that sold in the last 18 months and no farther than 1 km from a foreclosed sale. Columns 1–3 include ZIP code and year-quarter fixed effects. Columns 4 and 5 include ZIP code-by-year-quarter fixed effects. The instruments in Columns 5 and 6 are the year on year change in deposits, scaled by assets in the previous quarter, and tier 1 capital to assets, observed in the previous quarter. Standard errors are clustered at the REO matched sale and year-by-quarter level. Open in new tab Table 7 Spillovers . (1) . (2) . (3) . (4) . (5) . (6) . . OLS . IV . VARIABLES . no controls . hedonic controls . monthly house prices . zipcode*year-quarter . fixed effects . zipcode*bank fixed effects . liquidation value of REO asset, logs 0.402*** 0.250*** 0.253*** 0.253*** 0.298* 0.381* (0.00689) (0.00628) (0.00606) (0.00515) (0.179) (0.239) Lot size, square feet, logs 0.216*** 0.217*** 0.223*** 0.217*** 0.201*** (0.00438) (0.00461) (0.00474) (0.0249) (0.0348) Total number of bedrooms, logs 0.0299*** 0.0262*** 0.0351*** 0.0378*** 0.0350*** (0.00967) (0.00962) (0.00982) (0.0130) (0.0125) Total number of baths, logs 0.499*** 0.501*** 0.476*** 0.463*** 0.453*** (0.0130) (0.0131) (0.0117) (0.0466) (0.0635) Year built, logs 14.79*** 14.68*** 14.41*** 13.89*** 12.97*** (0.489) (0.498) (0.556) (2.220) (2.981) zipcode price index change 0.127 0.217 (0.252) (0.332) zipcode price index change, one month lag 1.886*** 1.828*** (0.272) (0.295) Observations 804,801 737,896 711,822 734,078 734,064 709,828 R-squared 0.571 0.667 0.666 0.714 0.714 0.689 . (1) . (2) . (3) . (4) . (5) . (6) . . OLS . IV . VARIABLES . no controls . hedonic controls . monthly house prices . zipcode*year-quarter . fixed effects . zipcode*bank fixed effects . liquidation value of REO asset, logs 0.402*** 0.250*** 0.253*** 0.253*** 0.298* 0.381* (0.00689) (0.00628) (0.00606) (0.00515) (0.179) (0.239) Lot size, square feet, logs 0.216*** 0.217*** 0.223*** 0.217*** 0.201*** (0.00438) (0.00461) (0.00474) (0.0249) (0.0348) Total number of bedrooms, logs 0.0299*** 0.0262*** 0.0351*** 0.0378*** 0.0350*** (0.00967) (0.00962) (0.00982) (0.0130) (0.0125) Total number of baths, logs 0.499*** 0.501*** 0.476*** 0.463*** 0.453*** (0.0130) (0.0131) (0.0117) (0.0466) (0.0635) Year built, logs 14.79*** 14.68*** 14.41*** 13.89*** 12.97*** (0.489) (0.498) (0.556) (2.220) (2.981) zipcode price index change 0.127 0.217 (0.252) (0.332) zipcode price index change, one month lag 1.886*** 1.828*** (0.272) (0.295) Observations 804,801 737,896 711,822 734,078 734,064 709,828 R-squared 0.571 0.667 0.666 0.714 0.714 0.689 This table examines the impact of bank liquidations on the prices of nearby sales. The dependent variable is the log price of a nonforeclosure transaction. “Nearby” is defined as those properties that sold in the last 18 months and no farther than 1 km from a foreclosed sale. Columns 1–3 include ZIP code and year-quarter fixed effects. Columns 4 and 5 include ZIP code-by-year-quarter fixed effects. The instruments in Columns 5 and 6 are the year on year change in deposits, scaled by assets in the previous quarter, and tier 1 capital to assets, observed in the previous quarter. Standard errors are clustered at the REO matched sale and year-by-quarter level. Open in new tab But time-varying local unobserved shocks that simultaneously affect liquidation values and the sale price of homes in the area is a significant challenge to causal inference within this empirical design. Positive shocks—local fundamentals—for example, that increase liquidation values could also increase the price of nearby arm’s length transactions, imparting an upward bias. The existing literature has used hedonic controls and spatial disaggregation to help address these concerns, and building on this existing work, Column 2 includes hedonic details about the house to control better for its “fundamental” value. These variables enter with their expected signs, and the point estimate on the log liquidation value of the nearest neighbor REO sale declines by about 36%, suggesting that these hedonic variables might reasonably control for the fundamental value of the asset. But, to limit the influence of broader economic shocks that might drive both liquidation values and arm’s length prices, Column 3 includes monthly changes in the Zillow ZIP code-level house price index. The point estimate on the liquidation value variable is unchanged. Column 4 nonparametrically absorbs time-varying shocks at the ZIP code level using ZIP code by year-quarter fixed effects. The point estimate on the liquidation value variable is again unchanged relative to Column 2. In results available on request, I have also allowed the price elasticity to vary depending on the distance between the arm’s length and foreclosed properties. For properties located in the nearest distance quartile—less than 70 m from the foreclosed property—the elasticity is about 34% higher relative to those properties in the furthest quartile—between 267 and 1,000 m. Because economic conditions are unlikely to change significantly between 70 and 267 m, these differences in elasticities likely reflect comparables pricing and disamenties from foreclosures rather than latent economic condition. Nevertheless, to be more convincing, I build on the previous evidence linking liquidity and solvency to liquidation values in order to estimate causally the impact of REO sales on nearby properties. Column 5 instruments the liquidation value of the REO asset with the deposit flows and tier 1 equity ratio of the selling bank, observed in the previous quarter. The 2SLS estimate in Column 5 is less precisely estimated than its ordinary least squares (OLS) counterpart in Column 4, and is about 18% larger, suggesting again that the influence of omitted local shocks might indeed be limited when using ZIP code by year-quarter fixed effects and controlling for the hedonic characteristics of the property. The 2SLS estimate can be biased if local economic shocks that affect liquidation and home values are also correlated with these balance sheet variables. For example, the balance sheet of banks that earn most of their revenues in a local area could be affected by the same unobserved shocks that also drive liquidation values and transactions prices. Put differently, in ZIP codes where liquidating banks had bigger market shares, unobserved shocks to the balance sheet of these banks could also affect the prices of nonbank transactions. Column 6 uses ZIP code-by-bank fixed effects to nonparametrically absorb the preexisting variation in individual bank market shares across ZIP codes. This specification also uses ZIP code-level monthly house price changes to proxy for time-varying local housing market conditions. The 2SLS estimate is somewhat larger than that in Column 5. Taken together then, suggestive evidence shows that balance sheet pressures might spill over more broadly onto the housing market. What would house prices have been if banks had operated with 2017 capitalization and liquidity outcomes during the sample period? Obviously, empirical studies focused on causal inference are ill-suited for these types of general equilibrium counterfactual experiments, but the “back-of-the-envelope” calculation from Table 4 suggests that liquidation values would have been about 2.3% higher under 2017 outcomes. Using the more conservative IV elasticity in Column 5 of Table 7, this suggests that the average sale price for arm’s length transactions in the sample would have been about 69 bps or about $\$$ 1,994 higher. 5. Conclusion This paper studies the relationship between bank balance sheets and the liquidation value of real estate collateral. It finds that liquidation values tend to decline and the speed of asset sales tend to increase when banks lose deposits or deplete their equity buffers. Bank illiquidity is also associated with an economically large price rebound. Evidence suggests that policies such as TARP and QE1 might have increased liquidation values, in part, by providing institutions with the balance sheet capacity to delay asset sales in down markets. Low liquidation values among bank-owned properties also spill over into non-bank-owned sales. All this suggests that the sharp and extended deflation in real estate prices common after crisis events reflect both the effects of household deleveraging as well as ongoing balance sheet adjustments at financial institutions. Empirical studies focused on causal inference cannot measure the general equilibrium effects of policy interventions. But the elasticities obtained in this paper could be useful for models that calibrate the effects of bank recapitalization and liquidity provisioning during times of distress. The elasticities also might be useful for models that study the efficacy of countercyclical capital regulations and enhanced liquidity standards. Also, the evidence in this paper, which echoes recent more general studies of the impact of government credit policy on housing markets, suggests that direct purchases of some of these liquidated properties by the government during times of distress might also contain price discounts and spillover effects.17 Future research can consider these questions in greater detail. Appendix I obtain data on liquidation values and residential real estate collateral from Ztrax (https://www.zillow.com/research/ztrax/). This database is collected by Zillow and includes transaction and assessment data for approximately 150 million parcels across 3,100 counties. Liquidation values are the price obtained by the bank, either through an auction sale or an arm’s length transaction. My sample runs from 2006 to 2015. Ztrax lists the name of selling bank. I use the text matching algorithm “matchit” in the Stata package to perform a fuzzy match between string bank names from Ztrax and the regulatory call report data (https://www.stata.com/meeting/switzerland16/slides/raffo-switzerland16.pdf). ZTRAX also identifies the delinquent properties that can then be matched to banks using “matchit.” ZTRAX can do this because most counties require the lien holder to “file a notice of delinquency” in order to begin proceedings to seize the collateral. ZTRAX records this date from the county records. The call report data are obtained from https://cdr.ffiec.gov/public/. I use the bank data from the “Consolidated Reports of Condition and Income for a Bank with Domestic Offices Only—FFIEC 041.” This excludes foreign operations that are included in FFIEC 031. For cases in which the selling bank is listed as the bank holding company, I use the ‘Consolidated Financial Statements for Holding Companies—FR Y-9C. The ratio of tier 1 capital to risk-weighted assets---rcon7206 The change in deposits scaled by assets: (rcon2200-rcon2200(four quarters prior))/rcon2170 The ratio of deposits to assets—rcon2200/rcon2170 The income to assets ratio—riad4340/rcon2170 The ratio of cash to assets—rcon0010/rcon2170 Total assets—rcon2170 Mortgage-backed securities--(rcong300+rcong304+rcong308+rcong312+rcong316+rcon320+ rcong303+rcong307+rcong311+rcong315+rcong319+rcong323) U.S. Treasury Holdings—rcon0211+rcon1287 Loans—rcon2122 RateWatch provides data on various deposit and interest rate products at the branch and bank-headquarters level for U.S. banking organizations. The data are available weekly at the branch-level and at the headquarters level. These data are averaged up to the quarterly level. The variable “acct_nbr_loc” is the RateWatch-designated account number for the branch offering the depository account. In specifications that do not distinguish between branch and headquarters (Table 4, Column 4), I link the nearest branch to the property and use that rate. If all rates are set at headquarters, then I use that rate. In specifications that use the headquarters rate (Table 4, Column 5), I use the rate from the “head branch” and restrict the sample to multimarket banks, that is, those active in all twelve states. The products used are the 6-month certification of deposit and the 48-month certification of deposit, all computed as changes relative to the previous quarter. Acknowledgments I thank my discussants Effi Benmelech, Rustom Irani, Justin Murfin, and Tommaso Oliveri; numerous seminar participants, including Harry DeAngelo, Stuart Gabriel, Larry Harris, and Gary Richardson; an anonymous referee; and the editor (Andrew Karolyi) for helpful comments. Daniel Chen, Jordan Lopez, and Justin Scott provided superb research assistance. Supplementary data can be found on The Review of Financial Studies Web site. Footnotes 1 Theoretical treatments of these ideas include Acharya, Shin, and Yorulmazer (2011), Allen and Gale (1994, 2005), Brunnermeier, and Sannikov (2014), Diamond and Rajan (2001, 2005), He and Krishnamurthy (2013), and Rochet and Vives (2004). See also the survey in Rajan and Ramcharan (2016). 2Annenberg and Kung (2014), Campbell, Giglio, and Pathak (2010), Gerardi et al. (2015), Immergluck and Smith (2006), Lin, Rosenblatt, and Yao (2009), and Mian, Sufi, and Trebbi (2015) study the spillover effects of foreclosures in the period immediately around the 2008–2009 crisis. Peek and Rosengren (2000) is a classic study of the real effects of the Japanese banking crisis. Separately, Chaney, Sraer, and Thesmar (2012) provide evidence of the importance of commercial real estate collateral for firm decisions. Also, Chu (2016) examines the impact of bank balance sheets on commercial real estate transactions. 3 Motivated by these macro-prudential arguments about the potential negative effects of balance sheet illiquidity, U.S. and international banking rules now regulate liquidity among large banks (BCBS 2013; Cecchetti and Kashyap (2018)). Also, capital requirements have increased, and countercyclical buffers are now in place to avoid procyclical asset liquidations. 4 Beginning with Pulvino (1998), a sizeable literature documents real fire sales among nonfinancial corporations. A recent example is Benmelech and Bergman (2008), who link the balance sheet of airlines to the value of collateral. An important literature beginning with Adrian and Shin (2010) provides time-series evidence linking financial institution’s balance sheets to financial asset prices. Adrian, Etula, and Muir (2014) and He, Kelly, and Manela (2016) provide more direct tests using standard asset price models for financial assets. Determining causality and identifying underlying mechanisms within the context of standard time-series asset pricing models is difficult, however. 5 Beyond real estate, Benmelech, Meisenzahl, and Ramcharan (2017) provide evidence that illiquidity among nondepository institutions can affect consumer durable goods credit. Irani and Meisenzahl (2017) study banks’ incentives to make syndicated loan sales, whereas Acharya and Mora (2015) provide evidence on liquidity stress in the traditional banking system during the 2007–2010 period. Also, Rajan and Ramcharan (2016) provide evidence linking banking sector distress during the Great Depression to real local asset values. 6 New liquidity regulations proposed in the sample period could also augment REO selling pressures. In 2010, regulators for the first time proposed formal liquidity regulations, such as the liquidity coverage ratio (LCR) and the net stable funding ratio (NSFR) (Basel 2013, 2014). In the case of the former, REO assets are not counted as high-quality liquid assets, making it more difficult for a bank to comply if REOs dominate its balance sheet. In the case of the NSFR, the weight on REO assets is 100%, whereas that of the equivalent performing mortgage is 65%. Thus, once real estate collateral comes on balance sheet, it absorbs greater liquidity relative to the loan, as the bank would have to seek even more stable sources of liquidity to fund the real estate asset. If liquidity is scarce, then the incentive for rapid asset sales increase. 7 See the discussion in Kashyap, Rajan, and Stein (2008) and analytical treatments of these ideas in Brunnermier and Pedersen (2008) and He and Krishnamurthy (2013). 8 An overview of the Basel 1 risk-weighting rules can be found here: https://www.occ.gov/publications/publications-by-type/comptrollers-handbook/index-comptrollers-handbook.html. 9 Once an REO comes on balance sheet, losses—the gap between the cost basis and the outstanding loan amount—are immediately booked and charged against equity. Any subsequent declines in the fair market value of the asset relative to the initial cost basis are also charged to loan loss allowances. Thus, holding onto the asset in a declining market ties up both equity and leads to further charges. For a bank with scarce equity then, selling can be optimal. Banking regulations also limit the REO holding period to 5 years in most cases, so banks must eventually dispose of the asset. And for much of the sample period, banking regulators discouraged banks from property management on an ongoing basis, encouraging them to dispose of REO assets quickly. The OCC’s regulations describe this process in greater detail: https://www.occ.treas.gov/publications/publications-by-type/comptrollers-handbook/other-real-estate-owned/pub-ch-oreo.pdf. 10 As banking sector losses began to increase rapidly in 2008, many in the U.S. Congress opposed federal attempts to assist the banking system, and uncertainty about the health and future form of the U.S. banking system was significant. For example, the TARP, which was initially intended to purchase bad assets, was originally rejected by the House of Representatives in September of 2008. Similarly, the FDIC’s deposit insurance fund fell from $\$$52.4 billion in late 2007 (a reserve ratio to deposits of 1.2%) to around $\$$13 billion in early 2009 (a reserve ratio of just 0.27%), leading to a number of emergency measures and eroding the perception of the government guaranty. See the discussion in Bair (2013). 11 A possible concern is that unobserved factors might both determine selection into asset sales by auction, where prices are about 3% lower and also balance sheet observables in the main pricing equation. In results available on request, these balance sheet observables do not predict whether the sale occurs through an auction. Also, because I observe liquidation values, both at auction and through other types of sales, I include an indicator variable in the main specification for selection into auction. This variable equals 1 when the price is obtained at auction and 0 otherwise. The effects of balance sheet observables on liquidation values do not change, suggesting that “selection into auction” is unrelated to the balance sheet outcomes of interest. 12 An overview of TARP can be found here: https://www.treasury.gov/initiatives/financial-stability/TARP-Programs/bank-investment-programs/cap/Pages/default.aspx. An overview of QE is here: https://research.stlouisfed.org/pageone-economics/uploads/newsletter/2011/201104.pdf. 13 In results available on request, I have also evaluated the effects of TAF, interacting the deposit change variable with an indicator for whether a bank uses TAF in the current quarter; the specification also includes a lag of the indicator variable linearly and interacted with the deposit flows variable. The results are insignificant. For an overview of TAF, please see (Armantier, Ghysels, Sarkar, & Shrader 2015). 14 In evaluating the effects of TARP, a sizeable literature has pursued a diverse range of empirical strategies. See, for example, Duchin and Sosyura (2014). 15 In the case of bank lending, the evidence is mixed. Rodnyanski and Darmouni (2016) find that QE1 and QE3—MBS purchases—led to more lending to firms. Using a similar methodology, Chakraborty, Goldstein, and MacKinlay (2016) find that banks more exposed to QE1 and QE2 through their MBS holdings increased mortgage lending, but decreased lending to firms. 16 The five states are Arizona, California, Colorado, Florida, Nevada, and Washington. 17 See Fieldhouse and Mertens (2018) and Fieldhouse, Mertens, and Ravn (2018) for detailed evidence on the government’s historic role in the U.S. mortgage market. References Acharya, V. , Shin H. , and Yorulmazer T. . 2011 . Crisis resolution and bank liquidity. Review of Financial Studies 24 : 2166 – 205 . Google Scholar Crossref Search ADS WorldCat Acharya, V. , and Mora N. . 2015 . A crisis of banks as liquidity providers. Journal of Finance 70 : 1 – 43 . Google Scholar Crossref Search ADS WorldCat Adrian, T. , and Shin, H. 2010 . Liquidity and Leverage. Journal of Financial Intermediation 19 : 418 – 437 . Google Scholar Crossref Search ADS WorldCat Adrian, T. , Etula, E. and Muir. T. 2014 . Financial Intermediaries and the Cross-Section of Asset Returns. Journal of Finance 69 : 2557 – 2596 . Google Scholar Crossref Search ADS WorldCat Allen, F. , and Gale D. . 1994 . Limited Market Participation and Volatility of Asset Prices. American Economic Review 84 : 933 – 55 . OpenURL Placeholder Text WorldCat Allen, F. , and Gale D. . 2005 . From Cash-In-The-Market Pricing to Financial Fragility. Journal of the European Economic Association 3 : 535 – 46 . Google Scholar Crossref Search ADS WorldCat Annenberg, E. , and Kung E. . 2014 . Estimates of the size and source of price declines due to nearby foreclosures . American Economic Review 104 : 2527 – 51 . Google Scholar Crossref Search ADS WorldCat Armantier, O. , Ghysels, E. , Sarkar, A. , & Shrader, J. ( 2015 ). Discount window stigma during the 2007-2008 financial crisis . Journal of Financial Economics , 118 ( 2 ), 317 – 335 . Retrieved from <Go to ISI>://WOS:000364604600006. doi: 10.1016/j.jfineco.2015.08.006 Google Scholar Crossref Search ADS WorldCat Crossref Benmelech, E. , Garmaise M. , and Moskowitz T. . 2005 . Do liquidation values affect financial contracts? Evidence from commercial loan contracts and zoning regulation. Quarterly Journal of Economics 120 : 1121 – 54 . OpenURL Placeholder Text WorldCat Benmelech, E. , and Bergman N. . 2008 . Liquidation Values and the Credibility of Financial Contract Renegotiation: Evidence from U.S. Airlines. Quarterly Journal of Economics 123 : 1635 – 1677 . Google Scholar Crossref Search ADS WorldCat Benmelech, E. , Meisenzahl R. , and Ramcharan R. . 2017 . The Real Effects of Liquidity During the Financial Crisis: Evidence from Automobiles . The Quarterly Journal of Economics 132 . OpenURL Placeholder Text WorldCat Benmelech, E. , and Bergman N. . 2011 . Bankruptcy and the collateral channel. Journal of Finance 66 : 337 – 78 . Google Scholar Crossref Search ADS WorldCat Bernanke, B. 2012 . The U.S. housing market: Current conditions and policy considerations . Report, Federal Reserve, Washington, DC . https://www.federalreserve.gov/publications/other-reports/files/housing-white-paper-20120104.pdf. Bernanke, B. 2010 . Monetary policy and the housing bubble. Speech at the Annual Meeting of the American Economic Association, Atlanta, Georgia , January 3 . https://www.federalreserve.gov/newsevents/speech/bernanke20100103a.htm. OpenURL Placeholder Text WorldCat BCBS . 2013 . Basel III: The Liquidity Coverage Ratio and Liquidity Risk Monitoring Tools . https://www.bis.org/publ/bcbs238.htm. Brunnermeier, M. , and Pedersen L. . 2009 . Market liquidity and funding liquidity. Review of Financial Studies 22 : 2201 – 38 . Google Scholar Crossref Search ADS WorldCat Brunnermeier, M. , and Sannikov Y. . 2014 . A macroeconomic model with a financial sector. American Economic Review 104 : 379 – 421 . Google Scholar Crossref Search ADS WorldCat Caballero, R. , and Krishnamurthy A. . 2008 . Collective risk management in a flight to quality episode. Journal of Finance 63 : 2195 – 230 . Google Scholar Crossref Search ADS WorldCat Campbell, J. , Giglio S. , and Pathak P. . 2010 . Forced sales and house prices. American Economic Review 101 : 2108 – 31 . Google Scholar Crossref Search ADS WorldCat Cecchetti, S. and Kashyap A. 2018 . What Binds? Interactions between Bank Capital and Liquidty Regulations . In The Changing Fortunes of Central Banking , eds. Hartmann P. Huang H. and Schoenmaker D. 192 – 202 . Cambridge : Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Chakraborty, I. , Goldstein I. , and MacKinlay A. . 2016 . Monetary stimulus and bank lending . Working Paper, University of Miami . Chaney, T. , Sraer D. , and Thesmar D. . 2012 . The collateral channel: How real estate shocks affect corporate investment. American Economic Review 102 : 2381 – 409 . Google Scholar Crossref Search ADS WorldCat Chari, V. , and Jaganathan R. . 1988 . Banking panics, information and rational expectations equilibrium. Journal of Finance 3 : 749 – 61 . Google Scholar Crossref Search ADS WorldCat Chu, Y. 2016 . Asset fire sales by banks: Evidence from commercial REO sales. Review of Corporate Finance Studies 5 : 76 – 101 . OpenURL Placeholder Text WorldCat Diamond, D. , and Kashyap A. . 2015 . Liquidity Requirements, Liquidity Choice and Financial Stability . Handbook of Macroeconomics 2 : 2263 – 303 . OpenURL Placeholder Text WorldCat Diamond, D. , and Dybvig P. . 1983 . Bank Runs, Deposit Insurance and Liquidity . Journal of Political Economy 91 : 401 – 19 . Google Scholar Crossref Search ADS WorldCat Diamond, D. , and Rajan R. . 2001 . Liquidity Risk, Liquidity Creation, and Financial Fragility: A Theory of Banking . Journal of Political Economy 109 : 287 – 327 . Google Scholar Crossref Search ADS WorldCat Diamond, D. , and Rajan R. . 2005 . Liquidity shortages and banking crises. Journal of Finance 60 : 615 – 47 . Google Scholar Crossref Search ADS WorldCat Duchin, R. , and Sosyura D. . 2014 . Safer ratios, riskier portfolios: Banks’ response to government aid. Journal of Financial Economics 113 : 1 – 28 . Google Scholar Crossref Search ADS WorldCat Fieldhouse, A.J. , and Mertens K. , 2018 . A Narrative Analysis of Mortgage Asset Purchases by Federal Agencies . NBER Working Paper No. 23165 . OpenURL Placeholder Text WorldCat Fieldhouse, A.J. , Mertens K. and Ravn M.O. The macroeconomic effects of government asset purchases: evidence from postwar US housing credit policy. The Quarterly Journal of Economics 133 : 1503 – 1560 . Crossref Search ADS WorldCat Foley-Fisher, N. , Ramcharan R. , and Yu E. . 2016 . The impact of unconventional monetary policy on firm financing constraints: Evidence from the maturity extension program. Journal of Financial Economics 122 : 409 – 29 . Google Scholar Crossref Search ADS WorldCat Favara, G. , and Imbs J. . 2015 . Credit supply and the price of housing . American Economic Review 105 : 958 – 92 . Google Scholar Crossref Search ADS WorldCat Gatev, E. , and Strahan P. . 2006 . Banks’ advantage in hedging liquidity risk: Theory and evidence from the commercial paper market. Journal of Finance 61 : 867 – 92 . Google Scholar Crossref Search ADS WorldCat Geanakopolos, J. 2010 . The leverage cycle. In NBER macroeconomics annual 2009 , vol. 24 , eds. Acemoglu D. Rogoff K. and Woodford M. 1 – 65 . Chicago, IL : University of Chicago Press . Gerardi, K. , Rosenblatt E. , Willen P. , and Yao V. . 2015 . Foreclosure externalities: New evidence. Journal of Urban Economics 87 : 42 – 56 . Google Scholar Crossref Search ADS WorldCat Gertler, M. , and Kiyotaki N. . 2010 . Financial intermediation and credit policy in business cycle analysis. Handbook of Monetary Economics 3 : 547 – 99 . Google Scholar Crossref Search ADS WorldCat Granja, J. , Matvos G. , and Seru A. . 2017 . Selling failed banks. Journal of Finance 72 : 1723 – 84 . Google Scholar Crossref Search ADS WorldCat He, Z. , and Krishnamurthy A. . 2013 . Intermediary asset pricing. American Economic Review 103 : 732 – 70 . Google Scholar Crossref Search ADS WorldCat He, Z. , Kelly B. , and Manela A. . 2016 . Intermediary asset pricing: New evidence from many asset classes. Journal of Financial Economics 126 : 1 – 35 . Google Scholar Crossref Search ADS WorldCat Immergluck, D. , and Smith D. . 2006 . The Impact of Single-Family Foreclosures on Neighborhood Crime. Housing Studies 21 : 851 – 66 . Google Scholar Crossref Search ADS WorldCat Irani, R. , and Meisenzahl R. . 2017 . Loan sales and bank liquidity management: Evidence from a U.S. credit register. Review of Financial Studies 30 : 3455 – 501 . Google Scholar Crossref Search ADS WorldCat Kashyap, A.K. , Rajan R.G. , and Stein J.C. . 2008 . Rethinking Capital Regulation. 2008 Economic Symposium, Maintaining Stability in a Changing Financial System . Federal Reserve Bank of Kansas City . Lin, Z. , Rosenblatt E. , and Yao V. . 2009 . Spillover effects of foreclosures on neighborhood property values. Journal of Real Estate Finance and Economics 38 : 387 – 407 . Google Scholar Crossref Search ADS WorldCat Mian, A. , Sufi A. , and Trebbi F. . 2015 . Foreclosures, house prices, and the real economy. Journal of Finance 70 : 2587 – 634 . Google Scholar Crossref Search ADS WorldCat Murfin, J. , and Pratt R. . 2016 . Comparables pricing. Review of Financial Studies 32 : 688 – 737 . Google Scholar Crossref Search ADS WorldCat Office of the Comptroller of the Currency . 2010 . Examination handbook 120 appendix B, Risk-based capital: Risk-weighted assets . Report, Office of the Comptroller of the Currency, Washington, DC . https://www.occ.gov/static/news-issuances/ots/exam-handbook/ots-exam-handbook-120ab.pdf. Office of the Comptroller of the Currency . 2013 . Other real estate owned. Report, Office of the Comptroller of the Currency , Washington, DC . https://www.occ.treas.gov/publications/publications-by-type/comptrollers-handbook/other-real-estate-owned/pub-ch-oreo.pdf. Peek, J. , and Rosengren E. . 2000 . Collateral damage: Effects of the Japanese bank crisis on real activity in the United States. American Economic Review 90 : 30 – 45 . Google Scholar Crossref Search ADS WorldCat Pulvino, T. 1998 . Do fire-sales exist? An empirical study of commercial aircraft transactions. Journal of Finance 53 : 939 – 78 . Google Scholar Crossref Search ADS WorldCat Rajan, R. , and Ramcharan R. . 2015 . The anatomy of a credit crisis: The boom and bust in farmland prices in the United States in the 1920s. American Economic Review 105 : 1439 – 77 . Google Scholar Crossref Search ADS WorldCat Rajan, R. , and Ramcharan R. . 2016 . Local financial capacity and asset values: Evidence from bank failures. Journal of Financial Economics 120 : 229 – 51 . Google Scholar Crossref Search ADS WorldCat Rochet, J. , and Vives X. . 2004 . Coordination failures and the lender of last resort: Was Bagehot right after all? Journal of the European Economic Association 2 : 1116 – 47 . Google Scholar Crossref Search ADS WorldCat Rodnyanski, A. , and Darmouni O. . 2016 . The effects of quantitative easing on bank lending behavior. Review of Financial Studies 30 : 3858 – 87 . Google Scholar Crossref Search ADS WorldCat Shleifer, A. and Vishny R. 1992 . Liquidation Values and Debt Capacity: A Market Equilibrium Approach. Journal of Finance 47 : 1343 – 66 . Google Scholar Crossref Search ADS WorldCat Shleifer, A. and Vishny R. . 1997 . The Limits to Arbitrage. Journal of Finance 52 : 35 – 55 . Google Scholar Crossref Search ADS WorldCat Shleifer, A. and Vishny R. . 2011 . Fire sales in finance and macroeconomics. Journal of Economic Perspectives 25 : 29 – 48 . Google Scholar Crossref Search ADS WorldCat © The Author(s) 2019. 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)
TI - Banks’ Balance Sheets and Liquidation Values: Evidence from Real Estate Collateral
JF - The Review of Financial Studies
DO - 10.1093/rfs/hhz056
DA - 2020-02-01
UR - https://www.deepdyve.com/lp/oxford-university-press/banks-balance-sheets-and-liquidation-values-evidence-from-real-estate-HYoYuVcdiA
SP - 504
VL - 33
IS - 2
DP - DeepDyve
ER -