To See Is to Know: Simultaneous Display of Market Data for Retail Investors

To See Is to Know: Simultaneous Display of Market Data for Retail Investors Abstract I test whether the display format of market data affects the trading performance and behavior of retail investors. To do so, I exploit a large brokerage dataset covering a period during which the market information provided to the broker’s customers changed in format, but not in content. I find that a simultaneous display of cross-stock market data reduces the cognitive cost of monitoring the market and thus helps investors obtain better execution prices. In particular, investors better mitigate non-execution and adverse-selection risks when trading with limit orders. Hence, the display format of market data matters for the individual investor. 1. Introduction Identifying the determinants of households’ financial decisions is a fundamental question in the field of household finance (Campbell, 2006; Calvet, Campbell, and Sodini, 2007) and beyond the influence of a few key drivers such as IQ (Grinblatt, Keloharju, and Linnainmaa, 2012; Bhattacharya et al., 2012), financial sophistication (Calvet, Campbell, and Sodini, 2007), and financial wealth (Calvet and Sodini, 2014), the financial decision making of households is still not well understood. In particular, understanding the behavior of active retail investors and explaining the large cross-sectional distribution in individual trading performance are questions subject to an ongoing debate in the field. In this paper, I test whether a more efficient format of market data provided to active retail investors affects their trading behavior and performance. This is a question of interest because evidence from the laboartory suggests the display format of financial information may have first-order effects on trading behavior, perhaps even in a real-life setting. In addition, regulators and policymakers across the world increasingly care not only about how much information to provide to households, but also about which display format to use. In principle, a more efficient display should lead to better trading decisions and higher performance. However, a better display may also stimulate behavioral biases such as overconfidence and tempt active investors to increase speculative trading and incur more losses, thus leading to a detrimental effect of the news display. To explore this tradeoff, I exploit a unique setting in which the quantity of market information provided to investors remained fixed, while the display format of that information exogenously changed at some point in time. This setting is the June 2003 introduction of a trade order management software (hereafter, Trader+) by a large brokerage house. The software displayed market data in a more efficient way because it simultaneously gathered all relevant information items (market data, centralized limit order book, and investors’ orders) into a user-customized screen. Such simultaneous presentation of information items allows investors to understand how different stock price movements are related to one another and helps investors better assimilate that information when monitoring their orders. For instance, Hodge, Kennedy, and Maines (2004), who study a search-facilitating technology introduced by the SEC in 2003 (and adopted in 2009), argue that the “simultaneous presentation of related information directs users’ attention toward examining relations among the information items (Russo [1977]). Simultaneous presentation also reduces the cognitive costs of integrating the information”. Importantly, the software left the quantity of the data being processed unchanged. Indeed, the same information was available to the brokerage customers, in a more dispersed form, on other trading channels (such as using the brokerage website to submit an order).1 The key mechanism at play in this paper is that Trader+ decreased the cognitive cost of monitoring limit orders. Recall that any trade necessarily involves the choice between market and limit orders.2 If investors use limit orders, then they must monitor and revise/cancel them, after submission, to mitigate adverse selection and non-execution risks. That is, investors should monitor the market and their limit orders to improve their trading performance on those orders. If a display format of market data makes such market monitoring more efficient then, everything else being constant, it should also improve investors’ trade performance on their limit orders. To test these implications, I use a difference-in-differences (DID) identification strategy in which treated investors, who switch to the new software, are matched (with a propensity-score algorithm) to similar control investors who do not use the software. I find that the display format of market data does matter for the individual investor. As in Linnainmaa (2010) and Grinblatt, Keloharju, and Linnainmaa (2012), for each buy and sell limit order on a given stock in my sample, I compute the signed return from the execution price to the closing price of that stock on the same day. Indeed, this return has been used in the literature as a proxy for the risk of adverse selection faced by investors when trading with limit orders. It is therefore a performance measure of one order’s execution quality and should capture “the active management of individuals.”3 Intuitively, the higher the risk of adverse selection, the lower this return. My DID estimates show that following the introduction of Trader+, investors’ intraday returns on their limit orders jump by 8 basis points. The economic magnitude of this result is large in light of previous studies on the trading performance of investors. For instance, Grinblatt, Keloharju, and Linnainmaa (2012) find that intraday returns of Finnish investors with a high IQ (in the top 5% of the distribution) outperform those with a low IQ by 11 basis points. Similarly, Kuo, Lin, and Zhao (2014) find that investors with high cognitive abilities outperform the limit order intraday returns of those with low cognitive abilities by 3.9 basis points. These comparisons suggest that the display format of market data has an effect on trading performance similar to that of individual cognitive abilities. My explanation for this increase in trade performance is consistent with market microstructure theories of order choice, in which the “cognitive cost” of monitoring limit orders is reduced by the new display. In the paper, I show that the key feature of the new display is the simultaneous cross-stock limit order book. I provide additional evidence supporting this hypothesis. Because the new simultaneous display reduces the cost of monitoring, it increases the expected utility of trading with a limit order relative to the utility of trading with a market order. Accordingly, I find that treated investors trade more limit orders than before when using the software. Additionally, because my sample is populated by active traders, I expect treated investors to be able to spot and monitor short-term trading opportunities that were likely to be too difficult to monitor before. This change in behavior should lead to a decrease in individual trading horizons, which is confirmed by the data. Overall, trading activity of the treated group increases, relative to the control group, within 1 month of the software’s introduction. I also consider, and rule out, alternative explanations for my findings based on trading speed, overconfidence, and investors’ self-selection. This paper directly contributes to two different literatures. First, this paper is part of the few recent works that study how alternative presentation formats of financial information influence investors’ financial decisions. To the best of my knowledge, this is the first paper to show that the display format of market data affects the real trading choices and performance of individual investors.4 Second, this paper contributes to the recent literature that uses insight from market microstructure theory to understand individual investors’ decisions (Linnainmaa, 2010; Kelley and Tetlock, 2013; Barber et al., 2014). The paper also contributes to the market microstructure literature that focuses on order submissions strategies (see Foucault, 1999; Hollifield, Miller, and Sandas, 2004). In the vein of Foucault, Roell, and Sandas (2003) and Liu (2009), my economic mechanism relies on extensions of standard models of order choice that embed the cognitive costs of monitoring and managing limit orders. In particular, this paper is directly related to Fong and Liu (2010), which uses aggregated institutional and retail order flow data to establish a link between monitoring costs and limit order performance. My paper goes one step further by focusing on retail investors and by linking the display format of market data to retail trading performance. Hence, I can participate in the ongoing debate about what drives the large cross-section variation in retail trading skills that has been previously identified (Barber and Odean, 2013). My results imply that part of this unexplained variation in performance may be due to differences in the display format of the data being processed by retail investors. This finding is important because the drivers of retail performance previously documented in the literature (such as IQ, financial sophistication, or financial wealth) cannot be easily manipulated by policymakers, but data display format can be. This article proceeds as follows. Section 2 discusses my testable empirical hypothesis. Section 3 presents the brokerage dataset and Section 4 motivates and discusses my identification strategy. The results are given in Section 5. Before concluding, I discuss the potential alternative explanations for my results in Section 6. 2. Testable Hypothesis In this paper, I argue that the new display of market data that comes with Trader+ is a positive exogenous shock on the market monitoring capacities of individual investors. Essentially, the new display allows investors to process, more efficiently, the same quantity of market data as before. The simultaneous display effect operates via two different channels. First, before submitting any order, investors are more likely to better process the current market conditions. This improved market timing should benefit both market and limit orders so intraday returns should increase for both types of orders. Second, after submitting a limit order, a trader faces non-execution and adverse selection risks. Both of these risks can be mitigated, via Trader+, by active order management (also called monitoring activity).5 Hence, limit orders gains from both better market timing and better order management. This reasoning implies that Trader+ should, in turn, induce investors to use limit orders more often than they previously did before because it is more profitable to do so. Indeed, an investor submits a limit order if the expected utility of using a limit order is greater than the certain utility of using a market order. The investor sends a market order otherwise. This latter hypothesis has several testable implications. First, investors switching to Trader+ should be more likely to place limit orders on stocks with smaller bid–ask spreads. Indeed, when the cost of processing limit order book data is reduced, limit order traders are more likely to post smaller spreads to attract uninformed investors and cancel/revise their orders if market conditions change (Liu, 2009). Second, investors should be able to manage, better than before, situations where they must monitor multiple orders on multiple stocks at the same time. If this reasoning is correct, then we should observe an increase in the probability of submitting orders on more than one stock during the same day. Furthermore, short-term strategies in which investors trade very frequently (such as “day-trading,” where investors revert their positions at the daily level) are more difficult to monitor than long-term ones. If this is correct, a more efficient monitoring activity should allow active investors to spot more short-term trading opportunities than before. This effect implies that the proportion of short-term trading strategies should increase (leading to higher trading activity) when investors monitor their limit orders more efficiently.6 In summary, I argue that the new simultaneous display of market data should decrease the cost of monitoring the market and the cost of monitoring pending limit orders. This mechanism implies the following testable implications: A more efficient monitoring activity increases the trading performance of limit orders, because investors gain from reduced adverse selection, reduced non-execution risks, and better market timing. A more efficient monitoring activity has a positive effect on the performance of market orders, as market orders benefit from better market timing. A more efficient monitoring activity increases the probability of submitting a limit order, as limit orders become more profitable than market orders. A more efficient monitoring activity increases the probability of trading multiple stocks at the same time. A more efficient monitoring activity increases trading volume and reduces trading horizon, because investors exploit short-term strategies that were too costly to execute and monitor before. 3. Data In this section, I describe the data and define the variables that I will use in order to test the empirical implications described in the previous section. 3.1. The Brokerage Dataset The data used in this paper comes from a leading French online broker.7 The raw dataset contains at the daily level all of the executed trades sent by each of the 145,801 customers of the broker from 1999 up to 2010, which represents >15 million trades. In this paper, I focus on active liquidity-providing retail investors, which represent a small part of the entire retail population but generate most of the orders in my database.8 Each trade comes with the following information: the asset type (equity, bonds, etc.), the trading exchange identifier (the ISIN), the trading date, the quantity in the number of stocks, the total amount traded in euro, the order type (limit order, market order, and other minor orders types), the trading place, the trading channel used to submit the order, and a margin trade indicator. Most of the trades are executed on the NYSE Euronext Paris trading exchange. Depending on the period, the data also include the fees paid at the single trade level.9 Basic demographic information includes date of birth, gender, French department of residence, and opening and closing date of any brokerage account in the data. Last, portfolio holdings are available at the monthly level. I match the trades in my dataset with market data provided by Eurofidai, the European financial data institute. Trades are matched by ISIN code, trading day, and trading exchange code. Trades for which no information is available from Eurofidai are discarded from the sample. Summary statistics of the raw brokerage dataset are provided in Table I. This table shows the corresponding number of trades, percentage, and cumulative percentage for several categorical variables that describe the nature of my data well. Panel A shows for instance that trades are in the majority of cases limit orders (60%) and market orders (27%). The other minor orders types are rarely used by investors. The information on order type is completely missing for 1999 and 2000 and may be sometimes missing up to 2004. Panel B of the table shows that most of individual trading activity consists of buying or selling common stocks. This dataset, therefore, shows patterns of individual investment behavior that are similar to other recent databases used in the literature. For instance, Finnish investors in Linnainmaa (2010) also use limit orders for most of their trades on the Helsinki Stock Exchange. Panel C shows the trading channels available for retail investors to submit a trade. Investors could submit an order by using the telephone to speak to a broker official (Telephone), by calling a voicemail service and typing instruction using the telephone’s keys (Phone+), by using a web navigator to connect to the broker website (Web), by using an old French Videotex online service accessible through the telephone lines (Minitel), or by using online basic computer software (Online+). Trader+ is a trading software that was introduced in June 2003. It will be fully discussed and presented in Section 4. Panel C of the table highlights that a large majority of trades are submitted using the internet or trading software (Online+ and Trader+). Table I. Summary statistics of the brokerage dataset This table describes the main characteristics of the brokerage dataset used in this paper. The table gives the corresponding number of trades, percentage, and cumulative percentage for each category of information (investors’ order choices, investors’ use of trading channels, investors’ use of asset classes, and investors’ use of trading exchanges). The sample period is from 2002 to 2010. Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      Table I. Summary statistics of the brokerage dataset This table describes the main characteristics of the brokerage dataset used in this paper. The table gives the corresponding number of trades, percentage, and cumulative percentage for each category of information (investors’ order choices, investors’ use of trading channels, investors’ use of asset classes, and investors’ use of trading exchanges). The sample period is from 2002 to 2010. Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      3.2. Variable Definition I describe in this section how I measure investors’ trading performance (Section 3.2.a) and investor trading horizon (Section 3.2.b). 3.2.a. Measuring trading performance I measure trading performance as follows. For each trade in my sample, I compute the signed difference between the closing price of the stock bought (or sold) and the execution price of the order, divided by the execution price of the order:   Ri,t,k,s=signi,t,k,s*Closet,s−Pricei,t,k,sPricei,t,k,s, where Closet,s is the closing price on day t, of the traded stock s, and Pricei,t,k,s is the execution price of order number k submitted on day t, for stock s, by individual i. Signi,t,k,s is a dummy variable that equals 1 for a buy order and −1 for a sell order. This expost performance measure is well suited for assessing the performance of a limit order, because it captures the adverse selection risk faced by investors using those orders (see Harris and Hasbrouck, 1996). Indeed, Hollifield et al. (2006) define the risk of adverse selection as the expected loss (or gain) due to future expected changes in stock value given execution. Similarly, Liu (2009) obtain a proxy for this expectation by comparing the current price after the execution of a limit order to the price at which the order has been executed. A similar approach is also adopted in Grinblatt, Keloharju, and Linnainmaa (2012) and Linnainmaa (2010). For instance, Grinblatt, Keloharju, and Linnainmaa (2012) compare the performance between high and low IQ investors and argue that this return is essentially a measure of one order’s execution quality and should capture “the active management of individuals.” “High IQ investors,” they say, “may be better or quicker at processing information into a useful signal, or excel at distinguishing useful information from noise.” Because the researchers are able to detect statistically significant differences between the returns of the high IQ and low IQ groups, their hypothesis cannot be rejected. My approach follows the same reasoning. If a more efficient display of market data allows investors to monitor their limit orders more efficiently, then investors should improve upon the execution quality of their limit orders relative to investors using a less efficient data display. This improvement is captured by my intraday returns. Execution quality can also be assessed in other ways. In Appendix A.1, I use an alternative way to compute short-term returns which closely follows Linnainmaa (2011). Also, in Appendix A.2, I follow a standard procedure in the brokerage industry and compare the average execution price obtained on any order with the volume-weighted average price (VWAP) of the same stock the same day (see Hendershott, Jones, and Menkveld, 2011). To test these hypothesis, I need an improvement in the display of market data for retail investors. This point will be discussed in Section 4. 3.2.b. Measuring long-term returns and investors’ trading horizon To proxy for investors’ trading horizon and have a long-term measure of trading performance, I adopt a general methodology used in Schlarbaum, Lewellen, and Lease (1978); Puckett and Yan (2011); and Chakrabarty, Moulton, and Trzcinka (2017). The main idea is to take full advantage of the granularity of my data and to aggregate investor i’s single trades into round-trip trades. To do so, I first compute the daily net quantity traded by investor i, on stock s and day t as: Qi,t,s=Qtiti,t,sbuy−Qtiti,t,ssell, where Qtiti,t,sbuy (respectively, Qtiti,t,ssell) represents the actual quantity of stock s bought (respectively, sold) by investor i on day t. Then, I sort all the daily net quantities in my dataset by trader, stock, and trading day, and I keep track of the cumulative stock quantity held by a trader day after day. A round-trip starts and ends with a zero net cumulative quantity. Round-trips are thus trading positions that are fully unwound: stocks previously bought are entirely sold and stocks previously sold short are completely bought back.10 This methodology is appealing in the context of my paper because it requires very few assumptions on the data (Schlarbaum, Lewellen, and Lease, 1978). To see whether trading horizon decreases after the introduction of Trader+, I study how Trader+ affects the proportion of short-term round-trip trades that are started in a given day by a given trader. Also, this aggregation of individual trades into round-trips allows me to identify when an investor initiates or closes a position on a given stock. I will use this information in Section 5.1. The gross return on a round-trip is computed as R=EuroSell−EuroBuyEuroBuy, where EuroBuy (EuroSell) is the total euro cash-flow paid when buying stocks (received when selling stocks) during a particular round-trip executed by any investor on a given stock. To get round-trip returns net of commissions, trading fees for the buy-leg are added to EuroBuy and trading fees for the sell-leg are subtracted to EuroSell. Last, risk-adjusted returns are obtained by subtracting to the round-trip return the return that would have been obtained, passively, by investing in the French market index (CAC40) over the same holding period. Finally, following Shapira and Venezia (2001), I convert these gross and net returns into daily returns using the formula: Rdaily=(1+R)1T−1, where T, the duration of a round-trip, is computed as the number of days between the opening date and the closing date of a round-trip. 4. Methodology and Identification Strategy 4.1. Trader+ I explain in this section why the display of market data in Trader+ should help investors more efficiently monitor their limit orders. The new information display of market data, available on Trader+, allowed Trader+ users to monitor their limit orders more efficiently than before. Indeed, prior to June 2003, the broker customers submitted their orders mainly through two trading channels: using the existing trading software available (Online+) and connecting to the brokerage web interface (Web). Figure 1 provides a screenshot of each of these trading channels and highlights how identical information items (such as the current state of the limit order book or the recent market movements statistics) are displayed differently on those trading channels. In both cases, one can see that the market data are dispersed through different webpages (for Web) or through different tabs (for Online+). Figure 1. View largeDownload slide Main trading channels used by the broker customers before Trader+. This figure shows two screenshots of the main trading channels used by investors in my sample, before the introduction of Trader+. (a) Investors used to send their orders using Online+, a basic trading software that has been available since 1999, or (b) using the Web brokerage interface (Web). The same quantity of market information is provided on each trading channel. Figure 1. View largeDownload slide Main trading channels used by the broker customers before Trader+. This figure shows two screenshots of the main trading channels used by investors in my sample, before the introduction of Trader+. (a) Investors used to send their orders using Online+, a basic trading software that has been available since 1999, or (b) using the Web brokerage interface (Web). The same quantity of market information is provided on each trading channel. Instead, Trader+ displayed market data much more efficiently than those trading channels because it simultaneously gathered all relevant information items into a user-customized screen. A screenshot of Trader+ is shown in Figure 2. As one can see, investors using Trader+ could see on their computer screen, not only the limit order book or the most recent market statistics, but also their pending orders, the stock intraday graphics, and market data at the stock level.11 The key point is that these information items were also available, in a more dispersed form, on the other trading channels. In other words, the quantity of information remains the same, while the display of that information varies when Trader+ is introduced. Figure 2. View largeDownload slide Screenshot of Trader+. This is a screenshot of the main Trader+ interface in June 2003. Users can follow the status of all their submitted orders, along with all the characteristics of these orders (security id, buy/sell indicator quantity, price, execution/cancellation). There is an aggregated limit order book that simultaneously shows the best bid and ask prices and quantites for several stocks (ACCOR, AGF, AIR LIQUIDE, etc.). Users can also get news related to multiple securities at the same time. Additional market data are provided on the top left of the screen and a complete order submission form is shown on the bottom left of the screen. Figure 2. View largeDownload slide Screenshot of Trader+. This is a screenshot of the main Trader+ interface in June 2003. Users can follow the status of all their submitted orders, along with all the characteristics of these orders (security id, buy/sell indicator quantity, price, execution/cancellation). There is an aggregated limit order book that simultaneously shows the best bid and ask prices and quantites for several stocks (ACCOR, AGF, AIR LIQUIDE, etc.). Users can also get news related to multiple securities at the same time. Additional market data are provided on the top left of the screen and a complete order submission form is shown on the bottom left of the screen. The simultaneous presentation of cross-stock limit order book data is the critical feature that allows investors to better understand how stock prices and market movements are related to one another and helps investors better assimilate that information when they monitor their limit orders. I therefore use the introduction of Trader+ as a positive shock to the investors’ limited monitoring capacities. This assertion is consistent with Hodge, Kennedy, and Maines (2004), who consider a search-facilitating technology [the eXtensible Business Reporting Language (XBRL) technology] introduced by the SEC in 2003 and state that “This simultaneous presentation helps users to evaluate items in relation to each other and to integrate the related information when making decisions.” In summary, the display of information of Trader+ should allow investors to monitor more efficiently their limit orders, which gives me the opportunity to test the empirical implications derived in Section 2. 4.2. Identification Strategy To test my empirical predictions, I use in this paper a DID methodology. The key identifying assumption in this setup (see Angrist and Pischke, 2008) is that trends in outcomes for treated and control investors would have been the same in the absence of treatment. This fundamental, but untestable, identifying assumption can nonetheless be evaluated by comparing graphically the trends in outcomes for both groups in multiple periods before and after the treatment exposure. I use the 10-year range of my dataset to support this “parallel trend” assumption. DID estimates can be obtained using an OLS regression, which has the advantage of allowing for a correct specification of standard errors (see Bertrand, Duflo, and Mullainathan, 2004). I therefore run the following regression for outcomes at the investor-day-trade level12:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (1) Monitoringi,t is the variable of interest. It is a dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+. This specification thus allows for a staggered entry into treatment at the investor level and is more precise than a standard DID framework that imposes the same treatment period for every treated investor. The regression specification includes a full range of individual fixed-effects δi, stock-fixed effects γs, and day fixed-effects θt every day from January 2002 to January 2005. Standard errors are robust to heteroscedasticity and are clustered at the individual level (White, 1980; Rogers, 1994). The fixed effects δi capture the differences between investors that are fixed over time (possibly correlated with the treatment variable), whereas the day-fixed effects capture the time factors that are common to both treated and control investors within a given trading day. Stock-fixed effects control for the fact that investors could possibly trade different stocks before and after the introduction of Trader+. The DID estimate is the coefficient β. It shows the effect of a more effective monitoring activity (induced by the aggregated display of information of Trader+) on the outcome Y. The DID framework is a powerful tool for causal inference but my estimates may be biased if investors are allowed to enter or exit the market after (or before) the treatment period. To control for attrition, I require that both treated and control investors have a stock holding position at least between January 2002 and January 2005, at least a single trade between January and the end of May 2003, and at least one trade after June 2003. That is, I follow the same population of traders over time. Treated investors have submitted at least one trade with Trader+ after June 2003, while control investors have not (and never will in my sample). 4.3. The Control Group Investors who switch to Trader+ may be different, in many ways, from other investors who never use the software. To obtain a control group that provides a credible counterfactual, I use a nearest-neighbor propensity score algorithm in order to find the best control investor for each treated investor in my sample. The propensity-score algorithm matches treated and control investors who share the same probability of switching to Trader+ after June 2003. To avoid any look-ahead bias, the estimated probability is computed using only pre-treatment data. Intuitively, this approach relies on finding the right set of covariates that determines the switch to treatment. In this respect, the seminal paper on online investors by Barber and Odean (2002) is very useful. The authors find that young active male traders with high incomes are more likely to switch to online trading than other traders. Those who switch also report more trading experience and a particular preference for speculative trades. Prior to going online, moreover, investors experienced unusually strong performance. Choi, Laibson, and Metrick (2002) study in a different setting the impact of a web-based trading channel on two large corporate 401(k) plans. They obtain that traders that are used to phone trading are less likely to try the web. While these results may not be completely generalizable to this study, as investors in my sample are already online investors, these papers emphasize that trading behavior before the treatment period can help identify the variables key to the selection process. I therefore compute various covariates according to investor demographics, general trading behavior, account size, and trading channel preferences, and I estimate, with pre-treatment data only, a cross-section logit regression where the dependent variable is one for a treated investor and zero otherwise. All these covariates are obtained using individual trading data from January 2003 to the end of May 2003, but my results are robust to the choice of different pre-event windows for the matching procedure. These covariates are grouped in four classes: demographics, trading behavior, account size, and trading channels. I use two demographic variables: female is a dummy variable that is 1 for females and 0 for males, and age corresponds to the investor’s age in 2003. The variables that capture investor trading behavior, in the pre-treatment period, are the individual number of orders executed (number of executed trades), the individual number of different asset classes traded (number of asset types), the individual median euro amount traded (median amount traded), the individual median daily-return on executed orders (median daily return), the individual percentage of orders that are limit orders (% limit orders), and the individual percentage of orders submitted with a margin account (% margin). As a simple proxy for trading horizon, I also include the proportion of round-trip trades, initiated in May 2003, that are reversed within 2 days (% round trip). The account size variables provide information about the total euro value of an investor’s common stock holdings on its spot-market account (in ten of thousands euros) in March 2003. Finally, trading channels variables provide, for each trading channel, the corresponding individual percentages of trades submitted by an investor through that channel. Results of the logit regression are given in Table II. What determines the switch to the software appears clearly: active traders with high past performance have a larger probability of switching to treatment. Furthermore, investors who use online trading channels, such as the web or the old computer software that has been available since 1999 (variable “Online+,” highly significant), are also more likely to use Trader+. Table II. Estimation of the probability of using Trader+ at least once after June 2003 (using pre-treatment data only) This table gives the estimates from a cross-section logistic regression where the dependent variable is a dummy variable that equals 1 if an investor is considered as treated: he has submitted at least one trade using Trader+ during the period June 2003–December 2010. Investors in my sample are required to have an open common stock position at least between 2002 and 2005, they have submitted at least one trade before June 2003, and at least one trade after June 2003. I compute various covariates according to investor demographics, general trading behavior, account size, and trading channel preferences in the pre-treatment period.   Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444    Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444  Table II. Estimation of the probability of using Trader+ at least once after June 2003 (using pre-treatment data only) This table gives the estimates from a cross-section logistic regression where the dependent variable is a dummy variable that equals 1 if an investor is considered as treated: he has submitted at least one trade using Trader+ during the period June 2003–December 2010. Investors in my sample are required to have an open common stock position at least between 2002 and 2005, they have submitted at least one trade before June 2003, and at least one trade after June 2003. I compute various covariates according to investor demographics, general trading behavior, account size, and trading channel preferences in the pre-treatment period.   Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444    Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444  The matching algorithm, based on the above logit estimation, performs well as Table III shows. Table III compares the mean covariate values between treated and matched control investors in my sample. It appears that all the variables that determine the investor propensity to use Trader+ (according to Table II) are very similar between treated and matched control investors. Indeed, treated and matched controls have a similar pre-treatment performance, submit on average the same number of trades between January and May 2003 (35 versus 29), the average order amount is almost equal (3200€ versus 2900€), and they manage their orders using the existing online trading tools (online+: 54% versus 56%, web: 34% versus 33%). Table III. Comparisons between treated investors and control investors obtained via propensity matching This table shows the average covariates between treated and control investors before and after the propensity matching procedures. All computations are performed with trading data coming from the pre-treatment period. All variables are defined in Table II. Standard deviations are between parentheses.   Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748    Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748  Table III. Comparisons between treated investors and control investors obtained via propensity matching This table shows the average covariates between treated and control investors before and after the propensity matching procedures. All computations are performed with trading data coming from the pre-treatment period. All variables are defined in Table II. Standard deviations are between parentheses.   Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748    Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748  Last, in the Appendix, Figure A3 shows for both the treatment and the control group, the total number of orders (each year from 1999 to 2010) broken down by order type and trading channel. These figures summarize well the outcome of the matching procedure. For instance, one can see that from the figure that: (i) the number of orders submitted each year until 2003 is very similar between treated and control investors and (ii) both groups submit limit orders and use online tools most of the time. 5. Results 5.1. Trading Performance I test in this section my predictions regarding the effect of the new display on trading performance. Does a more effective display of market data (due to Trader+) allow investors to better monitor their limit orders and increase, in turn, their returns on those orders? Panel A in Table IV provides a first answer. For each trading channel, this table provides a statistical description of the distribution of investors’ intraday returns. I use all the orders submitted by both treated and control investors after June 2003. What emerges from the table is that returns obtained by traders using Trader+ are greater, in most quantiles, than returns achieved through other trading channels. Moreover, the table shows a progressive shift in returns from negative values toward positive ones as investors move from hard-to-read displays of market data (e.g., minitel) to more efficient ones (online trading channels). These facts are thus consistent with the idea that online tools display market data in a way that is “more processable” for investors (Russo, 1977). Table IV. Summary statistics on investors’ trading channels and investors’ order choice Panel A shows the main quantiles, the mean, and the standard deviation (SD) of investors’ intraday returns, tabulated by trading channel. Panel B is a two-way table that provides, for each trading channel, the corresponding row percentage (top), column percentage (middle), and frequency of orders of a given order type (bottom). The sample contains all the orders of both treated and control investors after June 2003. Panel A: Intraday returns versus trading channel     p10  p25  p50  Mean  p75  p90  SD  Trader+  −2.011  −0.923  −0.225  −0.256  0.446  1.461  1.657  Online+  −1.969  −0.962  −0.263  −0.304  0.391  1.302  1.580  Web  −2.200  −1.105  −0.361  −0.388  0.353  1.386  1.688  Minitel  −2.029  −1.088  −0.400  −0.428  0.259  1.130  1.488  Phone+  −2.088  −1.162  −0.471  −0.494  0.198  1.063  1.451  Telephone  −2.193  −1.173  −0.378  −0.425  0.322  1.313  1.645  Total  −2.058  −0.995  −0.275  −0.315  0.397  1.378  1.639    Panel A: Intraday returns versus trading channel     p10  p25  p50  Mean  p75  p90  SD  Trader+  −2.011  −0.923  −0.225  −0.256  0.446  1.461  1.657  Online+  −1.969  −0.962  −0.263  −0.304  0.391  1.302  1.580  Web  −2.200  −1.105  −0.361  −0.388  0.353  1.386  1.688  Minitel  −2.029  −1.088  −0.400  −0.428  0.259  1.130  1.488  Phone+  −2.088  −1.162  −0.471  −0.494  0.198  1.063  1.451  Telephone  −2.193  −1.173  −0.378  −0.425  0.322  1.313  1.645  Total  −2.058  −0.995  −0.275  −0.315  0.397  1.378  1.639    Panel B: Trading channel versus order choice     Limit order  Market order  Others  Missing  Total  Trader+  74.6  17.6  7.7  0.2  100.0  37.8  34.2  30.2  2.4  35.6  341,582  80,420  35,371  723  458,096  Online+  70.6  13.7  10.2  5.5  100.0  33.2  24.7  37.0  78.5  33.0  299,756  57,996  43,313  23,241  424,306  Web  66.7  23.4  9.3  0.6  100.0  27.4  37.1  29.6  7.0  28.9  247,632  87,113  34,667  2073  371,485  Minitel  54.9  6.7  20.0  18.4  100.0  0.7  0.3  2.0  7.3  0.9  6429  779  2342  2150  11,700  Phone+  43.5  33.4  11.4  11.7  100.0  0.5  1.4  1.0  4.0  0.8  4385  3370  1146  1178  10,079  Telephone  40.9  54.1  2.4  2.6  100.0  0.4  2.2  0.2  0.9  0.7  3907  5170  234  253  9564  Total  70.3  18.3  9.1  2.3  100.0  100.0  100.0  100.0  100.0  100.0  903,691  234,848  117,073  29,618  1,285,230    Panel B: Trading channel versus order choice     Limit order  Market order  Others  Missing  Total  Trader+  74.6  17.6  7.7  0.2  100.0  37.8  34.2  30.2  2.4  35.6  341,582  80,420  35,371  723  458,096  Online+  70.6  13.7  10.2  5.5  100.0  33.2  24.7  37.0  78.5  33.0  299,756  57,996  43,313  23,241  424,306  Web  66.7  23.4  9.3  0.6  100.0  27.4  37.1  29.6  7.0  28.9  247,632  87,113  34,667  2073  371,485  Minitel  54.9  6.7  20.0  18.4  100.0  0.7  0.3  2.0  7.3  0.9  6429  779  2342  2150  11,700  Phone+  43.5  33.4  11.4  11.7  100.0  0.5  1.4  1.0  4.0  0.8  4385  3370  1146  1178  10,079  Telephone  40.9  54.1  2.4  2.6  100.0  0.4  2.2  0.2  0.9  0.7  3907  5170  234  253  9564  Total  70.3  18.3  9.1  2.3  100.0  100.0  100.0  100.0  100.0  100.0  903,691  234,848  117,073  29,618  1,285,230    Table IV. Summary statistics on investors’ trading channels and investors’ order choice Panel A shows the main quantiles, the mean, and the standard deviation (SD) of investors’ intraday returns, tabulated by trading channel. Panel B is a two-way table that provides, for each trading channel, the corresponding row percentage (top), column percentage (middle), and frequency of orders of a given order type (bottom). The sample contains all the orders of both treated and control investors after June 2003. Panel A: Intraday returns versus trading channel     p10  p25  p50  Mean  p75  p90  SD  Trader+  −2.011  −0.923  −0.225  −0.256  0.446  1.461  1.657  Online+  −1.969  −0.962  −0.263  −0.304  0.391  1.302  1.580  Web  −2.200  −1.105  −0.361  −0.388  0.353  1.386  1.688  Minitel  −2.029  −1.088  −0.400  −0.428  0.259  1.130  1.488  Phone+  −2.088  −1.162  −0.471  −0.494  0.198  1.063  1.451  Telephone  −2.193  −1.173  −0.378  −0.425  0.322  1.313  1.645  Total  −2.058  −0.995  −0.275  −0.315  0.397  1.378  1.639    Panel A: Intraday returns versus trading channel     p10  p25  p50  Mean  p75  p90  SD  Trader+  −2.011  −0.923  −0.225  −0.256  0.446  1.461  1.657  Online+  −1.969  −0.962  −0.263  −0.304  0.391  1.302  1.580  Web  −2.200  −1.105  −0.361  −0.388  0.353  1.386  1.688  Minitel  −2.029  −1.088  −0.400  −0.428  0.259  1.130  1.488  Phone+  −2.088  −1.162  −0.471  −0.494  0.198  1.063  1.451  Telephone  −2.193  −1.173  −0.378  −0.425  0.322  1.313  1.645  Total  −2.058  −0.995  −0.275  −0.315  0.397  1.378  1.639    Panel B: Trading channel versus order choice     Limit order  Market order  Others  Missing  Total  Trader+  74.6  17.6  7.7  0.2  100.0  37.8  34.2  30.2  2.4  35.6  341,582  80,420  35,371  723  458,096  Online+  70.6  13.7  10.2  5.5  100.0  33.2  24.7  37.0  78.5  33.0  299,756  57,996  43,313  23,241  424,306  Web  66.7  23.4  9.3  0.6  100.0  27.4  37.1  29.6  7.0  28.9  247,632  87,113  34,667  2073  371,485  Minitel  54.9  6.7  20.0  18.4  100.0  0.7  0.3  2.0  7.3  0.9  6429  779  2342  2150  11,700  Phone+  43.5  33.4  11.4  11.7  100.0  0.5  1.4  1.0  4.0  0.8  4385  3370  1146  1178  10,079  Telephone  40.9  54.1  2.4  2.6  100.0  0.4  2.2  0.2  0.9  0.7  3907  5170  234  253  9564  Total  70.3  18.3  9.1  2.3  100.0  100.0  100.0  100.0  100.0  100.0  903,691  234,848  117,073  29,618  1,285,230    Panel B: Trading channel versus order choice     Limit order  Market order  Others  Missing  Total  Trader+  74.6  17.6  7.7  0.2  100.0  37.8  34.2  30.2  2.4  35.6  341,582  80,420  35,371  723  458,096  Online+  70.6  13.7  10.2  5.5  100.0  33.2  24.7  37.0  78.5  33.0  299,756  57,996  43,313  23,241  424,306  Web  66.7  23.4  9.3  0.6  100.0  27.4  37.1  29.6  7.0  28.9  247,632  87,113  34,667  2073  371,485  Minitel  54.9  6.7  20.0  18.4  100.0  0.7  0.3  2.0  7.3  0.9  6429  779  2342  2150  11,700  Phone+  43.5  33.4  11.4  11.7  100.0  0.5  1.4  1.0  4.0  0.8  4385  3370  1146  1178  10,079  Telephone  40.9  54.1  2.4  2.6  100.0  0.4  2.2  0.2  0.9  0.7  3907  5170  234  253  9564  Total  70.3  18.3  9.1  2.3  100.0  100.0  100.0  100.0  100.0  100.0  903,691  234,848  117,073  29,618  1,285,230    A more formal test is shown in panel A of Table V. I first use three different samples of orders. I report in the column labeled “All Orders,” the DID coefficient of interest (see Section 4), when I include in my sample all the orders of both treated and control investors. In the columns “Limit Order” and “Market Order,” the sample is restricted to executed limit orders only or executed market orders only. In all cases, the sample period is between January 2002 and January 2005, and standard errors are clustered at the individual level.13 Table V. Does the display format of market data matters for retail investors? This table show the results of the following DID OLS regressions:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (2) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use). δi is an individual-fixed effect, γs is a stock fixed-effect, and θt is a day fixed-effect. In panel A, the sample contains the orders of both the treated and the matched control group and the sample period is from January 2002 to January 2005. In Panel B, the sample only contains trades from treated investors so the parameter of interest is identified via the staggered entry into treatment of treated investors over time. In panel B, the sample period is from 2002 to 2010. Returns are in percentage and standard errors (in parenthesis) are clustered at the individual level. *, **, *** denote significance at the 10%, 5%, and 1% levels. Panel A     Trading returns   (1)  (2)  (3)  All orders  Limit order  Market order  Monitoring  0.063***  0.079***  0.036  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856    Panel A     Trading returns   (1)  (2)  (3)  All orders  Limit order  Market order  Monitoring  0.063***  0.079***  0.036  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856    Panel B     Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring  0.047***  0.058***  0.027  (0.01)  (0.01)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  2,464,477  1,799,058  429,981  R-square  0.030  0.041  0.048  Number of investors  7187  6612  4938    Panel B     Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring  0.047***  0.058***  0.027  (0.01)  (0.01)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  2,464,477  1,799,058  429,981  R-square  0.030  0.041  0.048  Number of investors  7187  6612  4938    Table V. Does the display format of market data matters for retail investors? This table show the results of the following DID OLS regressions:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (2) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use). δi is an individual-fixed effect, γs is a stock fixed-effect, and θt is a day fixed-effect. In panel A, the sample contains the orders of both the treated and the matched control group and the sample period is from January 2002 to January 2005. In Panel B, the sample only contains trades from treated investors so the parameter of interest is identified via the staggered entry into treatment of treated investors over time. In panel B, the sample period is from 2002 to 2010. Returns are in percentage and standard errors (in parenthesis) are clustered at the individual level. *, **, *** denote significance at the 10%, 5%, and 1% levels. Panel A     Trading returns   (1)  (2)  (3)  All orders  Limit order  Market order  Monitoring  0.063***  0.079***  0.036  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856    Panel A     Trading returns   (1)  (2)  (3)  All orders  Limit order  Market order  Monitoring  0.063***  0.079***  0.036  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856    Panel B     Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring  0.047***  0.058***  0.027  (0.01)  (0.01)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  2,464,477  1,799,058  429,981  R-square  0.030  0.041  0.048  Number of investors  7187  6612  4938    Panel B     Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring  0.047***  0.058***  0.027  (0.01)  (0.01)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  2,464,477  1,799,058  429,981  R-square  0.030  0.041  0.048  Number of investors  7187  6612  4938    The DID coefficient, which is the coefficient on the variable is Monitoring, captures the effect on investors’ trading performance induced by a more efficient monitoring activity. As predicted, I find a strong increase in the trading performance of executed limit orders (8 basis points) and a positive, albeit non-significant, effect on market orders. These results confirm the first two empirical implications as the coefficient of interest is positive for both limit and market orders. While my display effect is statistically strong for limit orders, it is not for market orders. A major data limitation, namely the absence of intraday execution timestamps, might play a role here. Indeed, as argued before, my display effect is likely smaller for market order. Hence, computing returns around a short window of time after execution, rather than using end-of-day prices, might have revealed a stronger market order effect. To further examine which type of limit order has the greatest improvement in performance, I now exploit the timing of the round-trip strategies implemented by investors in my sample. Specifically, I identify the days in the data where a given investor starts or closes a round-trip position on any given stock and I create accordingly three mutually exclusive dummy variables that reflect this classification. OpenPosi,t,s is a dummy variable that is 1 if a new position on stock s is opened on day t by individual i, ClosePosi,t,s is a dummy variable that is 1 if an existing open position on stock s is closed on day t by individual i, and Posi,t,s is a dummy that equals 1 in any other case. When an individual executes a round-trip within the same day, both OpenPos and ClosePos are equal to one. In that case, I consider that this is a closing position and set OpenPos to zero. Then, I interact these three dummy variables with Monitoringi,t to understand if the improvement in performance due to the new simultaneous display is higher when opening, closing, or trading within an existing round-trip strategy. Results are shown in Table VI. As before, I first consider the full set of trades, and then I estimate my DID regression separately for the sample of limit orders and for the sample of market orders in my data. The results show that, for limit orders, the simultaneous cross-stock limit order book display is most useful when closing a position or when managing an existing round-trip strategy. This result is consistent with the intuition that when liquidating a large existing position, a trader faces a much higher adverse selection risk. Indeed, euro amounts at stake are generally higher than usual when liquidating a position: “Holding everything else equal, larger orders are likely to have lower execution probabilities and to face higher picking-off risk” (Hollifield et al., 2006). The magnitude and the large statistical significance of the interacted variable Monitoringi,t*ClosePosi,t,s thus reflects the fact that the new display allows investors to better mitigate that risk.14 Table VI. Monitoring and round-trip performance In this table, I interact the treatment variable Monitoringi,t with three mutually exclusive dummy variables: OpenPosi,t,s is a dummy variable that is 1 if a new position on stock s is opened on day t by individual i, ClosePosi,t,s is a dummy variable that is 1 if an existing open position on stock s is closed on day t by individual i, and Posi,t,s is a dummy that equals 1 in any other case. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Returns are in percentage and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels.   (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring * ClosePos  0.043*  0.13***  −0.051  (0.02)  (0.03)  (0.05)  Monitoring * Pos  0.074***  0.072***  0.057  (0.02)  (0.02)  (0.04)  Monitoring * OpenPos  0.053**  0.047*  0.068  (0.02)  (0.03)  (0.05)  ClosePos  0.053***  −0.026  0.15***  (0.01)  (0.02)  (0.02)  OpenPos  −0.11***  −0.12***  −0.10***  (0.01)  (0.02)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.044  0.066  0.066  Number of investors  3404  3069  1856    (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring * ClosePos  0.043*  0.13***  −0.051  (0.02)  (0.03)  (0.05)  Monitoring * Pos  0.074***  0.072***  0.057  (0.02)  (0.02)  (0.04)  Monitoring * OpenPos  0.053**  0.047*  0.068  (0.02)  (0.03)  (0.05)  ClosePos  0.053***  −0.026  0.15***  (0.01)  (0.02)  (0.02)  OpenPos  −0.11***  −0.12***  −0.10***  (0.01)  (0.02)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.044  0.066  0.066  Number of investors  3404  3069  1856  Table VI. Monitoring and round-trip performance In this table, I interact the treatment variable Monitoringi,t with three mutually exclusive dummy variables: OpenPosi,t,s is a dummy variable that is 1 if a new position on stock s is opened on day t by individual i, ClosePosi,t,s is a dummy variable that is 1 if an existing open position on stock s is closed on day t by individual i, and Posi,t,s is a dummy that equals 1 in any other case. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Returns are in percentage and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels.   (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring * ClosePos  0.043*  0.13***  −0.051  (0.02)  (0.03)  (0.05)  Monitoring * Pos  0.074***  0.072***  0.057  (0.02)  (0.02)  (0.04)  Monitoring * OpenPos  0.053**  0.047*  0.068  (0.02)  (0.03)  (0.05)  ClosePos  0.053***  −0.026  0.15***  (0.01)  (0.02)  (0.02)  OpenPos  −0.11***  −0.12***  −0.10***  (0.01)  (0.02)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.044  0.066  0.066  Number of investors  3404  3069  1856    (1)  (2)  (3)  All orders  Limit orders  Market orders  Monitoring * ClosePos  0.043*  0.13***  −0.051  (0.02)  (0.03)  (0.05)  Monitoring * Pos  0.074***  0.072***  0.057  (0.02)  (0.02)  (0.04)  Monitoring * OpenPos  0.053**  0.047*  0.068  (0.02)  (0.03)  (0.05)  ClosePos  0.053***  −0.026  0.15***  (0.01)  (0.02)  (0.02)  OpenPos  −0.11***  −0.12***  −0.10***  (0.01)  (0.02)  (0.02)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.044  0.066  0.066  Number of investors  3404  3069  1856  Monthly trends in returns for both treated and control investors, before and after June 2003, are shown in Figures 3 and 4. Figure 3 refers to the sample containing all the orders of both the treatment and control group, while Figure 4 shows the trend in returns separately for the restricted samples of market orders only and limit orders only. In both cases, these figures support the parallel trend assumption. Recall that the DID estimates could be biased if, for instance, treated investors had experienced a transitory shock in returns (positive or negative) before treatment. Indeed, my regression in this case would erroneously attribute to the treatment an effect due to mean-reversion. Instead, Figure 4 suggests that investors’ returns between the treatment and the control group are visually (and statistically) indistinguishable prior to the introduction of Trader+, which goes against a dynamic selection issue. Figure 3. View largeDownload slide Monitoring activity and performance. I compute and plot the monthly average of daily returns achieved by treated, control, and other investors in my sample. The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between January 2002 and January 2005, and have submitted at least one trade after June 2003. Treated investors are investors who submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Other investors are investors in the pool of potential control investors. Figure 3. View largeDownload slide Monitoring activity and performance. I compute and plot the monthly average of daily returns achieved by treated, control, and other investors in my sample. The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between January 2002 and January 2005, and have submitted at least one trade after June 2003. Treated investors are investors who submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Other investors are investors in the pool of potential control investors. Figure 4. View largeDownload slide Monitoring activity and performance: market orders versus limit orders. I compute and plot the monthly average of daily returns achieved by treated, control, and other investors in my sample. The computation only includes executed limit orders in the top figure (a), and only includes executed market orders in the bottom figure (b). The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between January 2002 and January 2005 and have submitted at least one trade after June 2003. Treated investors are investors who submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Other investors are investors in the pool of potential control investors. Figure 4. View largeDownload slide Monitoring activity and performance: market orders versus limit orders. I compute and plot the monthly average of daily returns achieved by treated, control, and other investors in my sample. The computation only includes executed limit orders in the top figure (a), and only includes executed market orders in the bottom figure (b). The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between January 2002 and January 2005 and have submitted at least one trade after June 2003. Treated investors are investors who submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Other investors are investors in the pool of potential control investors. 5.2. Trading Behavior I test in this section my predictions related to the change trading behavior induced by the new display. My hypothesis is that the simultaneous display of cross-stock limit order book data should allow investors to be more aggressive in the placement of their limit orders and thus the probability of observing a trade on a stock with small bid–asks should increase. Indeed, this is in line with the intuition in Liu (2009) that when the cost of monitoring is sufficiently reduced, limit order traders are more likely to post smaller spreads to attract uninformed investors and cancel/revise their orders if market conditions change. To test this idea, I sort stocks in my sample in five quintiles according to their average daily bid–ask spreads during the pre-treatment period from January 2003 to June 2003 (average bid–ask spreads are provided by Eurofidai and come from the Euronext Paris exchange). I then create a dummy variable that is 1 if a trade is executed on a stock that belongs to the lowest quintile, the one that includes stocks with very low bid–ask spreads. I use this dummy variable as a dependent variable in my main DID specification. Results are shown in the first three columns of panel A in Table VII. As previously, I also run a separate regression for each order type (limit versus market order). The output of the regression confirms my hypothesis: investors are indeed more likely to select stocks with lower bid–ask spreads. Consistent with the economic mechanism above, this pattern is not present for market orders. Table VII. Trading behavior This table show the results of the following DID OLS regressions:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (3) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use). δi is an individual-fixed effect, γs is a stock fixed-effect, and θt is a day-fixed effect. I estimate how a more efficient display of market data affects investor’s trading behavior. SmallBidAsk is a dummy variable that equals 1 if a trade is submitted on a stock that is in the lower quintile of the distribution of daily bid–ask spreads. Similarly, LargeStock is a dummy variable that equals 1 if a trade is submitted on a stock that is in the upper quintile of market capitalization. In both cases, stocks are sorted in quintiles using pre-treatment market data. LO is a dummy variable that equals 1 if an order is a limit order and zero otherwise. Log amount is the log of the trade amount (in euro) and margin is a dummy variable that equals 1 if the trade is a margin trade. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Panel A     (1)  (2)  (3)  SmallBidAsk—all  SmallBidAsk—LO  SmallBidAsk—MO  Monitoring  0.016***  0.024***  0.00040  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  524,643  318,894  102,140  R-square  0.21  0.24  0.23  Number of investors  3399  3064  1850    Panel A     (1)  (2)  (3)  SmallBidAsk—all  SmallBidAsk—LO  SmallBidAsk—MO  Monitoring  0.016***  0.024***  0.00040  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  524,643  318,894  102,140  R-square  0.21  0.24  0.23  Number of investors  3399  3064  1850    Panel B     (1)  (2)  (3)  (4)  LO  LargeStock—LO  Log amount  Margin  Monitoring  0.081***  0.043***  0.18***  0.059***  (0.02)  (0.01)  (0.03)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  No  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Number of observations  531,584  318,288  520,954  531,584  R-square  0.49  0.29  0.67  0.63  Number of investors  3404  3064  3386  3404    Panel B     (1)  (2)  (3)  (4)  LO  LargeStock—LO  Log amount  Margin  Monitoring  0.081***  0.043***  0.18***  0.059***  (0.02)  (0.01)  (0.03)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  No  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Number of observations  531,584  318,288  520,954  531,584  R-square  0.49  0.29  0.67  0.63  Number of investors  3404  3064  3386  3404    Table VII. Trading behavior This table show the results of the following DID OLS regressions:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (3) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use). δi is an individual-fixed effect, γs is a stock fixed-effect, and θt is a day-fixed effect. I estimate how a more efficient display of market data affects investor’s trading behavior. SmallBidAsk is a dummy variable that equals 1 if a trade is submitted on a stock that is in the lower quintile of the distribution of daily bid–ask spreads. Similarly, LargeStock is a dummy variable that equals 1 if a trade is submitted on a stock that is in the upper quintile of market capitalization. In both cases, stocks are sorted in quintiles using pre-treatment market data. LO is a dummy variable that equals 1 if an order is a limit order and zero otherwise. Log amount is the log of the trade amount (in euro) and margin is a dummy variable that equals 1 if the trade is a margin trade. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Panel A     (1)  (2)  (3)  SmallBidAsk—all  SmallBidAsk—LO  SmallBidAsk—MO  Monitoring  0.016***  0.024***  0.00040  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  524,643  318,894  102,140  R-square  0.21  0.24  0.23  Number of investors  3399  3064  1850    Panel A     (1)  (2)  (3)  SmallBidAsk—all  SmallBidAsk—LO  SmallBidAsk—MO  Monitoring  0.016***  0.024***  0.00040  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  524,643  318,894  102,140  R-square  0.21  0.24  0.23  Number of investors  3399  3064  1850    Panel B     (1)  (2)  (3)  (4)  LO  LargeStock—LO  Log amount  Margin  Monitoring  0.081***  0.043***  0.18***  0.059***  (0.02)  (0.01)  (0.03)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  No  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Number of observations  531,584  318,288  520,954  531,584  R-square  0.49  0.29  0.67  0.63  Number of investors  3404  3064  3386  3404    Panel B     (1)  (2)  (3)  (4)  LO  LargeStock—LO  Log amount  Margin  Monitoring  0.081***  0.043***  0.18***  0.059***  (0.02)  (0.01)  (0.03)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  No  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Number of observations  531,584  318,288  520,954  531,584  R-square  0.49  0.29  0.67  0.63  Number of investors  3404  3064  3386  3404    I then consider the effect of the new display on order choice. Panel B in Table IV provides a very direct test of the hypothesis that Trader+ should cause investors to use limit orders more often than market orders. Panel B is a frequency table that gives the corresponding number of trades and row and column percentages for each possible combination of trading channel and order type. I use all the orders submitted by both treated and control investors after June 2003. The table shows that there is a strong empirical association between the use of limit orders and the use of trade order management software. Investors submit and execute 37% of all their limit orders using Trader+, the rest being submitted with Online+ (33%) or using the Web (27%). It is interesting to see how the choice between limit and market orders is completely reversed when one compares orders submitted by telephone with orders submitted by Trader+: 54% of the orders submitted by telephone are market orders while this figure drops to 34% for Trader+. This empirical pattern illustrates well the idea that telephone users may incur very high monitoring costs that render limit orders too difficult to monitor, and therefore less profitable, on average, than market orders. Turning to a more formal test, I report in column “LO” of Table VII, the estimates of a DID regression where the dependent variable is a dummy variable that equals 1 for a limit order and 0 otherwise.15 The effect of a more efficient monitoring activity is identified, again, by the DID coefficient on the variable Monitoring. The coefficient shows that following their switch to Trader+, the probability of submitting a limit order increases for treated investors by about 8 percentage points, relative to control investors. In terms of economic significance, I find that the effect is strong enough to be visible at the week level after June 2003. Figure A4 in the Appendix shows the proportion of limit orders executed by treated and control investors as the proportion of trades submitted by treated investors with Trader+ increases. As expected, limit orders become more and more important in the order flow of the treated group after June 2003, which confirms visually the previous results. The rest of Table VII shows that, with the new simultaneous display, investors are more likely to trade large stocks (which is expected because large stocks have smaller bid–ask spreads), trade a large amount (consistent with the intuition that investors with better monitoring capacities can put more capital at risk), and use leverage more. Table VIII explores other dimensions of trading behavior. I argue that the cross-stock display of limit order book data should help investors to manage multiple stocks and multiple orders at the same time. If this is true, then I should observe an increase in the probability of trading multiple stocks during the same day. Additionally, conditional on trading more than one stock on a given day, the probability of trading stocks from different sectors should be higher as well. This implication comes from the intuition that when investors incur very high monitoring costs and want to trade multiple stocks simultaneously, then it is more efficient for them to choose more correlated stocks. That is the case because, due to the high correlation, market movements of one stock can be sufficiently informative about the market movements of the others. Reducing the cost of monitoring should allow investors to place orders on less correlated stocks. To test these ideas, I create a dummy variable at the investor-day level that is one if an individual trades strictly more than one stock on a given day, and similarly I create a dummy variable that is one if an individual trades stocks that belong to more than two (or three) different sectors. The results shown in Table VIII support the idea the cross-display of information helps investors to handle several orders at the same time. Table VIII. Trading volume This table show the results of the following DID OLS regressions:   Yi,t=α+β*Monitoringi,t+θt+δi+ϵi,t, (4) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use) and δi is an individual fixed-effect. θt is a day fixed-effect in panel A and a month-fixed effect in regression 3, 4 and 5 in Panel B. In Panel A, Nb Stocks >1 is a dummy variable that equals 1 if an individual trades more than two stocks on the same day. Nb. Sectors >n is a dummy variable that equals 1 if stocks traded by individual i on day t belong to more than n different sectors. In Panel B, RoundTripi,t is a dummy variable that equals 1 if individual i submits at least one round-trip on day t. Nb Alli,m, Nb Limiti,m, and Nb Marketi,m refer to the number of orders, limit orders, and market orders executed by investor i in month m. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Returns are multiplied by 100 and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Panel A     (1)  (2)  (3)  Number of stocks > 1  Number of sectors > 1  Number of sectors > 2  Monitoring  0.030***  0.033***  0.018***  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  252,432  252,432  252,432  R-square  0.16  0.16  0.16  Number of investors  3397  3397  3397    Panel A     (1)  (2)  (3)  Number of stocks > 1  Number of sectors > 1  Number of sectors > 2  Monitoring  0.030***  0.033***  0.018***  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  252,432  252,432  252,432  R-square  0.16  0.16  0.16  Number of investors  3397  3397  3397    Panel B     (1)  (2)  (3)  (4)  (5)  RiskAdjusted  RoundTrip  Number of all  Number of limit  Number of market  Monitoring  0.081***  0.018***  4.75***  4.71***  0.64**  (0.02)  (0.00)  (1.28)  (1.29)  (0.26)  Id FE  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Number of observations  287,777  294,690  60,798  55,559  55,559  R-square  0.38  0.10  0.53  0.53  0.46  Number of investors  3464  3469  3386  3331  3331    Panel B     (1)  (2)  (3)  (4)  (5)  RiskAdjusted  RoundTrip  Number of all  Number of limit  Number of market  Monitoring  0.081***  0.018***  4.75***  4.71***  0.64**  (0.02)  (0.00)  (1.28)  (1.29)  (0.26)  Id FE  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Number of observations  287,777  294,690  60,798  55,559  55,559  R-square  0.38  0.10  0.53  0.53  0.46  Number of investors  3464  3469  3386  3331  3331    Table VIII. Trading volume This table show the results of the following DID OLS regressions:   Yi,t=α+β*Monitoringi,t+θt+δi+ϵi,t, (4) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use) and δi is an individual fixed-effect. θt is a day fixed-effect in panel A and a month-fixed effect in regression 3, 4 and 5 in Panel B. In Panel A, Nb Stocks >1 is a dummy variable that equals 1 if an individual trades more than two stocks on the same day. Nb. Sectors >n is a dummy variable that equals 1 if stocks traded by individual i on day t belong to more than n different sectors. In Panel B, RoundTripi,t is a dummy variable that equals 1 if individual i submits at least one round-trip on day t. Nb Alli,m, Nb Limiti,m, and Nb Marketi,m refer to the number of orders, limit orders, and market orders executed by investor i in month m. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Returns are multiplied by 100 and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Panel A     (1)  (2)  (3)  Number of stocks > 1  Number of sectors > 1  Number of sectors > 2  Monitoring  0.030***  0.033***  0.018***  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  252,432  252,432  252,432  R-square  0.16  0.16  0.16  Number of investors  3397  3397  3397    Panel A     (1)  (2)  (3)  Number of stocks > 1  Number of sectors > 1  Number of sectors > 2  Monitoring  0.030***  0.033***  0.018***  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Stock FE  No  No  No  Day FE  Yes  Yes  Yes  Number of observations  252,432  252,432  252,432  R-square  0.16  0.16  0.16  Number of investors  3397  3397  3397    Panel B     (1)  (2)  (3)  (4)  (5)  RiskAdjusted  RoundTrip  Number of all  Number of limit  Number of market  Monitoring  0.081***  0.018***  4.75***  4.71***  0.64**  (0.02)  (0.00)  (1.28)  (1.29)  (0.26)  Id FE  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Number of observations  287,777  294,690  60,798  55,559  55,559  R-square  0.38  0.10  0.53  0.53  0.46  Number of investors  3464  3469  3386  3331  3331    Panel B     (1)  (2)  (3)  (4)  (5)  RiskAdjusted  RoundTrip  Number of all  Number of limit  Number of market  Monitoring  0.081***  0.018***  4.75***  4.71***  0.64**  (0.02)  (0.00)  (1.28)  (1.29)  (0.26)  Id FE  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Number of observations  287,777  294,690  60,798  55,559  55,559  R-square  0.38  0.10  0.53  0.53  0.46  Number of investors  3464  3469  3386  3331  3331    Next, I also expect an increase in short-term trading activity for investors who switch to Trader+, as they should be able to implement short-term trading strategies that are unprofitable without efficient monitoring of limit orders. Results are presented in panel B of Table VIII. First, as discussed before, to assess whether investors are more likely to engage in daily-trading, I generate a dummy variable that equals 1 if there is at least one round-trip completed on the same day across all stock traded by investor i that day (column “RoundTrip”) and 0 otherwise. The coefficient on Monitoring is 0.02. This means that after the switch to Trader+, treated investors increase the proportion of short-term round-trips by almost 2 percentage points, which is in line with my predictions. Turning to the trading volume variables in Columns 3, 4, and 5, we see that trading activity at the individual-month level for treated investors strongly increases when monitoring costs are lower, and this increase stems from limit order trading. Indeed, the table shows that both increases in the number of limit and market orders are highly significant, but the coefficient on the number of limit orders (4.7 additional orders per month, column “number of limit”) is about seven times the coefficient on the number of market orders. This increase in trading activity is likely driven by two factors. First, better monitoring reduces non-execution risk, so part of this increase in limit order trades comes from this channel. Second, short-term trades, implemented thanks to the better display, also generate additional trading volume. Appendix A.4 provides additional discussion and graphical illustration for the findings above. In summary, the results discussed in this section suggest that allowing investors to monitor their limit orders more efficiently, thanks to a better display of market data, significantly affects trading volume in terms of (i) the frequency of trading (investors submit more orders and especially more limit orders) and (ii) the investor horizon (investors execute more short-term strategies than they previously did). 6. Alternative Explanations I find in this paper that retail traders who switch to a new order management software subsequently earn higher returns on their limit orders, use limit orders more, trade more, and have a shorter trading horizon. I propose an explanation in which investors have bounded rationality. The introduction of Trader+ allows them to better monitor the market and their limit orders thanks to a more efficient display of market data. The objective of this section is to debate whether self-selection can be an issue in my DID setting. Appendices A.5.a and A.5.b consider, and rule out, other alternative explanations based on trading speed or overconfidence. 6.1. Dynamic Self-Selection A possible concern in my identification strategy is self-selection. As stated before, DIDs are robust to self-selection issues if selection into treatment is determined by time-invariant variables (either observable or unobservable). Indeed, those important variables (possibly correlated with the treatment) will be controlled for by the individual-fixed effects in my regressions. However, DIDs are still biased in the presence of dynamic self-selection because the “parallel trend assumption” would not be verified in that case. In my setting this would be the case if, for instance, investors suddenly understood the importance of monitoring their limit orders, perhaps as a reaction to the brokers’ ad in favor of the software, and subsequently began to monitor them more efficiently by simply exerting more attentional effort. In this example, investors eventually switched to Trader+ simply because they had the opportunity to do so, but the effect of Trader+ is nil. In other words, treated investors in my sample would have obtained higher returns even without using Trader+. Dynamic self-selection seem not to be a serious issue in my setting for four reasons. The first reason is provided by the dynamics of individual entry into treatment. Subfigure (a) in Figure 5 provides, starting from June 2003, the number of investors who switch into treatment status each month (i.e., investors who submit their first Trader+ trade that month). From the graph, one can see that almost all treated investors switch immediately to Trader+ in June 2003. This pattern, and the jump in returns (shown in Figure 3) that immediately follows the introduction of Trader+, is thus only consistent with self-selection mechanisms that contemporaneously affect all investors in the treatment group exactly in June 2003. While I cannot definitely reject this possibility, this phenomenon seems hard to justify. Figure 5. View largeDownload slide Is self-selection an issue? I identify year-cohorts of treated investors, according to the year in which investors submit their first trade using Trader+. Subfigures (b) and (c) compare the trends in daily returns between the first cohort of treated investors (that switched to Trader+ in 2003) and investors treated 1 year later (in 2004) or 2 year later (in 2005). Figure 5. View largeDownload slide Is self-selection an issue? I identify year-cohorts of treated investors, according to the year in which investors submit their first trade using Trader+. Subfigures (b) and (c) compare the trends in daily returns between the first cohort of treated investors (that switched to Trader+ in 2003) and investors treated 1 year later (in 2004) or 2 year later (in 2005). Second, if the use of Trader+ has no effect, then there should be no specific pattern between investors’ increases in returns and investors’ switches to Trader+. Instead, the data show that improvements in trading performance systematically follow the switch to Trader+. This is causality in the sense of Granger: consequences follow causes, not the opposite. These patterns can be seen in subfigures (b) and (c) on Figure 5. I consider here a larger sample of investors, which also includes those that opened a trading account after 2003. I then classify all the investors who eventually use Trader+ in my dataset into various cohorts, according to the year in which they submitted their first trade using Trader+. I obtain eight cohorts from 2003 to 2010. In subfigures (b) and (c) in Figure 5, I compare the trends in outcomes between the 2003 cohort and the 2004 cohort and between the 2003 cohort and the 2005 cohort. Figure 5 shows a very interesting pattern. For instance, in subfigure (c), returns for investors in the 2003 cohort and investors in the 2005 cohort clearly follow the same trend until the 2003 cohort enters into treatment status (by definition, between June and December 2003). We then see a sharp jump in returns for this cohort relative to the 2005 one. This difference (the causal effect of Trader+) persists over time until the 2005 cohort switches to treatment status as well. After 2005, trends in returns are again almost indistinguishable. These patterns suggest that not (yet) treated investors may provide an interesting alternative control group to assess how robust my results are. Building on this idea, I show more formally that my estimates are similar when I run a specification in the spirit of Giroud (2013), where my sample contains treated investors only so that the control group at any point in time consists of treated investors that have not (yet) switched yet to Trader+. These not (yet) treated investors arguably possess these same unobservable (possibly time-changing) characteristics that will push them to switch to Trader+ in the future, and therefore could provide a better counterfactual. Results are shown in panel B of Table V. In this regression, my sample contains treated investors only (a little >7000 in total) and the parameter of interest is thus identified by the staggered entry into treatment of treated investors between 2003 and 2010. The fact that my estimates are little changed when I use this completely different specification provides support in favor of a causal interpretation of my main estimates. Third, I can also exploit the 10-year time range of my dataset to test the critical “parallel trend” in another way. Recall that this assumption means that, in the absence of treatment, the trends between the treatment and the control group would have been the same. Of course, it is not possible to test this identifying assumption, but as a more indirect test, one can assess whether the parallel trend assumption holds for all the periods before the introduction of the treatment. To this end, recall that my matching procedure matches treatment and control individuals using trading data only between January and May 2003. That is, trends between treatment and control groups before January 2003 provide an out-of-sample test of the parallel trend assumption. As shown in Figure 3, the trends between the two groups are very similar even several years before the matching and the introduction of Trader+. Fourth, I complement the trends visible in Figure 3 and Figure 5, with a more formal test of whether there are identical counterfactual trends for the treatment and the control group. Indeed, one possible way to quantify the likeliness that this assumption will hold is to introduce lead and lag treatment dummy variable in the main DID regressions. That is, instead of using the variable Monitoring defined in the previous sections, I can now introduce a “lead” and b “lag” treatment dummy variables as: ∑j=−abβj*Treatmenti,t(t=k+j), where k is the month in which a given individual enters into the treatment group (by submitting his first Trader+ trade). I expect the leads to have no statistical significance, and the lags to allow for a more flexible estimation of the treatment dynamics. As suggested by the previous graphs, the effects due to Trader+ are concentrated in the few months around its introduction. Therefore, to better capture the treatment dynamics, I focus on a short time window around June 2003: I include one lead (a dummy that is 1 in the month right before k), one lag (a dummy that is 1 in the month right after k), and a dummy variable that is always one starting 2 months after k. The results of the regressions are shown in Table IX. As expected, Table IX shows that the lead dummy has no statistical effect on the independent variables of interest. Also, while the coefficients on the dummy treatment variables (in particular for the trading performance measures) begin to increase during the month when a given individual starts to use Trader+, the coefficients become statistically significant just 1 month after (i.e., when t=k+1). Interestingly, the coefficient on the variable Treatment(>1 m) is often of the same magnitude as the one at time t=k+1. This pattern means that the treatment effect is persistent over time, which is of course consistent with the visual trends examined before. Importantly, these dynamics confirm that the treatment effect is likely to be additive and thus that the DID setting is appropriate in this paper. Table IX. Dynamic treatment effects To understand the treatment effect dynamics, I use a lead-lag treatment specification instead of using the treatment indicator Monitoringi,t. In this table [Treatment(=−1)]i,t is equal to 1 only in the month prior to the month of first use of Trade+ by investor i and 0 otherwise. [Treatment(=0)]i,t is equal to 1 only in the month of first use of Trade+ by investor i and 0 otherwise and, similarly, [Treatment(t>1)]i,t is one starting 2 months after the first Trader+ trade by a investor i and 0 otherwise. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Returns are in percentage and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels.   Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Treatment(−1 m)  0.0025  0.024  −0.036  (0.03)  (0.03)  (0.06)  Treatment(0 m)  0.029  0.041  0.017  (0.02)  (0.03)  (0.05)  Treatment(+1 m)  0.074***  0.10***  −0.00017  (0.02)  (0.03)  (0.06)  Treatment(>1 m)  0.075***  0.097***  0.054  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856    Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Treatment(−1 m)  0.0025  0.024  −0.036  (0.03)  (0.03)  (0.06)  Treatment(0 m)  0.029  0.041  0.017  (0.02)  (0.03)  (0.05)  Treatment(+1 m)  0.074***  0.10***  −0.00017  (0.02)  (0.03)  (0.06)  Treatment(>1 m)  0.075***  0.097***  0.054  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856  Table IX. Dynamic treatment effects To understand the treatment effect dynamics, I use a lead-lag treatment specification instead of using the treatment indicator Monitoringi,t. In this table [Treatment(=−1)]i,t is equal to 1 only in the month prior to the month of first use of Trade+ by investor i and 0 otherwise. [Treatment(=0)]i,t is equal to 1 only in the month of first use of Trade+ by investor i and 0 otherwise and, similarly, [Treatment(t>1)]i,t is one starting 2 months after the first Trader+ trade by a investor i and 0 otherwise. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Returns are in percentage and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels.   Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Treatment(−1 m)  0.0025  0.024  −0.036  (0.03)  (0.03)  (0.06)  Treatment(0 m)  0.029  0.041  0.017  (0.02)  (0.03)  (0.05)  Treatment(+1 m)  0.074***  0.10***  −0.00017  (0.02)  (0.03)  (0.06)  Treatment(>1 m)  0.075***  0.097***  0.054  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856    Trading returns   (1)  (2)  (3)  All orders  Limit orders  Market orders  Treatment(−1 m)  0.0025  0.024  −0.036  (0.03)  (0.03)  (0.06)  Treatment(0 m)  0.029  0.041  0.017  (0.02)  (0.03)  (0.05)  Treatment(+1 m)  0.074***  0.10***  −0.00017  (0.02)  (0.03)  (0.06)  Treatment(>1 m)  0.075***  0.097***  0.054  (0.02)  (0.02)  (0.04)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Number of observations  531,584  323,380  103,349  R-square  0.043  0.066  0.065  Number of investors  3404  3069  1856  In summary, I find no evidence that dynamic self-selection is a serious concern in my paper. 7. Conclusion I test whether the display of market data affects the trading performance of retail investors. I use the introduction of a trade order management software, that gathered all the relevant (and otherwise dispersed) market information into a single screen, as a shock that allows investors to better monitor their limit orders. I find a strong increase in limit order trading performance for investors who use the software, suggesting that for retail investors the display format of financial information is critical. This result is also of interest for policymakers because regulators are increasingly concerned with not only how much financial information is provided to investors, but also how it is displayed to them. One such example is the European Commission’s PRIIPs proposal (2012, 2014) aimed at simplifying the information provided to investors in financial products. This paper shows that the display of information can indeed have first-order effects on investment behavior in real-life settings. To conclude, identifying the determinants of households’ financial decisions is still an open and important question in the field. Like IQ, financial sophistication, and financial wealth, the display of information can be an important driver of such decisions. Appendix A Appendix A.1 proposes an alternative measure of daily returns, Appendix A.2 provides the additional evidence, based on VWAP prices, that intraday returns increase with the new display. Appendix A.3 provides evidence that my results are robust to the inclusion of transaction costs and discusses whether long-term behavior is impacted by the display. Appendix A.4 shows additional results to complement my results on trading behavior. Appendices A.5.a and A.5.b complement the alternative explanations section of the paper by discussing the possible issues of trading speed and overconfidence. Last, Appendix A.6 provides two additional charts to illustrate some features of the data discussed in the text. A.1. Measuring Short-Term Returns For some of my tests, I need a measure of performance at the investor-day level. To this end, I follow Linnainmaa (2011) and I compute the investor i’s short-term gross profits on day t as:   profiti,t= {∑s=1n(Pt,si,sell−Pt,si,buy)qt,si,buy if qt,si,sell=qt,si,buy∑s=1n(Pt,si,sell−Pt,si,buy)qt,si,sell+(Ct,s−Pt,si,buy)(qt,si,buy−qt,si,sell) if qt,si,sell<qt,si,buy∑s=1n(Pt,si,sell−Pt,si,buy)qt,si,buy+(Pt,si,sell−Ct,s)(qt,si,sell−qt,si,buy) if qt,si,sell>qt,si,buy, where n is the number of stock traded by investor i on day t, Pt,si,sell and Pt,si,buy are i’s investor-weighted average selling and buying prices of stock s on day t. Ct,s is the closing price of stock s on day t and qt,si,sell and qt,si,buy are the i’s investor quantities sold and bought of stock s on day t. I subtract from the gross profits the daily total trading fees paid by investor i on day t to obtain a net daily profit measure.16 Then, I compute gross (net) returns by dividing the gross (net) profits by the investor i’s order size on day t. Ri,t=profiti,txti, where   xti=∑s=1nmax(qt,si,buy,qt,si,sell)Pt,si,buyqt,si,buy+Pt,si,sellqt,si,sellqt,si,buy+qt,si,sell. That is, the trade size is the maximum of the number of shares bought and sold times the volume-weighted average transaction price computed from all investor i’s trades on day t. The advantage of this approach, as suggested by Linnainmaa (2011), is that complete round-trips are taken into account while remaining positions on day t and stock s (i.e., such that qt,si,sell≠qt,si,buy) are marked-to-market to the closing price of the stock on that day. A.2. Volume-Weighted Average Price Table AI provides additional evidence on the effect of Trader+ on trading performance. I follow a standard in the brokerage industry and I compare the execution price of any order to the VWAP of the stock the same day. The VWAP is a well-known benchmark that is used to assess the quality of an execution strategy. Buying below the VWAP or selling above the VWAP is indicative of superior order placement and better execution. I expect Trader+ users to obtain better prices relative to this benchmark. To test this hypothesis, in panel A of Table AI I compute the signed difference (in EUR) between the execution price of an order and the VWAP that day: Qi,t,k,s=signi,t,k,s*(VWAPt,s−Pricei,t,k,s) The variable is signed, so positive values indicate better execution quality. As we can see from the table, individuals are more likely to get execution prices in line with the market, after they switch to Trader+. Consistent the prior results, the bulk of this increase in execution quality seem to come from limit orders. On average, the new display allows investors to execute their buy limit orders at prices about 0.023 EUR below the VWAP. Market orders also show a positive improvement, but the effect is not statistically significant. In panel B, I perform the same exercise using indicator variables instead of the euro difference. I create a dummy variable that is 1 if the execution price is better than the VWAP and 0 otherwise. The results I get in this alternative specification tell the same story: investors monitor the market more efficiently, manage their order better, and therefore obtain better execution prices. A.3. Trading Costs and Long-Term Behavior In this section, I perform two robustness tests. I first investigate whether the inclusion of transaction costs affects the improved trading performance of treated investors discussed in the previous sections. Then, I assess whether long-term measures of trading performance are affected by the new display. Results are shown in Table AII. In the first panel, labeled “Short-Term P&L,” I include transaction costs in the computations of daily returns. I perform this test because treated investors trade more when they start using the new display. In other words, while transactions costs will not affect my economic mechanism per se (because I consider whether orders are executed at better prices relative to a benchmark such as the closing price), it may be the case that the improvement in returns due to the new display is actually offset by the accumulated impact of transactions fees. Columns 1 and 2 of Table AII show that this is not the case. Gross and, in particular, net returns when all trades are aggregated at the investor-day level still increase with the new display, which is consistent with the previous evidence shown in the paper. Actually, returns net of transactions costs are even higher because treated investors pay smaller transaction fees, on average.17 Because the distribution of daily profits and losses has high kurtosis and is negatively skewed in the data, I use dummy variables to understand how specific areas of the daily P&L distribution are affected by the change in behavior of treated investors. In particular, I generate a dummy variable that is 1 if the gross (net) daily profits and losses (in Euros) of investor i on day t are lower (or higher) than a specific threshold.18 Columns 3 (gross profits), 4 (net profits), 5 (net profits) and 6 (net profits) give the DID estimates when the dependent variable is one of the dummy variables described above. The probability of ending the day with a positive gross or net profit increases with the new display, although both tails of the distribution seem to become fatter (Columns 5 and 6). Overall, this evidence is consistent with the fact that the improvement in trading returns survives the inclusion of transaction costs. The second part of Table AII, which focuses on long-term measures of performance, shows that there is some slight improvement in long-term returns as well. However, this improvement is not statistically significant. This null result is actually reassuring; it means that treated investors use the display to execute their orders more efficiently than before, but they do not alter their trading behavior in a way that is detrimental in the long run, relative to the control group. Two important caveats are in order. First, as stated in the “introduction” section, the scope of this paper is neither to promote active investing nor to enforce the use of any trading software for retail investors. Indeed, in this sample as in many other studies, active investing is a losing proposition (Barber et al., 2008). Rather, this paper studies the causal effect of a change in the display of information while keeping the information set fixed. It shows that a simultaneous display has desirable implications relative to a display that is dispersed for tasks that are costly in cognitive resources. In this paper, I considered the specific task of monitoring limit orders because the market microstructure literature had clean predictions to test my simultaneous-display effect. Second, my display effect is estimated from daily returns in a DID setting. This means that what matters for identification is the change in performance for treated investors relative to the change for control investors. Trading performance has to be assessed in a relative, not absolute, way. For instance, in the same vein, Kuo, Lin, and Zhao (2014) studies the relative performance between two set of individual investors that both experience negative trading returns, on average (p. 25, table 5). A.4. Trading Volume To illustrate the findings discussed in Section 5.2, I plot the normalized number of trades executed (total number of trades divided by the number of investors, in a given month) and the percentage of levered orders executed. Subfigure (a) (labeled “Trading Volume”) shows how sudden the rise in trading activity is right after the introduction of Trader+. This effect is interesting. Indeed, as a comparison, it takes 18 months for investors in Choi, Laibson, and Metrick (2002) to increase trading by 50%, relative to control investors, after their switch to a web-based trading channel. Figure A2 illustrates more precisely the changes in trading horizon. I recover the empirical distribution of the duration of investors’ round-trips before and after June 2003 in both groups. The densities are shown in the top subfigure in Figure A2. Before treatment, the distribution of round-trips lengths between treated and control investor is very similar. However, after treatment, the number of short-term trading strategies clearly increases for treated investors relative to the control investors. The bottom subfigure focuses on the log of the trading horizon for both groups. Again, and similarly to the patterns identified above, one can see that the log horizon of the treated group falls, relative to the control group, right after June 2003. A.5. Alternative Explanations A.5.a. Trading speed Another possibility is that investors achieve higher trading returns because the decision to use the software is correlated with a positive variation in trading speed that allows investors to place their order faster than other investors. This possibility is related to several recent papers that focus on automation and speed in financial markets (see Garvey and Wu, 2010; Hendershott and Moulton, 2011).19 Trading speed could play a role in my setting if investors switch to Trader+ when the quality of their Internet connection increases or if orders sent through Trader+ are routed more quickly to the market. If this alternative explanation is correct, the improvement in performance should even be greater for market orders than for limit orders, because market orders benefit the most from trading speed increases. Indeed, investors with a speed-advantage should use market orders more often than limit orders, in order to gain from speed-related trading strategies, such as picking off stale limit orders not being monitored by other slow investors (Garvey and Wu, 2010). Tables V and VII show instead that for the treated group, after the switch to Trader+, (i) the lion’s share of improvements in returns comes from limit orders and (ii) investors use limit orders more than before. These changes in behavior thus suggest that my results are not driven by variations in trading speed. A.5.b. Overconfidence One last possible alternative explanation for my results is that Trader+ may have increased investors’ overconfidence. Indeed, overconfidence has become the leading behavioral explanation for the sharp increase in trading, and the lower subsequent performance, of retail investors that switched to online trading at the beginning of 2000.20 A mechanism similar to the one in Barber and Odean (2002) would suggest that investors with past successes prior to the introduction of Trader+ are more likely to become overconfident and also more likely to switch to the software because overconfidence is enhanced by the “feeling of empowerment” (or the “illusion of control”) enjoyed when using a trading software. Therefore, treated investors in this paper could become overconfident about their ability to monitor their limit orders. This reasoning has two implications. First, overconfidence arguments predict a decrease in trading performance. Table V shows the opposite result. Alternatively, more overconfident investors switching to the software should achieve even lower returns than less overconfident investors, because investors more prone to overconfidence are more likely to commit behavioral trading mistakes. To test this hypothesis, I compute the Sharpe ratio of past trading returns, from June 2002 to June 2003, for each investor in my sample.21 I then classify each investor into three groups defined by the terciles of the distribution of Sharpe ratios and I run a DID regression for each group. This specification allows me to compare the changes in trading performance of (more or less) overconfident treated investors, after their switch to Trader+, compared with control investors with the same level of overconfidence. Results are shown in Table AIII. I find that investors with higher past returns (in the top group) are those who benefit the most from the software (18 basis points per limit order trade), which contradicts the previous prediction based on overconfidence arguments. Therefore, overconfidence does not seem to play a role in this paper. Interestingly, one can also see from the table that improving the display of information can only be beneficial for retail investors (at least in terms of their trading returns). That is, the treatment effect I document benefit retail investors with some degree of trading skills, but is not detrimental to those with a low level of such skills. This result has some important policy implications. It means that policymakers have an effective way of improving the trading decisions of most households by simply manipulating the display of financial information provided to them. Table AI. Monitoring and execution quality This table show the results of the following DID OLS regressions:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (A.1) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use). δi is an individual-fixed effect, γs is a stock fixed-effect, and θt is a day fixed-effect. In Panel A, I test whether a more efficient display of market data helps investors to obtain better execution prices on their orders relative to the volume weighted average price (VWAP) benchmark. Hence, the dependent variable is measured in Euros and is (VWAP–Price) for buy orders and (Price–VWAP) for sell orders. I run one regression including all orders, and two separate regressions that restrict the sample to limit orders or market orders only. In Panel B, I replace the euro differences above by a dummy variable that is 1 if the difference is positive and 0 otherwise. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Panel A     All   Limit order   Market order   (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell  Monitoring  0.017***  0.015***  0.023***  0.014***  0.0080  0.011  (0.01)  (0.00)  (0.01)  (0.00)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  266,861  246,938  163,372  148,807  52,994  46,668  R-square  0.11  0.11  0.12  0.14  0.15  0.15  Number of investors  3288  3219  2840  2793  1538  1523    Panel B      All  Limit order  Market order  (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell    Monitoring  0.046***  0.034***  0.051***  0.031***  0.031**  0.011  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  272,297  252,006  166,754  152,057  54,015  47,463  R-square  0.094  0.10  0.11  0.12  0.13  0.13  Number of investors  3298  3230  2853  2810  1553  1527  Panel A     All   Limit order   Market order   (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell  Monitoring  0.017***  0.015***  0.023***  0.014***  0.0080  0.011  (0.01)  (0.00)  (0.01)  (0.00)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  266,861  246,938  163,372  148,807  52,994  46,668  R-square  0.11  0.11  0.12  0.14  0.15  0.15  Number of investors  3288  3219  2840  2793  1538  1523    Panel B      All  Limit order  Market order  (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell    Monitoring  0.046***  0.034***  0.051***  0.031***  0.031**  0.011  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  272,297  252,006  166,754  152,057  54,015  47,463  R-square  0.094  0.10  0.11  0.12  0.13  0.13  Number of investors  3298  3230  2853  2810  1553  1527  Table AI. Monitoring and execution quality This table show the results of the following DID OLS regressions:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (A.1) Monitoringi,t is a (treatment) dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+ (including the first day of use). δi is an individual-fixed effect, γs is a stock fixed-effect, and θt is a day fixed-effect. In Panel A, I test whether a more efficient display of market data helps investors to obtain better execution prices on their orders relative to the volume weighted average price (VWAP) benchmark. Hence, the dependent variable is measured in Euros and is (VWAP–Price) for buy orders and (Price–VWAP) for sell orders. I run one regression including all orders, and two separate regressions that restrict the sample to limit orders or market orders only. In Panel B, I replace the euro differences above by a dummy variable that is 1 if the difference is positive and 0 otherwise. The sample contains the orders of both the treated and the control group. The sample period is from January 2002 to January 2005. Standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Panel A     All   Limit order   Market order   (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell  Monitoring  0.017***  0.015***  0.023***  0.014***  0.0080  0.011  (0.01)  (0.00)  (0.01)  (0.00)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  266,861  246,938  163,372  148,807  52,994  46,668  R-square  0.11  0.11  0.12  0.14  0.15  0.15  Number of investors  3288  3219  2840  2793  1538  1523    Panel B      All  Limit order  Market order  (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell    Monitoring  0.046***  0.034***  0.051***  0.031***  0.031**  0.011  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  272,297  252,006  166,754  152,057  54,015  47,463  R-square  0.094  0.10  0.11  0.12  0.13  0.13  Number of investors  3298  3230  2853  2810  1553  1527  Panel A     All   Limit order   Market order   (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell  Monitoring  0.017***  0.015***  0.023***  0.014***  0.0080  0.011  (0.01)  (0.00)  (0.01)  (0.00)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  266,861  246,938  163,372  148,807  52,994  46,668  R-square  0.11  0.11  0.12  0.14  0.15  0.15  Number of investors  3288  3219  2840  2793  1538  1523    Panel B      All  Limit order  Market order  (1)  (2)  (3)  (4)  (5)  (6)  VWAP buy  VWAP sell  VWAP buy  VWAP sell  VWAP buy  VWAP sell    Monitoring  0.046***  0.034***  0.051***  0.031***  0.031**  0.011  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  (0.01)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Yes  Yes  Yes  Day FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  272,297  252,006  166,754  152,057  54,015  47,463  R-square  0.094  0.10  0.11  0.12  0.13  0.13  Number of investors  3298  3230  2853  2810  1553  1527  Table AII. Robustness: trading costs and long-term behavior In this table, I study the effect of the new display when trading costs and long-term performance measures are taken into account. Panel “Short-Term P&L” considers (gross and net) trading profits and losses at the daily horizon. The sample period is from January 2002 to January 2005 and returns (in Columns 1 and 2) are in percentage. In Panel “Long-Term P&L” I consider all the round-trip trades initiated by investors in my sample between January 2002 to January 2005. I include round-trips that are left open by investors in my sample by closing them at the latest date available in the sample. Round-trip returns are converted into daily returns and shown in percentage, and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Short-term P&L     (1)  (2)  (3)  (4)  (5)  (6)  Gross return  Net return  Profit  >0€  Net  >0€  Net  >50€  Net<− 50€  Monitoring  0.084***  0.17***  0.034***  0.022***  0.013***  0.019***  (0.018)  (0.035)  (0.004)  (0.004)  (0.003)  (0.007)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  287,457  266,138  289,876  268,032  268,032  268,032  R-square  0.098  0.19  0.059  0.070  0.070  0.16  Number of investors  3461  3411  3465  3417  3417  3417    Short-term P&L     (1)  (2)  (3)  (4)  (5)  (6)  Gross return  Net return  Profit  >0€  Net  >0€  Net  >50€  Net<− 50€  Monitoring  0.084***  0.17***  0.034***  0.022***  0.013***  0.019***  (0.018)  (0.035)  (0.004)  (0.004)  (0.003)  (0.007)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  287,457  266,138  289,876  268,032  268,032  268,032  R-square  0.098  0.19  0.059  0.070  0.070  0.16  Number of investors  3461  3411  3465  3417  3417  3417    Long-term P&L     (1)  (2)  (3)    Gross return  Net return  Adj. return  Monitoring  0.016  0.0070  0.0093  (0.012)  (0.017)  (0.011)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Time FE  Yes  Yes  Yes  Number of observations  114,720  102,534  114,452  R-square  0.17  0.19  0.12  Number of investors  3243  3109  3237    Long-term P&L     (1)  (2)  (3)    Gross return  Net return  Adj. return  Monitoring  0.016  0.0070  0.0093  (0.012)  (0.017)  (0.011)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Time FE  Yes  Yes  Yes  Number of observations  114,720  102,534  114,452  R-square  0.17  0.19  0.12  Number of investors  3243  3109  3237    Table AII. Robustness: trading costs and long-term behavior In this table, I study the effect of the new display when trading costs and long-term performance measures are taken into account. Panel “Short-Term P&L” considers (gross and net) trading profits and losses at the daily horizon. The sample period is from January 2002 to January 2005 and returns (in Columns 1 and 2) are in percentage. In Panel “Long-Term P&L” I consider all the round-trip trades initiated by investors in my sample between January 2002 to January 2005. I include round-trips that are left open by investors in my sample by closing them at the latest date available in the sample. Round-trip returns are converted into daily returns and shown in percentage, and standard errors (in parenthesis) are clustered at the individual level. *, **, and *** denote significance at the 10%, 5%, and 1% levels. Short-term P&L     (1)  (2)  (3)  (4)  (5)  (6)  Gross return  Net return  Profit  >0€  Net  >0€  Net  >50€  Net<− 50€  Monitoring  0.084***  0.17***  0.034***  0.022***  0.013***  0.019***  (0.018)  (0.035)  (0.004)  (0.004)  (0.003)  (0.007)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  287,457  266,138  289,876  268,032  268,032  268,032  R-square  0.098  0.19  0.059  0.070  0.070  0.16  Number of investors  3461  3411  3465  3417  3417  3417    Short-term P&L     (1)  (2)  (3)  (4)  (5)  (6)  Gross return  Net return  Profit  >0€  Net  >0€  Net  >50€  Net<− 50€  Monitoring  0.084***  0.17***  0.034***  0.022***  0.013***  0.019***  (0.018)  (0.035)  (0.004)  (0.004)  (0.003)  (0.007)  Id FE  Yes  Yes  Yes  Yes  Yes  Yes  Stock FE  No  No  No  No  No  No  Time FE  Yes  Yes  Yes  Yes  Yes  Yes  Number of observations  287,457  266,138  289,876  268,032  268,032  268,032  R-square  0.098  0.19  0.059  0.070  0.070  0.16  Number of investors  3461  3411  3465  3417  3417  3417    Long-term P&L     (1)  (2)  (3)    Gross return  Net return  Adj. return  Monitoring  0.016  0.0070  0.0093  (0.012)  (0.017)  (0.011)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Time FE  Yes  Yes  Yes  Number of observations  114,720  102,534  114,452  R-square  0.17  0.19  0.12  Number of investors  3243  3109  3237    Long-term P&L     (1)  (2)  (3)    Gross return  Net return  Adj. return  Monitoring  0.016  0.0070  0.0093  (0.012)  (0.017)  (0.011)  Id FE  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  Time FE  Yes  Yes  Yes  Number of observations  114,720  102,534  114,452  R-square  0.17  0.19  0.12  Number of investors  3243  3109  3237    Table AIII. Heterogeneous treatment effects I classify each investor in my matched sample (including both treated and control investors) into three groups defined by the terciles of the distribution of individual Sharpe ratios (computed from trading returns) before June 2003. This table shows the results of the previous DID OLS regressions, estimated separately for the three tercile groups. Bottom group     (1)  (2)  (3)  (4)  All orders  Limit  Market  Daily return  Monitoring  0.044  0.039  0.086  0.035  (0.03)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  109,118  64,056  23,027  68,808  R-square  0.082  0.11  0.12  0.12  Number of investors  567  538  330  567    Middle group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.067***  0.063**  0.062  0.12***    (0.02)  (0.03)  (0.04)  (0.03)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  187,183  106,275  39,710  92,739  R-square  0.038  0.063  0.076  0.062  Number of investors  567  549  392  567    Top group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.14***  0.18***  0.098  0.091**    (0.04)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  159,500  97,835  29,693  73,461  R-square  0.032  0.057  0.077  0.049  Number of investors  566  534  388  566  Bottom group     (1)  (2)  (3)  (4)  All orders  Limit  Market  Daily return  Monitoring  0.044  0.039  0.086  0.035  (0.03)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  109,118  64,056  23,027  68,808  R-square  0.082  0.11  0.12  0.12  Number of investors  567  538  330  567    Middle group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.067***  0.063**  0.062  0.12***    (0.02)  (0.03)  (0.04)  (0.03)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  187,183  106,275  39,710  92,739  R-square  0.038  0.063  0.076  0.062  Number of investors  567  549  392  567    Top group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.14***  0.18***  0.098  0.091**    (0.04)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  159,500  97,835  29,693  73,461  R-square  0.032  0.057  0.077  0.049  Number of investors  566  534  388  566  Table AIII. Heterogeneous treatment effects I classify each investor in my matched sample (including both treated and control investors) into three groups defined by the terciles of the distribution of individual Sharpe ratios (computed from trading returns) before June 2003. This table shows the results of the previous DID OLS regressions, estimated separately for the three tercile groups. Bottom group     (1)  (2)  (3)  (4)  All orders  Limit  Market  Daily return  Monitoring  0.044  0.039  0.086  0.035  (0.03)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  109,118  64,056  23,027  68,808  R-square  0.082  0.11  0.12  0.12  Number of investors  567  538  330  567    Middle group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.067***  0.063**  0.062  0.12***    (0.02)  (0.03)  (0.04)  (0.03)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  187,183  106,275  39,710  92,739  R-square  0.038  0.063  0.076  0.062  Number of investors  567  549  392  567    Top group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.14***  0.18***  0.098  0.091**    (0.04)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  159,500  97,835  29,693  73,461  R-square  0.032  0.057  0.077  0.049  Number of investors  566  534  388  566  Bottom group     (1)  (2)  (3)  (4)  All orders  Limit  Market  Daily return  Monitoring  0.044  0.039  0.086  0.035  (0.03)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  109,118  64,056  23,027  68,808  R-square  0.082  0.11  0.12  0.12  Number of investors  567  538  330  567    Middle group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.067***  0.063**  0.062  0.12***    (0.02)  (0.03)  (0.04)  (0.03)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  187,183  106,275  39,710  92,739  R-square  0.038  0.063  0.076  0.062  Number of investors  567  549  392  567    Top group      (1)  (2)  (3)  (4)    All orders  Limit  Market  Daily return    Monitoring  0.14***  0.18***  0.098  0.091**    (0.04)  (0.04)  (0.07)  (0.04)  Id FE  Yes  Yes  Yes  Yes  Stock FE  Yes  Yes  Yes  No  Day FE  Yes  Yes  Yes  Yes  Number of observations  159,500  97,835  29,693  73,461  R-square  0.032  0.057  0.077  0.049  Number of investors  566  534  388  566  Figure A1. View largeDownload slide Monitoring activity and trading volume. I compute and plot the monthly normalized number of trades (number of trades/number of investors in a given month) and the monthly percentage of leveraged orders (through margin account). The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between 2002 and 2005, and have submitted at least one trade after June 2003. Treated investors are investors who submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Other investors are investors in the pool of potential control investors. Figure A1. View largeDownload slide Monitoring activity and trading volume. I compute and plot the monthly normalized number of trades (number of trades/number of investors in a given month) and the monthly percentage of leveraged orders (through margin account). The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between 2002 and 2005, and have submitted at least one trade after June 2003. Treated investors are investors who submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Other investors are investors in the pool of potential control investors. Figure A2. View largeDownload slide Monitoring activity and trading horizon. I sort all trades in my database by investor, stock, and trading day and I keep track of the number of stocks held by an investor at any time. A round-trip on a given stock starts and ends with a zero net inventory position on that stock. The duration of a round-trip is then the number of days between those two events. Using data in the pre-treatment period (January 2002–June 2003) and the post-treatment period (June 2003–January 2005), I generate in subfigure (a) the densities of investors’ round-trip durations (in days). In subfigure (b) I show the log duration of their round-trips. Figure A2. View largeDownload slide Monitoring activity and trading horizon. I sort all trades in my database by investor, stock, and trading day and I keep track of the number of stocks held by an investor at any time. A round-trip on a given stock starts and ends with a zero net inventory position on that stock. The duration of a round-trip is then the number of days between those two events. Using data in the pre-treatment period (January 2002–June 2003) and the post-treatment period (June 2003–January 2005), I generate in subfigure (a) the densities of investors’ round-trip durations (in days). In subfigure (b) I show the log duration of their round-trips. Figure A3. View largeDownload slide Orders submitted by treated and matched control investors, 1999–2010. For each year (on the x-axis) this figure shows (on the y-axis) the total number of trades submitted by treated or matched control investors that year. This number is then broken down according to order type (limit order, market order, minor order types, or missing), or trading channel (Online+, Website, Trader+, other minor trading channels, and missing). Each bar thus is a sum of stacked bars that give the corresponding number of orders of a given order type or trading channel. Figure A3. View largeDownload slide Orders submitted by treated and matched control investors, 1999–2010. For each year (on the x-axis) this figure shows (on the y-axis) the total number of trades submitted by treated or matched control investors that year. This number is then broken down according to order type (limit order, market order, minor order types, or missing), or trading channel (Online+, Website, Trader+, other minor trading channels, and missing). Each bar thus is a sum of stacked bars that give the corresponding number of orders of a given order type or trading channel. Figure A4. View largeDownload slide Monitoring activity and order choice. I compute and plot the weekly percentage of executed orders that are limit orders submitted by treated and control investors in my sample. The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between January 2002 and January 2005, and have submitted at least one trade after June 2003. Treated investors are investors that have submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Figure A4. View largeDownload slide Monitoring activity and order choice. I compute and plot the weekly percentage of executed orders that are limit orders submitted by treated and control investors in my sample. The vertical black line identifies the introduction of Trader+ by the brokerage house in June 2003. All investors have submitted at least one trade before June 2003, have at least one stock in their portfolios between January 2002 and January 2005, and have submitted at least one trade after June 2003. Treated investors are investors that have submitted at least one trade using Trader+ after June 2003. Control investors have never used Trader+ and are matched one-to-one to treated investors with a propensity score methodology. Footnotes 1 This distinction is important for two reasons. First, informationally equivalent displays of information should be equally processed by rational investors. Second, it makes my paper different from previous related works [such as Barber and Odean (2002) or the XBRL literature] in which a technology shock increased the quantity of information available to investors. 2 Limit orders specify a number of shares and a price: the maximum price at which an investor is willing to buy or the minimum price at which an investor is willing to sell. Limit orders accumulate in the limit order book in descending buy–price order or ascending sell–price order (price priority) and join the queue composed of other limit orders that have the same price (time priority). Market orders specify a number of shares to buy or sell, but no particular price. Market orders are automatically filled at the most attractive price posted by previous limit orders in the limit order book. 3 See Grinblatt, Keloharju, and Linnainmaa (2012). For instance, Hollifield et al. (2006) define the risk of adverse selection as the expected loss (or gain) due to future expected changes in stock value given execution. Accordingly, Liu (2009) obtain a proxy for this expectation by comparing the current price after the execution of a limit order to the price at which the order has been executed. 4 Two interesting recent working papers using real-world data are related to my paper. Shaton (2017) exploits a regulatory change in the display of the performance of Israeli retirement funds to understand how that affects retail retirement–investment decisions when investors have limited attention [see also Stango and Zinman (2014)]. Levi (2014) uses a random field experiment to study the importance of ‘information architecture’ in the context of spending decisions (restaurants, utilities, etc.). 5 Indeed, limit orders can be cancelled and/or revised after their submission. A higher limit price priority will increase the execution probability and thus decrease the risk of non-execution. A lower limit price priority will instead decrease the likelihood of execution and thus decrease the risk of adverse selection (Fong and Liu, 2010). As monitoring is a costly cognitive process, investors cannot continuously monitor their limit orders. 6 This mechanism is actually mentioned in Barber and Odean (2002): “Investors may trade more when they go online simply because of greater ease of access. For rational investors this implies that there were potentially profitable trades that the investors declined to make before going online because the expected profits did not warrant the effort of calling a broker.” 7 A subsample of this database has already been used, to address other research questions, in Foucault, Sraer, and Thesmar (2011) and Barrot, Kaniel, and Sraer (2016). 8 In the data, about 7800 unique investors used Trader+ at least once from 2003 to 2010. 9 Some variables can have missing data at some point in time. These missing data are due to changes in how information was stored internally in the servers of the brokerage house. For instance, intraday execution timestamps were not stored in their servers until 2011. 10 I also include in my analysis all the round-trips that are not closed by investors by closing them at the earliest trading date between (i) the date at which an investor closed his trading account, (ii) the last trading date in the sample (end of December 2010), and (iii) the last valid last trading date of the security traded. Additionally, as in Chakrabarty, Moulton, and Trzcinka (2017), I exclude round-trips that occur between a stock split or a stock reverse-split. 11 Additionally, investors can fully customize their Trader+ graphical-user interface according to their needs and preferences, choosing the relative ordering and appearance of each information items (display color, size, etc.). Investors could also pre-fill and store their trading orders. 12 Similarly, the DID regression at the investor-month level is: Yi,m=α+β*Monitoringi,m+θm+δi+ϵi,m. 13 Because the raw brokerage dataset also contains a small quantity of accounting operations (such as account transfers, dividend payments, and others) I apply some filters to the data. In particular, I drop trades with an amount <250 EUR, I drop trades whose prices are outside of the daily maxima and only consider trades executed on euro-denominated stocks listed at the main French exchange Euronext Paris (to avoid the influence of the brokerage FX rate). In my regressions, performance variables are trimmed at the 1% and 99% level. 14 Appendix A.2 provides additional evidence based on the volume weighted average price (VWAP), while Appendix A.3 shows that my results are robust to the inclusion of transaction costs and that the improvements in short-term returns are not offset by larger losses over longer time horizons. 15 Because the DID captures the change in the relative difference between treated and control group, I recode any missing values for the order choice variable as zeros. 16 Trading fees include all trading commissions and other commissions such as margin trading fees. Trading fees are available at the trade level after 2006, and at the monthly level before that. Hence, to obtain trading fees at the trade level before 2006 I divide the total euro amount paid in fees by the total euro amount traded by an investor in a given month to obtain a monthly trading fee in percentage. 17 This point highlights why gross returns, rather than net returns, are more appropriate to evaluate the causal effect of the new display. In the data, there is a great deal of variation in trading fees across investors because fees are often negotiated on a case-by-case basis with the brokerage house. 18 For instance, I use the −50 Euro threshold because it corresponds approximately to the 25th percentile of the profits and losses distribution. Hence, my dummy variable is likely to capture tail risk. 19 It should be recalled that Trader+ is not an algorithmic trading tool: it takes no trading decisions alone. Behind each single order in my database there is the voluntarily mouse-click of an investor, so the improvements in order execution shown in the data are the outcomes of the investors’ trade decisions. 20 See Choi, Laibson, and Metrick (2002) and Barber and Odean (2002). 21 I compute E(r)/σ(r), where E(r) is the average return and σ(r) the standard deviation of individual i’s returns between June 2002 and June 2003. To have a meaningful ratio, I only consider investors with at least 15 trades during this time window. References Angrist J. D., Pischke J.-S. ( 2008): Mostly Harmless Econometrics: An Empiricist’s Companion , Princeton University Press, Princeton, New Jersey. Barber B. M., Lee Y.-T., Liu Y.-J., Odean T. ( 2008): Just how much do individual investors lose by trading?, Review of Financial Studies  22, 609– 632. Google Scholar CrossRef Search ADS   Barber B. M., Lee Y.-T., Liu Y.-J., Odean T. ( 2014): The cross-section of speculator skill: evidence from day trading, Journal of Financial Markets  18, 1– 24. Google Scholar CrossRef Search ADS   Barber B. M., Odean T. ( 2002): Online investors: do the slow die first?, Review of Financial Studies  15, 455– 488. Google Scholar CrossRef Search ADS   Barber B. M., Odean T. ( 2013): Chapter 22—the behavior of individual investors, in: George M. H., Constantinides M., Stulz R. M.. (eds.), Handbook of the Economics of Finance , Vol. 2, Part B, Elsevier, pp. 1533– 1570. Barrot J.-N., Kaniel R., Sraer D. ( 2016): Are retail traders compensated for providing liquidity?, Journal of Financial Economics  120, 146– 168. Google Scholar CrossRef Search ADS   Bertrand M., Duflo E., Mullainathan S. ( 2004): How much should we trust differences-in-differences estimates?, The Quarterly Journal of Economics  119, 249– 275. Google Scholar CrossRef Search ADS   Bhattacharya U., Hackethal A., Kaesler S., Loos B., Meyer S. ( 2012): Is unbiased financial advice to retail investors sufficient? Answers from a large field study, Review of Financial Studies  25, 975– 1032. Google Scholar CrossRef Search ADS   Calvet L. E., Campbell J. Y., Sodini P. ( 2007): Down or out: assessing the welfare costs of household investment mistakes, Journal of Political Economy  115, 707– 747. Google Scholar CrossRef Search ADS   Calvet L. E., Sodini P. ( 2014): Twin picks: disentangling the determinants of risk-taking in household portfolios, Journal of Finance  69, 867– 906. Google Scholar CrossRef Search ADS   Campbell J. Y. ( 2006): Household finance, Journal of Finance  61, 1553– 1604. Google Scholar CrossRef Search ADS   Chakrabarty B., Moulton P. C., Trzcinka C. ( 2017): The performance of short-term institutional trades, Journal of Financial and Quantitative Analysis  1– 26. Choi J., Laibson D., Metrick A. ( 2002): How does the internet affect trading? Evidence from investor behavior in 401(k) plans, Journal of Financial Economics  64, 397– 421. Google Scholar CrossRef Search ADS   Fong K. Y., Liu W.-M. ( 2010): Limit order revisions, Journal of Banking & Finance  34, 1873– 1885. Google Scholar CrossRef Search ADS   Foucault T. ( 1999): Order flow composition and trading costs in a dynamic limit order market, Journal of Financial Markets  2, 99– 134. Google Scholar CrossRef Search ADS   Foucault T., Roell A., Sandas P. ( 2003): Market making with costly monitoring: an analysis of the SOES controversy, Review of Financial Studies  16, 345– 384. Google Scholar CrossRef Search ADS   Foucault T., Sraer D., Thesmar D. J. ( 2011): Individual investors and volatility, Journal of Finance  66, 1369– 1406. Google Scholar CrossRef Search ADS   Garvey R., Wu F. ( 2010): Speed, distance, and electronic trading: new evidence on why location matters, Journal of Financial Markets  13, 367– 396. Google Scholar CrossRef Search ADS   Giroud X. ( 2013): Proximity and investment: evidence from plant-level data, The Quarterly Journal of Economics  128, 861– 915. Google Scholar CrossRef Search ADS   Grinblatt M., Keloharju M., Linnainmaa J. T. ( 2012): IQ, trading behavior, and performance, Journal of Financial Economics  104, 339– 362. Google Scholar CrossRef Search ADS   Harris L., Hasbrouck J. ( 1996): Market vs. limit orders: the superDOT evidence on order submission strategy, Journal of Financial and Quantitative Analysis  31, 213– 231. Google Scholar CrossRef Search ADS   Hendershott T., Jones C. M., Menkveld A. J. ( 2011): Does algorithmic trading improve liquidity?, Journal of Finance  66, 1– 33. Google Scholar CrossRef Search ADS   Hendershott T., Moulton P. C. ( 2011): Automation, speed, and stock market quality: the NYSE’s hybrid, Journal of Financial Markets  14, 568– 604. Google Scholar CrossRef Search ADS   Hodge F. D., Kennedy J. J., Maines L. A. ( 2004): Does search-facilitating technology improve the transparency of financial reporting?, The Accounting Review  79, 687– 703. Google Scholar CrossRef Search ADS   Hollifield B., Miller R. A., Sandas P. ( 2004): Empirical analysis of limit order markets, Review of Economic Studies  71, 1027– 1063. Google Scholar CrossRef Search ADS   Hollifield B., Miller R. A., Sandas P., Slive J. ( 2006): Estimating the gains from trade in limit-order markets, The Journal of Finance  61, 2753– 2804. Google Scholar CrossRef Search ADS   Kelley E. K., Tetlock P. C. ( 2013): How wise are crowds? Insights from retail orders and stock returns, Journal of Finance  68, 1229– 1265. Google Scholar CrossRef Search ADS   Kuo W.-Y., Lin T.-C., Zhao J. ( 2014): Cognitive limitation and investment performance: evidence from limit order clustering, Review of Financial Studies  28, 838– 875. Google Scholar CrossRef Search ADS   Levi Y. ( 2014): Information architecture and intertemporal choice: a randomized field experiment in the United States. Working paper, USC. Linnainmaa J. T. ( 2010): Do limit orders alter inferences about investor performance and behavior?, Journal of Finance  65, 1473– 1506. Google Scholar CrossRef Search ADS   Linnainmaa J. T. ( 2011): Why do (some) households trade so much?, Review of Financial Studies  24, 1630– 1666. Google Scholar CrossRef Search ADS   Liu W.-M. ( 2009): Monitoring and limit order submission risks, Journal of Financial Markets  12, 107– 141. Google Scholar CrossRef Search ADS   Puckett A., Yan X. S. ( 2011): The interim trading skills of institutional investors, Journal of Finance  66, 601– 633. Google Scholar CrossRef Search ADS   Rogers W. ( 1994): Regression standard errors in clustered samples, Stata Technical Bulletin 3. Russo J. E. ( 1977): The value of unit price information, Journal of Marketing Research  14, 193– 201. Google Scholar CrossRef Search ADS   Schlarbaum G. G., Lewellen W. G., Lease R. C. ( 1978): Realized returns on common stock investments: the experience of individual investors, Journal of Business  51, 299– 325. Google Scholar CrossRef Search ADS   Shapira Z., Venezia I. ( 2001): Patterns of behavior of professionally managed and independent investors, Journal of Banking & Finance  25, 1573– 1587. Google Scholar CrossRef Search ADS   Shaton M. ( 2017): The display of information and household investment behavior, Finance and Economics Discussion Series 2017-043, Board of Governors of the Federal Reserve System ( U.S.). Stango V., Zinman J. ( 2014): Limited and varying consumer attention: evidence from shocks to the salience of bank overdraft fees, Review of Financial Studies  27, 990– 1030. Google Scholar CrossRef Search ADS   White H. ( 1980): A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity, Econometrica  48, 817– 838. Google Scholar CrossRef Search ADS   Published by Oxford University Press on behalf of the European Finance Association 2018. This work is written by a US Government employee and is in the public domain in the US. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Review of Finance Oxford University Press

To See Is to Know: Simultaneous Display of Market Data for Retail Investors

Review of Finance , Volume Advance Article – Feb 24, 2018

Loading next page...
 
/lp/ou_press/to-see-is-to-know-simultaneous-display-of-market-data-for-retail-9SvK0i20wK
Publisher
Oxford University Press
Copyright
Published by Oxford University Press on behalf of the European Finance Association 2018. This work is written by a US Government employee and is in the public domain in the US.
ISSN
1572-3097
eISSN
1573-692X
D.O.I.
10.1093/rof/rfy007
Publisher site
See Article on Publisher Site

Abstract

Abstract I test whether the display format of market data affects the trading performance and behavior of retail investors. To do so, I exploit a large brokerage dataset covering a period during which the market information provided to the broker’s customers changed in format, but not in content. I find that a simultaneous display of cross-stock market data reduces the cognitive cost of monitoring the market and thus helps investors obtain better execution prices. In particular, investors better mitigate non-execution and adverse-selection risks when trading with limit orders. Hence, the display format of market data matters for the individual investor. 1. Introduction Identifying the determinants of households’ financial decisions is a fundamental question in the field of household finance (Campbell, 2006; Calvet, Campbell, and Sodini, 2007) and beyond the influence of a few key drivers such as IQ (Grinblatt, Keloharju, and Linnainmaa, 2012; Bhattacharya et al., 2012), financial sophistication (Calvet, Campbell, and Sodini, 2007), and financial wealth (Calvet and Sodini, 2014), the financial decision making of households is still not well understood. In particular, understanding the behavior of active retail investors and explaining the large cross-sectional distribution in individual trading performance are questions subject to an ongoing debate in the field. In this paper, I test whether a more efficient format of market data provided to active retail investors affects their trading behavior and performance. This is a question of interest because evidence from the laboartory suggests the display format of financial information may have first-order effects on trading behavior, perhaps even in a real-life setting. In addition, regulators and policymakers across the world increasingly care not only about how much information to provide to households, but also about which display format to use. In principle, a more efficient display should lead to better trading decisions and higher performance. However, a better display may also stimulate behavioral biases such as overconfidence and tempt active investors to increase speculative trading and incur more losses, thus leading to a detrimental effect of the news display. To explore this tradeoff, I exploit a unique setting in which the quantity of market information provided to investors remained fixed, while the display format of that information exogenously changed at some point in time. This setting is the June 2003 introduction of a trade order management software (hereafter, Trader+) by a large brokerage house. The software displayed market data in a more efficient way because it simultaneously gathered all relevant information items (market data, centralized limit order book, and investors’ orders) into a user-customized screen. Such simultaneous presentation of information items allows investors to understand how different stock price movements are related to one another and helps investors better assimilate that information when monitoring their orders. For instance, Hodge, Kennedy, and Maines (2004), who study a search-facilitating technology introduced by the SEC in 2003 (and adopted in 2009), argue that the “simultaneous presentation of related information directs users’ attention toward examining relations among the information items (Russo [1977]). Simultaneous presentation also reduces the cognitive costs of integrating the information”. Importantly, the software left the quantity of the data being processed unchanged. Indeed, the same information was available to the brokerage customers, in a more dispersed form, on other trading channels (such as using the brokerage website to submit an order).1 The key mechanism at play in this paper is that Trader+ decreased the cognitive cost of monitoring limit orders. Recall that any trade necessarily involves the choice between market and limit orders.2 If investors use limit orders, then they must monitor and revise/cancel them, after submission, to mitigate adverse selection and non-execution risks. That is, investors should monitor the market and their limit orders to improve their trading performance on those orders. If a display format of market data makes such market monitoring more efficient then, everything else being constant, it should also improve investors’ trade performance on their limit orders. To test these implications, I use a difference-in-differences (DID) identification strategy in which treated investors, who switch to the new software, are matched (with a propensity-score algorithm) to similar control investors who do not use the software. I find that the display format of market data does matter for the individual investor. As in Linnainmaa (2010) and Grinblatt, Keloharju, and Linnainmaa (2012), for each buy and sell limit order on a given stock in my sample, I compute the signed return from the execution price to the closing price of that stock on the same day. Indeed, this return has been used in the literature as a proxy for the risk of adverse selection faced by investors when trading with limit orders. It is therefore a performance measure of one order’s execution quality and should capture “the active management of individuals.”3 Intuitively, the higher the risk of adverse selection, the lower this return. My DID estimates show that following the introduction of Trader+, investors’ intraday returns on their limit orders jump by 8 basis points. The economic magnitude of this result is large in light of previous studies on the trading performance of investors. For instance, Grinblatt, Keloharju, and Linnainmaa (2012) find that intraday returns of Finnish investors with a high IQ (in the top 5% of the distribution) outperform those with a low IQ by 11 basis points. Similarly, Kuo, Lin, and Zhao (2014) find that investors with high cognitive abilities outperform the limit order intraday returns of those with low cognitive abilities by 3.9 basis points. These comparisons suggest that the display format of market data has an effect on trading performance similar to that of individual cognitive abilities. My explanation for this increase in trade performance is consistent with market microstructure theories of order choice, in which the “cognitive cost” of monitoring limit orders is reduced by the new display. In the paper, I show that the key feature of the new display is the simultaneous cross-stock limit order book. I provide additional evidence supporting this hypothesis. Because the new simultaneous display reduces the cost of monitoring, it increases the expected utility of trading with a limit order relative to the utility of trading with a market order. Accordingly, I find that treated investors trade more limit orders than before when using the software. Additionally, because my sample is populated by active traders, I expect treated investors to be able to spot and monitor short-term trading opportunities that were likely to be too difficult to monitor before. This change in behavior should lead to a decrease in individual trading horizons, which is confirmed by the data. Overall, trading activity of the treated group increases, relative to the control group, within 1 month of the software’s introduction. I also consider, and rule out, alternative explanations for my findings based on trading speed, overconfidence, and investors’ self-selection. This paper directly contributes to two different literatures. First, this paper is part of the few recent works that study how alternative presentation formats of financial information influence investors’ financial decisions. To the best of my knowledge, this is the first paper to show that the display format of market data affects the real trading choices and performance of individual investors.4 Second, this paper contributes to the recent literature that uses insight from market microstructure theory to understand individual investors’ decisions (Linnainmaa, 2010; Kelley and Tetlock, 2013; Barber et al., 2014). The paper also contributes to the market microstructure literature that focuses on order submissions strategies (see Foucault, 1999; Hollifield, Miller, and Sandas, 2004). In the vein of Foucault, Roell, and Sandas (2003) and Liu (2009), my economic mechanism relies on extensions of standard models of order choice that embed the cognitive costs of monitoring and managing limit orders. In particular, this paper is directly related to Fong and Liu (2010), which uses aggregated institutional and retail order flow data to establish a link between monitoring costs and limit order performance. My paper goes one step further by focusing on retail investors and by linking the display format of market data to retail trading performance. Hence, I can participate in the ongoing debate about what drives the large cross-section variation in retail trading skills that has been previously identified (Barber and Odean, 2013). My results imply that part of this unexplained variation in performance may be due to differences in the display format of the data being processed by retail investors. This finding is important because the drivers of retail performance previously documented in the literature (such as IQ, financial sophistication, or financial wealth) cannot be easily manipulated by policymakers, but data display format can be. This article proceeds as follows. Section 2 discusses my testable empirical hypothesis. Section 3 presents the brokerage dataset and Section 4 motivates and discusses my identification strategy. The results are given in Section 5. Before concluding, I discuss the potential alternative explanations for my results in Section 6. 2. Testable Hypothesis In this paper, I argue that the new display of market data that comes with Trader+ is a positive exogenous shock on the market monitoring capacities of individual investors. Essentially, the new display allows investors to process, more efficiently, the same quantity of market data as before. The simultaneous display effect operates via two different channels. First, before submitting any order, investors are more likely to better process the current market conditions. This improved market timing should benefit both market and limit orders so intraday returns should increase for both types of orders. Second, after submitting a limit order, a trader faces non-execution and adverse selection risks. Both of these risks can be mitigated, via Trader+, by active order management (also called monitoring activity).5 Hence, limit orders gains from both better market timing and better order management. This reasoning implies that Trader+ should, in turn, induce investors to use limit orders more often than they previously did before because it is more profitable to do so. Indeed, an investor submits a limit order if the expected utility of using a limit order is greater than the certain utility of using a market order. The investor sends a market order otherwise. This latter hypothesis has several testable implications. First, investors switching to Trader+ should be more likely to place limit orders on stocks with smaller bid–ask spreads. Indeed, when the cost of processing limit order book data is reduced, limit order traders are more likely to post smaller spreads to attract uninformed investors and cancel/revise their orders if market conditions change (Liu, 2009). Second, investors should be able to manage, better than before, situations where they must monitor multiple orders on multiple stocks at the same time. If this reasoning is correct, then we should observe an increase in the probability of submitting orders on more than one stock during the same day. Furthermore, short-term strategies in which investors trade very frequently (such as “day-trading,” where investors revert their positions at the daily level) are more difficult to monitor than long-term ones. If this is correct, a more efficient monitoring activity should allow active investors to spot more short-term trading opportunities than before. This effect implies that the proportion of short-term trading strategies should increase (leading to higher trading activity) when investors monitor their limit orders more efficiently.6 In summary, I argue that the new simultaneous display of market data should decrease the cost of monitoring the market and the cost of monitoring pending limit orders. This mechanism implies the following testable implications: A more efficient monitoring activity increases the trading performance of limit orders, because investors gain from reduced adverse selection, reduced non-execution risks, and better market timing. A more efficient monitoring activity has a positive effect on the performance of market orders, as market orders benefit from better market timing. A more efficient monitoring activity increases the probability of submitting a limit order, as limit orders become more profitable than market orders. A more efficient monitoring activity increases the probability of trading multiple stocks at the same time. A more efficient monitoring activity increases trading volume and reduces trading horizon, because investors exploit short-term strategies that were too costly to execute and monitor before. 3. Data In this section, I describe the data and define the variables that I will use in order to test the empirical implications described in the previous section. 3.1. The Brokerage Dataset The data used in this paper comes from a leading French online broker.7 The raw dataset contains at the daily level all of the executed trades sent by each of the 145,801 customers of the broker from 1999 up to 2010, which represents >15 million trades. In this paper, I focus on active liquidity-providing retail investors, which represent a small part of the entire retail population but generate most of the orders in my database.8 Each trade comes with the following information: the asset type (equity, bonds, etc.), the trading exchange identifier (the ISIN), the trading date, the quantity in the number of stocks, the total amount traded in euro, the order type (limit order, market order, and other minor orders types), the trading place, the trading channel used to submit the order, and a margin trade indicator. Most of the trades are executed on the NYSE Euronext Paris trading exchange. Depending on the period, the data also include the fees paid at the single trade level.9 Basic demographic information includes date of birth, gender, French department of residence, and opening and closing date of any brokerage account in the data. Last, portfolio holdings are available at the monthly level. I match the trades in my dataset with market data provided by Eurofidai, the European financial data institute. Trades are matched by ISIN code, trading day, and trading exchange code. Trades for which no information is available from Eurofidai are discarded from the sample. Summary statistics of the raw brokerage dataset are provided in Table I. This table shows the corresponding number of trades, percentage, and cumulative percentage for several categorical variables that describe the nature of my data well. Panel A shows for instance that trades are in the majority of cases limit orders (60%) and market orders (27%). The other minor orders types are rarely used by investors. The information on order type is completely missing for 1999 and 2000 and may be sometimes missing up to 2004. Panel B of the table shows that most of individual trading activity consists of buying or selling common stocks. This dataset, therefore, shows patterns of individual investment behavior that are similar to other recent databases used in the literature. For instance, Finnish investors in Linnainmaa (2010) also use limit orders for most of their trades on the Helsinki Stock Exchange. Panel C shows the trading channels available for retail investors to submit a trade. Investors could submit an order by using the telephone to speak to a broker official (Telephone), by calling a voicemail service and typing instruction using the telephone’s keys (Phone+), by using a web navigator to connect to the broker website (Web), by using an old French Videotex online service accessible through the telephone lines (Minitel), or by using online basic computer software (Online+). Trader+ is a trading software that was introduced in June 2003. It will be fully discussed and presented in Section 4. Panel C of the table highlights that a large majority of trades are submitted using the internet or trading software (Online+ and Trader+). Table I. Summary statistics of the brokerage dataset This table describes the main characteristics of the brokerage dataset used in this paper. The table gives the corresponding number of trades, percentage, and cumulative percentage for each category of information (investors’ order choices, investors’ use of trading channels, investors’ use of asset classes, and investors’ use of trading exchanges). The sample period is from 2002 to 2010. Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      Table I. Summary statistics of the brokerage dataset This table describes the main characteristics of the brokerage dataset used in this paper. The table gives the corresponding number of trades, percentage, and cumulative percentage for each category of information (investors’ order choices, investors’ use of trading channels, investors’ use of asset classes, and investors’ use of trading exchanges). The sample period is from 2002 to 2010. Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      Panel A: Order choice     Frequency  Percentage  Cumulative percentage  Limit order  6,207,531  60.2  60.2  Market order  2,828,769  27.4  87.6  Minor order types  907,754  8.80  96.4  Missing  372,388  3.61  100  Total  10,316,442  100      Panel B: Asset classes      Frequency  Percentage  Cumulative percentage    Common stocks  8,754,630  84.9  84.9  Warrants  838,691  8.13  93.0  Minor asset types  592,533  5.74  98.7  ETF  130,588  1.27  100  Total  10,316,442  100      Panel C: Trading channels      Frequency  Percentage  Cumulative percentage    Web  5,268,500  51.2  51.2  Online+  2,113,790  20.5  71.7  Trader+  2,049,605  19.9  91.6  Telephone  341,289  3.31  94.9  Minitel  219,998  2.14  97.0  Phone+  177,697  1.73  98.8  Manual  128,103  1.24  100  Total  10,298,982  100    Number of distinct investors  110,591      Number of treated investors  7810      Number of buy order  5,344,967      Number of sell orders  4,971,475      3.2. Variable Definition I describe in this section how I measure investors’ trading performance (Section 3.2.a) and investor trading horizon (Section 3.2.b). 3.2.a. Measuring trading performance I measure trading performance as follows. For each trade in my sample, I compute the signed difference between the closing price of the stock bought (or sold) and the execution price of the order, divided by the execution price of the order:   Ri,t,k,s=signi,t,k,s*Closet,s−Pricei,t,k,sPricei,t,k,s, where Closet,s is the closing price on day t, of the traded stock s, and Pricei,t,k,s is the execution price of order number k submitted on day t, for stock s, by individual i. Signi,t,k,s is a dummy variable that equals 1 for a buy order and −1 for a sell order. This expost performance measure is well suited for assessing the performance of a limit order, because it captures the adverse selection risk faced by investors using those orders (see Harris and Hasbrouck, 1996). Indeed, Hollifield et al. (2006) define the risk of adverse selection as the expected loss (or gain) due to future expected changes in stock value given execution. Similarly, Liu (2009) obtain a proxy for this expectation by comparing the current price after the execution of a limit order to the price at which the order has been executed. A similar approach is also adopted in Grinblatt, Keloharju, and Linnainmaa (2012) and Linnainmaa (2010). For instance, Grinblatt, Keloharju, and Linnainmaa (2012) compare the performance between high and low IQ investors and argue that this return is essentially a measure of one order’s execution quality and should capture “the active management of individuals.” “High IQ investors,” they say, “may be better or quicker at processing information into a useful signal, or excel at distinguishing useful information from noise.” Because the researchers are able to detect statistically significant differences between the returns of the high IQ and low IQ groups, their hypothesis cannot be rejected. My approach follows the same reasoning. If a more efficient display of market data allows investors to monitor their limit orders more efficiently, then investors should improve upon the execution quality of their limit orders relative to investors using a less efficient data display. This improvement is captured by my intraday returns. Execution quality can also be assessed in other ways. In Appendix A.1, I use an alternative way to compute short-term returns which closely follows Linnainmaa (2011). Also, in Appendix A.2, I follow a standard procedure in the brokerage industry and compare the average execution price obtained on any order with the volume-weighted average price (VWAP) of the same stock the same day (see Hendershott, Jones, and Menkveld, 2011). To test these hypothesis, I need an improvement in the display of market data for retail investors. This point will be discussed in Section 4. 3.2.b. Measuring long-term returns and investors’ trading horizon To proxy for investors’ trading horizon and have a long-term measure of trading performance, I adopt a general methodology used in Schlarbaum, Lewellen, and Lease (1978); Puckett and Yan (2011); and Chakrabarty, Moulton, and Trzcinka (2017). The main idea is to take full advantage of the granularity of my data and to aggregate investor i’s single trades into round-trip trades. To do so, I first compute the daily net quantity traded by investor i, on stock s and day t as: Qi,t,s=Qtiti,t,sbuy−Qtiti,t,ssell, where Qtiti,t,sbuy (respectively, Qtiti,t,ssell) represents the actual quantity of stock s bought (respectively, sold) by investor i on day t. Then, I sort all the daily net quantities in my dataset by trader, stock, and trading day, and I keep track of the cumulative stock quantity held by a trader day after day. A round-trip starts and ends with a zero net cumulative quantity. Round-trips are thus trading positions that are fully unwound: stocks previously bought are entirely sold and stocks previously sold short are completely bought back.10 This methodology is appealing in the context of my paper because it requires very few assumptions on the data (Schlarbaum, Lewellen, and Lease, 1978). To see whether trading horizon decreases after the introduction of Trader+, I study how Trader+ affects the proportion of short-term round-trip trades that are started in a given day by a given trader. Also, this aggregation of individual trades into round-trips allows me to identify when an investor initiates or closes a position on a given stock. I will use this information in Section 5.1. The gross return on a round-trip is computed as R=EuroSell−EuroBuyEuroBuy, where EuroBuy (EuroSell) is the total euro cash-flow paid when buying stocks (received when selling stocks) during a particular round-trip executed by any investor on a given stock. To get round-trip returns net of commissions, trading fees for the buy-leg are added to EuroBuy and trading fees for the sell-leg are subtracted to EuroSell. Last, risk-adjusted returns are obtained by subtracting to the round-trip return the return that would have been obtained, passively, by investing in the French market index (CAC40) over the same holding period. Finally, following Shapira and Venezia (2001), I convert these gross and net returns into daily returns using the formula: Rdaily=(1+R)1T−1, where T, the duration of a round-trip, is computed as the number of days between the opening date and the closing date of a round-trip. 4. Methodology and Identification Strategy 4.1. Trader+ I explain in this section why the display of market data in Trader+ should help investors more efficiently monitor their limit orders. The new information display of market data, available on Trader+, allowed Trader+ users to monitor their limit orders more efficiently than before. Indeed, prior to June 2003, the broker customers submitted their orders mainly through two trading channels: using the existing trading software available (Online+) and connecting to the brokerage web interface (Web). Figure 1 provides a screenshot of each of these trading channels and highlights how identical information items (such as the current state of the limit order book or the recent market movements statistics) are displayed differently on those trading channels. In both cases, one can see that the market data are dispersed through different webpages (for Web) or through different tabs (for Online+). Figure 1. View largeDownload slide Main trading channels used by the broker customers before Trader+. This figure shows two screenshots of the main trading channels used by investors in my sample, before the introduction of Trader+. (a) Investors used to send their orders using Online+, a basic trading software that has been available since 1999, or (b) using the Web brokerage interface (Web). The same quantity of market information is provided on each trading channel. Figure 1. View largeDownload slide Main trading channels used by the broker customers before Trader+. This figure shows two screenshots of the main trading channels used by investors in my sample, before the introduction of Trader+. (a) Investors used to send their orders using Online+, a basic trading software that has been available since 1999, or (b) using the Web brokerage interface (Web). The same quantity of market information is provided on each trading channel. Instead, Trader+ displayed market data much more efficiently than those trading channels because it simultaneously gathered all relevant information items into a user-customized screen. A screenshot of Trader+ is shown in Figure 2. As one can see, investors using Trader+ could see on their computer screen, not only the limit order book or the most recent market statistics, but also their pending orders, the stock intraday graphics, and market data at the stock level.11 The key point is that these information items were also available, in a more dispersed form, on the other trading channels. In other words, the quantity of information remains the same, while the display of that information varies when Trader+ is introduced. Figure 2. View largeDownload slide Screenshot of Trader+. This is a screenshot of the main Trader+ interface in June 2003. Users can follow the status of all their submitted orders, along with all the characteristics of these orders (security id, buy/sell indicator quantity, price, execution/cancellation). There is an aggregated limit order book that simultaneously shows the best bid and ask prices and quantites for several stocks (ACCOR, AGF, AIR LIQUIDE, etc.). Users can also get news related to multiple securities at the same time. Additional market data are provided on the top left of the screen and a complete order submission form is shown on the bottom left of the screen. Figure 2. View largeDownload slide Screenshot of Trader+. This is a screenshot of the main Trader+ interface in June 2003. Users can follow the status of all their submitted orders, along with all the characteristics of these orders (security id, buy/sell indicator quantity, price, execution/cancellation). There is an aggregated limit order book that simultaneously shows the best bid and ask prices and quantites for several stocks (ACCOR, AGF, AIR LIQUIDE, etc.). Users can also get news related to multiple securities at the same time. Additional market data are provided on the top left of the screen and a complete order submission form is shown on the bottom left of the screen. The simultaneous presentation of cross-stock limit order book data is the critical feature that allows investors to better understand how stock prices and market movements are related to one another and helps investors better assimilate that information when they monitor their limit orders. I therefore use the introduction of Trader+ as a positive shock to the investors’ limited monitoring capacities. This assertion is consistent with Hodge, Kennedy, and Maines (2004), who consider a search-facilitating technology [the eXtensible Business Reporting Language (XBRL) technology] introduced by the SEC in 2003 and state that “This simultaneous presentation helps users to evaluate items in relation to each other and to integrate the related information when making decisions.” In summary, the display of information of Trader+ should allow investors to monitor more efficiently their limit orders, which gives me the opportunity to test the empirical implications derived in Section 2. 4.2. Identification Strategy To test my empirical predictions, I use in this paper a DID methodology. The key identifying assumption in this setup (see Angrist and Pischke, 2008) is that trends in outcomes for treated and control investors would have been the same in the absence of treatment. This fundamental, but untestable, identifying assumption can nonetheless be evaluated by comparing graphically the trends in outcomes for both groups in multiple periods before and after the treatment exposure. I use the 10-year range of my dataset to support this “parallel trend” assumption. DID estimates can be obtained using an OLS regression, which has the advantage of allowing for a correct specification of standard errors (see Bertrand, Duflo, and Mullainathan, 2004). I therefore run the following regression for outcomes at the investor-day-trade level12:   Yi,t,k,s=α+β*Monitoringi,t+θt+γs+δi+ϵi,t,k,s, (1) Monitoringi,t is the variable of interest. It is a dummy variable that equals 1, for investor i, for all days t that follow his first trade executed through Trader+. This specification thus allows for a staggered entry into treatment at the investor level and is more precise than a standard DID framework that imposes the same treatment period for every treated investor. The regression specification includes a full range of individual fixed-effects δi, stock-fixed effects γs, and day fixed-effects θt every day from January 2002 to January 2005. Standard errors are robust to heteroscedasticity and are clustered at the individual level (White, 1980; Rogers, 1994). The fixed effects δi capture the differences between investors that are fixed over time (possibly correlated with the treatment variable), whereas the day-fixed effects capture the time factors that are common to both treated and control investors within a given trading day. Stock-fixed effects control for the fact that investors could possibly trade different stocks before and after the introduction of Trader+. The DID estimate is the coefficient β. It shows the effect of a more effective monitoring activity (induced by the aggregated display of information of Trader+) on the outcome Y. The DID framework is a powerful tool for causal inference but my estimates may be biased if investors are allowed to enter or exit the market after (or before) the treatment period. To control for attrition, I require that both treated and control investors have a stock holding position at least between January 2002 and January 2005, at least a single trade between January and the end of May 2003, and at least one trade after June 2003. That is, I follow the same population of traders over time. Treated investors have submitted at least one trade with Trader+ after June 2003, while control investors have not (and never will in my sample). 4.3. The Control Group Investors who switch to Trader+ may be different, in many ways, from other investors who never use the software. To obtain a control group that provides a credible counterfactual, I use a nearest-neighbor propensity score algorithm in order to find the best control investor for each treated investor in my sample. The propensity-score algorithm matches treated and control investors who share the same probability of switching to Trader+ after June 2003. To avoid any look-ahead bias, the estimated probability is computed using only pre-treatment data. Intuitively, this approach relies on finding the right set of covariates that determines the switch to treatment. In this respect, the seminal paper on online investors by Barber and Odean (2002) is very useful. The authors find that young active male traders with high incomes are more likely to switch to online trading than other traders. Those who switch also report more trading experience and a particular preference for speculative trades. Prior to going online, moreover, investors experienced unusually strong performance. Choi, Laibson, and Metrick (2002) study in a different setting the impact of a web-based trading channel on two large corporate 401(k) plans. They obtain that traders that are used to phone trading are less likely to try the web. While these results may not be completely generalizable to this study, as investors in my sample are already online investors, these papers emphasize that trading behavior before the treatment period can help identify the variables key to the selection process. I therefore compute various covariates according to investor demographics, general trading behavior, account size, and trading channel preferences, and I estimate, with pre-treatment data only, a cross-section logit regression where the dependent variable is one for a treated investor and zero otherwise. All these covariates are obtained using individual trading data from January 2003 to the end of May 2003, but my results are robust to the choice of different pre-event windows for the matching procedure. These covariates are grouped in four classes: demographics, trading behavior, account size, and trading channels. I use two demographic variables: female is a dummy variable that is 1 for females and 0 for males, and age corresponds to the investor’s age in 2003. The variables that capture investor trading behavior, in the pre-treatment period, are the individual number of orders executed (number of executed trades), the individual number of different asset classes traded (number of asset types), the individual median euro amount traded (median amount traded), the individual median daily-return on executed orders (median daily return), the individual percentage of orders that are limit orders (% limit orders), and the individual percentage of orders submitted with a margin account (% margin). As a simple proxy for trading horizon, I also include the proportion of round-trip trades, initiated in May 2003, that are reversed within 2 days (% round trip). The account size variables provide information about the total euro value of an investor’s common stock holdings on its spot-market account (in ten of thousands euros) in March 2003. Finally, trading channels variables provide, for each trading channel, the corresponding individual percentages of trades submitted by an investor through that channel. Results of the logit regression are given in Table II. What determines the switch to the software appears clearly: active traders with high past performance have a larger probability of switching to treatment. Furthermore, investors who use online trading channels, such as the web or the old computer software that has been available since 1999 (variable “Online+,” highly significant), are also more likely to use Trader+. Table II. Estimation of the probability of using Trader+ at least once after June 2003 (using pre-treatment data only) This table gives the estimates from a cross-section logistic regression where the dependent variable is a dummy variable that equals 1 if an investor is considered as treated: he has submitted at least one trade using Trader+ during the period June 2003–December 2010. Investors in my sample are required to have an open common stock position at least between 2002 and 2005, they have submitted at least one trade before June 2003, and at least one trade after June 2003. I compute various covariates according to investor demographics, general trading behavior, account size, and trading channel preferences in the pre-treatment period.   Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444    Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444  Table II. Estimation of the probability of using Trader+ at least once after June 2003 (using pre-treatment data only) This table gives the estimates from a cross-section logistic regression where the dependent variable is a dummy variable that equals 1 if an investor is considered as treated: he has submitted at least one trade using Trader+ during the period June 2003–December 2010. Investors in my sample are required to have an open common stock position at least between 2002 and 2005, they have submitted at least one trade before June 2003, and at least one trade after June 2003. I compute various covariates according to investor demographics, general trading behavior, account size, and trading channel preferences in the pre-treatment period.   Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444    Propensity score logit  Demographics: Female dummy  –0.361***  (0.079)  Demographics: Age in 2003  0.002  (0.002)  Trading behavior (January–May 2003): number of executed trades  0.005***  (0.001)  Trading behavior (January–May 2003): number of asset types  0.079***  (0.031)  Trading behavior (January–May 2003): median daily performance  0.056***  (0.016)  Trading behavior (January–May 2003): median amount traded  0.050***  (0.007)  Trading behavior (January–May 2003): % limit orders  –0.106  (0.065)  Trading behavior (January–May 2003): % margin  0.932***  (0.077)  Trading behavior (May 2003): % round trips  0.744***  (0.190)  Account size (March 2003): market value in EUR  0.037***  (0.006)  Use of trading channels (January–May 2003): % Phone+  0.483***  (0.207)  Use of trading channels (January–May 2003): % minitel  0.056  (0.213)  Use of trading channels (January–May 2003): % Online+  1.762***  (0.156)  Use of trading channels (January–May 2003): % Web  1.078***  (0.159)  Constant  –3.636***  (0.198)  Observations  13,444  The matching algorithm, based on the above logit estimation, performs well as Table III shows. Table III compares the mean covariate values between treated and matched control investors in my sample. It appears that all the variables that determine the investor propensity to use Trader+ (according to Table II) are very similar between treated and matched control investors. Indeed, treated and matched controls have a similar pre-treatment performance, submit on average the same number of trades between January and May 2003 (35 versus 29), the average order amount is almost equal (3200€ versus 2900€), and they manage their orders using the existing online trading tools (online+: 54% versus 56%, web: 34% versus 33%). Table III. Comparisons between treated investors and control investors obtained via propensity matching This table shows the average covariates between treated and control investors before and after the propensity matching procedures. All computations are performed with trading data coming from the pre-treatment period. All variables are defined in Table II. Standard deviations are between parentheses.   Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748    Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748  Table III. Comparisons between treated investors and control investors obtained via propensity matching This table shows the average covariates between treated and control investors before and after the propensity matching procedures. All computations are performed with trading data coming from the pre-treatment period. All variables are defined in Table II. Standard deviations are between parentheses.   Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748    Before matching   After matching   Raw control  Raw treated  Matched control  Matched treated  Demographics   Female  0.19  0.13  0.13  0.13  (0.40)  (0.33)  (0.33)  (0.33)   Age in 2003  48.2  47.6  47.8  47.6  (13.7)  (12.5)  (13.2)  (12.5)  Trading behavior (January–May 2003)   Number of executed trades  11.9  35.4  28.7  35.4  (32.7)  (71.4)  (71.6)  (71.4)   Number of asset types  1.48  1.66  1.66  1.66  (0.78)  (0.97)  (0.97)  (0.97)   Median amount  1.92  3.22  2.98  3.22  (2.99)  (7.54)  (5.85)  (7.54)   % Round trips  0.015  0.052  0.049  0.052  (0.094)  (0.18)  (0.17)  (0.18)   Median daily return  −1.54  −1.09  −1.03  −1.09  (2.05)  (1.54)  (1.63)  (1.54)   % Limit orders  0.58  0.59  0.60  0.59  (0.44)  (0.41)  (0.42)  (0.41)   % Margin  0.14  0.31  0.32  0.31  (0.31)  (0.40)  (0.41)  (0.40)  Account size in March 2003   Market value (EUR)  1.81  2.49  2.47  2.49  (3.58)  (4.42)  (6.20)  (4.42)  Use of trading channels (January–May 2003)   % Phone+  0.075  0.039  0.044  0.039  (0.25)  (0.18)  (0.19)  (0.18)   % Minitel  0.086  0.033  0.032  0.033  (0.27)  (0.17)  (0.16)  (0.17)   % Online+  0.31  0.54  0.56  0.54  (0.45)  (0.47)  (0.48)  (0.47)   % Web  0.41  0.34  0.33  0.34  (0.47)  (0.45)  (0.45)  (0.45)   Number of unique investors  11,696  1748  1748  1748  Last, in the Appendix, Figure A3 shows for both the treatment and the control group, the total number of orders (each year from 1999 to 2010) broken down by order type and trading channel. These figures summarize well the outcome of the matching procedure. For instance, one can see that from the figure that: (i) the number of orders submitted each year until 2003 is very similar between treated and control investors and (ii) both groups submit limit orders and use online tools most of the time. 5. Results 5.1. Trading Performance I test in this section my predictions regarding the effect of the new display on trading performance. Does a more effective display of market data (due to Trader+) allow investors to better monitor their limit orders and increase, in turn, their returns on those orders? Panel A in Table IV provides a first answer. For each trading channel, this table provides a statistical description of the distribution of investors’ intraday returns. I use all the orders submitted by both treated and control investors after June 2003. What emerges from the table is that returns obtained by traders using Trader+ are greater, in most quantiles, than returns achieved through other trading channels.