* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download 2. The Bonds and Bills Market - The University of Texas at Dallas
Survey
Document related concepts
Investment management wikipedia , lookup
Systemic risk wikipedia , lookup
International monetary systems wikipedia , lookup
Public finance wikipedia , lookup
Interest rate ceiling wikipedia , lookup
Financial economics wikipedia , lookup
Stock trader wikipedia , lookup
Trading room wikipedia , lookup
Financial literacy wikipedia , lookup
Global financial system wikipedia , lookup
Financial crisis wikipedia , lookup
Systemically important financial institution wikipedia , lookup
Transcript
Financial Networks and Trading in Bond Markets G. Geoffrey Booth Eli Broad Graduate School of Management, Michigan State University Umit G. Gurun School of Management, University of Texas at Dallas Harold H. Zhang School of Management, University of Texas at Dallas This paper examines the role of financial networks in influencing asset prices and trading performance. Consistent with theoretical studies on the role of communication networks in information dissemination, we posit that financial institutions with more extensive financial networks (global financial institutions) can more efficiently acquire and process information pertaining to asset trading in global financial markets than financial institutions with more limited financial networks (local financial institutions). The information advantage affords the global financial institutions more favorable transaction prices and better trading performance relative to their local counterparts. Using transaction-level Turkish government bond trading data, we find that financial institutions with global financial networks exhibit a stronger tendency to trade in the more liquid bonds. Further, they consistently trade at more favorable prices and enjoy better trading performance than local financial institutions. Together, these results suggest that global financial institutions have information advantages and benefit from bond trading in an open emerging market. Key Words: Financial Networks, Information, Bond Markets 1. Introduction Although it is well established that information moves security prices, how information flows through financial markets and is incorporated in the prices of financial assets is not as well understood. Traditional asset pricing approaches assume that individual agents behave anonymously with new information becoming known by all the agents in the market simultaneously, thereby making the information common knowledge. Information, however, can also gradually spread throughout the market by word-of-mouth and observational learning. Because of differences in institutional structures and traders’ information processing abilities, it is unlikely that information diffusion will be amorphous. Instead, information is likely to spread more rapidly within trading firms than between trading firms, not only because of the presence of an intra-firm network but also because of financial incentives provided to traders that are related to firm profitability. As a result, traditional approaches disregard the possibility that agent behavior (individually and collectively) may be influenced by a communication network. Models of trading dynamics recognize the presence of asymmetric information. The distinction between informed and uninformed traders leads to a number of useful insights. For instance, informed traders tend to respond more quickly to news, tend to trade in more liquid markets, and tend to show better performance than uninformed traders. Yet it is not entirely clear who the informed traders are or how they become informed. In this regard several empirical studies show that individuals who reside and work in the same location tend to make similar financial decisions, which suggests the presence of internal group communication.1 The idea is that traders who are spatially and electronically close are exposed to similar information that is diffused via networks within the same group once the information is received by one or more of the traders. For example, investors tend to invest locally (e.g., Grinblatt and Keloharju, 2001; Ivković and Weisbenner, 2005; Massa and Simonov, 2006), as do professional money managers (Coval and Moskowitz, 2001). Investors also tend to follow their colleagues and neighbors (e.g., Dufflo and Saez, 2003; Hong, Kubik, and Stein, 2005; Ivković and Weisbenner, 2007). Moreover, Hong, Kubik, and Stein (2004) develop a model in which stock market participation is influenced by social interaction, and Xia shows (2007) that the influence of information on transaction prices depends on the structure of the network. 1 1 We address this gap in the literature by comparing the information networks of financial institutions that trade bonds in an emerging market. We select this type of market because, as Biais and Green (2005) point out, bond markets often provide little pre-trading transparency, which creates opportunities for informed traders to take advantage of their superior information. We classify the sample institutions as those that have offices in the local economy only and those that have offices both in the local economy and in major bond trading markets such as New York City and London. We define a financial network to be a set of offices that are linked together by an electronic communication system. Consistent with the implications of theoretical studies on the role of networks in information dissemination, we posit that a financial institution with a global (more extensive) financial network can more efficiently acquire and process information related to global financial market movements than an institution with only a local (less extensive) network. This information advantage is expected to allow global financial institutions to trade more nimbly and perform better relative to local financial institutions. We test our hypotheses by empirically investigating day trading in the government bonds of Turkey, an open emerging market in which financial institutions with different scopes of financial networks are permitted to participate with limited government interference. We find that financial institutions with global financial networks trade at more favorable prices and demonstrate better performance in the Turkish government bond market relative to financial institutions with only local networks. Our empirical findings thus support the conjecture that financial institutions with a more extensive financial network such as those with a global network exhibit an information advantage and benefit from utilizing their superior information relative to financial institutions with a less extensive financial network such as those with only a local network. 2. The Bonds and Bills Market 2.1 The Market 2 Turkey’s public bond market, the Bonds and Bills Market, is an important investment and trading venue for financial institutions. Using total market capitalization standardized by GDP as a measure of importance, according to World Bank Database on Financial Development and Structure, Turkey ranked 9th out of 30 major world bond markets, with its bond market being 2.3 times as large as its equity market (see Beck et al. (2000) for details of this database). Almost every month, the Turkish Treasury auctions bonds with maturities ranging from one month to 10 years. After the primary market allocation, these bonds are traded on an automated secondary market, the Bonds and Bills Market. This market also facilitates repurchase agreements, but these transactions are executed separately and excluded from our analysis. The institutions that are authorized to trade on the Bonds and Bills Market are Istanbul Stock Exchange (ISE) member banks and member brokerage houses. These financial institutions typically trade on their own accounts. Sometimes they fill retail buy orders from their inventory, but if their inventory is insufficient they may have to go to market to meet demand. Each institution employs multiple traders who form an information network. They are in constant contact with each other throughout the trading day, permitting them to be better aware of the local buy and sell order flow. For instance, it is not uncommon for traders to inform the participants in their network that they have learned that a particular financial institution is a net buyer today or that another financial institution is trying to liquidate a sizeable position. Some institutions have home offices in multiple markets while others have branches; such organizational structures create multi-market trader networks that facilitate the transmission of information relevant to the local market. Bond market participants are a diverse mix of small and large Turkish financial institutions and large international financial institutions. These institutions have different arrangements to disseminating information. International banks, for instance, have their bond trading floors connected by a “hoot”. Nowadays a “hoot” refers to an electronic communication system, but originally it was a device devoted to a single trading floor. “Hoot” transmissions tend to flow from New York and London to other markets. 3 In contrast, bond traders of Turkish banks (especially large Turkish banks) gather information by making phone calls to fellow bond traders in overseas financial centers. Of course, information is also available to all traders whose firms have access to public information networks such as Reuters, Bloomberg, and similar providers. Different financial institutions, however, may still have different information processing capabilities, which may lead to differences in interpretation of publicly released information and in turn trading performance. 2.2 The Trading System The Bonds and Bills Market is a limit order book market that uses an electronic system to match, administer, and report transactions. The market operates in two sessions: from 9:30 a.m. to 12:00 noon, and from 1:00 p.m. to 5:00 p.m. Bonds with same-day and next-day settlement trade until 2:00 p.m., which is the settlement time for the day; between 2:00 p.m. and closing, only bonds with next-day settlement trade. Thus, the number of transactions declines noticeably after 2:00 p.m. Orders are processed and executed according to price and time priority in an automated trading system. The ISE uses an order-driven electronic continuous market with no intermediary such as a market maker and no floor brokers. The majority of the orders are routed electronically via member firms to the central limit order book through an order processing system that does not require any re-entry by the member firms. In very rare cases, member firms call representatives at the exchange to have their orders entered for them. Member firms can execute market orders and limit orders, as well as orders that require further conditions for execution (e.g., Fill-or-kill and Stop-loss). Member firms are not allowed to enter orders when the market is not open; however, they are allowed to withdraw their existing orders. It is not unusual to see traders filling out their order screen prior to opening time and submitting multiple orders at the open. Price information on the 20 best bids and offers is continuously available to member firms. The system does not display quantity demanded or offered at each of these prices, but past transactions can be viewed by all members. The tick size is 1 Turkish lira (TL) for a 100,000 TL face value bond, with 4 minimum (maximum) order size set to 100,000 TL (10 million TL); there exists no formal upstairs market for block trades. An incoming market order is executed automatically against the best limit orders in the book. Execution within the inside quotes is allowed. Once a transaction takes place, a confirmation notice is sent to the parties involved in the transaction. The other market participants do not learn the identities of the parties, but they do observe that a transaction took place at a specific price and quantity. All information pertaining to price, yield, and volume of best orders as well as details of the last transaction and a summary of all transactions are disseminated to data vendors, including Bloomberg, Reuters, and some local firms, immediately after each transaction. In addition, all trades are reported to the clearing organization, the ISE Settlement and Custody Bank Inc. (Takas Bank), at the end of the day to facilitate bookkeeping. We do not have information on what percentage of the transactions take place in ISE; however, anecdotal evidence suggests that ISE consolidates more than 97% of the turnover value of the Bonds and Bills Market’s transactions. The remaining portion is captured by OTC markets. The Turkish government typically plays a minimal role in the Bonds and Mills Market. Nevertheless, after the 2001 banking crisis, the Undersecretariat of the Treasury initiated a primary dealer system that requires some market’s members that participate in the primary market auction to provide liquidity by quoting a bid and an ask (not necessarily the best bid or ask) in the secondary market. The quotes are identified as being given by a primary dealer. The rationale for this innovation is that these members would accommodate liquidity needs that may arise during times of crisis, although anecdotal evidence indicates that such action by the primary dealers has yet to occur. The number of primary dealer members (typically between eight and 14) and its composition (foreign or domestic) is determined by the Undersecretariat. In 2006, the primary dealer system consisted of 12 primary dealers. The most recently issued bond is designated as the active (or benchmark) bond. 3. Data and Summary Statistics 5 Our sample consists of 1,716,917 tick-by-tick time-stamped transactions beginning May 1, 2001 and ending June 15, 2005 (1,039 trading days). For each transaction, we have detailed information on the time of order placed and filled, transaction price, and trade size for 177 Turkish lira-denominated Treasury bills and notes. More important, our data set also contains the identities of the traders on both sides of a transaction from their unique identification code. The starting date of the sample is two months after the Turkish financial crisis attributed to liquidity shortages in the banking system that ended in February 2001. Data availability dictates the sample’s ending date. One hundred seventy distinct financial institutions participated in the Bonds and Bills Market. We classify these into local versus global financial institutions. A financial institution is classified as “global” if it has branches or offices in major financial markets outside Turkey; otherwise, it is classified as “local”. Based on information collected from the Istanbul Stock Exchange and data from the Turkish Bank Association (http://www.tbb.org.tr/net/subeler) on overseas branches and offices, we classify 146 as local financial institutions and 24 as global financial institutions. We use the ISE asset size categories to divide local financial institutions into 116 small and 30 large financial institutions.2 The roster of global financial institutions includes large foreign banks such as Deutsche Bank, Citibank, and JP Morgan Chase as well as large domestic financial institutions such as Yapı Kredi Bankası A.Ş., Vakıflar Bankası A.Ş., and Akbank A.Ş. The foreign global financial institutions have home offices in New York and throughout Europe, with the latter including offices in London, Amsterdam, Paris, and Frankfurt to name but a few. Six of the global financial institutions are Turkish and together account for more than 22% of the global financial institutions’ participation in the Turkish We also divide financial institutions according to their trading volume using their past month’s transactions and find that the two proxies for size are highly correlated. This is not surprising since anecdotal evidence suggests that large financial institutions participate in treasury auctions more frequently and have the ability to obtain more bonds. 2 6 market. These financial institutions have branch and liaison offices not only in New York and Europe but also in Bahrain, Tokyo, and Moscow.3 Using the trader identification code, our global versus local classification, and ISE asset size categories, we classify each transaction as being made by a local small, local large, or global financial institution. Table 1 reports selective summary statistics for our data. Panel A shows the average daily trading volume in U.S. dollars (USD) of the local small, local large, and global financial institutions sorted by seller- and buyer-initiated trades and their counterparties.4 Of the USD 640 million of total daily volume, trading between local large and global financial institutions is the highest, with average daily trading volume reaching USD 133.8 million for seller-initiated trades and USD 139 million for buyerinitiated trades. Trading among global financial institutions is the second highest, with seller- and buyerinitiated daily trading volume of USD 76.5 million and USD 95.6 million, respectively. In Panel B, we report the average size and number of tick-by-tick transactions for local small, local large, and global financial institutions without reference to the initiator. The average volume per transaction is highest, at USD 0.9 million, between local large and global financial institutions. The transaction volume for trades among global financial institutions is the second highest at USD 0.65 million, while trades between local small and global financial institutions ranks third. In terms of the number of transactions, trades between local large and global financial institutions account for 36.9% of all trades, trades among local large financial institutions account for 21.7%, and trades among global financial institutions account for 16.1%. 4. Empirical Analysis and Results Our analysis consists of three distinct but related analyses. We first examine whether global financial institutions are more likely to trade more liquid bonds, a practice that allows them to more easily 3 We classify as local financial institutions several small Turkish banks with offices in nearby nations that provide primarily retail banking services (deposits and remittances). We performed our analyses with these banks classified as global financial institutions. Our results were not affected. 4 The USD volume is obtained by using the daily closing TL/USD exchange rate for that day. Turkey dropped six zeros from its currency at the end of 2004. We incorporate this change in our calculations. During the sample period, the average exchange rate was TL 1.46 = USD 1 with a standard deviation of TL 0.11. 7 hide informed strategic trades. We then investigate whether global financial institutions consistently transact at more favorable prices. Finally, we explore the day-trading profitability by different financial institutions. The evidence supports the notion that financial institutions with global information networks are more informed than those institutions with only local networks. 4.1 Strategic Trading Chowdry and Nanda (1991) show that informed investors tend to trade in more liquid markets, presumably because of their need to hide strategic transactions that convey information. Combining this observation with Pasquariello and Vega’s (2007) suggestion that the most liquid bonds are the ones that are most recently issued, we hypothesize that if global financial institutions have an informational advantage over local financial institutions, they tend to trade active bonds, i.e., the most recently issued bonds, relative to the other bonds, which we refer to as passive bonds. In Table 2 we report different financial institutions’ average daily transactions—measured by volume of trade (in U.S. dollars) and number of trades—in active bonds and passive bonds and the daily ratio of trading in active bonds to passive bonds. Consistent with Pasquariello and Vega (2007), active bonds have the highest transaction volume/number compared to the rest of the bonds. As a robustness check, we calculate the transaction volume at day t-1 and designate the bond with highest score as the active bond for the next trading day. Our conclusions are not affected by this alternative definition of “active” bond. Next, we split the sample transactions into cases where global financial institutions are on both sides of the transaction, a global financial institution is on one side and a local financial institution on the other, and local financial institutions are on both sides. The results for trading volume and number of transactions are similar, although those for trading volume are more compelling. The ratio of active to passive bonds is greatest in terms of trading volume (in U.S. dollars) when global financial institutions are on both sides of the transaction and smallest when local financial institutions are on both sides of the trade. For example, the ratio of active to passive bonds in terms of trading volume is 2.33 for global 8 financial institutions and 1.38 for local financial institutions. The difference in the ratios of active to passive bonds for global/global and local/local financial institutions is significant (p = 0.000), suggesting that global financial institutions have a stronger preference for trading active bonds than local financial institutions. The results based on the number of transactions are similar. While the ratio of active to passive bonds for trades involving only global financial institutions is 1.34, the corresponding ratio for only local financial institutions is 1.19. The difference is again significant (p = 0.000). Our empirical evidence on preferences of global and local financial institutions with respect to active versus passive bonds thus supports the conjecture that global financial institutions have an information advantage over local financial institutions. This evidence may also be consistent, however, with alternative explanations such as the liquidity concerns of global financial institutions trading large quantities. 4.2 Pricing Advantage We define a financial institution as better “informed” if it consistently buys (or sells) before other financial institutions before the market price rises (or declines). Similar to Massa and Simonov (2003), we determine the degree of informativeness of different financial institutions by examining the delayed price changes in the same bond in a D-minute interval following each transaction initiated by the financial institution. For a given bond k, we estimate the following time-series regression to identify the delayed price changes associated with trades initiated by investor i: Pik 0 ik Tikj k I k ik , D (1) D where Pik is the delayed price change of bond k in D minutes following the trade, Ik is a binary indicator that takes a value of one for bond k (the bond fixed effect), and Tik is the signed transaction volume in million Turkish liras (positive for purchases and negative for sales). The notation for the subscript indicates that investor i initiated trade of bond k. θik measures the effect on the delayed price 9 D D change ( Pik ) of a trade in bond k initiated by investor i. For Pik to be defined, there must be a transaction in bond k within D minutes after the trade. If there is more than one trade, we use the last D transaction within the interval, and if no transaction takes place, we set Pik = 0. We use a 10-minute time interval (D = 10) as the baseline case in our regression to identify the price impact of informed trades. Given the total number of trades and trading days in our sample, on average there are about 30 trades per 10-minute interval. We also perform our analysis using 20- and 30minute time intervals, which increases the average number of trades per interval to approximately 60 and 90, respectively. This analysis allows us to assess how quickly information dissipates in this market. Excluding the top and bottom deciles of trades from each day to control for the potential impact of extreme trading does not change our results. We estimate equation (1) for each financial institution and each bond. The coefficient θik represents the degree of informativeness of financial institution i, which initiated the trade on bond k. A positive θik means that the trade initiator, financial institution i, consistently bought (sold) from (to) other financial institutions before an increase (decrease) in the price of bond k. A significant value for θik implies that financial institution i is more informed than its trading counterparts with respect to bond k. The larger the coefficient’s value, the greater the informational content of the trade. To examine how the relative informativeness of trades varies with type of financial institution, we estimate the following equation that relates trade informativeness to financial institutions characteristics: ik 0 1 SDUM ik 2 GDUM ik it , (2) where SDUM is a dummy variable that takes a value of one for large financial institutions and zero otherwise, and GDUM is a dummy variable that takes a value of one for global financial institutions and zero otherwise. The local small financial institution is our benchmark group. Since all global financial institutions are large in size, the size dummy SDUM takes a value of one if a financial institution is a global financial institution. Therefore, the coefficient on GDUM measures the additional effect associated 10 with being a global financial institution beyond the size effect. We estimate equation (2) using robust clustered standard error regression. In equation (2), we include all θs estimated in equation (1). Limiting observations to θs that are statically significant at p = 0.000 does not change our results. We include bond fixed effect to control for possible unobserved heterogeneity among different bonds. Table 3 presents the estimation results of equation (2). Panel A presents the results for the full sample. For the baseline 10-minute time interval, the estimated coefficient on the size effect SDUM suggests that for every one million Turkish lira that large financial institutions trade, the bond price changes by TL 0.14 more than if the one million Turkish lira were traded by local small financial institutions. Given the average bond price of TL 78 for the entire sample, the estimated bond price change suggests the large financial institutions enjoy an 18 basis point price advantage. Using the average exchange rate during the sample period, 1.46 TL/USD, this effect corresponds to 10 U.S. cents. The coefficient estimate on GDUM is also positive and statistically significant. Specifically, for every one million Turkish lira traded by global financial institutions, the price changes TL 0.09 or 6 U.S. cents more in a 10-minute interval than if the one million Turkish lira were traded by a typical large financial institution. Global financial institutions thus enjoy an additional price advantage of 11 basis points. These results indicate that, on average, global financial institutions enjoy better pricing in bond trading than local financial institutions, which is consistent with having an information advantage over local financial institutions. For the 30-minute interval, the coefficient estimate on GDUM is positive but insignificant. This suggests that the information advantage afforded global financial institutions is shortlived and consistent with the type of market where the information is mostly related to order flows on global financial markets but not to fundamentals. The coefficient estimate on SDUM, however, remains positive and highly significant, indicating that large financial institutions benefit from their relative size when the trading interval is extended to 30 minutes. To evaluate whether our findings are due to the dichotomy of foreign versus domestic financial institutions, we perform the regression analysis using only observations on Turkish financial institutions. 11 Panel B reports the estimation results excluding foreign financial institutions. The coefficient estimates on SDUM and GDUM are very similar to those based on the entire sample. The pricing benefit is slightly higher at TL 0.12 (8 U.S. cents) for global financial institutions than for large financial institutions. The information advantage afforded global financial institutions disappears after 30 minutes, as when using the full sample. Further, the estimated size effect is also slightly larger than that in the full sample case when the pricing impact is measured 30 minutes after trade. These results indicate that our findings on favorable pricing in Turkish bond trading are more likely attributed to global financial institutions’ more extensive network (relative to local financial institutions) than to the foreign versus domestic dichotomy. Panel C reports the estimation results using the subsample of liquid bonds. We identify liquid bonds based on the Amihud (2002) liquidity measure, calculated for each bond as the absolute value of daily returns divided by daily dollar volume for the previous day, with low (below-median) Amihud measure bonds comprising the liquid bonds. Panel D presents the results for the liquid bond subsample excluding foreign financial institutions. In both panels the coefficient estimates on SDUM and GDUM remain quantitatively similar to the findings based on the full sample. This suggests that more favorable pricing by global financial institutions is unlikely to be attributed to differences in bond liquidity. Finally, in unreported tables, we replicate our analysis by excluding financial institutions that are designated as primary dealers. As explained above, primary dealers are required to quote a bid and ask price (though not necessarily the best bid and ask) in the limit order book of some bonds in the secondary market. If these primary dealers systematically include non-global members, then it is possible that our results are driven by liquidity provision required from such members. Our results are quantitatively similar when primary dealers are excluded from the sample. This is in line with the Turkish market not experiencing a buying/selling frenzy that requires the primary dealers to act as liquidity providers. 4.3 Day-Trading Profitability We next investigate how different financial institutions perform in terms of day-trading profitability in the bond market. To do so, we construct daily trading cycles for each bond k by financial 12 institution i. We assume that the initial the inventory is zero and that purchases increase inventory while sales decrease it. As the day unfolds the inventory level will typically hit zero several times. The time between the adjacent zeros is considered a cycle and the profits associated with the transactions in each cycle are calculated using the buy and sell prices and corresponding trade volumes. Using information on the percentage profit and funds invested in trading cycles for all bonds for a given trader i on day t, we construct the day-trading profitability measure for each investor i on day t as follows: PRFit K C k 1 c 1 K Profit Invitkc . C k 1 itkc (3) Invitkc c 1 PRF is the weighted average percentage profit per trading cycle for investor i on day t. We ignore direct transaction costs in our calculation because they are very small at approximately 0.001%. Moreover, since we are not privy to tick-by-tick TL/USD exchange rates, we calculate profits per cycle in Turkish lira. Although lira and dollar profits may differ, there is no compelling reason for this difference to create a short-term systematic bias. In Table 4, we report the PRF of local small, local large, and global financial institutions. All investor groups earn on average a negative profit on their day-trading activities. Trades attributed to global financial institutions, however, have a smaller average loss (-0.02%) than the trades of local large financial institutions (-0.03%) or local small financial institutions (-0.04%). Overall, 33.3% of day trades produce a positive profit. The fraction of day trades with positive PRF is 37.2% for global financial institutions, 34.4% for local large financial institutions, and 29.8% for local small financial institutions. 5 5 Analyzing the Taiwanese stock market, Barber et al. (2006) find that individual day traders routinely incur losses, while institutional day traders, on average, profit. The composition of trades in this market is dramatically different from the Bond and Bills Market, however. For instance, in the Turkish market all of the traders are institutions, while institutions only account for 10.5% of the Taiwanese trade value. 13 Although on average day-trading profitability is negative for all traders, financial institutions may still make a net profit using other investment strategies, such as buy-and-hold, on certain bonds. In our sample, the buy-and-hold strategy is used by both financial institutions that do not day trade (48% of the traders do not participate in day-trading activities) and by day traders as part of their overall trading strategy. Lending support to this view, day trades make up only 35% of trading volume (in U.S. dollars). In column 1 of Panel B, we report that global financial institutions earn significantly higher day-trading profits than local large financial institutions (p = 0.000), which in turn earn higher day-trading profits than local small financial institutions (p = 0.000). Barber et al. (2004) suggest that frequent day traders perform better than infrequent day traders. In addition, if investors are overconfident in their trading skills, prior profitability will likely influence their participation decisions in day trading. It is also plausible that traders learn from prior trading, adjusting their trading strategies based on prior day-trading successes and market conditions. Because a large proportion of financial institutions do not participate in day trading, we conduct joint estimation of the participation decision and day-trading profitability using Heckman’s (1979) self-selection model, where we specify the participation decision as a function of the trader’s prior day-trading successes and market conditions, as proxied by interest rate volatility. Specifically, we estimate the following model: PRFit 0 1 SDUM i 2 GDUM i 3 LPRFit 4 LPRFit * SDUM i 5 LPRFit * GDUM i 6VOLINTt 1 it PARTit 1 LPRFit 2VOLINTt 1 uit 0, (4) where PRFit is the percentage day-trading profit of trader i on day t, LPRFit is the percentage profit of previous day-trading, PARTit is the participation value, which takes value of one if financial institution i participates in day-trading on day t, and VOLINTt-1 is the standard deviation of the interest rate on the 14 previous day using 30-minute observations. The interaction terms capture the impact of past trading performance and market conditions on the day-trading profitability of various financial institutions. 6 Table 5 reports the day-trading profitability results. Column (1) shows the estimation results for the Heckman model for all bonds traded during the sample period. The regression results suggest that, controlling for interest rate volatility and previous day-trading profitability, large financial institutions earn 0.005 percentage point higher profits than local small financial institutions on day-trading. Moreover, global financial institutions earn day-trading profits that are 0.007 percentage points higher than those of typical large financial institutions. Both estimates are significant (p = 0.000). The coefficient estimate on lagged day-trading profitability is positive and significant (p = 0.000), indicating some persistence in day-trading profitability. Consistent with the participation of more informed financial institutions in day trading, the lagged percentage profit in day-trading negatively affects the participation decision. A plausible reason for this finding is that an informed financial institution that has just traded may not immediately trade again unless the new information is received immediately after the previous trading cycle. Interest rate volatility has a significant negative effect (p = 0.000) on day-trading profitability. This implies that it is more difficult to profit from day-trading when the interest rate is more volatile. However, higher interest rate volatility has a significantly positive effect (p = 0.000) on the decision to participate in day-trading on the following day. One explanation is that informed financial institutions are more likely to day trade and higher volatility reflects more informed traders in the market. 7 6 Our conclusions remain unaffected if we use alternative selection equation specifications such as including (1) constant, (2) squared lag profitability, and (2) more lags of prior volatility. 7 One difficulty with measuring trading cycles on a daily basis is that a financial institution may have non-zero inventories of certain bonds at the end of the day. We handle this issue by excluding the ending cycle for these bonds. In addition, we conduct a robustness exercise by creating weekly trading cycles in the same manner that we constructed daily trading cycles. The signs and statistical significance of key variables’ weekly estimates are the same as those using the daily trading cycles. The only exception is interest rate volatility, which is no longer statistically significant at conventional levels. This suggests that our findings on the trading profitability of different types of financial institutions are robust to extending the trading cycle to a weekly basis. As anticipated, the magnitudes of coefficient estimates for key variables are larger for the weekly trading cycles than for the daily trading cycles. 15 To test whether different financial institutions perform differently for different realizations of past day-trading performance, we introduce the interaction terms LPRF*SDUM and LPRF*GDUM in our regression analysis. Column (2) presents the regression results. The coefficient on LPRF*SDUM is significantly negative (p = 0.000). This suggests that the persistence in day-trading profitability is weaker for large financial institutions than for local small financial institutions. However, the weaker persistence in day-trading profitability does not apply to global financial institutions. Balduzzi, Elton, and Green (2001) find that intermediate- and long-term bonds are more responsive to macroeconomic news than short-term bonds. We therefore also examine whether daytrading profitability behaves differently between short- and long-term bonds. Columns (3) and (4) in Table 5 report the estimation results for short-term bonds (remaining maturity is one year or less) while columns (5) and (6) show the results for long-term bonds (remaining maturity is longer than one year). For the short-term bonds, we find that consistent with the full sample, both SDUM and GDUM have a significantly positive effect. This indicates that both network size and a global network contribute to daytrading profitability on short-term bonds. Further, the coefficient estimate increases from 0.005 to 0.008 for SDUM and from 0.007 to 0.008 for GDUM, suggesting stronger effects on short-term bonds from size and a global network. The persistence in day-trading profitability is also stronger for short-term bonds, whereas the difference in the effect of interest rate volatility is very small. In contrast, for long-term bonds, neither a network’s size nor the existence of a global network has a significant effect on daytrading profitability. Further, interest rate volatility has a substantially larger negative effect on daytrading profitability for long bonds than for short bonds, consistent with the fact that long-term bonds are more sensitive to changes in the interest rate, which makes it more difficult to profit from day trading. Additional results on long-term bonds show that the coefficients on the interaction terms indicate that large financial institutions again exhibit weaker performance persistence in day-trading profitability than local small financial institutions, and that a global financial network no longer has any additional effect on the persistence on day-trading profitability. 16 Next, we test whether media emphasis affects day-trading profitability across different types of financial institutions, following Morris and Shin (2002) who suggests that bond yields react most to news emphasized by the media. Specifically, we estimate equation (4) after excluding Turkish scheduled macroeconomic announcements collected from Bloomberg. The news items include announcements related to inflation, gross national product, industrial production, current account, trade balance, unemployment, and capacity utilization. 8 The results, reported in Table 6, show that the coefficient estimates based on the no-domestic macroeconomic news subsample are qualitatively similar to those reported in Table 5. In particular, the effects of SDUM and GDUM remain stronger for the short-term bonds than for the long-term bonds. A question that remains is whether the differences in day-trading profitability are caused by differences in the liquidity of bonds traded by different financial institutions. To address this issue, we examine the day-trading profitability of different financial institutions of bonds with different liquidity. We partition bonds into liquid and illiquid bonds based on two different measures: (1) the active versus passive classification used in Section 4.1, with the active bonds classified as liquid bonds; and (2) the Amihud (2002) liquidity measure used in Section 4.2, with low (below-median) Amihud measure bonds taken to be the liquid bonds. Table 7 reports the estimation results of equation (4) for different groups of bonds with different liquidity. Our results suggest that for both liquid and illiquid bonds, the estimated coefficient on GDUM is positive and statistically significant. The estimates’ magnitudes are also comparable for liquid and illiquid bonds, and the results suggest similar persistence in day-trading profitability for bonds with different liquidity. Finally, just as we test whether our results on pricing are due to differences between foreign versus domestic financial institutions rather than global versus local networks, we analyze whether our day-trading profitability results correspond entirely to domestic financial institutions. To accomplish this, we estimate equation (4) using day trades of 146 local financial institutions and the six Turkish global We also exclude dates on which Turkey’s sovereign rating is changed by the three principal sovereign rating agencies (Standard & Poor’s (S&P), Moody’s Investor Services, and Fitch Investor Service). We obtain similar results. 8 17 financial institutions. Table 8 shows that the results based on domestic financial institutions offer similar conclusions as the results based on both foreign and domestic financial institutions reported in Table 5. For instance, again, large financial institutions earn higher day-trading profits than local small financial institutions, and global financial institutions earn higher day-trading profits than other financial institutions, though the magnitude of outperformance for global financial institutions is slightly smaller here than with foreign financial institutions included (0.004 percentage points versus 0.007 percentage points). Also consistent with the findings reported in Table 5, day-trading profitability exhibits weaker persistence for large financial institutions than for large financial institutions and little persistence for global financial institutions. Interest rate volatility remains negative and highly significant, indicating that a more volatile market makes it more difficult to profit from day-trading. 5. Concluding Remarks We empirically investigate the role of financial networks in influencing asset prices and trading performance. Consistent with theoretical studies on the role of communication networks on information dissemination, we posit that financial institutions with a more extensive financial network (global financial institutions) can more efficiently acquire and process information than institutions with only a local network (local financial institutions). We test this hypothesis by examining the information advantage and associated benefits of global versus local financial institutions in trading government bonds in an open emerging market. Using transaction-level data from the Turkish Bonds and Bills Market, we find that global financial institutions have a greater propensity to trade in the more liquid part of the market than do their local counterparts, consistent with the need of informed investors to strategically hide trades that might convey information. In addition, global financial institutions consistently buy (or sell) prior to market price increases (or decreases), that is, enjoy more favorable pricing in bond trading, relative to local financial institutions (both large and small), lending support to financial institutions with a global network having an information advantage. The pricing advantage of global financial institutions disappears 18 quickly, however, suggesting that the information advantage may be related to the order flows in global financial markets and not to fundamentals. Finally, analysis of day-trading profitability reveals that global financial institutions consistently and significantly outperform local financial institutions (both large and small), providing further support to the information advantage hypothesis. These findings are robust to numerous alternative samples and definitions. Most importantly, they continue to hold when we restrict our sample to only domestic financial institutions and to bonds with different liquidity, which suggests that our findings are unlikely to be due to the foreign versus domestic dichotomy or to liquidity. Overall our empirical analysis provides strong evidence that in a globally integrated financial market, financial institutions with more extensive global information networks can more efficiently acquire and process information such as order flows than local financial institutions can, and this information advantage allows the global financial institutions to enjoy more favorable pricing and consistently outperform their local counterparts. 19 References Amihud, Y., 2002. Illiquidity and stock returns: cross-section and time-series effects, Journal of Financial Markets 5 31–56. Balduzzi, P., E. Elton, C. Green, 2001. Economic News and Bond Prices: Evidence from the U.S. Treasury Market, Journal of Financial and Quantitative Analysis 36 523 – 543. Barber, B.M., Y.-T. Lee, Y.-J. Liu, T. Odean, 2004. Do Individual Day Traders Make Money? Evidence from Taiwan, Working Paper, The University of California - Berkeley, Berkeley, CA. Barber, B.M., Y.-T. Lee, Y.-J. Liu, T. Odean, 2006. Just How Much Do Individual Investors Lose by Trading, Working Paper, The University of California - Berkeley, Berkeley, CA. Beck, T., A. Demirgüç-Kunt, R. Levine, (2000), A New Database on Financial Development and Structure, World Bank Economic Review 14 597-605. Biais, B.M., R. Green, 2005. The Microstructure of the Bond Market in the 20th Century, Working Paper, Carnegie Mellon University, Pittsburgh, PA. Chowdhry, B., V. Nanda, 1991. Multimarket trading and market liquidity, Review of Financial Studies 4 483 – 511. Coval, J.D., T.J. Moskowitz, 2001. The geography of investment: Informed trading and asset prices, Journal of Political Economy 109 811 – 841. Duflo, E., E. Saez, 2003. The role of information and social interactions in retirement plan decisions: Evidence from a randomized experiment, Quarterly Journal of Public Economics 118 815 – 142. Grinblatt, M., M. Keloharju, 2001. How distance, language and culture influence stockholdings and trades, Journal of Finance 56 1053 – 1073. Heckman, J.J., 1979, Sample selection bias as a specification error, Econometrica 47 153 – 162. 20 Hong, H., J.D. Kubik, J.C. Stein, 2004. Social interaction and stock market participation, Journal of Finance 59 137 – 163. Hong, H., J.D. Kubik, J.C. Stein, 2005. Thy neighbor’s portfolio: Word-of-mouth effects in the holdings and trades of money managers, Journal of Finance 60 2801 – 2824. Ivković, Z., S. Weisbenner, 2005. Local does as local is: Information content of the geography of individual investors’ common stock investments, Journal of Finance 60 267 – 306. Ivković, Z., S. Weisbenner, 2007. Information diffusion effects in individual investors’ common stock purchases: Covet thy neighbors’ investment choices, Review of Financial Studies 20 1327 – 1357. Massa, M., A. Simonov, 2003. Reputation and interdealer trading: A microstructure analysis of the Treasury bond market, Journal of Financial Markets 6 99 – 141. Massa, M., A. Simonov, 2006. Hedging, familiarity and portfolio choice, Review of Financial Studies 19, 633 – 685. Morris, S., H. Shin, 2002. Social value of public information, American Economic Review 92 1521 – 1534. Pasquariello, P., C. Vega, 2007. Informed and strategic order flow in the bond markets, Review of Financial Studies 20 1975 – 2019. Xia, C., 2007. Communication and Confidence in Financial Networks, Working Paper, University of Minnesota, Minneapolis, MN. 21 Table 1 Transactions by Financial Institution Type This table presents summary statistics on transactions by type of financial institution (local small, local large, and global). Panel A reports the daily mean U.S. dollar trading volume when we focus attention on buyer- and seller-initiated trades. Panel B presents the mean per-transaction U.S. dollar volume and the total number of transactions. The sample period is May 1, 2001 to June 15, 2005 (1,039 trading days). Panel A. Daily Trading Volume (in million U.S. dollars) Seller Buyer Local Large 8.469 60.605 67.383 Global Local Small Local Large Global Seller Initiated Trades Local Small 1.939 9.473 9.730 Local Large 8.097 63.297 65.999 Global Local Small Local Large Global Buyer Initiated Trades Local Small 1.995 8.578 8.635 8.500 66.414 76.527 9.402 72.974 95.600 Panel B. Transaction Volume and Number Local Small Local Large Global Mean Per-Transaction Volume (in million U.S. dollars) Local Small Local Large Global 0.078 0.160 0.157 0.346 0.236 0.435 0.248 0.460 0.647 Local Small Local Large Global Number of Transactions Local Small Local Large Global 52,093 107,587 119,444 372,225 80,709 318,635 74,877 315,117 276,230 22 Table 2 Relative Trading Activity in Active and Passive Bonds This table summarizes active and passive bond transactions by volume in million U.S. dollars (Panel A) and number of trades (Panel B). We present our measures for transactions involving global financial institutions on both sides of the trade (buy and sell), global financial institutions on at least one side of the trade (buy or sell), and local financial institutions on both sides of the trade. The entries corresponding to “Active” and “Passive” rows represent the volume or number of daily transactions of Active and Passive bonds, whereas those of “Ratio” refer to the average daily ratio of active to passive bond transactions. An active bond is the most recently issued bond. p-values are reported for the null hypothesis that the sample mean of one of the categories involving global financial institutions is greater than the mean of the category involving only local financial institutions. The percentage of days Ratio>Local/Local Ratio measures the portion of days that the Global/Global or Global/Local categories are greater than the Local/Local category. Panel A: Volume of Transactions Active Passive Local/Local 85.77 76.64 Global/Local 162.30 146.79 Global/Global 96.42 75.78 1.38 0.04 1.38 0.04 2.33 0.30 (0.554) 0.14 53.99 (0.000) 3.10 55.73 Local/Local 315.06 311.84 Global/Local 418.66 341.45 Global/Global 135.98 130.02 1.19 0.03 1.46 0.03 1.34 0.04 (0.000) 14.66 73.53 (0.000) 4.00 52.94 Ratio Std. error of mean p-value t-value % of days Ratio>Local/Local Ratio Panel B: Number of Transactions Active Passive Ratio Std. error of mean p-value t-value % of days Ratio>Local/Local Ratio 23 Table 3 Informativeness of Different Financial Institutions Panel A reports the regression results of equation (2) for all financial institutions: ik 0 1 SDUM ik 2 GDUM ik it The dependent variable, θik, is the coefficient estimated . using equation (1) for the transactions of bond k initiated by financial institution i and filled by trader j (j i) for all trading days. Each column indicates the price change interval used (D = 10, 20, or 30 minutes). GDUMi (SDUMi) is a dummy variable that takes the value of one if financial institution i is a global (large) financial institution and zero otherwise. Standard errors are clustered by bonds. p-values associated with coefficient estimates are provided in brackets. Panel B shows the results excluding transactions of foreign financial institutions. Panel C presents the results after excluding transactions involving illiquid bonds. Panel D provides the results using transactions of local financial institutions involving liquid bonds. Panel A. All financial institutions Theta10 0.139 [0.0001] Theta20 0.104 [0.0204] Theta30 0.224 [0.0001] GDUM 0.085 [0.0083] 0.081 [0.0463] 0.065 [0.1277] Constant Bond Fixed Effects Included Included Included Included Included Included 5,594 0.41 5,593 0.26 5,594 0.21 SDUM N R2 Panel B. Excluding foreign financial institutions Theta10 0.151 [0.0001] Theta20 0.114 [0.021] Theta30 0.240 [0.001] GDUM 0.121 [0.002] 0.130 [0.013] 0.104 [0.070] Constant Bond Fixed Effects Included Included Included Included Included Included 4,855 0.39 4,855 0.25 4,855 0.20 SDUM N R2 24 Panel C. All financial institutions using liquid bonds only Theta10 0.138 [0.001] Theta20 0.116 [0.010] Theta30 0.232 [0.001] GDUM 0.086 [0.007] 0.080 [0.047] 0.058 [0.151] Constant Bond Fixed Effects Included Included Included Included Included Included 5,463 0.42 5,463 0.27 5,463 0.22 SDUM N R2 Panel D. Excluding foreign financial institutions and illiquid bonds Theta10 0.149 [0.001] Theta20 0.126 [0.0113] Theta30 0.247 [0.0001] GDUM 0.114 [0.004] 0.124 [0.016] 0.100 [0.076] Constant Bond Fixed Effects Included Included Included Included Included Included 4,753 0.40 4,753 0.26 4,753 0.20 SDUM N R2 25 Table 4 Day-Trading Profitability Panel A presents descriptive statistics on percentage day-trading profits (PRF) for global and local financial institutions. Panel B reports the Pearson chi-square values and their corresponding p-values of the hypothesis that the median of the first column is equal to the mean of the second column. PRF is calculated using the methodology described in Section 4.3. Panel A. Descriptive Statistics Local Small Local Large Global Obs 16,946 16,880 10,374 All 44,200 Mean -0.038 -0.027 -0.020 PRF Std. Dev. 0.118 0.109 0.115 Median -0.019 -0.013 -0.010 75% 0.007 0.013 0.023 -0.030 0.114 -0.014 0.013 Panel B. Test Statistics of Median Comparisons Local Small Local Large Local Large Global Global Local Small 26 PRF 59.61 [0.000] 18.63 [0.000] 101.96 [0.000] Table 5 Day-Trading Profits, Prior Profitability, Volatility, and Financial Networks This table reports the estimates for the Heckman selection model, equation (4), for the entire sample (Columns 1 and 2), for short-term bonds only (Columns 3 and 4), and for long-term bonds only (Columns 5 and 6). PRFit, the dependent variable, is the day-trading percentage profit of financial institution i on day t, LPRFit is lagged profitability, and VOLINTt is the standard deviation of the interest rate on the previous day using 30-minute observations. SDUMi (GDUMi) takes the value of one if the financial institution is a large (global) financial institution and zero otherwise. p-values associated with coefficient estimates are provided in brackets and are based on robust standard errors clustered by day. ***, **, and * represent the 0.1%, 1%, and 5% level of significance respectively, of the Inverse Mill’s ratios. -0.8837 [0.000] -0.0458 [0.000] (2) 0.0066 [0.000] 0.0030 [0.029] 0.1186 [0.000] -0.0162 [0.349] -0.0665 [0.000] -0.8732 [0.000] -0.0444 [0.000] Participation Equation LPRF -0.4293 [0.000] VOLINT 57.0596 [0.000] Inv. Mill Ratio Rho N Censored Uncensored GDUM SDUM LPRF (1) 0.0068 [0.000] 0.0051 [0.000] 0.0700 [0.000] LPRF *GDUM LPRF *SDUM VOLINT Constant 0.026*** 0.253 44,200 11,065 33,135 (3) 0.0079 [0.000] 0.0081 [0.000] 0.0813 [0.000] -0.8710 [0.000] -0.0444 [0.000] (4) 0.0073 [0.000] 0.0055 [0.002] 0.1384 [0.000] -0.0380 [0.082] -0.0713 [0.001] -0.8693 [0.000] -0.0425 [0.000] -2.3909 [0.000] -0.0381 [0.000] (6) 0.0035 [0.086] 0.0011 [0.555] 0.1021 [0.000] 0.0339 [0.146] -0.0733 [0.003] -2.3937 [0.005] -0.0372 [0.000] -0.4295 [0.000] 57.0753 [0.000] -0.2342 [0.000] 40.5614 [0.000] -0.2344 [0.000] 40.5730 [0.000] -0.0355 [0.556] 123.6571 [0.000] -0.0352 [0.559] 123.6822 [0.000] 0.026*** 0.256 44,200 11,065 33,135 0.019*** 0.182 32,321 10,915 21,406 0.020*** 0.185 32,321 10,915 21,406 0.027*** 0.273 22,319 6,363 15,956 0.027*** 0.269 22,319 6,363 15,956 27 (5) 0.0027 [0.162] 0.0029 [0.120] 0.0607 [0.000] Table 6 Day-Trading Profits, Prior Profitability, Volatility, and Financial Networks Excluding Turkish Macro News Announcement Days This table reports the estimates for the Heckman selection model, equation (4), after excluding the Turkish macroeconomic news announcement days, for all bonds (Columns 1 and 2), for short-term bonds only (Columns 3 and 4), and for long-term bonds only (Columns 5 and 6). PRFit, the dependent variable, is the day-trading percentage profit of financial institution i on day t, LPRFit is lagged profitability, and VOLINTt is the standard deviation of the interest rate on the previous day using 30-minute observations. SDUMi (GDUMi) takes the value of one if the financial institution is a large (global) financial institution and zero otherwise. p-values associated with coefficient estimates are provided in brackets and are based on robust standard errors clustered by day. ***, **, and * represent the 0.1%, 1%, and 5% level of significance respectively, of the Inverse Mill’s ratios. -1.2653 [0.000] Constant -0.0374 [0.000] Participation Equation LPRF -0.1157 [0.035] VOLINT 26.7774 [0.000] (2) 0.0054 [0.003] 0.0041 [0.013] 0.1177 [0.000] -0.0242 [0.252] -0.0555 [0.008] -1.2556 [0.000] -0.0360 [0.000] Inv. Mill Ratio Rho N Censored Uncensored GDUM SDUM LPRF (1) 0.0058 [0.000] 0.0059 [0.000] 0.0742 [0.000] LPRF *GDUM LPRF *SDUM VOLINT 0.014*** 0.131 36,483 14,179 22,304 (3) 0.0071 [0.000] 0.0090 [0.000] 0.0780 [0.000] -1.1068 [0.416] -0.0377 [0.000] (4) 0.0060 [0.007] 0.0071 [0.001] 0.1222 [0.000] -0.0525 [0.041] -0.0474 [0.073] -1.1065 [0.000] -0.0359 [0.000] -3.2845 [0.000] -0.0240 [0.000] (6) 0.0023 [0.366] 0.0028 [0.227] 0.1024 [0.000] 0.0638 [0.021] -0.0832 [0.005] -3.3229 [0.000] -0.0231 [0.000] -0.1157 [0.035] 26.7800 [0.000] -0.0475 [0.416] 17.6432 [0.000] -0.0475 [0.416] 17.6440 [0.000] 0.2020 [0.007] 37.6064 [0.000] 0.2020 [0.007] 37.6092 [0.000] 0.014*** 0.132 36,483 14,179 22,304 0.008*** 0.0777 27,054 12,106 14,948 0.008*** 0.079 27,054 12,106 14,948 0. 007** 0.0727 17,816 7,746 10,070 0.007** 0.074 17,816 7,746 10,070 28 (5) 0.0009 [0.731] 0.0047 [0.036] 0.0640 [0.000] Table 7 Day-Trading Profitability and Liquidity of Bonds This table reports the estimates for the Heckman selection model, equation (4), using liquid and illiquid bonds. The column heading (Active and Amihud or Passive and Amihud) represents the liquidity measure used to separate bonds. PRFit, the dependent variable, is the day-trading percentage profit of financial institution i on day t, LPRFit is lagged profitability, and VOLINTt is the standard deviation of the interest rate on the previous day using 30-minute observations. SDUMi (GDUMi) takes the value of one if the financial institution is a large (global) financial institution and zero otherwise. p-values associated with coefficient estimates are provided in brackets and are based on robust standard errors clustered by day. ***, **, and * represent the 0.1%, 1%, and 5% level of significance respectively, of the Inverse Mill’s ratios. GDUM SDUM LPRF LPRF *GDUM LPRF *SDUM VOLINT Constant Participation Equation LPRF VOLINT Inv. Mill Ratio Rho N Censored Uncensored Liquid Bonds Active Amihud (1) (2) 0.0059 0.0062 [0.001] [0.000] 0.0140 0.0004 [0.000] [0.775] 0.1751 0.1148 [0.000] [0.000] -0.0170 -0.0122 [0.452] [0.524] -0.0852 -0.0629 [0.000] [0.001] -2.0174 -1.5319 [0.000] [0.000] 0.0491 -0.0350 [0.000] [0.000] Illiquid Bonds Passive Amihud (3) (4) 0.0049 0.0080 [0.021] [0.004] -0.0061 0.0225 [0.001] [0.000] 0.0917 0.1388 [0.000] [0.000] -0.0273 -0.0351 [0.271] [0.219] -0.0440 -0.0808 [0.055] [0.008] -1.9046 1.7481 [0.000] [0.087] -0.0178 0.0538 [0.000] [0.000] -0.5611 [0.000] 23.1810 [0.000] -0.1775 [0.000] 43.2030 [0.000] 0.1909 [0.000] 18.3999 [0.000] -0.0965 [0.179] -19.4802 [0.000] 0.113*** -0.832 29,726 10,416 19,310 0.022*** 0.212 38,558 11,715 26,843 0.009*** 0.089 30,256 11,884 18,372 -0.132* -0.839 18,456 9,629 8,827 29 Table 8 Day-Trading Profit Analysis Excluding Foreign Financial Institutions This table reports the estimates for the Heckman selection model, equation (4), excluding foreign financial institutions. PRFit, the dependent variable, is the day-trading percentage profit of financial institution i on day t, LPRFit is lagged profitability, and VOLINTt is the standard deviation of the interest rate on the previous day using 30-minute observations. SDUMi (GDUMi) takes the value of one if the financial institution is a large (global) financial institution and zero otherwise. p-values associated with coefficient estimates are provided in brackets and are based on robust standard errors clustered by day. ***, **, and * represent the 0.1%, 1%, and 5% level of significance respectively, of the Inverse Mill’s ratios. GDUM SDUM LPRF (1) 0.0038 [0.001] 0.0050 [0.001] 0.0774 [0.000] -0.7613 [0.001] -0.0445 [0.000] (2) 0.0035 [0.004] 0.0029 [0.058] 0.1201 [0.000] -0.0265 [0.225] -0.0649 [0.000] -0.7523 [0.001] -0.0433 [0.000] -0.4779 [0.000] 57.6870 [0.000] -0.4779 [0.000] 57.7021 [0.000] 0.025*** 0.241 38,574 9,461 29,113 0.025*** 0.244 38,574 9,461 29,113 LPRF *GDUM LPRF *SDUM VOLINT Constant Participation Equation LPRF VOLINT Inv. Mill Ratio Rho N Censored Uncensored 30