IPO Madness, Index Rigging, and the Introduction of an Opening and Closing Call: The Case of Singapore

Transcription

1 IPO Madness, Index Rigging, and the Introduction of an Opening and Closing Call: The Case of Singapore Carole Comerton-Forde Finance Discipline University of Sydney Sydney, NSW, 2006 Australia and Securities Industry Research Centre of Asia Pacific Voice: Fax: C.Comerton-Forde@econ.usyd.edu.au Sie Ting Lau Department of Banking and Finance Nanyang Technological University Singapore Voice: Fax: astlau@ntu.edu.sg Thomas H. McInish University of Memphis Memphis, TN Voice: Fax: tmcinish@memphis.edu August 2003 Please address correspondence to: Thomas H. McInish Professor and Wunderlich Chair of Finance Department of Finance, Insurance, and Real Estate The University of Memphis Memphis, TN The authors wish to thank Reuters and the Securities Industry Research Centre of Asia-Pacific for access to the data used in this paper.

2 IPO Madness, Index Rigging, and the Introduction of an Opening and Closing Call: The Case of Singapore ABSTRACT This paper investigates the introduction of an opening and closing call on the Singapore Exchange in August 2000 and concludes that the calls improved trading at the beginning and ending of the trading day and aided the initiation of trading for initial public offerings (IPOs). The number of shares traded on the first and last trades of the day increased substantially, more than doubling at the close. For IPOs the number of shares traded on the opening trade increased more than five times. We also find the introduction of the calls increased open-to-open and close-to-close market model adjusted R-squares, a measure of market quality.

3 IPO Madness, Index Rigging, and the Introduction of an Opening and Closing Call: The Case of Singapore On August 21, 2000, the Singapore Exchange (SGX 1 ) initiated both an opening call and a closing call to enhance the operation of its screen-based, order-driven trading system. On the SGX all orders are limit orders that expire at the end of the trading day. Hence, when the exchange turns on its computer for trading at 09:00 there are no orders in the order book. This system made it difficult for the price discovery process to get started. Traders placed orders at wide spreads at first until they could see how other market participants behaved. Unusually high trading volume stresses exchanges and interferes with price discovery. One period of stress is the opening, which follows the overnight non-trading period. On the New York Stock Exchange, the opening trade often accounts for 5% of total daily volume even for active stocks and accounts for more than 20% of total daily volume for the least active stocks. Comparable figures for Singapore prior to the introduction of the call auction are 2% for active stocks and 12% for inactive stocks. 2 In Singapore the usual stress associated with market openings was exacerbated for IPOs on their first day of trading and often resulted in what was called IPO madness with hundreds of brokers and their clients placing orders for hot IPO stocks on the first day of trading. Another period of stress is at the end of the trading day when many traders seek to complete their orders and institutions attempt to match the closing price, which on the NYSE leads to the use of market-on-close orders. Cushing and Madhavan (2000) report that for the most actively traded stocks more than 20% of the volume during the last five minutes of trading on the NYSE is in blocks of 10,000 shares or more and blocks have a higher share of the order flow at the close than at any other time of the trading day. In Singapore, on May 31,

4 2000, several component stocks of the Morgan Stanley Capital International Index (Singapore) (MSCI) saw large price changes at the close, which was seen as an attempt to manipulate the closing index value. This incident is only one of many that involved component stocks of the Singapore Straits Times Index (STI) and the MSCI, the two most widely followed Singapore stock indices. These incidents resulted in requests for new procedures to prevent index rigging or index manipulation at the close. One way of dealing with this type of closing price problem identified by the SGX was the Hong Kong system of calculating the closing price as the median of the 5 nominal prices sampled every 15 second during the last minute of daily trading. However, the SGX chose to institute a closing call. We use the term index rigging to cover any activities designed to influence closing prices whether or not the stocks manipulated are part of an index. The Singapore Exchange trades derivatives on both the STI and MSCI indexes. Traders might seek to manipulate closing stock values as part of derivative strategies. Attempts to manipulate closing values might also be related to margin requirements. Mutual funds, bank trust departments or others might seek to manipulate closing prices as part of attempts at window dressing. Note that the incident cited in the previous paragraph occurred on the last trading day of May. The end of the month is a common time that various institutions are concerned about the appearance of their accounting books. Economides and Schwartz (1995) advocate the use of an opening and closing call to address order imbalances at the beginning and end of the trading day. Pagano and Schwartz (2003) investigate the initiation of a closing call on the Paris Bourse and conclude that the call 1 The Singapore Stock Exchange merged with the Singapore International Monetary Exchange (SIMEX), a futures exchange, on December 1, 1999, and adopted the name Singapore Exchange. 2 Of course, even with the availability of call trading, the majority of trading takes place in the continuous trading period. Kalay, Wei and Wohl (2002) examine the move of the Tel Aviv Stock exchange from call trading only to continuous trading with an opening call. After the switch, 90 percent of the total daily trading volume occurred during the continuous trading period. Examining a change in the trading procedures on the Riga Stock Exchange, Kairys, Kruza, and Kumpins (2000) show that the addition of a continuous trading session to a call-only market greatly increases trading volume for thick stocks, but reduces trading volume for thin stocks.

5 improved market quality. However, due to the fact that their study examined the introduction of a closing call rather than an opening call, Pagano and Schwartz (2003) did not specifically examine learning during the pre call period. This is reasonable for closing calls because the pre call period is short. But there is evidence that the ability to place non-binding quotes prior to calls, especially at the open, improves the trading process. Biais, Hillion, and Spatt (1999) examine quotes placed during the pre opening period on the Paris Bourse and conclude that even though there is no trading, participants learn from these quotes, and, consequently, there is an improvement in informational efficiency and a convergence toward equilibrium. Following the pre opening period, the Paris Bourse conducts an opening call. But Cao, Ghysels, and Hatheway (2000) study the pre opening period on Nasdaq, which is not followed by an opening call, and provide evidence that the non-binding quotes placed during the pre opening period convey information. Muscarella and Piwowar (2001) report that stocks that transfer from call trading to continuous trading experience improved prices and market quality, but that a transfer from continuous trading to call trading experienced decreased prices and market quality. These authors conclude that continuous markets offer better liquidity for thick stocks, but call markets do not offer better liquidity for thin stocks. We use the introduction of the opening and closing call on the SGX to investigate whether the introduction of these calls mitigated IPO madness and index rigging and improved market quality generally. We examine a sample of the 250 most actively traded stocks for 240 trading days before (the pre period) and 240 trading days after (the post period) the introduction of call trading. Fifty-five of these stocks were components of one of the two major indexes tracking Singapore stocks. In addition to this sample, we also examine 127 IPOs that occurred during the sample period. Following the introduction of an opening and closing call, we find that there was a substantial and statistically significant increase in the number of shares traded on the first trade of the day and on the last trade of the day. This is consistent with an improvement in liquidity due to the opening and closing calls and supports the view of Economides and

6 Schwartz (1995) that the use of calls can benefit markets. For the opening call, we find that the percentage of days with an opening call trade is more than 45% for the most active stocks, but less than 2% for the least active stocks. Hence, we reinforce the finding of Muscarella and Piwowar (2001) that calls are most suited for actively traded stocks and do not benefit thin stocks. Yet, for the pre opening period, for the most active stocks, we find that quote changes on more than 96% of days, and, for the least active stocks, we find quote changes on more than 26% of days. Hence, there is substantial opportunity for learning without trading as reported for Nasdaq stocks by Cao, Ghysels, and Hatheway (2000). Using the methodology of Pagano and Schwartz (2003), we find a statistically significant increase in market quality for the open-to-open period for the top four most actively traded quintiles of stocks and for the close-to-close returns for all five quintiles of stocks. These results present striking evidence of the usefulness of opening and closing calls in improving market quality. I. Data On August 21, 2000, the Singapore Exchange (SGX) instituted an opening call to begin trading each day and a closing call to end trading each day. We collect daily data for 240 trading days before (the pre period) and 240 trading days after (the post period) the initiation of call trading. We rank firms based on dollar trading volume during the pre period, excluding the opening and closing call, and collect data for the 250 most active firms. In total there were about 480 firms traded on the SGX during our sample period. Our sample of 250 firms does not include any IPOs. We examine 127 IPOs in our analysis of first day IPO openings. The remaining firms listed on the SGX (about 100) were excluded because they were not actively traded. For each firm for each day, we collect data on the number of quotes during the preopening and pre-closing periods and the size and trade price of the opening and closing call trades when these occur. To be included in our sample, we require at least four trades on a given day. For each firm for each day, we collect the time, price, and size of the first two and

7 last two trades during the intraday period (e.g., excluding the opening and closing call trades). For each IPO we collect the time, price, and size of each trade on the first day of trading. All of these data are obtained from Reuters through the Security Industry Research Centre of Asia-Pacific (SIRCA) in Australia. Further, we identify the firms that are components of the Straits Times Index, the most widely followed equity index in Singapore. Our sample of 250 firms included 50 of the 55 firms that are components of the STI. The remaining five firms are not actively traded. The MSCI comprises 28 firms, six of which are not in the STI. 3 Of these six, five are in our sample of 250 and the remaining one is not actively traded. In our empirical work, we use the 55 stocks that are components of either the STI or MSCI as our index sample. The Singapore Stock Exchange operates a fully automated, order-driven, screenbased trading system. The hours of operation of the Singapore Stock Exchange are 09:00 to 12:30 and 14:00 to 17:00. The pre-opening period is from 08:30 to 08:59 and the pre-closing period is from 17:00 to 17:05. Appendix A provides additional details concerning the operations of the Singapore Exchange. The Singapore Exchange initiated an opening call and a closing call on August 21, Appendix B provides details of how call trade prices are determined. II. Trading at the beginning and ending of the day As indicated previously, the SGX s reason for introducing opening and closing calls was to improve market openings and closings generally, and, specifically, to mitigate problems associated with the beginning of trading in IPOs and perceived attempts to manipulate closing index values. Table 1 presents data concerning the frequency of occurrence of call trades and changes in the best bid and ask and the size of the opening and closing trades. Statistics are presented for the entire sample and for quintiles based on trading 3 The six stocks that are only in the MSCI are: DBSL.SI, HPAR.SI, OVES.SI, RBSS.SI, STCM.SI, and UTOS.SI.

8 volume. We test for differences between the opening and closing values by jointly ranking the values and calculating a t-statistic for the difference in mean ranks for the two periods. Examining Table 1, Panel A, two findings are immediately apparent. First, if success is measured in terms of call trades, the closing call is much more successful than the opening call. Second, the opening and closing calls are much more successful for actively traded stocks than for less actively traded stocks. For the 250 firms over the 240 days in the sample, 15.61% of the days have an opening call trade, 34.71% of days have a closing call trade, and 11.00% of days have both an opening and closing call trade. For the quintile of most active stocks, 45.45% of days have an opening call trade, 79.21% of stocks have a closing call trade, and 41.69% of stocks have both an opening and closing call trade. For the quintile of least active stocks, 1.36% of days have an opening call trade, 6.90% of days have a closing call trade, and 0.31% of days have both an opening and closing call trade. For both the full sample and for each quintile, the number of days with a closing call trade is significantly greater than the number of days with an opening call trade (at the 0.01 level). Further, for both the opening and closing call, there is a substantial decline in the number of days with a call trade as the level of trading activity declines. If calls were more suited to trading of thinly traded equities, many thinly traded stocks might trade only during the call. If this were the case, the percentage of days with call trading could actually increase as trading activity declined. But for our sample, call trading is concentrated in the active equities. Hence, our results indicate that call trading is not particularly suited to the trading of thin issues at least on the SGX where there is also the possibility of trading in the continuous market. Table 1, Panel A, also presents the number of days with quote changes during the pre opening period and during the pre closing period. One finding stands out: there are many more days with pre opening, and to a lesser extent pre closing, quote changes than days with pre opening and pre closing call trades. For the full sample, 65.86% of the days have pre opening quote changes compared with 15.61% of days with opening call trades, 43.41% of the days have pre closing quote changes compared with 34.71% of days with closing call trades, and 42.54% of days have both pre opening and pre closing quote changes compared

9 with 11.00% of days with both opening and closing call trades. It is evident that there are many days on which there is no call trades, but there is, nevertheless, an opportunity for learning about supply and demand from quote changes. For the full sample and for each quintile, there are significantly more quote changes during the opening call than during the closing call. However, this is not surprising given that the pre opening period is thirty minutes and the pre closing period is only five minutes. Turning to Table 1, Panel B, volume (number of shares) traded at the closing call is significantly larger than for the opening call for quintile 1, but there is no statistical difference in these volumes for the remaining quintiles or for the sample as a whole. Thus, while there are significantly fewer opening call trades than closing call trades, when opening call trades do occur their average size in shares is equal to or larger than the size of closing call trades except of the most active quintile of stocks (quintile 1). Further, the number of days with call trades declines rapidly as stocks become thinner, but the falloff on call trade size is not nearly as substantial. Dividing the percentage of days with call trades for the least active quintile by the percentage for the most active quintile, we obtain (1.36/45.45 X 100=) 3% for the opening call and (6.90/79.21 X 100=) 9% for the closing call. But dividing the mean call volume for the least active quintile by the mean call volume for the most active quintile, we obtain (17,668/32,362 X 100 =) 55% for the opening trades and (10,125/30,389 X 100 =) 33% for the closing trades. Thus, even for the inactive stocks, when there is a call trade the size of the trade is substantial. The number of quote changes is larger in the pre opening period than in the pre closing period. For the full sample there are on average quite changes each during the pre opening and 8.94 quote changes during the pre closing. But adjusting for the length of the pre-opening and pre-closing periods (30 minutes and 5 minutes, respectively), there are more quotes per minute (not shown) in the closing period. For the full sample, the number of quote changes per minute is 0.46 for the pre open and 1.79 for the pre close. We explore whether liquidity, as proxied by trading volume, increased following the introduction of call trading. In Table 2, Panel A, we turn to an analysis of the number of

10 shares traded at the beginning and end of the trading day. For this analysis we categorize the 250 stocks in our sample according to whether or not they were components of the STI and MSCI. Fifty-five of the firms are index components. For the pre period both the first and last trades are intraday trades. For the post period, we use the number of shares traded at the call when there is a call trade and intraday trades when there is no call trade. For both the index and non-index firms, there is a significant increase in the number of shares traded on both the first and last trades of the day following the introduction of call trading. For the first trade of the day, the number of shares traded on the first trade increased from 18,252 to 26,360 for the index firms and from 14,514 to 24,173 for the non-index firms. For the last trade of the day the number of shares traded increased from 10,158 to 38,069 for the index firms and from 9,639 to 22,765 for the non-index firms. As a control we also examine the change in daily volume. 4 The number of shares traded each day increased from 1,500,617 to 1,534,316 for the index firms, which is not statistically significant at the 0.05 level. For the non-index firms the number of shares traded each day declined from 1,035,570 in the pre period to 738,054 in the post period, which approaches statistical significance (t = -1.84). We also explore whether the introduction of call trading reduced volatility. In Table 2, Panel B, we present statistics for the volatility of returns at the beginning and end of the trading day. At the beginning of the day, we present the return on the second trade of the day rather than the overnight return. When there is an opening call trade the second return of the day is calculated using the opening call price and the first intraday trade price. When there is no opening call trade, the second return of the day is calculated using the prices of the first two intraday trades. Similarly, the last return of the day is calculated using the closing call price and the last intraday price when there is a closing call trade and the last two intraday 4 We investigated presenting the number of shares traded on each trade relative to the total number of shares traded each day rather than presenting the actual number of shares traded. Our results indicate that our conclusions are the same whether we use absolute or relative values. Further, because the index firms are the larger firms in Singapore, we do not believe that it is feasible to construct a sample of matched firms as a control.

11 prices otherwise. The intraday return is calculated using the first and last prices of the day without regard to whether these prices are call trade prices. For each firm in our sample, we calculate the standard deviation of the returns across days and present the mean and standard deviations of these across firms in Table 2, Panel B. We refer to the mean of the standard deviations as volatility. For the second return of the day, there is no statistical difference in the pre and post period volatility for either the index (t = -1.61) or non-index (t = 0.94) firms. However, the intraday return declined significantly for the index firms (t = -4.14), but not for the non-index firms (t = 1.60). On balance, we believe that this evidence is consistent with the view that the introduction of the opening call did not affect day-to-day volatility at the opening. One of the purposes of the SGX in introducing a closing call was to reduce manipulation of closing prices of components of the market indexes. The effect of the manipulation is to create unusually large returns on the last trade of the day. However, because this manipulation was likely only sporadic and concentrated in just a few stocks, it is difficult to ascertain whether this objective has been attained. For the last return of the day there is a significant increase (at the 0.05 level) in the standard deviation of returns from to for the index firms (t = 2.02) and from to for the non-index firms (t = 7.62). And this increase in volatility occurred despite a decrease in intraday volatility from to for the index firms and a statistically insignificant increase in volatility from to for the non-index firms. These results are not consistent with

12 the outcome that would be expected if the closing call reduced market manipulation at the close sufficiently to affect the overall level of return variability across days. 5 We have seen that for our sample of 250 firms there was a significant increase in the number of shares traded on the first trade of the day for both index and non-index firms. However, this sample did not include any IPOs. As we stated previously, a major reason that the SGX introduced an opening call was to facilitate the beginning of trading for IPOs. We turn our attention to the first day of trading for a sample of 127 IPOs, 83 from the pre period and 44 from the post period. We present statistics for the first five trades of the day for the pre-period IPOs and for the first six trades of the day (including the opening call) for the postperiod IPOs. We know that volume differs across firms due to firm characteristics such as size and analysts following. Also, volume varies through time. To help control for these for a given firm, we compute the sum of the number of shares traded on its offering day (SUMD) and the sum of the number of shares traded on trade i (TRADE i ), where i represents the call trade (post period only), the first trade, the second trade, and so forth, in turn. For each trade i, we compute RATIO i = TRADE i /SUMD and report the mean of these in Table 3. The proportion of daily volume for the call trade ( ) is more than five times the proportion of daily volume for the first trade during the pre period and the difference is statistically significant (t = 13.02). Comparing the first intraday trade in the post period with the first intraday trade in the pre period, the second intraday trade in the post period with the second intraday trade in the pre period, and so forth, biases the test against finding that the post-period trade is significantly larger. The reason for this is that for each intraday trade in the post period there 5 We conduct a number of additional tests to determine whether the introduction of a closing call reduced unusually large returns at the close with similar outcomes. For each firm for each day, we calculate the return on the last trade (L), ignoring whether the trade was a call trade. For the pre period, we calculate the mean (M) and standard deviation (S) of the last return of the day for each firm. Then we standardize each firm s returns by calculating (L M)/S for each trade. We standardize the post period returns in a similar way. There are 52,953 pre period returns and 40,808 post period returns. We examine the absolute values of these standardized returns. The number of returns during the pre and post period that exceed the indicated number of standard deviations (in parentheses) per 10,000 observations is: (4), 59, 48; (6), 9, 9; (8), 2.6, 1.2; (10), 0.76, This provides evidence of a slight reduction in the number of outliers in the post period. We rank the absolute value of these standardized returns for all stocks for both periods jointly. Using a t test, we find no difference in the mean rank

13 has been one additional prior trade compared to the comparable pre period trade and we know that in general trade size decreases as the trading day progresses at least until the middle of the day. Nevertheless, in each case the post period trade volume ratio is significantly larger that the pre period trade volume ratio (the trade number is presented in parentheses, followed by the pre period and post period volume ratio): (1) , ; (2), , ; (3), , ; (4), , ; (5), , These results strongly support the view that the introduction of an opening call increased the ability of the SGX to provide liquidity at the beginning of IPO trading. We turn next to consideration of how the introduction of an opening call affected volatility for these IPOs. For each IPO for its first trading day, we calculate the standard deviation of returns for the period until 09:10 (STD1) and for the remainder of the trading day (STD2). In Table 3, Panel B, we report the ratio STD1/STD2 for the 83 IPOs in the pre period and for the 44 IPOs in the post period. We find no significant change in the ratio of beginning-of-the-day volatility to remainder-of-the-day volatility. The results of this test need to be interpreted with caution since the pre period and post period firms are different so that variability of returns may change from the pre to the post period due to factors for which we have not controlled. Our overall assessment is that the introduction of an opening call increased the volume that the market is able to handle when IPOs first begin trading and that beginning-ofthe-day volatility on IPOs first day of trading was not affected. III. Assessment of market quality We assess whether market quality improves with the introduction of opening and closing call trading using the methodology described in Pagano and Schwartz (2003), which, in turn, is based on earlier work by Cohen, Hawawini, Maier, Schwartz and Whitcomb (1983a, 1983b). We consider the opening call first. We calculate daily open-to-open returns for each firm and for the Singapore Straits Times Index, which we use as the market index. between the pre and post periods for either the entire sample (t = 1.18) or the index sample (t = 0.16).

14 For this analysis we drop the requirement of four trades per day. Instead days that have no trades are assigned a return of zero. Using these daily open-to-open returns for each firm, we estimate the market model for the pre-period and for the post-period. For our sample this process produces 250 regression results for the pre-period and 250 regression results for the post-period. The process is then repeated for differencing intervals of 2, 3, 4, 5, 6, 8, 10, 12, 15, and 20 days. We choose these differencing intervals because each is evenly divisible into our sample period of 240 pre and 240 post trading days. This insures that all of the estimates are over the same pre and post time intervals. To assess market quality, we compare the adjusted R-squares for the pre and post periods. If market quality improves with the introduction of call trading, the adjusted R- square will be higher in the post period, indicating a tighter fit between the individual firm returns and market returns. After the assessment of the opening call is complete, we repeat the process for the closing call using close-to-close returns for the STI. Next, we provide a more detailed explanation of our test. Let R j be the open-to-open return for a one-day differencing interval on stock j in the period before the introduction of the opening and closing call and R m be the corresponding return on the market. For each stock, (j = 1,, 250), we estimate the following regression model: R j = b 0 + b 1 R m + ε j, where b 0, b 1 and ε j are parameters to be estimated. Pagano and Schwartz (2003) call these first pass regressions. We report the mean of the adjusted R-squares from these regressions for the 50 most actively traded firms in row 1 of column 2 of Table 4. The means of the adjusted R- squares for the remaining differencing intervals for quintile 1 are reported next and then comparable results are reported for the post period. The analysis is repeated for quintiles 2-5. The full sample (labeled All) shows a statistically significant increase in R-square at the five percent level or better from the pre to the post period for quintiles 1-4. Moreover, many of the individual differencing intervals also show a statistically significant increase in R-square.

15 These results are consistent with the view that the introduction of opening-call trading improved market quality. The first pass estimation of the model for the most actively traded quintile yields eleven pre-period and eleven post-period adjusted R-squares for each firm for the open-toopen returns. Next, we use these adjusted R-squares to estimate second pass regressions for each firm. (Throughout we maintain the terminology of Pagano and Schwartz, 2003, except that we re-label the coefficients as b 0 b 3 ). Let A jle = the adjusted R-square for firm j, for differencing interval L, for the period E (either pre or post). For each firm, we estimate the following regression model: A jle = b 0 + b 1 ln(1 + L -1 ) + b 2 DS je + b 3 DI je + ε jle (1) where b 0, b 1, b 2, b 3 are coefficients to be estimated, DS je is a dummy variable equal to ln(1 + L -1 ) if the observation is for the post period and to zero otherwise, DI je is a dummy variable equal to 1 if the observation is for the post period and to zero otherwise, and ε jle is a random error term. For a given quintile, there are 100 observations of the dependent variable (50 pre and 50 post). Consider first a regression including only an intercept and the variable ln(1 + L -1 ). The intercept can be interpreted as the asymptotic level of R-square as L approaches infinity. Hence, we expect the intercept to be positive. Suppose that we add a dummy variable that equals one for the post-period R-squares and 0 otherwise. Then, the intercept captures the asymptotic level of R-square for the pre period and the coefficient of the dummy variable captures the asymptotic level for the post period. If the introduction of a call improves market quality, the coefficient of the dummy variable (b 3 in equation 1) should also be significantly positive. R2CONSTANT is b 0 for the pre period and b 0 + b 3 for the post period. Hence, if the introduction of an opening call improved market quality, we expect an increase in R2CONSTANT between the pre and post period. Pagano and Schwartz (2003) argue that market frictions cause the adjusted shortinterval R-squares to be depressed relative to the longer-interval R-squares. Note that the

16 values of ln(1 + L -1 ) decrease as the differencing interval lengthens. Hence, the coefficient b 1 is expected to be negative. Whether the introduction of call trading results in more improvement in short-interval or long-interval R-squares is an empirical question. To investigate this we add a dummy variable that is equal to ln(1 + L -1 ) if the observation is for the post period and to zero otherwise. Based on the results of Pagano and Schwartz (2003), we expect the coefficient b 2 to be significantly negative. R2SLOPE is b 1 for the pre period and b 1 + b 2 for the post period. Therefore, we expect R2SLOPE to be negative and smaller in the post period. For each quintile for each differencing interval length and for each quintile as a whole, Table 4 presents the means of the R-squares obtained from the estimation of the market model. For each quintile there are (11 differencing interval X 2 periods X 50 firms =) 1,100 adjusted R-squares. We jointly rank the R-squares and test for a significant difference in the mean rank for the pre and post period. This is equivalent to a Wilcoxon rank sum test. For the full sample (labeled All) there is a statistically significant increase in R-square from the pre to the post period for quintiles 1-4. However, only 23 of the t-statistics for the 55 differencing intervals (11 differencing intervals X 5 quintiles) show a statistically significant increase. In Table 5 we report the mean and standard deviation of R2CONSTANT for the 50 individual-firm regressions in each quintile. The results for the opening call based on daily open-to-open returns are reported in columns 2-4. Column 4 reports the results of a matched pairs t-test for the difference between the pre and post values. R2CONSTANT increases from to for quintile 1, from to for quintile 2, from to for quintile 3, and from to for quintile 4. All of these increases are statistically significant at the 0.05 level (the lowest t-statistic is 2.68 for quintile 1). R2CONSTANT decreases from the pre to the post period for quintile 5. The coefficient b 3 provides a direct test of whether market quality improves with the introduction of call trading. Table 6, Panel A, presents the results of the test of significance of b 3 for the 50 firms in each quintile for the opening call. We use a critical value of 1.64, which

17 allocates 5% to each tail. For the opening call for quintiles 1-4, 40% to 50% of the coefficients are significantly positive while fewer than 25% are significantly negative. However, for the least active stocks (quintile 5) only 8 of the 50 firms (16%) benefited from the introduction of the opening call. On balance, we conclude that the introduction of the opening call significantly improved market quality. We consider whether the improvement in adjusted R-square was similar for the various differencing interval lengths by examining the coefficients of R2SLOPE reported in Table 5. For quintiles 2-5, R2SLOPE becomes more negative in the post period and the change is statistically significant. For quintile 1, the absolute value of R2SLOPE is almost the same for the pre and post periods and the slight difference is not statistically significant. Hence, we conclude that the introduction of an opening call resulted in a greater increase in R-square for short differencing intervals relative to the increase for longer differencing intervals. Next, we turn to an analysis of close-to-close returns. The methodology is the same as that for the open-to-open returns. The results of the first pass regressions are reported in Table 7. The results of the full sample (labeled All) are reported in column 7 of the last row for each quintile. The adjusted R-squares increase from , , , , and to , , , , and , respectively, for quintiles 1-5. All of these increases are statistically significant and the lowest t-statistic of the five is 4.09 for quintile 3. We also test whether there is a significant increase in the adjusted R-square for each differencing interval length for each quintile. Thirty-seven of the 55 t-statistics show a statistically significant increase at the 0.05 level. Examination of the second pass results in Table 5 (columns 5-7) show that there is a positive and statistically significant increase in R2CONSTANT for each of the five quintiles. Further, Table 6, panel B, presents the results of the analysis of the t-statistics for the coefficient b 3, which, as indicated previously, provides a direct test of whether market quality improved, leads to the same conclusion. The percentage of statistically significant positive t-

18 statistics ranges from 54% for quintile 3 to 88% for quintile 1. In contrast, the number of statistically significant negative coefficients is 8% for quintiles 1 and 2, 10% for quintiles 4 and 5 and 20% for quintile 3. Hence, these results provide strong evidence that the introduction of a closing call improved market quality. Finally, we turn to the analysis of whether the change in R-square from the pre to the post period is consistent over short and longer differencing intervals. Our results, reported in Table 5, show that for the close-to-close returns in the pre period, R2SLOPE is significantly negative for each of the five quintiles. These results indicate that the introduction of a closing call resulted in a greater increase in short differencing interval R-square than in longer differencing interval R-square. These results for the closing call are similar to those for the opening call. IV. Summary and conclusions In August 2000 the Singapore Exchange, which operates an order-driven limit order book exchange, introduced an opening and closing call. Previously trading had begun with an empty limit order schedule. The purpose of introducing these calls was to address perceived problems with the opening, especially related to the initiation of trading in IPOs, and with the closing, especially related to attempts to manipulate stock closing prices. Opening calls are used on many exchanges including the NYSE, Paris Bourse, and the Australian Stock Exchange. But because most exchanges have had opening calls for many years, there have been few opportunities to study the introduction of an opening call. Pagano and Schwartz (2003) recently studied the introduction of a closing call on the Paris Bourse and concluded that there was an improvement in market quality. We find that there was a substantial improvement in liquidity at both the opening and closing. After the introduction of the opening call, there was a 50% increase in the number of shares traded at the open and an increase of more than 100% in the number of shares traded at the close. IPOs had a five-fold increase in volume on the opening trade and the opening call

19 also aided subsequent trading, which also showed increases in trading volume. There was a statistically significant increase between the pre and post period in the market model adjusted R-square, a measure of market quality. We also find significant opportunity for opening and closing to benefit from the introduction of the opening and closing calls even in the absence of call trading. There were opening call trades on about 16% of the days in our sample and closing call trades on about 35% of the days in our sample. But there were quote changes during the pre opening period on about 66% of the days in our sample and during the pre closing period on about 43% of the days in our sample.

20 References Biais, Bruno, Pierre Hillion, and Chester Spatt, 1999, Price discovery and leaning during the pre-opening in the Paris Bourse, Journal of Political Economy 107, Cao, Charles, Eric Ghysels, and Frank Hatheway, 2000, Price discovery without trading: Evidence from the Nasdaq preopening, Journal of Finance 55, Cohen, Kalman J., Gabriel A. Hawawini, Steven F. Maier, Robert A. Schwartz, and David K. Whitcomb, 1983a, Friction in the trading process and the estimation of systematic risk, Journal of Financial Economics 12, Cohen, Kalman J., Gabriel A. Hawawini, Steven F. Maier, Robert A. Schwartz, and David K. Whitcomb, 1983b, Estimating and adjusting for the intervalling-effect bias in beta, Management Science 29, Cushing, David and Ananth Madhavan, 2000, Stock returns and trading at the close, Journal of Financial Markets 3, Economides, Nicholas, and Robert A. Schwartz, 1995, Electronic call market trading, Journal of Portfolio Management (spring), Kairys, Joseph P., Jr., Raimonds Kruza, and Ritvars Kumpins, 2000, Winners and losers from the introduction of continuous variable price trading: Evidence from the Riga Stock Exchange, Journal of Banking and Finance 24, Kalay, Avner, Li Wei, and Avi Wohl, 2002, Continuous trading or call auctions: Revealed preferences of inventors at the Tel Aviv Stock Exchange, Journal of Finance 57, Madhavan, A. and V. Panchapagesan, 2000, Price discovery in auction markets: A look inside the black box, Review of Financial Studies 13, Muscarella, Chris J., and Michael S. Powowar, 2001, Market microstructure and securities values: Evidence from the Paris Bourse, Journal of Financial Markets 4, Pagano, M. and R. Schwartz, 2003, A closing call s impact on market quality at Euronext Paris, Journal of Financial Economics 68,

21 Appendix A. Institutional detail The Singapore Exchange Limited (SGX) is an automated order driven market. The trading day is divided into a number of different market phases. These phases are: Market Phases User Login Pre-Open Non-Cancel/Publish Normal Trading Lunch break (System shutdown) Normal Trading Pre-Close Non-Cancel/Publish Enquire System Shutdown Time 04:00 08:30 08:30 08:59 08:59 09:00 09:00 12:30 12:30 14:00 (12:35 13:00) 14:00 17:00 17:00 17:05 17:05 17:06 17:06 20:00 20:00 04:00 The SGX system is started at 04:00 and brokers generally start connecting at around 08:00. The only information available to market participants prior to the pre-opening period is the historical high, low and previous day s closing price. The Pre-Open routine operates from 08:30 to 09:00. It comprises two sessions, the Pre-Open Period and the Non-Cancel Period. The Pre-Open period operates between 08:30 and 08:59. During this time orders may be entered, amended and withdrawn, but not executed. The Non-Cancel Period, which operates between 08:59 and 09:00, allows orders to be entered, but not amended or withdrawn. At 09:00, overlapping orders are matched at a single price. Unmatched orders are carried forward into the normal trading session. Brokers are not allowed to enter undisclosed orders during the pre-close routine. Details of the calculation of the opening and closing price are provided in Appendix B. Normal trading takes place in two sessions between 09:00 and 12:30 and 14:00 to 17:00. During trading hours brokers can view the entire limit order book. For a fee, investors are also able to access this information. They are also able to observe the last traded price, the cumulative volume and value traded today, the high and low prices for the day, the first price

22 for the day, the change and percentage change in price from the previous day s close and the cumulative volume of off-market (married trades). 6 Orders are executed according to price and time priority. The minimum tick size varies depending upon the price of the stock. There are 5 tick categories ranging from 0.5 cent for stocks priced less than $1.00 to 10 cents for stocks priced above $ The trading system generates a warning if an order is placed more than 6 minimum ticks away from the market. If the broker chooses to over-ride this warning the SGX charges S$0.20 for the order. The system does not allow the use of market orders, but marketable limit orders are permitted. All orders expire at the end of each trading day. The SGX system supports orders with undisclosed quantities if the disclosed quantity is at least 50,000 shares. As the disclosed quantity is traded and falls to zero the undisclosed quantity is used to top up the disclosed quantity to the maximum of the initial disclosed amount. However, these topped up orders lose their time priority. Between 12:30 and 12:35 and 13:00 and 14:00 participants can view the market and cancel orders, however no new orders may be entered. The system is shut down between 12:35 and 13:00. The Pre-Close routine operates between 17:00 and 17:06. Again it comprises two sessions, the Pre-Close Period and the Non-Cancel Period. The Pre-Close period operates between 17:00 and 17:05 with the same rules as the Pre-Open period. The Non-Cancel Period runs from 17:05 to 17:06. Overlapping orders are executed and unmatched orders become void. Brokers are not allowed to enter undisclosed orders during the pre-close routine. Any undisclosed quantities carried forward from normal trading will not be matched. 6 Off market trades are also known as married trades. This refers to business that is transacted directly between two participants outside the SGX market for volumes not less than 50,000 shares or 150,000 SGD.

23 Appendix B. Computation of opening and closing prices The following methodology for calculating the opening and closing price was circulated to SGX Members on 25 July Any bid/offer at a given price may also be executed at a lower/higher price. The cumulative bid volume at any price is the bid quantity at that price plus the sum of bid quantities at all higher prices. The cumulative offer volume at any price is the offer quantity at that price plus the sum of offered quantities at all lower prices. Sell/buy pressure occurs when the cumulative offer/bid volume exceeds the cumulative bid/offer volume at a particular price. The tradable volume at any price is the smaller of the cumulative bid/offer volume. The price overlap is the range of prices where tradable volumes are possible. The equilibrium is the price range within the price overlap where buy pressure changes to sell pressure. The equilibrium price is either one of the following: The price within the equilibrium that has the largest trade volume, or If there is no unique price, the average of all prices within the equilibrium with the maximum trade volume, or The average is rounded to the next multiple of the minimum price multiple for this stock in the direction of the previous day s price. If there is no settlement price, the average price is rounded to the next highest price multiple. If there is only buy or sell pressure within the price overlap: The opening price will be one of the following: o with only buy pressure within the price overlap, the highest price within the overlap with non-zero trade volume will be the opening price;

24 o with only sell pressure within the price overlap, the lowest price within the overlap with a non-zero trade volume will be the opening price. If there is no buy and sell pressure within the price overlap, the opening price will be one of the following: The average of all prices within the overlap, or If the price average is not the correct multiple the average is rounded to the next multiple in the direction of the previous day s price. If there is no previous day s price, the average price is rounded to the next highest Example 1 Cumulative Bid Volume price multiple. Bid Qty Price Offer Qty Cumulative Offer Volume Tradable Price Buy/Sell Pressure Tradable Volume S S B B B In this example, trades are possible at prices between $9.30 and $9.70. However, the buy pressure changes to sell pressure from $9.50 to $9.60. Since the same number of trades can be executed at both prices, the opening price would be the one that is closer to the previous day s closing price. Example 2 Cumulative Bid Volume Bid Qty Price Offer Qty Cumulative Offer Volume Tradable Price Buy/Sell Pressure Tradable Volume B B B

25 For example 2, trades are possible at prices between $0.30 and $0.31. There is buy pressure from $0.30 and $0.31. The opening price would be $0.31, which is the highest price within the overlap with a non-zero trade volume.

26 Table 1. Frequency and size of call trades and number of quotes in the pre-opening and pre-closing periods. On August 21, 2000, the Singapore Exchange introduced an opening call and a closing call on all firms traded on the exchange. In Panel A, we present the percentage of days with either an opening or closing call and with both an opening and closing call. In addition, we present the percentage of days with a change in the best bid or ask price (quote change) in the pre opening period and in the pre closing period. In Panel B, we present the mean number of quote changes in the pre opening period and pre closing period and the mean volume traded in the opening and closing call. We began collecting data 240 trading days (approximately one calendar year) prior to the beginning of the call opening and closing (the pre period) and continue for 240 trading days after August 21, 2001 (the post period). We collect data for the 250 most active firms. We divide the sample into quintiles of 50 firms each (group 1 comprises the most active stocks) based on intraday trading volume (excluding the opening and closing trade) during the post period. To compare the means of two samples, we rank the observations in each sample jointly and perform a t-test on the mean rank for each sample. This is equivalent to a Wilcoxon rank sum test. Full sample Trading volume quintile (1 = most active) Number of firms: Panel A: Percentage of days with calls and quote changes Days (%) with call: Opening Closing Both t-stat. (open vs. close) * * * * * * Days (%) with quote changes Pre open Pre closing Both t-stat. (open vs. close) * * * * * * Panel B: Mean size of call trades and number of quote changes Mean call volume Opening Closing t-stat. (open vs. close) 27,968 30, ,362 64, * 33,691 37, ,780 24, ,122 14, ,668 10, Mean no. of quote changes Pre open Pre close * statistically significant at the 0.05 level