Pacific-Basin Finance Journal 7 Ž1999. 103–127 www.elsevier.comrlocatereconbase
Emerging market transaction costs: Evidence from Indonesia Catherine Bonser-Neal a , David Linnan b, Robert Neal a
a,)
Kelley School of Business, Indiana UniÕersity, 801 W. Michigan Street, Indianapolis, IN 46202, USA b School of Law, UniÕersity of South Carolina, USA
Abstract Despite investor interest in emerging stock markets, relatively little is known about the cost of transacting on these markets. This paper uses transactions data to estimate the execution costs of trading on one emerging market, the Jakarta Stock Exchange ŽJSX. in Indonesia. We find that JSX execution costs appear surprisingly similar to those of non-U.S. developed markets. Factors influencing these costs include the difficulty of the trade, the size of the firm traded, and the broker executing the trade. We also find that trades initiated by foreigners have a much larger price impact than trades initiated by local investors. q 1999 Elsevier Science B.V. All rights reserved. JEL classification: G15 Keywords: Emerging markets; Trading costs
1. Introduction The flow of funds to emerging markets has increased sharply in recent years. For example, the International Finance Ž1997. reports that aggregate net capital flows to emerging markets increased fourfold from US$71.1 billion in 1985 to US$284.6 billion in 1996 Ž1997 Emerging Markets Factbook .. Investor interest in
)
Corresponding author.
[email protected]
Tel.:
q 1-317-274-3348;
fax:
q 1-317-273-3312;
0927-538Xr99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 7 - 5 3 8 X Ž 9 9 . 0 0 0 0 3 - 7
e-mail:
104
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
these markets surged in response to their prospects for rapid economic growth, financial deregulation, and the benefits of international diversification. Despite the widespread interest in emerging stock markets, relatively little is known about the costs of trading on such markets. Previous studies of trading costs have instead focused on developed markets. For example, Bessembinder and Kaufman Ž1997., Chan and Lakonishok Ž1993, 1997. and Keim and Madhavan Ž1996, 1997. estimate trading costs in the United States, while Perold and Sirri Ž1995. estimate trading costs in nineteen non-U.S. developed stock markets. Knowledge of trading costs in emerging markets, however, is important for at least two reasons. First, trading costs have an important effect on emerging market investment strategies. If trading costs are low, then a variety of asset allocation strategies such as those discussed in Harvey Ž1994. are potentially attractive. If trading costs are high, however, then a buy-and-hold strategy may be the only feasible alternative. Second, an analysis of trading costs can provide insight into factors affecting the intraday price behavior in emerging markets. Previous studies have shown that intraday price behavior for U.S. stocks is affected by firm size, trade difficulty, and who conducts the trade. The impact of these variables, however, has not been studied in emerging markets. This paper estimates the costs of trading on one emerging stock market, the Jakarta Stock Exchange ŽJSX.. The JSX is the primary exchange in Indonesia and our analysis covers the period September 1992 through January 1995. This period encompasses an era of rapid economic growth, prior to the Asian Crisis of 1997. We focus on estimating the execution cost component of overall equity trading costs. The execution cost is the implicit cost incurred by investors as a result of the price impact of their trade. Keim and Madhavan Ž1998. conclude from their survey of equity trading costs that execution costs comprise an economically significant share of overall trading costs. Investors also incur explicit costs such as commissions when trading. We do not have data on commissions, however, and so our estimates will tend to understate the overall costs of trading on the JSX. Our methodology parallels Chan and Lakonishok Ž1993.. Using 594,686 transactions, we estimate JSX execution costs using open-to-trade, trade-to-close, and volume-weighted benchmarks. Following Chan and Lakonishok we also estimate the extent to which firm size, trade difficulty and broker identity influence JSX execution costs. Our data is unique, however, in that it also contains an identifier for whether the investor placing the trade is a local or foreign investor. Hence, we can test whether execution costs for trades initiated by foreigners differ from those initiated by locals. Our results show that typical orders have a significant price impact. If the impact is measured by the difference between the trade price and the volumeweighted average price over the day, the round-trip execution costs range from about 1.2% for difficult trades in small firms, to 0.54% for easy trades in large firms. In addition, we show that the price impact of a trade is negatively related to firm size, positively related to the difficulty of the trade, and that brokers with
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
105
large execution costs for purchases tend to have large execution costs for sales. We also find that trades initiated by foreigners have a significantly larger impact on the price than trades initiated by local investors. While this latter result at first may appear consistent with foreigners receiving inferior execution quality, these foreign trades tend to generate price continuations rather than price reversals. While differences in the sample periods and the benchmarks make comparisons across markets difficult, our estimates for the Indonesian market appear surprisingly similar to those for developed countries. For example, using a volumeweighted price benchmark, Chan and Lakonishok Ž1993. report that the average one-way market impact for difficult sales in small NYSE stocks is 0.18%. A roughly comparable sample of easy sales in large Indonesian stocks yields a price impact of 0.29%. Using an average price benchmark to estimate trading costs of similar trades in nineteen non-U.S. developed markets, Perold and Sirri Ž1995. report that the average market impact is 0.36%. Our approach of focusing on transaction costs differs from several recent studies of emerging markets. Bailey and Jagtiani Ž1994. and Domowitz et al. Ž1997. analyze the effects of foreign ownership restrictions in Thailand and Mexico, but do not estimate trading costs. Choe et al. Ž1998. examine Korean trades, but their focus is to determine whether foreign traders exhibit ‘herding’ behavior that destabilizes the Korean market. They find evidence of herding, but show that foreign trades are not destabilizing. Their estimates of the market impact of foreign trades, however, appear smaller than the estimates that we report. The paper is organized as follows. Section 2 discusses the institutional features of the JSX. Section 3 describes the characteristics of the data set and provides summary statistics. Section 4 presents estimates of execution costs on the JSX. This section also estimates the extent to which firm size, trade complexity, and broker and investor identities affect these execution costs, and compares the execution costs on the JSX with the costs for the U.S. and other developed stock markets. Section 5 concludes.
2. Characteristics of trading on JSX The JSX is the largest of the three exchanges operating in Indonesia during our sample period. Since April 1992, the JSX has been a privately operated exchange which is self-regulated subject to the authority of Bapepam, the national capital market regulatory agency. As of the end of our sample in January 1995, 229 firms were listed on the exchange with a market capitalization of roughly US$45.4 billion. Previous studies of the Indonesian market have focused on regulatory issues and the behavior of daily returns. Examples include Linnan Ž1994. and the stock price studies of Roll Ž1995. and Chang et al. Ž1995.. Transactions on the JSX are classified into six groups, or boards. The first and largest group is the regular board, comprising 91% of trades and 45.1% of trading
106
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
volume. During the period of our study, the JSX operated as a manual order-based trading system. 1 The regular ‘board’ is actually a whiteboard where the two best bids and offers are publicly posted. 2 Trades are effected by writing the customer’s transaction on the board and then updating the bid and ask orders. Unlike other exchanges, the JSX does not have active market makers. The posted bids and offers come directly from the underlying investors and not from market makers dealing for their own accounts. In addition, the identity of the buying and selling brokers is disclosed for each trade. Trades by foreign investors are written on the board in a separate color. All trades on the remaining boards require direct negotiation by exchange members. The second largest board, the crossing board, represents 7.1% of trades and 42% of the Rupiah volume of trading. These trades are similar to ‘upstairs’ trades on the NYSE. Transactions written on the crossing board are trades in which a single broker represents both sides of the transaction. These transactions, which are common for large orders, are executed in-house and then reported to the exchange. The remaining boards for block trades, odd lot trades, cash trades, and foreign board trades, together comprise less than 1.9% of trades and 12.9% of trading volume. Block trades are negotiated trades of 200,000 or more shares between exchange members. Odd lot trades are small trades with order sizes less than 500 shares. Cash trades are trades with nonstandard settlement terms. Finally, foreign board trades are negotiated trades between foreigners. These trades occur when the stock in question has reached the limits of permissible foreign ownership, equal to 49% of the outstanding shares during the period underlying our study. Upon reaching this ceiling, foreigners cannot buy more of a stock from a local, but they can purchase more from another foreigner, using either the foreign or crossing board. 3 Trading on the foreign board is thin, and accounts for only 1.7% of trades and 5% of trading volume by value. 4 For all trades, settlement occurs four days after the transaction.
3. Data description and summary statistics 3.1. Data description Our data consists of individual records for all transactions on the JSX from September 1, 1992 until January 31, 1995. Each record contains the stock code, 1
After our sample period, the JSX is converted to a computerized trading system. The minimum price interval for posted prices is 25 Rupiah, or slightly more than US$0.01. 3 Restrictions such as this are common in emerging markets. Bailey and Jagtiani Ž1994. examine restrictions in Thailand while Domowitz et al. Ž1997. analyze Mexico. 4 Restrictions on foreign ownership of nonbank firms were removed in Fall 1997 following the Asian currency crisis. 2
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
107
date, transaction price, number of shares, and Rupiah value of the trade. Data on commissions is not provided. Unlike the U.S. transaction data, however, each record contains the identity of the buying and selling brokers and an indicator of whether the customer represented by the broker is a domestic or foreigner investor. Our estimates of execution costs on the JSX are based solely on the regular board trades. We impose this restriction for two reasons. First, the regular board is the most active board, representing 91% of all transactions. It is also the board on which a typical investor is more likely to place a trade. Second, the transactions are not time-stamped; rather, trades are ordered in the sequence they occur. Since the time of the trades is unknown, it is not possible to construct an integrated sample of trades from all the boards. This restriction, however, means that our estimates reflect the cost of trading on the regular board only. Consequently, we do not estimate the costs of the large negotiated trades reported on the crossing board. Similarly, we do not estimate the cost of trades by foreign investors on the foreign board, which occur when foreign ownership exceeds 49%. Given our sample of regular board trades, we apply two additional screens to the data. First, we require stocks to trade for at least 20 days. We restrict our sample to ‘seasoned’ stocks to avoid any price effects associated with the initial public offering. Second, we require that a stock trades at least three times per day. This screen eliminates stocks with minimal liquidity and allows us to estimate the open-to-trade return and the trade-to-close return. Using these returns, we can then test whether the price impact of a trade is subsequently reversed. A consequence of this filter, however, is that the sample of stocks we analyze changes daily. For
Fig. 1. Average closing price of JSX stocks that traded at least three times per day.
108
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
example, if a stock trades four times on date t and two times on date t y 1, then it would be included in the sample only on date t. To estimate the execution costs of our filtered sample, we must first distinguish between buys and sells. This is difficult because bid and ask prices are unavailable and the transactions are not time-stamped. Since the data preserves the transaction sequence, however, we use the tick test to classify trades. If the price change between transactions is positive, then the transaction is coded as a buy-initiated trade. A negative price change yields a sell-initiated trade. In cases when the price change is zero, we compare the trade price P Ž t . with the trade price P Ž t y 2.. If that price change is zero, we compare P Ž t . with P Ž t y 3. and so on up to P Ž t y 5. until we obtain a non-zero price change. If the price change P Ž t . y P Ž t y 5. is still zero, then this trade is omitted. This procedure should be fairly accurate for the JSX data because virtually all buy orders are executed at the ask and sell orders at the bid. However, trades can be misclassified when large price moves cause the bid price at time t q 1 to exceed the ask price at time t. We discuss the potential for misclassification errors more fully in Section 3. Our final sample consists of 594,686 regular board trades. Fig. 1 graphs the average closing price for our sample from September 1, 1992 to January 31, 1995. The figure shows a significant upward trend in prices until early 1994, followed by a sharp downturn. 5 Over the entire sample, there are slightly more buys than sells, which is consistent with the slight average upward trend in prices over most of our sample period. Fig. 2 graphs the average transaction volume. This figure shows a marked increase in transactions volume in the latter half of our sample. 3.2. Summary statistics for JSX trades Summary statistics for the regular board trades in our sample are presented in Tables 1 and 2. Panel A of Table 1 partitions each firm into quartiles based on the Rupiah value of trading, a proxy for market value. 6 For each of these quartiles, trades are classified according to transaction size and according to whether they are buy- or sell-initiated trades. The largest size quartile accounts for 61.7% of trades and 72.7% of value, while the smallest quartile accounts for only 1.5% of trades and 0.73% of value. Panel B of Table 1 also reveals that the average number of shares per trade is small. Approximately 70% of trades are for less than 5000 shares, while only 1% of trades are for more than 50,000 shares.
5 Our sample displays greater volatility than the overall market. This greater volatility arises because the stocks which comprise our sample change from day to day. 6 The accuracy of this proxy may be affected by the ownership structure of the companies. Many listed firms are still controlled by their founding shareholder families and this reduces the number of shares available for the public to trade.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
109
Fig. 2. Daily trading volume of JSX stocks that traded at least three times per day.
Table 2 presents more detailed statistics on the individual transactions. Panel A shows that for all firms, the mean number of shares per trade equals 7100 for buys and 7500 for sells, and the median number of shares traded is 3000. More than 75% of trades involve less than 10,000 shares per trade. This distribution holds even for the largest firms. Panel B presents the distribution of the Rupiah value of regular board trades. For all trades, the average value per trade is 30.1 million Rupiah for buys and 29.7 million Rupiah for sells. Using the 1992 to 1994 average exchange rate value of Rupiah 2092.6 per dollar, this translates into dollar values of US$14,384 for buys and US$14,193 for sells. 7 The median values are considerably smaller, 10.8 million Rupiah ŽUS$5161. for buys and 10.3 million Rupiah ŽUS$4922. for sells. These medians are less than 10% of the NYSE medians reported by Chan and Lakonishok Ž1993.. More than 25% of trades are less than 5 million Rupiah ŽUS$2389., and less than 10% of trades involve values greater than 100 million Rupiah ŽUS$47,787.. For large firms traded, the values are slightly higher. Still, less than 10% of the trades involving large firms exceed 100 million Rupiah. Panel C of Table 2 presents statistics on trade size in relation to normal trading volume. This is a proxy for the difficulty of executing the trade. Normal trading volume for a particular day is defined as the average daily trading volume over the
7
The exchange rate average is computed from the period average exchange rates for 1992, 1993, and 1994 listed in the IMF’s International Financial Statistics.
110
Panel A. Number of trades and Rupiah value traded Žin parentheses. for all buys and sells in the sample, and in each category of firm size. Firm size is measured by the quartiles of average Rupiah trade value. Group 1 comprises firms in the smallest 25% of JSX Rupiah value of trading and group 4 comprises firms in the top 25% of JSX Rupiah trading value. 1 small
Buys Sells Total
2
3
4 Žlarge.
Total
a trades Ž000.
Rupiah Žtrillion.
a trades Ž000.
Rupiah Žtrillion.
a trades Ž000.
Rupiah Žtrillion.
a trades Ž000.
Rupiah Žtrillion.
a trades Ž000.
Rupiah Žtrillion.
5.47 3.66 9.13
0.07 0.06 0.13
32.80 25.68 58.48
0.60 0.47 1.07
85.22 75.97 161.19
2.00 1.67 3.67
195.82 172.49 368.31
6.93 6.04 12.97
319.32 277.81 597.13
9.61 8.24 17.85
Panel B. Frequency distribution of trade size Žnumber of shares traded. for all buys Žall sells in parentheses. in sample. a shares traded Ž000.
% of observations
Cumulative % of observations
0–2 2–5 5–10 10–20 20–30 30–50 50q
36 Ž35. 35 Ž35. 15 Ž15. 8 Ž8. 4 Ž4. 2 Ž2. 1 Ž1.
36 Ž35. 71 Ž70. 86 Ž85. 93 Ž93. 97 Ž97. 99 Ž99. 100 Ž100.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
Table 1 Sample characteristics of all JSX regular board trades by firms with at least three trades per day from September 1, 1992 to January 31, 1995
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
111
Table 2 Mean, standard deviation, and fractiles of distribution of JSX regular board trades All Buys
Sells
1 Žsmall.
2
Buys
Buys
Sells
Panel A: shares traded (in thousands) Ž7.5. Ž5.1. Mean 7.1 3.9 Ž17.5. Ž16.3. Standard 16.3 10.4 deviation Ž3.0. Ž2.0. Median 3.0 2.0 Ž1.0. Ž0.5. 10% 0.5 0.5 Ž1.5. Ž1.0. 25% 1.5 1.0 Ž8.0. Ž4.5. 75% 7.5 3.5 Ž15.0. Ž9.5. 90% 15.0 7.0 Ž25.0. Ž15.0. 95% 25.0 10.0 Ž60.0. Ž70.0. 99% 52.5 50.0 Panel B: Rupiah Õalue of trade (millions) Ž29.7. Ž15.1. Mean 30.1 12.3 Ž95.6. Ž57.9. Standard 98.7 46.4 deviation Ž10.3. Ž4.5. Median 10.8 4.4 Ž2.1. Ž1.1. 10% 2.2 1.2 Ž4.3. Ž2.1. 25% 4.5 2.2 Ž25.9. Ž10.3. 75% 26.6 9.9 Ž60.8. Ž24.0. 90% 61.3 20.8 Ž44.4. 95% 105.0 Ž103.8. 34.9 99% 325.1 Ž322.5. 170.0 Ž231.0.
3 Sells
4 Žlarge.
Buys
Sells
Buys
Sells
4.9 13.7
Ž5.2. Ž15.8.
6.6 14.8
Ž6.9. Ž15.3.
7.8 17.4
Ž8.1. Ž18.6.
2.5 0.5 1.0 5.0 10.0 13.0 47.5
Ž2.5. Ž0.5. Ž1.0. Ž5.0. Ž10.0. Ž15.0. Ž50.0.
2.5 0.5 1.0 6.0 13.5 25.0 50.0
Ž3.0. Ž0.5. Ž1.5. Ž7.0. Ž15.0. Ž25.0. Ž51.5.
4.0 1.0 1.5 9.0 17.0 25.0 60.0
Ž4.0. Ž1.0. Ž1.5. Ž10.0. Ž18.5. Ž25.0. Ž70.0.
18.4 73.8
Ž18.4. Ž72.0.
23.5 74.2
Ž22.0. Ž75.7.
35.4 111.5
Ž35.0. Ž106.2.
7.5 1.8 3.5 16.0 31.6 51.3 205.2
Ž7.1. Ž1.7. Ž3.3. Ž15.6. Ž32.3. Ž54.7. Ž211.3.
8.1 1.8 3.5 19.2 43.6 78.3 287.0
Ž7.6. Ž1.6. Ž3.2. Ž18.0. Ž40.9. Ž72.5. Ž270.0.
13.7 2.8 5.7 33.0 75.5 124.0 360.0
Ž13.2. Ž2.7. Ž5.4. Ž32.0. Ž75.0. Ž124.0. Ž355.0.
Panel C: trade size in relation to normal trading Õolume a Mean 0.207 Ž0.215. 1.809 Ž2.891. 0.649 Standard 3.944 Ž4.712. 8.926 Ž29.200. 8.468 deviation Median 0.025 Ž0.023. 0.377 Ž0.396. 0.095 10% 0.004 Ž0.004. 0.062 Ž0.063. 0.015 25% 0.009 Ž0.008. 0.140 Ž0.146. 0.035 75% 0.073 Ž0.070. 1.099 Ž1.333. 0.270 90% 0.226 Ž0.213. 3.333 Ž3.999. 0.794 95% 0.486 Ž0.457. 6.557 Ž8.830. 1.709 99% 2.499 Ž2.597. 23.808 Ž25.262. 7.317
Ž0.719. Ž6.983.
0.248 4.730
Ž0.260. Ž5.188.
0.071 1.059
Ž0.071. Ž0.664.
Ž0.091. Ž0.014. Ž0.033. Ž0.278. Ž0.916. Ž2.105. Ž9.523.
0.036 0.006 0.014 0.101 0.287 0.596 2.786
Ž0.036. Ž0.006. Ž0.013. Ž0.099. Ž0.279. Ž0.582. Ž2.819.
0.017 0.003 0.006 0.043 0.107 0.195 0.761
Ž0.016. Ž0.003. Ž0.006. Ž0.043. Ž0.108. Ž0.200. Ž0.784.
The sample comprises all JSX regular board trades of firms with at least three trades per day from September 1, 1992 to January 31, 1995. Summary statistics are reported for buys Žsells in parentheses. in the entire sample, and in each category of firm size Žaverage Rupiah value of shares traded.. a Normal daily volume is computed as the average daily volume over a prior 20-day interval.
previous twenty days. 8 For all firms, the median value of trade size to normal trading volume is 2.5% for purchases and 2.3% for sells. The mean value of trade 8
Using average trading volume over the previous 60 days produced very similar results.
112
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
size to normal trading volume, however, is much larger, averaging 20.7%. The difference between the mean and the median arises from a small number of larger trades. For small firms, the median value of trade size to normal trading volume is considerably larger, 37.7% for buys and 39.6% for sells. For the largest firms, the median ratio of trade size to normal trading volume is much smaller, 1.7% for buys and 1.6% for sells. 9 4. Execution costs on JSX Execution costs are estimated by measuring the price impact of a trade. Purchases and sales are examined separately to avoid the positive price impact of purchases offsetting the negative price impact of sales. Important to these calculations is the benchmark used to evaluate the impact of the trade. Following Chan and Lakonishok Ž1993., we examine the price impact in four ways. First, we calculate the open-to-trade return as a measure of the initial impact of the trade. Second, we calculate the trade-to-close return to determine whether the effect of the trade is reversed within the day. Our third measure of price impact is the sum of the open-to-trade and trade-to-close returns, or the open-to-close return. This measure combines the effects of the initial impact and any subsequent price reversal. 10 Finally, we compare the trade price with the volume-weighted average of all trade prices in that stock on the trade date. Each of these methods estimate the trading costs by taking the difference between the trade price and a benchmark price. Unfortunately, there is no consensus regarding the optimal benchmark. As Perold and Sirri Ž1995. note, not only can the act of trading affect the benchmark, but the choice of benchmark can affect the trading strategy. For example, consider a large trade near the end of the day. Since the trade will likely affect the closing price, it will also affect the trading costs of all trades that use the closing price benchmark. Alternatively, consider a large trade made shortly after the open. While the trade is unlikely to affect the open price, it gives a trader the ability to ‘game’ the benchmark. Suppose the benchmark is the opening price and the trader’s compensation is closely linked to his ability to minimize trading costs. If the price has declined since the open, all buy-initiated trades will have a negative trading costs, regard9 The difference in the median normalized trading volume ratios between small and large firms is largely a consequence of trading volume. If a stock has only three equal sized trades a day, the ratio, all else constant, will be 33%. However, if the stock trades 20 times a day, the ratio will fall to 5%. 10 Chan and Lakonishok Ž1993. and others define the open-to-trade percentage change as the ‘total’ impact, the trade-to-close as the ‘temporary’ impact, and the open-to-close as the ‘permanent’ impact. Classifying price impacts as ‘total’ or ‘temporary’ presumes that there is a price reversal after the initial impact, with the reversal being the ‘temporary’ component. In our data, however, we often observe a price continuation rather than a reversal. Consequently, we avoid labeling the open-to-trade price change as the ‘total’ impact, and the trade-to-close price change as the temporary impact.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
113
less of the trader’s ability. A third benchmark, the volume-weighted average price lies somewhere between the open and closing benchmarks. It is more difficult for a trader to game the volume-weighted benchmark than the opening benchmark, and a large trade at the end of the day will have less of an effect on the volume-weighted benchmark than on the closing benchmark. Berkowitz et al. Ž1988. also argue that this measure yields more reliable results because the variance of the volume-weighted average is less than the open or closing prices. Nevertheless, given the lack of consensus over the appropriate benchmark, we report all four separate estimates of execution costs in the sections below. Section 4.1 reports our estimates of execution cost using each of these measures, and Section 4.2 compares these estimates with those from other markets. Section 4.3 examines the effect of brokers on trading costs and Section 4.4 analyzes the differential effect of foreign and domestic trades. 4.1. How large are trading costs on the JSX? Table 3 summarizes the price impact of regular board trades. The estimates are based on stocks with at least three trades per day. The results show that purchases induce a 1.51% average price increase relative to the open. The data also suggest this positive price impact is not reversed after the trade, and in fact even increases. In particular, the average trade-to-close return is 0.31%, making the average cumulative open-to-close return 1.82%. 11 The dispersion of these return measures is large. For the open-to-trade returns, the 90th percentile is 4.88%, while the 10th percentile is 0.73. The price effects tend to be smaller and less disperse using the volume-weighted price as a benchmark. The price impact falls to 0.32% for purchases with a corresponding median of 0.22%. Similar to the findings of Chan and Lakonishok Ž1993. for the U.S., we find the price impact of sales to be smaller than for purchases. As discussed in Keim and Madhavan Ž1996., this difference in price impacts may be due to the cost of locating counterparties for the trades. Across all firms, the average open-to-trade return for sales is y0.50%, one-third of the impact of a purchase. The effect is partially reversed following the trade, as shown by the positive trade-to-close return of 0.13%. The cumulative effect of the trade is therefore y0.37%. The volume-weighted price impact is similar at y0.34%. As with the purchases, the distribution of returns for sales is relatively large. The price changes for sales falling in the 10th and 90th percentile of the open-to-trade returns are y3.39% and 2.18%, respectively. 11
An open-to-close return of 1.82% for stock purchases is surprisingly large. One potential reason is bid–ask bounce. In the next section we show that trades at the open are slightly more likely to be sells, while trades at the close tend are slightly more likely to be buys. Thus, the open-to-close returns incorporate part of the bid–ask spread.
114
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
Table 3 Mean, standard deviation, and fractiles of distribution of market impact for purchases Žpanel A. and sales Žpanel B. Return from: Opening price to trade Ž%.
Trade to closing price Ž%.
Opening to closing Ž%.
Same day volumeweighted price to trade Ž%.
Panel A: purchases Mean Standard deviation Proportion) 0 Median 10th percentile 25th percentile 75th percentile 90th percentile
1.51 3.18 62 0.96 y0.73 0.00 2.60 4.88
0.31 2.60 38 0.00 y1.85 y0.48 1.06 2.60
1.82 4.15 63 1.29 y1.79 0.00 3.39 6.25
0.32 1.67 63 0.22 y1.05 y0.18 0.90 1.83
Panel B: sales Mean Standard deviation Proportion) 0 Median 10th percentile 25th percentile 75th percentile 90th percentile
y0.50 2.91 22 0.00 y3.39 y1.58 0.00 2.18
0.13 2.19 34 0.00 y1.91 y0.66 0.86 2.17
y0.37 3.57 32 0.00 y4.04 y1.94 1.01 3.33
y0.34 1.44 31 y0.18 y1.68 y0.80 0.14 0.88
The sample comprises all JSX regular board trades of firms with at least three trades per day from September 1, 1992 to January 31, 1995. Price impact is measured as the return: from the opening price on the trade date to the trade; from the trade to the closing price on the trade date; from the opening to the closing price on the trade date; and from the volume-weighted average of all transaction prices in the stock on the trade date to the trade.
Following Chan and Lakonishok Ž1993., we analyze the price impact of purchases and sales Žin parentheses. according to the firm size and the complexity, or relative difficulty of the trade. We define complexity as the trade size relative to
Notes to Table 4: Price impact is measured as the return: from the opening price to the trade; from the trade to the closing price; from the opening to the closing; and from the volume-weighted price to the trade. Group 1 comprises firms in the top 25% of JSX Rupiah trading value and group 4 comprises firms in the top 25% of firm size. The complexity levels are divided into quartiles of the average daily trade volume over a prior 20-day period. The easiest complexity Žpanel D. are firms with the smallest 25% average daily volume and the most difficult Žpanel A. are the top 25%. The sample comprises all JSX regular board trades of firms with at least three trades per day from September 1, 1992 to January 31, 1995.
Table 4 Price impact for purchases and sales Žin parentheses., classified according to firm size Žaverage Rupiah value. into quarters, and trade complexity Žtrade size in relation to average daily volume over a prior 20-day period. Size group 2
Size group 3
Largest firms
All firms
Panel A: most difficult Opening price to trade Trade to closing price Opening to closing Volume-weighted price to trade
3.20 Žy0.42. 1.85 Ž1.12. 5.05 Ž0.70. 0.32 Žy0.85.
2.03 Žy0.15. 0.36 Ž0.32. 2.39 Ž0.17. 0.42 Žy0.43.
1.80 Žy0.45. 0.37 Ž0.22. 2.17 Žy0.22. 0.36 Žy0.36.
1.35 Žy0.39. 0.40 Ž0.15. 1.75 Žy0.24. 0.25 Žy0.26.
1.77 Žy0.36. 0.46 Ž0.25. 2.23 Žy0.10. 0.33 Žy0.36.
Panel B: complexity group 2 Opening price to trade Trade to closing price Opening to closing Volume-weighted price to trade
1.89 Žy0.16. 0.68 Ž0.79. 2.57 Ž0.63. 0.61 Žy0.61.
2.07 Žy0.32. 0.16 Žy0.04. 2.22 Žy0.36. 0.50 Žy0.43.
1.92 Žy0.44. 0.26 Ž0.11. 2.19 Žy0.34. 0.39 Žy0.40.
1.47 Žy0.42. 0.31 Ž0.09. 1.78 Žy0.32. 0.31 Žy0.28.
1.67 Žy0.41. 0.28 Ž0.09. 1.95 Žy0.33. 0.35 Žy0.33.
Panel C: complexity group 3 Opening price to trade Trade to closing price Opening to closing Volume-weighted price to trade
2.65 Žy0.13. 1.00 Ž1.87. 3.65 Ž1.74. 0.86 Žy1.21.
1.83 Žy0.53. 0.08 Žy0.02. 1.90 Žy0.55. 0.44 Žy0.50.
1.75 Žy0.49. 0.14 Ž0.06. 1.89 Žy0.42. 0.40 Žy0.42.
1.44 Žy0.50. 0.27 Ž0.08. 1.71 Žy0.41. 0.31 Žy0.29.
1.54 Žy0.50. 0.23 Ž0.07. 1.77 Žy0.42. 0.34 Žy0.33.
Panel D: easiest Opening price to trade Trade to closing price Opening to closing Volume-weighted price to trade
2.37 Žy0.66. 0.48 Ž0.78. 2.85 Ž0.12. 0.61 Žy0.91.
1.69 Žy0.59. 0.21 Ž0.43. 1.90 Žy0.17. 0.52 Žy0.56.
1.12 Žy1.07. 0.16 Ž0.07. 1.29 Žy1.00. 0.32 Žy0.45.
1.22 Žy0.52. 0.31 Ž0.12. 1.53 Žy0.40. 0.25 Žy0.29.
1.22 Žy0.65. 0.28 Ž0.12. 1.49 Žy0.53. 0.28 Žy0.33.
Panel E: all trades Opening price to trade Trade to closing price Opening to closing Volume-weighted price to trade
3.01 Žy0.41. 1.62 Ž1.07. 4.63 Ž0.66. 0.38 Žy0.84.
1.97 Žy0.29. 0.26 Ž0.21. 2.23 Žy0.08. 0.45 Žy0.45.
1.62 Žy0.63. 0.24 Ž0.12. 1.86 Žy0.51. 0.36 Žy0.41.
1.35 Žy0.47. 0.31 Ž0.11. 1.66 Žy0.37. 0.28 Žy0.28.
115
Smallest firms
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
Return Žin %. from:
116
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
the average daily trade volume over the last 20 days. Both firm size and trade complexity are divided into quartiles. All else constant, trades in small firms should have a greater impact because they are less liquid and likely represent a larger fraction of shares outstanding. Similarly, large or more difficult trades should also have larger price impacts. Chan and Lakonishok offer two reasons for these effects. First, at least in the short term, the supply and demand curves for individual stocks are less than perfectly elastic. Increases in demand require higher stock prices to elicit additional supply. Second, large trades are more likely to be information-based than small trades, and thus are likely to have a greater impact. Table 4 provides some evidence that firm size and trade complexity affect execution costs. Aggregating across all levels of trade complexity, Panel E shows the market impact of purchases tends to be greater in small firms than in large firms. For sales Žin parentheses. the relationship is less robust. While sales in smaller firms generate larger price changes using the volume-weighted-price benchmark, a clear pattern does not emerge when using the open-to-trade or trade-to-close benchmarks. The last column of Table 4 summarizes the effects of trade complexity on the price impact across all firms. For purchases, the price impact generally rises with the difficulty of the trade. For sales, however, the relation between trade complexity and price impact is ambiguous. Indeed, using open-to-trade and open-to-close returns we find that the relation between trade difficulty and price impact goes in the opposite direction: the less difficult sales are associated with the larger price impact. The results also demonstrate that substantial differences among the benchmarks remain. In general, the open-to-trade execution cost is higher than the volumeweighted execution cost, which, in turn, is larger than the trade-to-close execution cost. While our findings are consistent with Chan and Lakonishok Ž1993., we leave to future research an explanation for the differences among benchmarks. 4.2. How do execution costs on the JSX compare with other markets? It is difficult to compare our estimates of Indonesian trading costs with other studies because differences in methodology can have a large effect on estimated trading costs. For example, Bessembinder and Kaufman Ž1997. estimate the average one-way price impact of NYSE trades is 0.297%. Keim and Madhavan Ž1997. report a similar price impact, 0.31% for purchases and y0.34% for sells. Keim and Madhavan’s Ž1996. estimate of the impact of block trades is larger, 1.50% for purchases and y1.60% for sells. Using a benchmark computed from the average of the closing prices the day before and the day of the trade, Perold and Sirri Ž1995. find the average one-way impact of all trades in 18 developed countries is 0.42%. In contrast, Chan and Lakonishok Ž1993. report much smaller execution costs. Their average NYSE round trip execution costs are 0.16% for open-to-trade, 0.08% for trade-to-close, 0.24% for open-to-close, and 0.04% for volume-
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
117
weighted-price-to-trade. While the corresponding estimates for JSX stocks are much larger, a more balanced comparison controls for differences in liquidity and trade difficulty. One reasonable approach is to contrast Chan and Lakonishok’s Ž1993. estimates for difficult sales in US small stocks with our estimates for easy sales in large Indonesian stocks. Using a volume-weighted measure, the Indonesian execution costs are 0.29% compared to the NYSE execution costs of 0.18%. Similarly, using an average price benchmark, Perold and Sirri Ž1995. report that the average market impact of international trades between US$10,000 and US$100,000 in value is 0.36%. Though our results suggest that JSX execution costs are of similar magnitude to those in non-U.S. developed countries, several caveats apply. First, our sample period is one in which the market was growing rapidly. It is unclear whether our results generalize to the current period, especially given the turbulence following the Asian Crisis of 1997. Second, our relatively low estimates of transactions costs reflect could low liquidity in the market. In particular, suppose the absence of liquidity for trades of real economic size causes larger trades to remain unexecuted. In this case, the trading costs will be estimated only from the sample of small trades that are executed. This selection bias will make our estimates of execution costs appear artificially low. While our sampling criteria reduces this bias by excluding illiquid stocks, the selection bias makes cross-exchange comparisons difficult. A third caveat is that our estimates may be an inaccurate measure of the average trading costs of foreign investors. A foreign investor trades on the regular board if the stock has not reached the foreign ownership limit, but will trade on the foreign board if the stock has reached the ownership limit. If the liquidity of stocks on the two boards differs, then the costs of trading on the foreign board may not equal the cost of trading on the regular board. Because our estimates are based only on regular board trades, they may under- or overestimate the costs faced by a foreigner who trades on both the regular and foreign boards. Finally, our methodology may impart a downward bias to our transaction cost estimates. One potential bias arises from the misclassification of the buyrsell indicator. 12 For example, suppose a sell-initiated trade is incorrectly classified as a buy-initiated trade. Such a misclassification could arise if a large price increase caused the bid price at time t q 1 to exceed the ask price at time t. In this case, a sell-initiated trade at time t q 1 would generate a positive price change and thus be classified as a buy-initiated trade. This misclassification imparts a downward bias to our average transaction cost estimates because the true impact of the buy-initiated trade should use the higher ask price at time t q 1, rather than the lower bid
12 Misclassification of the buyrsell indicator is not an issue in Chan and Lakonishok Ž1993., Keim and Madhavan Ž1996, 1997. or in Perold and Sirri Ž1995. because their data identifies transactions as buy-initiated or sell-initiated.
118
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
price at time t q 1. As a result, our procedure may underestimate JSX execution costs. To assess the magnitude of this bias, we compared the execution costs for stocks with absolute daily returns exceeding 10% with those for all other stocks. The stocks with large price moves are more likely to generate misclassification errors. For trades classified as buys, we find the estimated transaction costs are lower for the large price change stocks. While this is consistent with the misclassification hypothesis, sell-initiated trades produce the opposite result. Hence, misclassification errors seem unlikely to impart an overall bias to the transaction cost estimates. A second potential bias arises from intraday price trends. Suppose the opening trade tends to be a sell while the closing trade tends to be a buy. This trend will affect the transaction cost estimates because the opening price benchmark is too low while the closing price benchmark is too high. The effect of this bias, however, is small. Classifying buy-initiated trades as q1 and sell-initiated trades as y1, we find that the mean buyrsell indicator is y0.015 at the open and rises to 0.106 at the close. Since the median bid–ask spread is about 1%, this implies an intraday drift of w0.106 y Žy0.015.xr2, which is about one-sixteenth the size of the spread. 13 All else constant, a positive intraday drift should also imply the impact for sales using the trade-to-close benchmark should be larger than for the open-to-trade benchmark. However, Table 3 shows the opposite result, with the open-to-trade impact larger than the trade-to-close impact. The effects of intraday price trends are therefore unlikely to materially bias our results. A final potential bias stems from information leakage. Keim and Madhavan Ž1996. show that information leakage reduces the measured impact of a trade if the pre-trade price reflects information about the trade. For example, if information about a trade is available prior to the open, then the open price will reflect this information. As a result, the open-to-trade and trade-to-close measures will understate the true impact of the trade. One source of information leakage comes from large orders which may take several days to execute and spillover onto the various negotiated boards. Unfortunately, we have no information on how investors break up their orders, and so this bias is difficult to assess empirically. 14 4.3. Do trading costs differ across brokers? Chan and Lakonishok Ž1993, 1997. and Keim and Madhavan Ž1997. find that the identity of the institution behind the trade influences the price impact. Using 13 The median spread is estimated by the median absolute price change, computed from all non-zero price changes. 14 In Section 4.4, we report evidence of strong persistence in the order flow: purchases are followed by additional purchases, and sells by additional sells. While this is consistent with orders being broken up, it could also arise from a clustering of orders from different investors.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
119
information on buying and selling broker identities on the JSX, we test whether brokers have similar effects on execution costs on JSX execution costs. To analyze this issue, we run the following regression: 4
4
188
R h s a q b D q Ý d i Ch ,i q Ý g j Sh , j q Ý l k Bh , k q u h . is2
js2
Ž 1.
ks2
In this notation, R h is the return for a trade h, measured by the open-to-trade, trade-to-close, and volume-weighted price benchmarks. D is a time dummy that equals 1 for the periods of a down trending market ŽSeptember 1, 1992 to December 14, 1992 and January 14, 1994 to November 23, 1995. and zero otherwise. Ci is a series of four complexity dummies. If the trade is of complexity category i, then the ith dummy is 1 and the other dummies are zero. S j is a series of four size dummies. If the trade is of size category j, then the jth dummy is 1 and the other dummies are zero. Bk is a series of 188 broker dummies. If the trade involves broker category k, then the k th dummy is 1 and the other dummies are zero. The time dummy is included to control for systematic differences in the mean level of returns in a down-trending market versus an up-trending market. The complexity and firm size dummy variables are included to control for the effects of trade difficulty and firm size on execution cost. For the complexity variable, C1 represents trades in the easiest quartile while C4 represents the trades in the most difficult quartile. Similarly, for the firm size variable, S1 represents trades of firms in the smallest quartile while S4 represents trades of firms in the largest quartile. Finally, individual broker dummies are included to determine whether execution costs are affected by the identity of the broker executing the trade, after controlling for firm size and trade complexity. Note that for each group of dummy variables Ži.e., size, trade complexity, and broker identity., one of the dummies in each group must be dropped from the regression to ensure nonsingularity. Therefore, C1 , S1 , and the first broker are dropped from the specification. The regressions are run separately for purchases and sales, and each regression is estimated using open-to-trade, trade-to-close, and volume-weighted-price-totrade price changes. The coefficient estimates and t-statistics are presented in Table 5 Žpurchases. and Table 6 Žsales.. When evaluating these results, it is important to keep in mind that the sample is very large, 318,053 purchases and 276,633 sells. The large sample size drives the standard errors toward zero and t-statistics will not correspond to their usual levels of significance. Using a 20:1 posterior odds ratio following Zellner Ž1984, p. 286., a 95% confidence region for samples of this size requires a t-statistic of roughly 4.5. Table 5 shows that both firm size and trade complexity enter significantly in determining execution costs of purchases. In general, the firm size and trade complexity coefficients are consistent with Table 4. The larger the firm size, the smaller the execution costs. The more difficult the trade, the higher the execution
120
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
Table 5 Regression estimates of the effects of firm size, trade complexity, and broker identity on transaction costs: purchases Independent variable
Constant Time dummy C2 C3 C4 S2 S3 S4 F-statistic for joint Significance of brokers
Dependent variable Open-trade
Trade-close
Volume-weighted price-trade
2.778 Ž27.39. y0.672 Žy56.62. 0.256 Ž16.70. 0.335 Ž21.79. 0.325 Ž20.01. y1.025 Žy21.15. y1.135 Žy24.13. y1.313 Žy27.95. 18.983 w0.0001x
1.212 Ž14.53. y0.076 Žy7.83. y0.019 Žy1.50. 0.026 Ž2.06. 0.123 Ž9.19. y1.315 Žy33.03. y1.319 Žy34.11. y1.263 Žy32.71. 27.229 w0.0001x
0.421 Ž7.80. 0.017 Ž2.68. 0.041 Ž5.05. 0.050 Ž6.16. 0.010 Ž1.20. 0.054 Ž2.09. y0.022 Žy0.88. y0.092 Ž3.70. 12.239 w0.0001x
The dependent variables are the returns from: the open to the trade, the trade to the close, and the volume-weighted price to the trade. The time dummy equals 1 when the market is trending down. The Ci variable is the dummy variable for the trades’ classification by complexity, where C4 includes the most difficult trades. The Si is the dummy variable for the trades’ classification by firm size, where S4 trades are trades involving the largest firms. The sample comprises all JSX regular board trades by firms with at least three trades per day from September 1, 1992 to January 31, 1995. The coefficients are multiplied by 100 and the t-statistic are in parentheses. The significance level for the F-statistic for the joint test that all broker codes are equal to zero is in brackets.
cost. Individual broker codes also enter significantly and the hypothesis that the coefficients are jointly equal to zero is strongly rejected. This suggests that execution costs vary systematically across brokers. Table 6 shows that firm size and trade complexity also significantly affect the execution costs of sells. Execution costs are generally higher for trades in small firms, and lower for trades in the largest quartile. The trade complexity coefficients, however, are ambiguous. In particular, there is no systematic tendency for execution costs to rise with the difficulty of the trade. As with the buy transactions, execution costs are affected by the identity of the broker behind the trade. To further examine the effect of broker identity on execution costs, Fig. 3 plots the 187 individual broker coefficients underlying the results in Tables 5 and 6. The vertical axis measures the execution cost for purchases, while the horizontal axis measures the cost for sells. The broker coefficients are obtained from the regressions using volume-weighted-price-to-trade price changes as the dependent variable. 15
15 Note that most of the brokers in Fig. 3 appear to have negatiÕe execution costs for purchases. The negative values are a consequence of excluding a category from each of the dummy variables, causing the conditional mean to differ from the unconditional mean.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
121
Table 6 Regression estimates of the effects of firm size, trade complexity, and broker identity on transaction costs: sales Independent variable
Constant Time dummy C2 C3 C4 S2 S3 S4 F-statistic for joint Significance of brokers
Dependent variable Open-trade
Trade-close
Volume-weighted price-trade
y0.342 Žy3.36. y0.658 Žy55.42. 0.176 Ž11.74. 0.293 Ž19.23. 0.418 Ž25.65. 0.206 Ž3.87. 0.140 Ž2.68. 0.446 Ž8.53. 16.986 w0.0001x
0.883 Ž11.45. 0.085 Ž9.45. y0.021 Žy1.86. y0.001 Žy0.11. 0.148 Ž12.00. y0.801 Žy19.57. y0.861 Žy21.68. y0.824 Žy20.79. 9.53 w0.0001x
y0.794 Žy15.64. y0.018 Žy3.12. 0.003 Ž0.37. 0.017 Ž2.19. 0.035 Ž4.27. 0.392 Ž14.54. 0.452 Ž17.30. 0.573 Ž21.94. 5.974 w0.0001x
The dependent variables are the returns from: the open to the trade, the trade to the close, and the volume-weighted price to the trade. The time dummy equals 1 when the market is trending down. The Ci variable is the dummy variable for the trades’ classification by complexity, where C4 includes the most difficult trades. The Si is the dummy variable for the trades’ classification by firm size, where S4 trades are trades involving the largest firms. The sample comprises all JSX regular board trades by firms with at least three trades per day from September 1, 1992 to January 31, 1995. The coefficients are multiplied by 100 and the t-statistic are in parentheses. The significance level for the F-statistic for the joint test that all broker codes are equal to zero is in brackets.
To investigate whether there is a systematic tendency for broker execution quality on the buy side to mirror execution quality on the sell side, we estimate the correlation between individual broker coefficients from the purchase and sales regressions. The resulting correlation is y0.52 and statistically significant at the
Fig. 3. Average execution costs by broker. Each date point in the plot is the estimated execution cost for an individual broker. The estimates are obtained from the broker regression coeficients in Eq. Ž1.. The buy-initiated coefficient is plotted on the vertical axis and the sell-initiated coefficient on the horizontal axis.
122
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
one percent level. This suggests that brokers that offer good execution on purchases also offer good execution on sells. Similarly, brokers that offer poor execution on purchases also offer poor execution on sells. The dispersion of execution costs among brokers is large, with some brokers having costs that are 1% larger than the typical broker. This is much larger than the negotiated commissions, which are typically about 0.5%. 16 Differences in broker size may be one reason execution quality differs across brokers. In particular, large brokers with extensive research capabilities and access to information may offer superior execution quality compared to those without such services. To test this hypothesis, we categorize the 188 brokers into four groups, based upon the value of their trading, and reestimated regression Ž1. replacing the individual broker dummy variables with dummy variables representing the four groups. The first group includes the ten brokers with the largest trading volume. These brokers are generally joint-venture brokers that have established a relationship with a large foreign broker, such as Jardine Fleming or Morgan Grenfell. Many joint-venture brokers have a predominantly foreign customer base and offer research and other services. The second group includes the next largest 20 brokers, a mixture of joint venture brokers and the larger local brokers. The third group includes the smaller local brokers, ranked 31–139, while the fourth group includes the smallest brokers ranked 140 and below which typically trade for their own account. Table 7 reports the regression results using the volume-weighted price as the benchmark. Overall, the results suggest that broker size matters. All else constant, the largest broker group achieves superior execution for both buys and sells. Specifically, for purchases, the B1 coefficient is negative and significant, indicating that trades by the largest brokers paid a lower than average price. Similarly, on the sell side, trades by the largest brokers receive a price that is slightly higher than average. The dispersion of these results across brokers is, however, sufficiently large so that the execution quality of the best small brokers may exceed those of the largest brokers. 4.4. Do trading costs differ for foreign inÕestors? Since our data set distinguishes between foreign and domestic trades, we can test whether the investor identity affects execution costs. One reason execution costs might differ is that foreign and local investors may have different information sets. For example, consistent with the model presented in Brennan and Cao 16
Conversations with market participants suggest that commissions average roughly 0.5%. Some brokers which offer research and other services charge more, while others with more limited services charge less. Broker commissions are subject to negotiation, and they are limited to a maximum of 1% for each transaction.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
123
Table 7 Effects of broker size on transaction costs Independent variable
Constant Time dummy C2 C3 C4 S2 S3 S4 B1 B3 B4
Dependent variable Purchases
Sales
Volume-weighted price to trade
Volume-weighted price to trade
0.358 Ž13.90. 0.013 Ž2.07. 0.045 Ž5.55. 0.052 Ž6.36. 0.004 Ž0.44. 0.060 Ž2.34. y0.018 Žy0.71. y0.098 Žy3.96. y0.133 Žy14.12. 0.025 Ž3.50. y0.030 Žy1.45.
y0.881 Žy32.78. y0.019 Žy3.26. 0.0003 Ž0.494. 0.020 Ž2.59. 0.040 Ž4.96. 0.39 Ž14.68. 0.45 Ž17.39. 0.582 Ž22.34. 0.077 Ž8.61. 0.008 Ž1.19. 0.037 Ž2.03.
The dependent variables are the returns from the volume-weighted price to the trade. The broker dummies Bi classify brokers into groups according to the trading volume, where B1 includes the brokers with the highest trading volume and B4 includes brokers with the lowest trading volume. The other variables are defined in Tables 5 and 6. The coefficients are multiplied by 100 and the t-statistics are in parentheses.
Ž1997., locals may have superior information regarding the earnings prospects of Indonesian firms. In this case, price impact of the local trades would exceed that of the foreigners. 17 Differences in information about future order flow could also cause execution costs for foreign and local investors to be unequal. For example, institutional investors often spread their trades across time ŽChan and Lakonishok, 1995., and more likely to engage in momentum investment strategies ŽGrinblatt et al., 1995.. If institutional investors comprise a large portion of foreign trades, then foreign trades could signal additional information about future order flow. Alternatively, Aitken Ž1996. and Buckberg Ž1996. suggest that foreign trades contain information regarding future order flow because of asset allocation strategies. They argue that emerging markets are treated as an asset class, whereby foreign investors first allocate a share of their portfolio to emerging markets and then allocate that share among the individual emerging market countries. In either case, foreign trades may signal additional order flow and thereby have a greater price impact than trades from domestic investors.
17 However, it is possible that some of the foreign trades are actually trades initiated by domestic investors using foreign accounts. Conversations with market observers suggest the magnitude of ‘false foreigner’ trading is relatively small.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
124
To test for a differential impact of foreign trades, we estimate the regression: 4
4
R h s a q b D q Ý d i Ch ,i q Ý g j Sh , j q l Fh q u h , is2
Ž 2.
js2
where Fh is a dummy variable that takes the value one if trade h was initiated by a foreign investor with a local investor taking the opposite side, and zero otherwise. The other variables are defined as before. Because we are interested in the dynamics of how information contained in foreign trades is impounded in the price, we focus on the open-trade and trade-close measures of returns. All else constant, the open-to-trade return measures the initial impact of the trade. Similarly, the trade-to-close return measures whether the price reverts following the trade or whether the trade reveals additional information to the market. Table 8 presents the regression results. Controlling for trade difficulty and firm size, we find that foreign-initiated trades have a significant price impact. For example, Table 3 shows the average price impact for all sales using the open-totrade benchmark is y0.50%. Using the open-to-trade benchmark, Table 8 shows the incremental effect of foreign-initiated sales is y0.52%. Thus, the impact of a foreign-initiated sell is substantially larger than the impact of the average sell. Table 3 also shows a trade-to-close price reversal for sales of 0.13%. In contrast, Table 8 shows the trade-to-close return for foreign-initiated sales is y0.23%. Thus, foreign-initiated sales induce further price declines rather than price reversals. These results indicate that our estimates of the impact of foreign trades are larger than those reported in Choe et al. Ž1998.. Precise comparisons, however, are difficult because Choe, Kho, and Stulz, use an intra-day event study methodology
Table 8 Effects of investor type on transaction costs Independent variable Dependent variable Purchases
Constant Time dummy C2 C3 C4 S2 S3 S4 Foreign
Sales
Open-trade
Trade-close
Open-trade
Trade-close
2.881 Ž61.03. y0.705 Žy60.95. 0.286 Ž18.55. 0.363 Ž23.50. 0.324 Ž19.90. y0.998 Žy20.64. y1.143 Žy24.37. y1.319 Žy28.27. 0.131 Ž9.95.
1.436 Ž36.96. y0.373 Žy7.08. 0.996 Ž24.99. y0.062 Žy6.48. y0.663 Žy57.20. 0.078 Ž8.85. y0.007 Žy0.52. 0.160 Ž10.67. y0.029 Žy2.57. 0.057 Ž4.51. 0.245 Ž16.22. y0.023 Žy1.97. 0.201 Ž15.00. 0.315 Ž19.84. 0.101 Ž8.42. y1.324 Žy33.29. 0.169 Ž3.11. y0.83 Žy20.28. y1.308 Žy33.90. 0.064 Ž1.22. y0.899 Žy22.68. y1.210 Žy31.52. 0.330 Ž6.33. y0.890 Žy22.53. 0.331 Ž30.51. y0.525 Žy38.44. y0.234 Ž22.59.
The dependent variables are the returns from: the open to the trade and the trade to the close. The dummy variable Foreign equals 1 if a foreign investor initiates the trade with a local investor, and zero otherwise. The other variables are defined in Tables 5 and 6. The coefficients are multiplied by 100 and the t-statistics are in parentheses.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
125
that aggregates foreign buys and foreign sales and does not control for differences in trade characteristics. The large initial impact from foreign trades and the subsequent price continuation has several implications. First, it appears inconsistent with the conjecture that locals have superior information regarding JSX stocks. If the conjecture were true, then the open-to-trade impact of the local trades should be greater than the foreign trades. 18 Second, the large open-to-trade impact of foreign trades is not simply a consequence of foreign trades receiving poor execution for trades. 19 If the impact were solely due to inferior execution, then the trade-to-close return should show a larger price rebound, instead of a continuation. Finally, the price continuation following foreign trades raises the possibility that these trades signal future investment flows. If international institutional investors tend to dominate foreign trading, then the price continuations may be a consequence of the herding behavior discussed in Grinblatt et al. Ž1995. and in Sias and Starks Ž1997.. Consistent with this idea, Choe et al. Ž1998. report strong evidence of herding among foreign investors in Korean stocks. One possible explanation for the price continuations is positively correlated intra-day order flow. To investigate this possibility, we estimate the conditional likelihood that a foreign-initiated buy will be followed by a buy, and similarly for sells. These estimates are then compared to the conditional likelihoods for domestic-initiated trades. The conditional probability that a foreign buy is followed by a buy is 0.91, compared with 0.72 for a domestic buy. For sells, the conditional likelihoods are 0.89 for foreign trades and 0.88 for domestic trades. Under the assumption of independence, the t-statistics for these differences between foreign and domestic conditional likelihoods are 24.7 for buys and 14.7 for sells. While this suggests foreign trades convey additional information, the trade-to-trade persistence in order flow is extremely high for both foreign and domestic trades.
5. Conclusion This paper estimates the costs of transactions in an emerging market, the Jakarta Stock Exchange in Indonesia. Our analysis yields three interesting results. First, the trading costs on the JSX are surprisingly modest. For example, using the volume-weighted average price as a benchmark, the one-way transactions cost of 18
It is possible that foreign trades could have a greater impact even if locals have superior fundamental information. This would require, however, the foreignrdomestic signal to be more important than fundamental information. 19 Agency problems can arise when an investor cannot adequately monitor the quality of trades executed by his broker. This creates an incentive for a broker to provide poor execution quality. Neal and Reiffen Ž1996. discuss a formal model of this agency problem.
126
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
selling large Indonesian stocks is roughly 0.29% of the value of the stock traded. While these estimates are much larger than those from the NYSE, they are similar to those reported in Perold and Sirri Ž1995. for some European exchanges. Second, we find that execution costs are affected by broker identity. There is a wide dispersion of execution quality across brokers, but large joint-venture brokers have better execution quality. However, since we have no data on commissions, we cannot assess whether superior performance is reflected in higher commissions. Finally, we find that trades initiated by foreigners have significantly greater execution costs, even after controlling for the difficulty of the trade and the size of the firm being traded. The finding that the price impact is not subsequently reversed suggests that the information content of foreign-initiated trades differs from domestic-initiated trades.
Acknowledgements The authors gratefully acknowledge the helpful comments of the editor Warren Bailey, Ananth Madhavan Žthe referee., and seminar participants at the Federal Reserve Bank of Kansas City, the 1996 Global Finance Conference, and the 1997 Financial Management Association Meetings. Doug Rolph and Stacy Bear provided excellent research assistance.
References Aitken, B., 1996. Have institutional investors destabilized emerging markets?. International Monetary Fund Working Paper a 96r34, April. Bailey, W., Jagtiani, J., 1994. Foreign ownership restrictions and stock prices in the Thai capital market. Journal of Financial Economics 36, 57–87. Berkowitz, S., Logue, D., Noser, E., 1988. The total cost of transactions on the NYSE. The Journal of Finance 43, 97–112. Bessembinder, H., Kaufman, H., 1997. A comparison of trade execution costs for NYSE and NASDAQ-listed stocks. Journal of Financial and Quantitative Analysis 32, 287–310. Brennan, M., Cao, H., 1997. International portfolio investment flows. Journal of Finance 52, 1851–1880. Buckberg, E., 1996. Institutional investors and asset pricing in emerging markets. International Monetary Fund Working Paper a 96r2, January. Chan, L., Lakonishok, J., 1993. Institutional trades and stock price behavior. Journal of Financial Economics 33, 173–199. Chan, L., Lakonishok, J., 1995. The behavior of stock prices around institutional trades. The Journal of Finance 50, 1147–1174. Chan, L., Lakonishok, J., 1997. Institutional equity trading costs: NYSE vs. Nasdaq. The Journal of Finance 52, 713–735. Chang, R.P., Rhee, S.G., Soedigno, S., 1995. Price volatility of Indonesian stocks. Pacific Basin Finance Journal 3, 337–355. Choe, H., Kho, B., Stulz, R., 1998. Do foreign investors destabilize stock markets? The Korean Experience in 1997. Working paper, Ohio State University.
C. Bonser-Neal et al.r Pacific-Basin Finance Journal 7 (1999) 103–127
127
Domowitz, I., Glen, J., Madhavan, A., 1997. Market segmentation and stock prices: evidence from an emerging market. The Journal of Finance 52, 1059–1085. Grinblatt, M., Titman, S., Wermers, R., 1995. Momentum investment strategies, portfolio performance, and herding: A study of mutual fund behavior. American Economic Review 85, 1088–1105. Harvey, C.R., 1994. Conditional asset allocation in emerging markets. NBER Working paper No. 4623, January. International Finance, 1997. Emerging Markets Factbook, Washington, DC. Keim, D., Madhavan, A., 1996. The upstairs market for large-block transactions: analysis and measurement of price effects. Review of Financial Studies 9, 1–36. Keim, D., Madhavan, A., 1997. Transactions costs and investment style: an inter-exchange analysis of institutional equity trades. Journal of Financial Economics 46, 265–292. Keim, D., Madhavan, A., 1998. The cost of institutional equity trades: an overview. Rodney L. White Center For Financial Research Working Paper a 008-98, March. Linnan, D., 1994. Indonesian capital market development and privatization. In: McLeod, R.H. ŽEd.., Indonesia Assessment 1994: Finance as a Key Sector in Indonesia’s Development, Canberra and Singapore: Research School of Pacific and Asian Studies, Australian National University and Institute of Southeast Asian Studies, pp. 223–247. Neal, R., Reiffen, D., 1996. The effect of integration between brokersrdealers and specialists. In: Lo, A. ŽEd.., The Industrial Organization and Regulation of the Securities Industry, Univ. of Chicago Press, Chicago, pp. 177–206. Perold, A.F., Sirri, E.R., 1995. The cost of international equity trading. Working Paper, Securities and Exchange Commission. Roll, R., 1995. An empirical survey of indonesian equities 1985–1992. Pacific Basin Finance Journal 3, 159–192. Sias, R., Starks, L., 1997. Return autocorrelation and institutional investors. Journal of Financial Economics 46, 103–131. Zellner, A., 1984. Bayesian Econometrics and Statistics, Wiley, New York.