Intraday Information, Trading Volume, and Return
Volatility: Evidence from the Order Flows
on the Taiwan Stock Exchange
Edward H. C how
∗
Department of Finance
National Chengchi University
Yi-Tsung Lee
Department of Accounting
National Chengchi University
Yu-Jane Li u
Department of Finance
National Chengchi University
Keywords: Information, Trading volume, Return volatility, Order flow JEL classification: G14, G15, D82
∗ Correspondence: Edward H. Chow, Department of Finance, National Chengchi University, Taipei
116, Taiwan. Tel: (02) 2939-3091 ext. 81206; Fax: (02) 2939-3394; E-mail: echow@nccu.edu.tw. Chow would like to acknowledge the financial support of the National Science Council of the Republic of China (NSC852416H004032). We are grateful to the Securities and Futures Commission, the Taiwan Stock Exchange Corporation, and especially Nai-Kuan Hwang for their assistance in data collection.
ABSTRACT
Relative to the transaction data investigated in the literature, the complete order
flow data we have from the Taiwan Stock Exchange(TSE) is particularly appropriate for examining the intraday relationship between information, trading volume and volatility. We find that traders tend to concentrate their orders, but only the trading volume and volatility of the small stocks are positively associated with the concentration of orders. The liquidity orders do not influence volume and volatility as much as the information orders. The information carried by the large information orders tends to be private rather than public. But the trading volume at the open is not unusually high for our sample, notwithstanding very large order flows. We think that the TSE’s unique order-driven call market without specialists which imposes price limits and allows only limit orders makes traders very conservative at the open. In addition, the market-wide information affects the trading volume and volatility of the large firms through private information. The firm-specific information is not as clear a determinant of trading volume and volatility as the market-wide information. The diversity in the interpretation of information and speculative orders are positively related to trading volume and volatility.
1. INTRODUCTION
A pattern of unusually large trading volume and volatility at the market open and in particular at the close has been found in different markets.1 The pattern is baffling be-cause it cannot be satisfactorily explained by equilibrium asset pricing theories such as the capital asset pricing model. While the contemporaneous correlation between trad-ing volume and return volatility has been well documented,2 little empirical research has been done on the determinants of intraday trading volume and return volatility. As Bessembinder et al. (1996) put it: “Despite the importance of the topic, surprisingly little empirical research has addressed the determinants of trading volume.” This pa-per aims to fill the deficiency by empirically evaluating the effects of information on intraday trading volume and volatility.
In particular, we examine order flows as they relate to information, trading vol-ume, and volatility. A distinct feature of our inquiry is the use of order flow data in constructing variables for the determinants of trading volume and volatility. This is a clear departure from extant empirical research which mainly use bid-ask spread to infer information contents (see, e.g., Huang and Stoll (1994)). For example, Chan, Christie, and Schultz (1995) compare the intraday pattern of bid-ask spreads for NAS-DAQ securities with that of the NYSE to conclude that tests for the importance of information asymmetries in determining intraday spreads, as modeled by Admati and Pfleiderer (1988), Foster and Viswanathan (1990), and Madhavan (1992), must first consider the impact of institutional factors. Chan, Chung, and Johnson (1995) study the intraday behavior of bid-ask spreads for actively traded CBOE options and for their NYSE-traded underlying stocks. They confirm previous findings of McInish and Wood (1992) and Brock and Kleidon (1992) that stocks have a U-shaped spread pat-tern. They suggest that both the degree of competition in market making and the extent
1For instance, there have been findings for the New York Stock Exchange (Wood et al. (1985) and
Lockwood and Linn (1990)), the Toronto Stock Exchange (McInish and Wood (1990)), the Tokyo Stock Exchange (Chang, Fukuda, Rhee, and Takano (1993)), and the Kuala Lumper Stock Exchange (Chang, Kang, and Rhee (1993)).
2
See, for example, Karpoff (1987), Gerety and Mulherin (1992) for the positive relation between trading volume and price volatility, Mitchell and Mulherin (1994) for the relation between volume and public information flows, French and Roll (1986) and Jones et al. (1994) for the effects of public and private information on volatility, and Anderson (1996) for an information flow interpretation of trading volume and return volatility.
of informed trading are important for understanding the intraday behavior of spreads. However, contrary to the empirical evidence that the intraday bid-ask spread pat-tern and trading volume on the NYSE are U-shaped, many information-based models on spreads predict that higher volume is associated with lower spreads (Copeland and Galai (1983), Glosten and Milgrom (1985), Kyle (1985), Easley and O’Hara (1987, 1992), Admati and Pfleiderer (1988), Hasbrouck (1988), Foster and Viswanathan (1990), and Madhavan (1992)). Thus, the feasibility of extracting information con-tent from the bid-ask spread is dubious for our purpose. The order flow data avoid this problem and allow us to design alternative information measures that were not explored before.
Examining order flow data can potentially reveal valuable information that is not available from transaction data. The observed transaction price and volume do not necessarily expose trading behavior adequately. For example, trading volume can be high when order is imbalance or balanced, but the information content in the two cases should be quite different. Order imbalance should be low when traders have very disparate views about market condition, when information asymmetry is severe, or when there is high liquidity demand and supply. Griffiths et al. (2000) find that aggressive buy orders are more likely to be motivated by information and tend to occur when bid-ask spreads are narrow. Hasbrouck and Seppi (2001) and Brown et al. (1997) study the interaction between order imbalance and stock prices. Empirically bid-ask spread is low when order imbalance is low. So when order imbalance is low, there is more private information. And the informed submit orders when order imbalance is low in order to hide their information.
There have been more and more studies that use order imbalance for their probe. For example, Chan and Fong (2000) find that after controlling for the return impact of order imbalance, the volatility-volume relationship becomes much weaker. In our paper we believe that low order imbalance can result from a low degree of conformity of traders’ opinions. Our intuition is that in the absence of information asymmetry, order is more imbalanced because traders’ opinions are more one-sided. Low order imbalance can be associated with high trading volume and with high or low volatility. When volatility is high, our finding implies that there is much information in the mar-ketplace. When volatility is low, there is a lack of information. In this case the trading volume could be due to liquidity demand and supply. Thus, by studying order flow one can better discern the relationship between trading volume and information.
The TSE’s microstructure arguably provides us with a cleaner testing environment than the NYSE. The TSE is an order-driven computerized trading system. There are no specialists on the TSE. Orders are matched by computers through a call mechanism throughout the trading day. The time between two calls is about 45 seconds. Except for the intraday price limit that restricts transaction price from moving more than two ticks from one call to another there are no other artificial arrangements to affect the transaction prices. As such the intraday trading volume and volatility on the TSE are not affected by the complicated interaction between specialists and the informed and the noise traders as considered by, e.g., Stoll and Whaley (1990) and Brock and Kleidon (1992), or by the differences in trading mechanism at the open and the close as examined by Amihud and Mendelson (1987, 1991), Amihud et al. (1990), and Chow et al. (1996).
Since the TSE is a centralized, computerized, order-driven call market, it can readily record the entire order flow in the computer. Our order flow data are more comprehensive than other order flow data that have been examined in the literature in the sense that we have the complete order flow. In contrast, the data used by Biais et al. (1995) cannot be observed for order placement or cancellation outside of the best five bid and offer prices. In addition, the Institute for the Study of Security Markets (ISSM) data summarizing the NYSE transactions includes the prevailing quotations, but not orders away from the quote. The electronic book on the NYSE contains only about 30 percent of the executed trading volume. The electronic order book and order flow data from the NYSE (such as the Trades, Orders, Reports, and Quotes (TORQ) data base) do not include orders such as some market makers’ orders and most large orders. Given the advantage of our data we define variables based on the entire order flow, which avoids the bias caused by examining only a part of the whole order flow.
Our primary interest is in the investigation of the relationship between informa-tion asymmetry and trading volume and volatility. Admati and Pfleiderer (1988) sug-gest that the informed and noise traders tend to concentrate their transactions. The informed benefit more from their private information when noise traders trade, and the competition between the informed lowers the transaction cost for the noise traders if they trade along with the informed. Slezak (1994) argues that if market closure delays the resolution of information uncertainty which imposes excessive risk on the informed, the informed have the motivation to trade before the market close in order to shift the risk to the noise traders. As the market reopens, those who cannot trade overnight trade according to the information revealed during the market closure. Since
trading releases private information, one would expect volatility to be high at the mar-ket open and the close as well.3
For our purpose we control for the effects on trading volume and volatility of other information characteristics. In specific, we distinguish between market-wide and firm-specific information. Consistent with Subrahmanyam (1991) and Gorton and Pennacchi (1993), Bessembinder et al. (1996) find that traders with firm-specific formation trade primarily on the spot equity market, while those with market-wide in-formation trade on both the spot and index futures markets. The relationship between market-wide and firm-specific information and volatility and trading volume depends on whether the information is public or private. We are not able to separate the private information from the public information. But to the extent that a public information is not a surprise, private market-wide and firm-specific information tend to induce higher trading volume than the public ones and the volatility could be high for either public and private information or both.4
Another control variable is the diversity in the assessment of common informa-tion. In the theories of Varian (1986), Harris and Raviv (1993) and Shalen (1993) trading occurs because traders differ either in their prior beliefs regarding value or in their interpretation of common signals. Therefore, trading volume increases with the dispersion of traders’ private evaluation. Bessembinder et al. (1996), using S&P 500 index futures open interest as a proxy for the divergence of traders’ opinions, find ev-idence in support of the theory. If wide disparity in the interpretation of information is caused by the arrival of public and/or private information, volatility should also be high in this case.
The last control variable is speculative liquidity trading, which should increase trading volume and volatility. Note that by definition this type of trading is not based on information.
We find that information and liquidity orders tend to be concentrated and affect the trading volume and volatility of small firms. The market-wide information is pos-itively related to trading volume and volatility, but more so for large firms than small
3
In a model in which the informed traders possess different information, Foster and Viswanathan (1996) suggest that the initial correlation among the informed traders’ information signals should be considered to explain the volume-volatility relationship using structural models of speculative trading.
4
The relationship between public and private information and volatility is by no means resolved in the literature. French and Roll (1986) conclude that private, rather than public, information is the primary source of volatility. Berry and Howe (1994) find no significant correlation between intraday public in-formation arrival and volatility, while Jones et al. (1994) provide evidence that public inin-formation is the major source of short-term return volatility.
firms, while the speculative orders are associated with trading volume and volatility quite equally for large and small firms. It is also found that the trading volume is not unusually high at the market open despite very large order flows, which we think is caused by the distinct trading mechanism of the Taiwan Stock Exchange (TSE). Unlike the NYSE which has specialists to solicit orders to balance the orders for the opening call, the TSE relies exclusively on computers to match orders mechanically. The TSE’s call mechanism allowing only limit orders makes traders particularly conservative in placing orders at the market open because of the trading recess. Our evidence shows that the small orders are even more conservative than the large orders, since the small orders are likely to possess less information than large orders.
Our methodology considers trading volume and volatility jointly because they both are closely related to information. We classify our sample into four categories: high volume/high volatility (HH), high volume/low volatility (HL), low volume/high volatility (LH), and low volume/low volatility (LL). It is commonly argued (e.g., French and Roll (1986), Jones et al. (1994)) that it takes trading to release private information, while public information can be instantly reflected in the price without much trading.5 Both private and public information induce volatility. Thus, HH is likely to be associated with private information, LH with public information, HL and LL with null information. By examining the effects of the variables on the probabil-ity that the four groups occur, one can infer whether the information content of the aforementioned variables is public or private.
For example, we proxy the information orders with the orders of greater than median size and the liquidity orders with the orders of smaller than median size. We find that for small firms the number of large orders is significantly directly related to the probability of HH and HL occurring. This means that large orders contain either private or little information but not public information. For small firms the small orders are directly related to HH and HL to a lesser degree than the large orders. For large firms the small orders are actually positively related to LL and negatively associated with HH. The finding means that the small orders contain either less information than the large orders or very little information.
We have some evidence that market-wide information is positively related to HH and negatively related to LL for large firms, implying that the bulk of market-wide in-formation affecting trading volume and volatility is private. However, for small firms
5While public information surprise would induce trading, we assume that it occurs randomly in our
the market-wide information is directly associated with LH and negatively associated with HL, meaning that public rather than private market-wide information determines the trading volume and volatility of small firms. In addition, for large firms the rela-tionship between firm-specific information and trading volume and volatility is similar to that of the market-wide information. However, firm-specific information is not a significant determinant of trading volume and volatility for small firms.
Speculative orders (order flows that are orthogonal to the other explanatory vari-ables) are positively related to HH and negatively related to LL, which is consistent with our expectation. Moreover, the less disparate the traders’ opinions are (mani-fested by the larger the order imbalance), the smaller the trading volume and volatility, a situation in accord with the low degree of information asymmetry where little private information is present.
This article is organized as follows. Section 2 reviews the TSE’s trading mecha-nism relevant to our study. Section 3 describes the data, defines variables and outlines our test methodology. Empirical results are presented in section 4. Section 5 discusses the empirical findings. Section 6 offers concluding remarks.
2. TH ETRADINGMEC HANISM ONTHETAIWA NSTOCK
EXCHANG E
The two major U.S. markets, the NYSE and NASDAQ, are an agency/auction market and a dealer market, respectively. NASDAQ is a computerized, quote-driven system whose market regularity depends on the competition among market makers (dealers). NASDAQ matches orders continuously as orders enter into the system. NYSE is an order-driven floor trading system which relies on specialists to act in a passive capacity to stabilize the market when there is a large order imbalance. NYSE adopts a contin-uous auction trading mechanism for trades initiated after the market open each day. However, the open price is determined by specialists through a call market.
In contrast, the TSE is an order-driven system without designated market makers. Stock brokers submit orders to a computer system through the Computer Assisted Trading System (CATS) from their offices. CATS then matches orders without the physical presence of brokers. Like the opening procedure on the NYSE, the opening prices on the TSE are determined through a call market by the exchange. Between 8:30 and 9:00 A.M. orders can be submitted to CATS for the opening call. However, unlike
the NYSE, after the open, orders on the TSE are matched about every 45 seconds throughout the trading day. During a 45-second call period, orders are sequentially accepted by the computer, then matched based on the price-time priority rule through the same call market mechanism as that at the open to maximize the trading volume. After each call the transaction price, transaction volume, the highest bid price and the lowest asking price among unmatched orders are released to the public. At the close the exchange conducts the last call for each stock. Remaining orders after the last call are removed from the book.
The exchange has a daily price limit of 7%. In addition, except for the opening price which is allowed to change up to 7% from the closing price of the previous trad-ing day, all the subsequent trades are subject to an intraday price limit which mandates that the stock price move within two ticks of the price determined in the previous call. Although officially the market orders are not allowed on the TSE, to secure match-ing priority, traders can submit limit orders hittmatch-ing the daily price limits (limit orders with prices beyond the price limits are not accepted by the computer). This type of price-limit orders can achieve the effect of market orders.6 Thus, on the TSE traders can aggressively submit price-limit orders to enhance the probability of trade and yet have some price protection from the intraday price limit. Unless the previous transac-tion price is within two ticks of the daily price limit, price-limit orders would not be matched at the daily price limit. This type of order submitting behavior induced by the TSE’s microstructure, in addition to information, could affect order flows and, in turn, trading volume and volatility.
Currently the total capitalization of the TSE is the fourth largest in Asia. How-ever, the annual turnover rate on the TSE has consistently been the highest in the world, coupled with unusually high volatility. In 1990, the TSE collapsed by posing a drop of more than 40% in value, while in 1996 the TSE’s value rose by 34%, the highest jump among the equity markets in the world. The high turnover and volatility may have something to do with the fact that trading on the TSE is dominated by individual traders. Individual traders, constituting about 90% of the trading volume, are consid-ered to be more speculative and short-term in their investments than institutions. It is commonly believed that the equity market in Taiwan is excessively speculative and is manipulated by individuals with large endowments. As an example, in June 1990 when the monthly turnover rate in Taiwan was 70%, compared to 3.7% in the US, in its
com-6If market orders are allowed, according to Schwartz (1991), they are equivalent to limit orders written
ments on Taiwan’s market, Pacific Rim Stock Markets (Baring Securities, July 1990) stated: “Last quarter we warned investors that Taiwan’s monthly stock market turnover had reached a level that equaled its annual GNP. This represented an excessive level of “froth” that will probably never be seen again by any world stock market...as the stock market capitalization has fallen substantially over the past quarter, monthly turnover still represents an exceptionally frothy 70% of market capitalization. Hence, Taiwan remains a high risk stock market and we continue to recommend caution.”
3. DATA, DEFINITION OF VARIABLE S, AND TESTABL E
HYPOTHESES
3.1
Data
The tick-by-tick transaction and order flow data of the TSE for the six months July through December 1994 are used for our analysis. The data consist of every original order for the buy and sell side as well as every transaction recorded by the computer by time, price and volume. Since the data volume is too large to include all stocks on the TSE, 60 stocks are chosen from about 300 stocks listed on the TSE, consisting of the 30 largest- and 30 smallest-capitalization stocks as of the beginning of our sample period. Stocks that had IPOs six months prior to our sample period, that distributed cash or stock dividends, and that had seasoned offerings during our sample period are excluded.7 All of the stocks in our sample are actively listed throughout the sample period. Preferred stocks and financially distressed stocks are not included.
The TSE trades from 9:00 A.M. till 12:00 P.M. on weekdays, and from 9:00 A.M. to 11:00 A.M. on Saturday. To avoid the potential effect of the length of trading time on the return-generating process, Saturday observations are not included in our sample. In total there are 104 trading days. To give readers some idea of the data employed in our analysis, we report in Table 1 summary statistics of the high- and low-capitalization stocks.
7
We have to exclude stocks that paid cash or stock dividends because of a peculiar phenomenon in Taiwan. Before a stock goes ex-dividend, there is usually expectation on the part of the investors, founded or unfounded, that the stock price would soon rise back to its pre-dividend level. Investors’ trading behavior is affected by the expectation. In order to avoid the effect of this kind of phenomenon on stock price, we exclude stocks that pay cash or stock dividend from our sample. It is possible that this kind of sample selection criterion could introduce bias in our result. But we do not feel that there is reason to believe that the bias would outweigh the disadvantage from the ex-dividend phenomenon.
Table 1 Summary Statistics of Large and Small Stocks
The variables are denominated in New Taiwanese Dollars (NT). Capitalization is in thousands of NT. Daily number of shares (orders) submitted is scaled by the total number of shares outstanding. The unscaled numbers are in parentheses, for which the daily number of shares outstanding is in thousands of shares.T -tests are conducted for the test of equal means between large and small stocks. Significance levels of 10%, 5%, and 1% are indicated by *, **, and ***.
Whole
sample Large Small T -test
No. of stocks 60 30 30
No. of trading days 104 104 104
Avg. price 36.24 42.56 29.91 *
Avg. daily return 0.0025 0.0030 0.0020
Std.-dev. of daily 0.0232 0.0222 0.0241
return
Daily trading volume 3,636.1 5,286.1 1,986.2 **
in 1,000 shares.
Daily trading volume 603.4 824.7 382.0 **
in no. of orders
Daily turnover 0.0147 0.0085 0.0209 ***
Capitalization 14,156,532 25,513,596 2,799,478 ***
Daily no. of shares 0.0660 0.0369 0. 0952 ***
submitted (13,827) (18,861) (8,743) (***)
Daily no. of orders 0.0071 0.0037 0.0106 ***
submitted (1,237) (1,560) (915) (***)
The large stocks’ average capitalization is NT25.5 billion, vs. the NT2.8 bil-lion of the small stocks. The large stocks’ share price and daily trading volume in terms of shares traded and number of orders traded are greater than those of the small stocks. However, the daily turnover of the large stocks (0.0085) is lower than that of the small stocks (0.0209), and the volatility is somewhat lower as well. The overall daily turnover is 0.0147, amounting to about 400% per year, i.e., on average a share changes hands four times a year. The numbers of shares and orders submitted for trad-ing are much larger than the tradtrad-ing volume. From Table 1 one can calculate that about 28% of orders submitted for trading for the large stocks result in trades, compared to only 23% for the small stocks.
3.2
Definition of variables
3.2.1 Information vs. Liquidity orders
Following what is generally presumed in the literature, we assume that large-size or-ders contain more information and are more likely to reflect traor-ders’ private informa-tion than small-size ones. We classify orders into the two categories based on whether or not an order size (in number of shares) is greater than the intraday median size of the orders.8 Throughout this paper, unless otherwise noted, our variables are calculated for each 6-minute interval. This is because there are about eight 45-second call-back periods in the interval, which in most cases are sufficient to accommodate large intra-day price moves so that the intraintra-day price limit is not binding. Including the opening call as a separate interval, there are in total 31 intraday intervals. Note that only new order flows that occur in a particular 6-minute interval are used to calculate the values of different variables. Orders whose sizes are greater (smaller) than the median size are called information (liquidity) orders, denoted as MOIijt(MOLijt) for stocki, in-traday intervalj, and day t. MOI and MOL are then divided by their intraday 6-minute averages to obtain OIijtand OLijt, which are the basis of our analysis of the intraday pattern of information and liquidity orders.
If information and liquidity traders concentrate their trading at the close and the open, we would expect the ratio of information and liquidity orders to be greater at the two points of time. Trading volume and volatility will be positively correlated with OI and OL if order concentration induces trading and volatility. Of course, the set of information orders according to our classification might include some liquidity traders, and likewise the set of liquidity orders might include some information orders. But as long as the crossovers occur randomly, our measures should be quite indicative of what we intend to capture.
8
Whether or not large orders contain more information than other orders is not yet conclusive in the literature. For example, Barclay et al. (1993) suggest that medium-sized rather than large orders contain more information. However, many studies find that large orders contain more information than small orders. Lee et al. (1999) find that on the Taiwan Stock Exchange large orders have much more information contents than small orders. Easley and O’Hara (1987) find that informed traders prefer to trade larger amounts at any given price. Hasbrouck (1988) provides strong evidence that large trades convey more information than samll trades. Since our paper does not aim to analyze the information content of different order sizes, we somewhat arbitrarily adopt the view that large orders have more information content. In light of our finding, to be shown later in our paper, that large orders have more impacts on the high volume/high volatility group, it seems that our classification achieves what it is meant to.
3.2.2 The degree of divergence in opinions
The degree of divergence in opinions measures the diverse views on the direction of market movement. Our proxy is order imbalance in every six-minute interval, which is calculated as the absolute value of the difference between the total shares of buy orders and those of sell orders, scaled by the total number of shares outstanding. The variable is denoted as OBijt, meaning the order imbalance of stocki in interval j on day t. The greater the order imbalance, the smaller the opinion difference. If trading is caused by diverse interpretation of market conditions, we expect trading volume to be high when OB is low. And if the diverse interpretation is induced by new information, then the volatility should be high as well.
3.2.3 The scope of information
The scope of information refers to the classification of information as market-wide or firm-specific. Orders driven by market-wide information should be quite evenly dis-tributed across different stocks, while those caused by firm-specific information should be concentrated in specific stocks. It is natural to assume that market-wide informa-tion would generate orders in proporinforma-tion to a firm’s market capitalizainforma-tion. Hasbrouck (1996) uses the orders from program trading to proxy for the effect of public informa-tion on order flow, because program orders cover multiple stocks and therefore should reflect market-wide information. In the case of index arbitrage, the broad-market index is usually weighted by market capitalization.
We do not have measures similar to that used by Hasbrouck (1996) because there is neither program trading on the TSE nor trading based on equity indexes such as index futures. But there is no obvious reason to believe that the market-wide infor-mation would have consistent asymmetric impacts on the order flow of stocks, dispro-portionate to the stocks’ capitalizations. Note that all stocks are traded actively on the TSE. In our sample the average turnover rate per six-minute interval is about 0.07% of outstanding shares. Table 1 shows that the daily turnover of small stocks actually is significantly higher than that of large stocks. There is thus very little non-synchronous trading problem or cross-autocorrelation between stock returns. We do not have the means to ascertain how market-wide information would generate orders differently in large and small stocks. But given that small stocks are also very active in terms of their trading volume, it seems that our assumption that market-wide information would generate orders in proportion to a firm’s market capitalization is innocuous. We define MNijtas the capitalization-weighted number of shares in intervalj for stock i on day t, calculated by multiplying the total amount of orders (in shares) of all stocks in the
sample in the intervalj on day t by stock i’s proportion of the total capitalization of the stocks in the sample.
Letting stocki’s orders in interval j on day t be Nijt, the firm-specific information variable (ACNijt) is proxied by the absolute value of CNijt = (Nijt− MNijt)/MNijt, which is the absolute deviation of stock i’s orders in interval j on day t from the weighted number of shares. Both MNijt and ACNijt can be public and private.9 Whether or not MNijt and ACNijt are larger at the open and the close is not clear a priori. But if trading volume and volatility are caused by market-wide and firm spe-cific information, we would expect the two variables to be positively correlated with trading volume and volatility.
3.2.4 Speculative orders
Orders that are not related to the above variables are called speculative orders, which are proxied by the residuals of the regression of the total order flows, TODijt, on the variables defined above. The regression residuals are denoted as RTODijt. In general, orders are generated out of information, liquidity, speculation and market making. There are no designated market makers on the TSE and to the best of our knowledge there are no dealers who act like liquidity providers. Hence the portion of order flows arising from inventory management should be quite insignificant. Since we have defined variables that are proxies for orders generated out of information and liquidity, we define the residual order flow that is not related to information and liquidity as speculative orders. This, of course, is not the perfect proxy for speculative orders. Basically, it is just something that is left in the order flow that is not captured by liquidity and information variables. Since speculation tends to induce trading volume and volatility, if RTOD captures the effect of speculative trading well, RTOD ought to be positively related to volume and volatility.
3.3
Testing methodology
We now suggest a classification scheme to test the relationship between information, trading volume, and return volatility. Our design will consider volume and volatility
9
It is less likely for market-wide information to be private than for firm-specific information. But one cannot completely rule out the possibility that some market-wide information might be private. If we define private information strictly as insider information, then it is indeed very unlikely that market-wide information is private. However, if we allow private information to be acquired through superb research that could predict policy moves or market trends, then the domain of private market-wide information is expanded to include such information.
together rather than separately. Trading volume for an interval is taken as the number of shares traded over the interval.10 Volatility is measured as the absolute price change from the end of the previous interval to the end of the current interval divided by the average of the two prices. Each intervali on a particular trading day t of stock j is clas-sified as one of the four volume/volatility groups: high volume or high volatility (HH), high volume/low volatility (HL), low volume/high volatility (LH), or low volume/low volatility (LL). An interval is in the HH group if its trading volume is higher than the intraday median volume and its volatility is greater than 80% of intraday volatilities, in the HL group if its trading volume is higher than the intraday median volume and its volatility is smaller than (including) 60% of intraday volatilities, and so forth for the LH and LL groups. For any 6-minute intervalj of stock i the percentage (denoted as Pijk) of the number of trading days belonging to one of the four groups,k (k = HH, HL, LH, LL), is calculated. For example, suppose that there are 52 days of stocki for which the last six-minute interval is in the HH group, then the calculated percentage, Pi31HH is 50% (52/104) for the stock.
We are not able to screen the volatility based on the median because it is likely that the median takes the value of zero. Table 2 shows, for the high-capitalization and low-capitalization groups, respectively, the percentage of the 30 stocks’ trading days that the volatility value is zero at a certain percentile of intraday volatility observations ranked in ascending order. One can see that at the 60th percentile, on average 25% of trading days have volatility values of zero. But the percentage drops to about 10% at the 70th percentile, and 3% at the 80th percentile. We therefore choose to classify volatility according to the rule explained above.11
The summary statistics of our classification are shown in Table 3. Except for share price, the values of other variables such as volatility and trading volume are statistically different across groups and the order of magnitude is as expected for each group. Except for volatility and turnover, the values of the variables of the large stocks are in general greater than those of the small stocks. Note that in Table 1 we already observed that the small stocks’ turnover is higher, which is accompanied by higher volatility. TheP ’s in Table 3 show that ninety percent of the intraday intervals are classified into one of the four groups. The total percentage in groups LL and HH (49%) are greater than that in groups HL and LH (41%), meaning that more often
10
Our results are invariant to different definitions of trading volume: number of shares traded or turnover.
Table 2 Percentage of Trading Days of the Large and Small Stocks Whose 6-Minute Volatility Is Zero for a Percentile of Intraday Volatilities Ranked by Their Absolute Values
The intraday 6-minute interval volatilities are ranked for each trading day in ascending order. The volatility is measured as the absolute value of the 6-minute return. There are in total 3,120
(104×30) trading days for the 30 stocks in each capitalization group. The percentage reported
in the table is that of the 3,120 observations that are zero for a particular percentile of ranked intraday volatilities.
Percentage of trading days Percentile of
zero returns Large stocks Small stocks
10 99.5 99.4 20 96.2 95.2 30 88.6 83.0 40 76.3 62.1 50 56.0 37.0 60 32.8 17.8 70 14.1 6.4 80 3.9 1.8 90 0.5 0.5
than not, trading volume is positively related to volatility. The LH group occurs least frequently, indicating that low volume does not often happen with high volatility. But since the percentage for the HL group is quite high (32%), it is quite often that high volume is accompanied by low volatility.
We are now ready to state the various hypotheses regarding the relationship be-tween information, trading volume, and volatility in terms of the regression equation below. For each volume/volatility group, Pijk is regressed on OB, OI or OL, MN, ACN, and RTOD.
Pijk= b0ik+ b1ik(OIij or OLij) + b2ikOBij+ b3ikMNij
+b4ikACNij+ b5ikRTODij+ eij, (1) wherebik’s are the estimated slopes for the groupk of stock i.
Table 3 Summary Statistics of Volume/Volatility Groups
High volume/high volatility, high volume/low volatility, low volume/high volatility, and low volume/low volatility groups are denoted as HH, HL, LH, and LL.T -tests are for the tests of equal means between large and small stocks. F -tests are for the tests of equal means across the four groups. Significance levels of 10%, 5%, and 1% are indicated by *, **, and ***.
Group Large stocks Small stocks T -test
Average per HH 42.45 29.83 2.47** share price HL 42.40 29.99 2.42** LH 42.64 29.87 2.44** LL 42.60 30.00 2.45** F -value 0.00059 0.00163 Average HH 0.0480 0.0469 0.09 six-minute HL 0.0031 −0.002 2.38*** return LH 0.0322 0.0591 −1.76* LL −0.0058 −0.0089 1.38 F -value 11.80*** 24.95*** Average HH 0.7990 0.8787 −2.10* six-minute HL 0.1015 0.1237 −3.30*** volatility (%) LH 0.6996 0.7695 −2.38** LL 0.0846 0.1056 −3.22*** F -value 516.74*** 535.37*** Average trading HH 280.85 129.39 2.27** volume in 1,000 HL 195.93 93.83 2.44** shares LH 55.19 25.02 2.57** LL 64.31 25.18 2.34** F -value 7.46*** 42.83*** Average trading HH 45.04 26.83 2.44** volume in HL 31.55 19.44 2.54** number of LH 13.15 8.24 2.77** orders LL 13.75 7.69 2.54** F -value 11.81*** 47.40*** Pij(%) HH 15.57 15.3 0.68 HL 32.76 31.49 3.64** LH 9.69 9.29 1.61 LL 33.56 33.38 0.36 F -value 1,951.58*** 1,875.57*** Average HH 0.00053 0.00137 −4.19*** turnover HL 0.00038 0.00101 −4.08*** LH 0.00010 0.00027 −5.29*** LL 0.00010 0.00026 −4.79*** F -value 5.82** 36.14***
Note that one regression is run by one of the four classifications ofPijkfor each of the 60 stocks in our sample.12 The explanatory variables are the six-minute inter-val inter-values averaged across trading days. Each regression has 31 intraday observations including the open and its subsequent 30 six-minute intervals. Since the correlation coefficient of OI or OL is usually above 0.9, either OI or OL is included in the re-gressions.13 The estimation results for the other explanatory variables are very similar whether OI or OL is included. In the following we report the results based on the regressions inclusive of OI. For the regressions with OL we only report the estima-tion results for the variable OL. The explanatory variables do not change with the volume/volatility category.
Since our variables are measured over a six-minute interval, if the error terms in the regression are autocorrelated, there could be bias in the estimated variance-covariance matrix of the error terms. We thus employ the method of White (1980) and Hansen (1982) to adjust the variance-covariance matrix for general conditional heteroskedasticity and autocorrelations in the error terms. Further to ensure that the Hansen-White procedures yield a positive semi-definite variance-covariance matrix in finite samples, the Newey and West (1987) weight is applied wherever necessary. We ran seperate regressions under the assumptions that the error terms are not autocorre-lated and that there is first-order autocorrelation. Both yield results similar to those of OLS. We thus only report the results assuming that the autocorrelation is of the first-order.
By examining the slope coefficients of each group, we can determine the rela-tionship between the independent variables in (1) and different combinations of vol-ume and volatility. Moreover, the regressions allow us to discern whether or not the relationship is related to public or private information. Since it is generally believed that private information affects volatility through trading, while public information is reflected in the price instantly, without much trading, the HH group is likely to be caused by private information, the LH group by public information, and the HL and LL groups by little information. Thus, one can induce whether the effects of the ex-planatory variables in (1) on the trading volume and volatility are caused by public or private information.
12
Since our paper is primarily interested in jointly explaining the volume-volatility relationship, we do not regress volatility and volume separately on explanatory variables. It is possible that the four regressions are correlated in some way. But for any particular stock during any six-minute segment, the volume/volatility pair could fall in any one of the four groups, determined by the order flow of the six-minute period. Thus, it seems that even if there is some correlation between groups, we do not have reason to believe that our methodology results in systematic bias in our findings. We thank the referee for pointing out this potential problem to us.
4. EM PIRICA L RE SULTS
Figures 1 and 2 show, for the large- and small-capitalization stock groups separately, the intraday six-minute average trading volume, average number of shares submitted, and average return volatility. For the two figures and other figures that follow we scale the variables to fit into the figures. Thus, the figures are only meaningful to the extent of their intraday pattern, not their magnitudes. The variables are averaged across trading days and stocks. The first observations in the figures are those for the opening call, and the rest are for the remaining six-minute intervals. The volatility for the opening call is measured from the close of the previous day to the open of the next day. Notwithstanding the volatility at the open, similar to the findings in the existing literature, volatility exhibits a clear U-shape. Although the trading volume clearly increases toward the end of the trading day, it is not distinctly high at the beginning of the trading day. However, the total order flow in terms of number of shares submitted has a clear U-shape as well. Thus, we have observed that large order flows do not necessarily result in large trading volume.
Table 4 depicts the descriptive statistics of the variables employed in this paper for four time sections of the trading day: open, 9:00 to 9:06, 9:06 to 11:54, and 11:54 to 12:00. Like in Figures 1 and 2, these variables are measured over six-minute intervals. Large and small stocks have very similar patterns. Except for the trading volume (VOL), ACN, and RTOD, all the other variables have values higher near the open and the close than during the middle of the trading day. As we have seen from Figures 1 and 2, the trading volume is only distinctly higher near the close, but not the open. ACN is higher during the middle of the day than near the open and close. RTOD is low near the open but very high at the close. Figures 3 and 4 show for the large and small stocks the intraday pattern of OB, OI, OL, MN, ACN, and RTOD. The pattern is consistent with the observations made from Table 4.
Figures 5 and 6 present for the large and small stocks the intraday pattern of the four volume/volatility groups in terms of the average percentage (Pij) across stocks of the number of six-minute intervals falling into one of the four groups. The patterns for the large stocks and the small stocks are very much alike. The HH group has a U-shaped pattern, while the pattern of the LL group is almost the mirror image of that of the HH group. The pattern of HL is monotonically increasing, but that of LH is just
9 8 7 6 5 4 3 2 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 TOD VOL ARET
Figure 1 Intraday Patterns of Total Order Flow (TOD), Volume (VOL), and Volatility (ARET) for Large Firms
25 20 15 10 5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 TOD VOL ARET
Figure 2 Intraday Patterns of Total Order Flow (TOD), Volume (VOL), and Volatility (ARET) for Small Firms
Ta b le 4 Summary Statistics fo r D iffe re nt Intr aday T ime Sec tio ns All the v ariables except for MN, A CN and v olatili ty are scaled b y the number o f shares outstand ing. The number “1,000” indicates that the v ariab les are m ultipl ied by 1,000. P anel A . L ar ge St ocks Open 9: 00–9:06 9:0 6–11:54 Mean Median St d. de v. Mean Medi an Std. de v. Mean Median Std. de v. Mean No. o f shares traded (× 1, 000 ) 0.20607 0.04679 0.6888 0.22714 0.04147 0.7503 0.26 61 0.06242 0.9025 0.76851 No. o f o rders sub-mitt ed (× 1, 000 ) 0.50506 0.29538 0.69728 0.10207 0.03955 0.23412 0.10273 0.03832 0.20497 0.21546 No. o f shares sub-mitt ed (× 1, 000 ) 4.6994 2.444 7.3873 1.0905 0.3462 2.5266 1.03 76 0.2958 2.9183 2.0787 V o la ti lity (%) 0 .77161 0.4662 1.2125 0.42454 0.32733 0.5257 0.26795 0 0 .3638 0.49545 OB (× 1, 00 0) 1.5847 0.5619 3.725 0.3786 0.10592 1.083 0.32 81 0.08221 1.2217 0.4729 OI 4.483 3.769 3.179 − 0.046 − 0.269 0.958 − 0.187 − 0.368 0 .728 OL 4.225 4.014 2.19 − 0.191 − 0.254 0.543 − 0.169 − 0.222 0 .463 0.698 MN 3434.78 2026.19 4472.29 839.85 468.84 1225.75 688.62 351.33 10 97.81 1481.53 A C N 0 .71804 0.62795 0.68745 0.98108 0.77153 1. 42121 0.98817 0.76555 1.52452 0.9089 RT O D (× 1, 000 ) − 0.0424 − 0.0125 0.1142 − 0.0529 − 0.044 9 0 .0854 0.00 25 − 0.0005 0.1209 0.0654
Ta b le 4 Summary Statistics fo r D iffer ent Intraday T ime Sections (continued) All the v ariables except for MN, A CN and v olatili ty are scaled b y the number o f shares outstand ing. The number indicates that the v ariab les are m ultipl ied by 1,000. P anel B . S mal l Stock s O p en 9:00–9:0 6 9:06–11:54 Mean Median Std. de v. Mean Median St d. de v. Mean Median Std. de v. No. o f shares traded (× 1, 000 ) 0.7184 0.2 2 .4421 0.7139 0.182 1 1 .8602 0 .6672 0.2816 1.3104 No. o f o rders sub-mitt ed (× 1, 000 ) 1.3456 0.9596 1.4503 0.3081 0.155 8 0 .5475 0 .2942 0.1817 0.3813 No. o f shares sub-mitt ed (× 1, 000 ) 13.596 7.8653 28.152 3.271 1.236 9 6 .871 2.579 1.2521 4.428 OB (× 1, 00 0) 5.709 2.1342 2 1 .51 1 .2642 0.422 7 3 .863 0 .8481 0.3451 2.026 V o la ti lity (%) 0 .9256 0.5420 1.3285 0.5001 0.396 8 0 .6203 0 .2975 0.2265 0.3874 OI 3.694 3.237 2 .385 − 0.059 − 0.296 0.913 − 0.17 − 0.35 0.723 OL 3.289 3.101 1 .567 − 0.218 − 0.314 0.555 − 0.145 − 0.223 0.512 MN 376.958 309.039 23 4.407 92.305 69.37 0 72. 572 7 5 .275 51.818 67.294 A C N 2 .51213 1.53104 4.13957 3.10185 0.92774 6. 31277 3.08045 1.14053 5.66102 RT O D (× 1, 000 ) − 0.069 − 0.054 0 .084 0.024 − 0.042 0.276 − 0.0008 − 0.0026 0.179
4 3.5 3 2.5 2 1.5 1 0.5 0 -0.5 -1 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 OB OI OL MN ACN RTOD
Intraday Six-Minute Intervals
Figure 3 Intraday Patterns of the Information, Liquidity, and Speculation Variables for Large Stocks
7 6 5 4 3 2 1 0 -1 -2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 OB OI/2 OL/2 MN ACN RTOD
Intraday Six-Minute Intervals
Figure 4 Intraday Patterns of the Information, Liquidity, and Speculation Variables for Small Stocks
0.6 0.5 0.4 0.3 0.2 0.1 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Pijk HH HL LH LL
Intraday Six-Minute Intervals
Figure 5 Intraday Volume/Volatility Classification Patterns for Large Stocks
0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Pijk HH LL LH HL
Intraday Six-Minute Intervals
the opposite. The patterns of the four groups start off from the open of the day quite near to each other, then diverge to different directions throughout the trading day. That is, near the open there is a similar chance for a six-minute interval to fall in one of the four groups, but near the close the HH and HL cases have a much higher probability of occurring than the LH and LL cases. Thus, volume is more likely than volatility to be extremely high at the close, while the chance of the four groups occurring at the open is even.
Table 5 reports the regression results of equation (1). The summary statistics of the regression coefficients and theirt-statistics (in parentheses) are included. The numbers of significant positive and negative coefficients are counted for significance levels of 10% and 5% for two-sided tests. Since the regression is run for each stock and each volume/volatility group, there are altogether 240 regressions. To conserve space, we report the summary statistics of the estimated coefficients only. Table 5 shows the number of significant coefficients for the two size-groups and the four vol-ume/volatility groups. From the number of significant coefficients and their signs we are able to judge if our testable hypotheses are supported by the regression results.
The results for OI indicate that the information orders are positively related to the trading volume and volatility, although OI also seems to be related to high trading volume and low volatility. For the small stocks, according to the significance level of 10% for two sided tests, there are 15(3) and 21(1) significantly positive (negative) coefficients for HH and HL groups, while 0(16) and 2(15) for the LH and LL groups. In addition, 25 and 26 out of 30 stocks are positive for the HH and HL groups and the same number of stocks are negative for LH and LL groups. The means and medians of t-statistics of the four groups are also significant. This means that some of the large trading volume happens when large information orders are present. But sometimes OI carries information (HH is positively associated with OI), while other times little information (HL is directly related to OI as well). In addition, the relationship between OI and the trading volume and volatility is not clear for the large stocks. For small stocks since there is positive association between OI, OL and the HL group, some of the trading volume is a result of liquidity trading,
Regarding the small stocks, OL’s relationship to the trading volume and volatility is very similar to but somewhat less significant than that of OI. But for the large stocks it is more likely that OL is negatively related to trading volume and volatility, because there are more stocks (25) that are negatively related to the probability of HH group than are positively related (5). Fifteen (5) out of 30 stocks are negatively (positively)
Table 5 Summary of the Relationship between Information, Volatility and Trad-ing Volume Based on the FollowTrad-ing Regression
Pijk= b0ik+ b1ik(OIij or OLij) + b2ikOBij+ b3ikMNij
+b4ikACNij+ b5ikRTODij+ εijk.
In the above regression Pijkis stocki’s percentage of the jthintraday 6-minute volume and volatility falling in one of the four volume/volatility groups (k = HH, HL, LH, LL for the high volume/high volatility, high volume/low volatility, low volume/high volatility and low volume/low volatility group). The subscriptj, j = 0, 1, 2, ..., 30, denotes the 31 intervals from open to close. Variables OIjk, OLjk, OBjk, MNjk, ACNjk, RTODjk are as defined in the paper and do not change withk. We include either OI or OL in the regression because the correlation coefficient of OI and OL is usually over 0.9. The regression is run for each stock in our sample. In the following we summarize the estimated coefficients of the 120 regressions for the large and small stocks respectively for the case of including OI only. Since, the results for the case of including OL are similar, for this case we only report the estimation results for OL. Numbers in parentheses are the summary statistics of t-statistics, and NP and NN indicate the number of significant positive and negative coefficients for the indicated significance level of two-sided tests. Numbers in parentheses below the numbers of significant coefficients are the numbers of positive and negative coefficents.
Variable Size Group Estimated coefficients NP NN
Mean Median σ Max. Min. 10% 5% 10% 5%
OI Large HH (0.14)0.048 (−0.047 0.346 0.783 −0.625−0.48) (2.49) (6.95) (−3.96) 7 (14)6 6 (16)5 HL (0.41)0.058 (0.29)0.037 (2.53) (7.96) (0.288 0.782 −0.537−2.97) 6 (16)6 7 (14)5 LH (−0.033 −0.001 0.156 0.272 −0.363−0.15) (−0.02) (2.04) (4.29) (−5.38) 4 (15)3 6 (15)6 LL (−0.025 0.051 0.398 0.886 −0.803−0.09) (0.38) (2.41) (3.41) (−7.89) 6 (17)6 7 (13)6 Small HH (2.85)0.252 (1.78)0.268 (3.66) (10.87) (0.284 0.810 −0.258 15−4.03) (25)15 3 (5)1 HL (2.64)0.263 (2.66)0.241 (2.06) (7.95) (0.203 0.731 −0.093 21−0.99) (26)19 0 (4)0 LH (−0.150 −0.130 0.131 0.057 −0.478−1.93) (−1.92) (1.72) (1.03) (−5.62) 0 (5)0 16 (25)15 LL (−0.356 −0.397 0.336 0.352 −0.932−1.94) (−1.67) (2.52) (4.86) (−8.40) 2 (4)2 15 (26)13
Table 5 Summary of the Relationship between Information, Volatility and Trad-ing Volume Based on the FollowTrad-ing Regression (continued)
Variable Size Group Estimated coefficients NP NN
Mean Median σ Max. Min. 10% 5% 10% 5%
OL Large HH (−0.128 −0.115 0.274 0.556 −0.879−1.32) (−1.64) (2.92) (9.44) (−4.86) 5 (5)4 15 (25)13 HL (0.29)0.053 (−0.001 0.222 0.673 −0.431−0.03) (2.29) (6.25) (−3.76) 9 (15)7 6 (15)5 LH (−0.046 −0.023 0.124 0.205 −0.393−0.52) (−0.59) (1.94) (4.27) (−4.45) 4 (12)2 8 (18)8 LL (1.06)0.118 (1.19)0.109 (2.50)0.291 (4.84)0.798 (−0.632−6.91) 13 (21)12 3 (9)2 Small HH (1.03)0.136 (0.96)0.130 (2.33)0.220 (5.48)0.634 (−0.315−6.27) 11 (20)10 3 (10)2 HL (2.85)0.237 (2.54)0.231 (1.88)0.163 (6.44)0.682 (−0.072−0.54) 23 (28)21 0 (2)0 LH (−0.146 −0.118 0.099 0.026 −0.367−2.23) (−2.22) (1.42) (0.42) (−5.02) 0 (2)0 21 (28)17 LL (−0.209 −0.255 0.260 0.409 −0.731−1.35) (−1.43) (1.56) (2.92) (−3.58) 2 (5)2 13 (25)13 OB Large HH (−381.3 −423.2 1025.3 4138.1 −1500.4−1.48) (−1.53) (1.84) (2.89) (−6.99) 1 (5)1 13 (25)10 HL (−243.3 −178.2 651.5 1323.3 −1173.5−1.07) (−1.07) (1.63) (2.20) (−4.96) 2 (7)1 11 (23)10 LH (0.65)112.4 (0.80)85.5 (1.47)355.4 (4.79)885.1 (−866.3−3.16) 5 (21)5 2 (9)1 LL (1.03)403.5 (1.22)514.2 1087.1 3233.4(1.53) (3.69) −2577.3 10(−1.96) (22)8 2 (8)1 Small HH (−221.7 −162.3 256.5 223.2 −979.3−3.26) (−2.49) (4.61) (7.65) (−18.40) 2 (3)2 17 (27)17 HL (−116.0 −76.0 200.5 245.9 −602.7−1.51) (−1.75) (2.31) (2.77) (−7.58) 1 (9)1 16 (21)12 LH (0.80)42.3 (0.77)34.4 (1.21)71.6 (3.98)193.0 (−116.3−1.42) 8 (21)4 0 (9)0 LL (2.03)202.5 (1.90)173.4 (3.51) (13.32) (331.0 1049.1 −474.1−5.69) 18 (22)13 4 (8)3
Table 5 Summary of the Relationship between Information, Volatility and Trad-ing Volume Based on the FollowTrad-ing Regression (continued)
Variable Size Group Estimated coefficients NP NN
Mean Median σ Max. Min. 10% 5% 10% 5%
MN Large HH 0.00010(1.24) 0.00026 0.00079 0.00131(1.55) (2.76) (7.73) −0.00209 14 13 4(−5.97) (21) (9)4 HL −0.00012 0.00008 0.00065 0.00100 −0.00231 9(0.25) (0.40) (2.75) (7.91) (−6.97) (17)7 5 (13)4 LH 0.00013(0.21) 0.00003 0.00033 0.00125(0.20) (2.13) (5.58) −0.00032 5(−5.68) (17)5 3 (13)3 LL −0.00009 −0.00033 0.00082 0.00193 −0.00130 4(−1.12) (−1.20) (2.50) (4.05) (−7.66) (9)2 14 (21)10 Small HH 0.00024(−0.01) 0.00089 0.00392 0.00818(0.50) (2.83) (7.37) −0.00756 8(−6.66) (17)6 9 (13)9 HL −0.00272 −0.00257 0.00308 0.00461 −0.01090 2(−1.73) (−1.89) (1.93) (2.55) (−5.58) (6)2 16 (24)15 LH 0.00230(2.09) 0.00202 0.00218 0.00759(2.44) (1.73) (5.92) −0.00089 19 16 0(−1.37) (25) (5)0 LL 0.00104(0.46) −0.00003 0.00516 0.01070 −0.01213 10 10 3(−0.01) (3.11) (7.92) (−8.08) (15) (15)3 ACM Large HH (−0.420 −0.128 0.928 0.610 −3.579−0.93) (−1.24) (1.58) (1.66) (−5.49) 1 (10)0 10 (20)7 HL (−0.484 −0.240 0.753 0.874 −2.500−1.17) (−1.35) (1.98) (3.58) (−6.52) 2 (8)2 11 (22)8 LH (0.55)0.153 (0.48)0.058 (1.67)0.324 (4.06)1.125 (−0.265−2.94) 7 (20)5 3 (10)2 LL (1.35)0.751 (1.29)0.391 (1.81)1.191 (4.10)4.342 (−0.589 14 12 1−3.65) (23) (7)1 Small HH (−0.007−0.12) (0.23)0.011 (2.09)0.105 (3.88)0.226 (−0.438−7.01) 4 (17)3 5 (13)4 HL (−0.025 −0.013 0.070 0.109 −0.188−0.73) (−0.60) (1.39) (1.35) (−3.81) 0 (11)0 8 (19)8 LH (0.77)0.012 (0.61)0.015 (2.20)0.043 (7.89)0.133 (−0.078−2.92) 9 (19)8 2 (11)1 LL (0.22)0.030 (0.53)0.018 (2.53)0.127 (5.67)0.519 (−0.137−7.81) 6 (16)4 4 (14)3
Table 5 Summary of the Relationship between Information, Volatility and Trad-ing Volume Based on the FollowTrad-ing Regression (continued)
Variable Size Group Estimated coefficients NP NN
Mean Median σ Max. Min. 10% 5% 10% 5%
RTOD Large HH 468.503(1.66) 414.988 674.439 3044.440(1.71) (1.57) (6.36) −627.150 16 12 0(−1.02) (26) (4)0 HL 136.746(0.69) 131.719 506.846 1217.210(0.54) (1.84) (4.93) −840.160 9(−2.64) (18)8 2 (12)2 LH −133.537 −108.193 242.132 372.920 −727.740 0(−0.98) (−1.02) (1.26) (1.34) (−4.04) (7)0 11 (23)10 LL −357.910 −377.366 537.060 695.420 −1384.600 2(−1.22) (−1.07) (2.08) (4.49) (−5.48) (8)1 13 (22)11 Small HH 136.288(1.92) 109.606 211.100 720.260(1.97) (2.30) (6.42) −369.320 16 15 3(−2.53) (25) (5)3 HL 119.598(1.31) 92.433 170.023 509.280(1.40) (1.84) (5.06) −145.590 14 11 2(−2.66) (23) (7)2 LH −69.300 −63.291 85.417 60.180 −283.470 2(−1.07) (−1.01) (1.92) (2.76) (−6.08) (7)1 10 (23)8 LL −173.765 −144.034 271.594 456.400 −859.340 4(−1.86) (−1.77) (2.56) (3.12) (−7.66) (6)4 16 (24)11
significant at the 10% significance level for the two-sided tests for the HH group, while 13 (3) are positively (negatively) significant for the LL group. The means of the t-statistics for the large stocks for the HH and LL groups are −1.32 and 1.06. Thus, trading volume and volatility tend to be lower when there are more small orders for the large stocks. Taking the findings for OI and OL together, it seems that the large information traders and small liquidity traders drive trading volume and volatility for the small stocks but not for the large stocks. In addition, the small orders contain less private information than the large orders because OL is less directly related to the trading volume and volatility than OI.
In terms of the impact of market-wide relative to firm-specific information, our evidence indicates that the former has greater impacts on trading volume and volatility than the latter. Contrary to the findings for OI and OL, the market-wide information proxied by MN is positively related to trading volume and volatility more so for the large stocks than the small stocks. The HH and LL groups have 14(14) significantly positive (negative) coefficients for the large stocks at the 10% significance level of two-sided tests. The number of positive (negative) coefficients for the HH group is 21(21)
for the large stocks. The means (medians) of the t-statistics for the HH and LL groups are 1.24(1.55) and−1.12(−1.2). We interpret this as weak evidence of the presence of private market-wide information. There could be many reasons for this result. We offer one plausible explanation as follows. When some traders acquire private market-wide information, they tend to trade large stocks on the information. This may be because they are institutional traders who are required to trade large stocks more than small stocks because of their institutions’ asset allocation guidelines. Take foreign institutional investors as an example. Their global asset allocation strategy dictates that they only trade large stocks for liquidity reason and benchmarking. As a result, when they have private market-wide information, they trade on large stocks more than small stocks.
In the case of small stocks, MN is negatively related to HL and positively related to LH. There are 16 and 19 significant coefficients for the two groups respectively. Since high volatility occurs when the trading volume is low, the public information affects the trading volume and volatility of small stocks.
The relationship between ACN and the trading volume and the volatility is not clear for the small stocks. Less than one third of the coefficients of the four groups are significant. For the large firms, there is some indication that ACN is negatively related to the trading volume and volatility. And the information is more likely to be public than private because ACN is negatively associated with HL and positively associated with LH.
The degree of consensus proxied by the magnitude of order imbalance, OB, tends to be negatively related to trading volume and volatility. There are 13(17) signifi-cantly negative and 10(18) signifisignifi-cantly positive coefficients for the large (small) stocks for the HH and LL groups respectively at the 10% significance level of two-sided tests. The means of the t-statistics for the HH and LL groups are−1.48(−3.26) and 1.03(2.03) for the large stocks (small stocks). There are 25(27) negative coefficients for the HH group of the large (small) stocks, 22(22) positive coefficients for the LL group. The finding suggests that a higher degree of consensus in the traders’ interpre-tation of information is associated with low trading volume and volatility. It is also consistent with the situation in which information asymmetry is mild due to lack of private information. Otherwise, the volatility would be high if there was information asymmetry. Since this is evidence of low information asymmetry, our finding also suggests that the bulk of trading volume is due to liquidity demand and supply.
vari-ables are positively associated with trading volume and volatility. Fourteen (15) and 11(13) out of 30 large (small) stocks for the HH and LL groups have significantly pos-itive and negative coefficients. The means of the t-statistics of the HH and LL groups for the large (small) stocks are 1.51(1.72) and−1.19(−1.51). For the HH group there are 26(25) positive coefficients for the large (small) stocks, and 22(24) negative co-efficients for the small stocks. Thus, order flows that are not accounted for by the information and liquidity variables also drive trading volume and volatility. As argued earlier, these orders are likely speculation oriented and constitute a significant part of TSE’s order flows.
Another observation born out of our analysis is that the explanatory variables do not explain the HL and LH groups as much as the HH and LL groups. This is consistent with the findings in the literature that high trading volume and volatility happen together and are driven by information, liquidity and speculation. In sum, we have found that information orders and liquidity orders drive the trading volume and volatility of small firms, while market-wide information affect the volume and volatility of large firms. Low degree of consensus in the traders’ interpretation of information and speculative orders drive trading volume and volatility for both large and small stocks. There is no clear sign that the firm-specific information produces volume and volatility.
Since we have some evidence that large information orders are more related to trading volume and volatility than small liquidity orders for the small stocks, and that the liquidity orders could actually be negatively associated with volume and volatility for the large stocks, it ought be the case that OI is more aggressive in placing orders than OL. Our data confirm this conjecture. Table 6 depicts the order-submitting pat-terns of the large and small orders. It shows, for a particular intraday interval, the average percentage (across the trading days and stocks in the sample) of buy (sell) or-ders in number of shares (or oror-ders, in parentheses) placed for different price ranges: at the upper price-limit (lower price limit), within one percent away from the upper price-limit (lower price limit), between one percent away from the upper price limit (lower price limit) and two percent away from the upper price limit (lower price limit), between two percent away from the upper price limit (lower price limit) and three per-cent away from the upper price limit (lower price limit), and over three perper-cents away from the upper price limit (lower price limit).
Table 6 Intraday Order Submission Patterns of OI and OL
The orders are classified for different price intervals which include, for the buy orders, orders placed at the upper price limit, the interval between the upper price limit and 1% from the upper price limit, the interval between 1% from the upper price limit and 2% from the upper price limit, etc. The sell orders are classified similarly except that price intervals are measured from the lower price limit. For each intraday time interval, the percentage of orders falling in a price interval is calculated. The percentage numbers reported are the averages across 104 trading days and the 60 stocks in our sample based on the number of shares submitted. Numbers in parentheses are based on the number of orders submitted.
Price 1% from 1–2% 2–3% 3% and
limit price from from over
limit price price
limit limit Buy orders OI Open (7.97)7.03 (2.19)3.04 (2.56)3.3 (3.58)4.53 (83.7)82.1 9–9:06 (17.71)13.96 (4.14)6.02 (4.68)5.82 (5.09)7.16 (68.38)67.04 9:06–11:54 (11.89)10.14 (5.21)6.99 (6.41)8.35 (6.92)8.73 (69.57)65.79 11:54–12 (15.54)15.7 (4.3)6.19 (6.02)7.38 (5.75)7.15 (68.39)63.63 OL Open (7.41)7.88 (3.62)3.15 (3.48)3.26 (3.2)3.34 (82.29)82.37 9–9:06 (11.97)12.78 (3.85)3.75 (4.34)4.9 (4.36)4.6 (75.48)73.97 9:06–11:54 (9.58)9.94 (5.2)5.32 (6.15)6.42 (6.45)6.76 (72.62)71.56 11:54–12 (12.44)12.82 (4.08)4.28 (5.24)5.54 (5.34)5.68 (72.9)71.68 Sell orders OI Open (3.97)7.15 (1.67)2.11 (1.54)1.53 (1.63)1.52 (91.19)87.69 9–9:06 (14.16)11.55 (3.41)3.82 (3.46)3.76 (3.11)3.75 (75.86)77.12 9:06–11:54 (9.2)6.34 (3.22)2.83 (3.51)3.78 (3.91)4.45 (80.16)82.6 11:54–12 (20.07)17.04 (3.04)2.76 (2.64)2.82 (3.27)3.24 (70.98)74.14 OL Open (3.85)3.6 (3.42)2.64 (1.65)1.65 (1.77)1.7 (89.31)90.41 9–9:06 (11.76)11.98 (3.94)4.26 (3.61)3.37 (3.54)3.19 (77.15)77.2 9:06–11:54 (8.03)8.22 (3.48)3.27 (3.2)3.17 (3.58)3.5 (81.71)81.84 11:54–12 (13.87)14.22 (2.82)2.88 (2.94)3.03 (3.21)3.1 (77.16)76.77
