Intraday Trading by floor traders and customers infutures markets: whose trades drive the...

Introduction

Under the mixture-of-distributions hypothesis in Clark (1973) and Tauchen and Pitts (1983), there exists a positive relation between prices and trading volume because both respond to the same underlying latent information. On the other hand, there is a considerable interest in

the impact of trader type on volatility. Theoretic models such as De Long et al. (1990) and Campbell and Kyle (1993) show that trading by uninformed noise traders increases market volatility. Shalan (1993) reaches a similar conclusion that the least informed traders increase volatility, whereas informed trading reduces volatility by removing uncertainty in the marketplace. In addition, Harris and Raviv (1993) show that even in the absence of noise traders, traders with homogenous prior information but different processing procedure can result in a positive volatility-volume relation. (1)

This paper aims to shed light on the question of whose trades drive the volatility-volume relation. Such a study presents an empirical test of the financial theories discussed earlier. In addition, the effect of trading parties on market volatility is an important public policy issue. For example, the advocates of throwing sand in the gear argue that transaction tax should be imposed to reduce volatility. One can argue that such a tax be imposed specifically on retail customers if only volume from trading by retail customers is positively related to price volatility. (2)

The paper makes four improvements over the previous studies. First, the unique transaction dataset in this study codes both parties for each trade, e.g., a floor trader against a retail customer. In contrast, data in previous studies such as Manaster and Mann (1996) and Wiley and Daigler (1998) contain trader type only for one of the two parties involved in a trade. The use of only one party in identification can hide the true impact of trader types on volatility. This paper is the most vigorous examination that reveals true pairings of parties and their effect on market volatility.

Second, minute-by-minute transaction data are used in this study, whereas previous studies use daily data. Given the speed of trading and price adjustments in future markets, studies with low frequency data easily can fail to capture the volatility-volume relation that is evident only in intraday data. Therefore, results from this study can complement and contrast with those from the previous studies to assess whether the volatility-volume relation is robust to market microstructures such as the frequency of data.

The third improvement comes from the fact that Hausman specification test is used to test the presence of simultaneity between volatility and volume. In the presence of simultaneity bias, ordinary least squares (OLS) method is replaced by the generalized method of moments (GMM) approach with instrumental variables to study the contemporaneous intraday volatility-volume relation. The removal of simultaneity bias makes the results in the paper more reliable.

Finally, in addition to using a vector-autoregressive (VAR) model to answer the question of which types of trading Granger-cause volatility, volume is decomposed into expected and unexpected parts using autoregressive integrated moving average (ARIMA) models to study the contemporaneous intraday volatility-volume relation.

Furthermore, a robustness check is done to examine whether results vary in different trading sessions, e.g., morning session versus afternoon session.

Related Studies

A positive volatility-volume relation in futures markets has been documented extensively in the literature. For example, Grammatikos and Saunders (1986) find that price variability and trading volume are positively correlated in all futures markets studied. Martell and Wolf (1987), Foster (1995), and Wang and Yau (2000) report evidence of a positive relation between trading volume and price volatility in various futures markets. Garcia, Leuthold, and Zapata (1986) use a VAR framework to document a lead-lag relationship between trading volume and price variability in five grain futures markets.

Using an ARIMA model, Bessembinder and Seguin (1993) find that unexpected volume shocks have a larger effect on volatility in futures markets than expected volume. Fujihara and Mougoue (1997) find a non-linear causal relation between volatility and volume in petroleum futures markets. Luu and Martens (2003) use realized volatility from intraday data and find a bi-directional causal relation between volatility and volume in the US stock index futures market.

On the other hand, Chang, Pinegar, and Schachter (1997) document a positive association between volatility and volume by large speculators in five futures markets. Daigler and Wiley (1999) report evidence that trading by the least-informed traders (i.e., retail customers as a group) is positively associated with volatility in five futures markets. In contrast, trading by floor traders, i.e., the most informed traders, is negatively associated with volatility.

In contrast, Irwin and Yoshimaru (1999) find that trading by large managed futures funds and pools does not increase market volatility. Similarly, Wang (2002) find that trading by large speculators actually stabilizes price volatility, whereas trading by hedgers is associated with increases in volatility. Presumably, these large commercial traders are better informed than small speculators. All these studies use daily data.

Data and Methodology Data

The transaction data are audit data compiled by Commodity Futures Trading Commission (CFTC); see Sahin and Sarajoti (2005) and Locke and Mann (2005) for a detailed description of the data. The data contain every trade in 1995 for live cattle, frozen pork bellies, German mark, and Swiss franc futures contracts at the Chicago Mercantile Exchange (CME). As shown in Panel A of Table 1, regular trading session is longer for the two foreign exchange futures than for the two agricultural contracts. There was no after-hour electronic trading in 1995. Because each trade in the dataset is timed to the nearest minute, there are usually multiple trades within a minute.

The trading parties behind every trade are coded by four customer type indicators (CTI) defined by CFTC. They are CTI 1--floor trader proprietary trading for own accounts; CTI 2--the floor trader is trading for the trader's clearing member; CTI 3--the floor trader is trading for another trader who is present on the floor (such as a futures options trader wishing to hedge); and CTI 4--trading for off-the-floor customers. Each trade is recorded twice. For example, a trade between a floor trader and a customer is recorded as a Type 14 trade and also as a Type 41 trade. In this study, only trades from Types 11, 12, 13, 14, 22, 23, 24, 33, 34, and 44 are used. (3)

Following the approach in Bessembinder and Seguin (1993) and Luu and Martens (2003), volume from different maturities is aggregated over a 30-minute sampling interval to reach intraday trading volume (TV) in equation (1). (4)

[TV.sub.t,t+k] = [j.summation over (i=1)] ([30.summation over (k=1)] [volume.sub.k]) (1)

where k indicates minute and i is for maturity month. As a result, there are more such 30-minute intervals for the foreign exchange contracts than for the two agricultural contracts. Trading for the two foreign exchange futures contracts is equally active, which results in identical sample size. In contrast, trading is more active for live cattle futures than for frozen pork bellies futures, which is reflected by a larger sample size for live cattle futures.

Panel B of Table 1 shows the composition of types of trading. The most active types of trading are trader-customer (Type 14). The next two active types of trading are customer-customer (Type 44) and trader-trader (Type 11). These three types of trading account for about 90 percent and 70 percent of all volume in the agricultural futures markets and the foreign exchange futures markets, respectively. These three types are the focus of the paper.

Panel C of Table 1 shows the mean and standard deviations of volume for each type of trading during the sampling interval of 30 minutes. Trading between traders and customers (Type 14) has the highest average trading volume as well as the highest standard deviations. Trading between floor traders has the lowest average volume and the smallest deviations.

Realized volatility, i.e., squared returns, is used widely with high frequency data. Because all trades are time-stamped to the nearest minute, the volume-weighted price is calculated to obtain the market price for each minute, see, e.g., Manaster and Mann (1996). As discussed by Andersen and Bollerslev (1998), realized volatility (V) should be calculated with 5-minute rather than 1-minute returns because the latter is more noisy due to market microstructure induced price bounces. (5) Following this approach, realized volatility is calculated using the first prices from the two adjacent 5-minute intervals.

[r.sub.t], = log([p.sub.t]) - log([p.sub.t-5]) (2)

where [r.sub.t] and [P.sub.t] are return and volume-weighted price, respectively. Similar to the calculation of trading volume, realized volatility from each contract month is summed up to reach realized volatility within the 30-minute sampling interval.

[V.sub.t] = [j.summation over (i=1)] ([6.summation over (s=1)] [r.sub.s.sup.2]) (3)

As in Luu and Martens (2003), overnight returns are calculated using today's first price and the last price from the previous day. Panel D of Table 1 reports the average realized volatility. Notice that volatility is higher for the agricultural futures contracts than for the foreign exchange contracts.

Table 2 presents Pearson correlation coefficients between volatility and volume for the three major types of trading. Consistent with the findings in Grammatikos and Saunders (1986), correlation between volatility and volume is positive and statistically significant at the 1 percent level. Volatility is more correlated with Type 14 and Type 44 volume, whereas volatility is least correlated with Type 11 volume, i.e., trading between floor traders. Type 14 trading is correlated more with Type 44 trading than with Type 11 trading in the two agricultural markets. Trading between floor traders, however, is correlated more with Type 14 than with Type 44 in the two foreign exchange markets. Comparing with the results in Wiley and Daigler (1998), Pearson coefficients between the types of volume are much smaller. This is due to further decomposition of volume according to both parties involved in a trade, rather than based on CTI alone. Table 2 provides some anecdotal evidence for the importance of customer trading in relation to volatility.

Hypotheses and Methodology

Of course, correlation does not imply causality. For example, Chang, Chou, and Nelling (2000) show that hedgers increase trading in response to the increase in market volatility. For public policy debate, it is necessary to show that individual types of volume Granger cause price volatility. It is hypothesized that there exists no Granger causality relation between volatility and any of the three major types of volume.

HO1: There is no Granger-causality between price volatility and each of the three individual trading types.

The VAR approach provides a framework for testing Granger causality and has been used widely in the literature, e.g., Garcia, Leuthold, and Zapata (1986), Fujihara and Mougoue (1997), and Luu and Martens (2003). VAR modeling requires that all times series be stationary. Both the augmented Dickey-Fuller test and the Phillips-Perron test are conducted for up to 12 lags for three cases: zero mean, single mean, and deterministic trend. Although not reported in the paper, the null hypothesis of nonstationarity is rejected in all cases, which is consistent with the results in many previous studies, e.g., Wiley and Daigler (1998), and Wang and Yau (2000).

As a result, the following VAR (p) model is estimated, in which the Schwarz Bayesian information criterion is used to determine the optimal lag length (p) in the model with a maximum lag of 5.

[[OMEGA].sub.t] = [alpha] + [p.summation (k=1)] [[lambda].sub.k] [[OMEGA].sub.t-k] + [[epsilon].sub.t] (4)

where [OMEGA] is a vector for volatility and trading volume from the three major types of trading. Given the VAR model above, there are two tests for each of the three major types of trading. Hence, six tests are conducted to shed light on the Granger-causality between price volatility and volume from the three main types of trading.

In addition to Granger-causality testing, contemporaneous volatility-volume relation is also of interest. Equation (5) is used to represent the contemporaneous volatility-volume relation.

[V.sub.t] = [alpha] + [[beta].sub.1] [TV.sub.11,t] + [[beta].sub.2] [TV.sub.14,t] + [[beta].sub.3] [TV.sub.44,t] + [psi] Monday + [[epsilon].sub.t] (5)

where [TV.sub.11], [TV.sub.14], and [TV.sub.44] are trading volume from trading between floor traders, trading between floor traders and customers, and trading between customers, respectively. The dummy variable Monday accounts for the well-known Monday effect. Chang, Pinegar, and Schachter (1997) and Han, Kling, and Sell (1999) report evidence for the Monday effect. The following hypothesis is tested.

HO2: There is no contemporaneous relation between price volatility and each of the three major types of trading volume.

It is important to account for possible simultaneity between contemporaneous volatility and volume. As discussed above, both the mixture-of-distributions hypothesis and the informed trading models predict that volume and volatility are jointly determined. Empirical evidences in Koch (1993), Foster (1995), and Wang and Yau (2000) find that simultaneity bias is present. Therefore, Hausman's specification test is conducted to test the null hypothesis that volume is exogenous to volatility. Under the null hypothesis of no simultaneity, OLS estimates are consistent and efficient. Under the alternative hypothesis of simultaneity, however, OLS estimates are inconsistent but efficient. After obtaining a consistent estimate under the alternative hypothesis, e.g., a two-stage least squares estimator (2SLS), one can apply Hausman m-statistic to test the null hypothesis of no simultaneity. (6)

If the null hypothesis of no simultaneity cannot be rejected, equation (5) is estimated by GMM. GMM imposes weaker distribution assumptions and is especially suitable due to positive serial correlation in squared returns. On the other hand, if simultaneity exists, GMM is used with lagged values as instrumental variables for volume.

Bessembinder and Seguin (1993) decompose volume into expected and unexpected parts to test informational effects of volume on volatility. It is likely, under the rational expectation theory, that the effect from expected and unexpected parts can be asymmetric. As in Bessembinder and Seguin (1993), ARIMA models are used to decompose volume into expected and unexpected parts. Because all time series are stationary, the following ARIMA(p,0,q) model is estimated.

[TV.sub.t] = [theta] + [p.summation over (i=1)] [[alpha].sub.i] [TV.sub.t-i] + [q.summation over (j=0)] [[beta].sub.i][[epsilon].sub.t-j] (6)

The Schwarz Bayesian information criterion is used to determine the orders of the autoregressive and moving average parts in the ARIMA models. Equation (7) is used to test the effects of expected and unexpected volume on contemporaneous volatility.

[V.sub.t] = [alpha] + [3.summation over (1)] [[phi].sub.i]E([TV.sub.i,t]) + [3.summation over (1)] [[lambda].sub.i] U ([TV.sub.i,t]) + [psi] Monday + [[epsilon].sub.t]

where V and TV are for volatility and trading volume, respectively. E(TV) is expected volume, which is the fitted value from the corresponding ARIMA models. U(TV) is unexpected volume, which is the residual of the corresponding ARIMA models. Equation (7) also is estimated by GMM.

Results

VAR Model

Table 3 presents the results from the Granger-causality test for volatility and volume. In the two agricultural markets, a bi-directional causality is found between volatility and each of the three major types of trading volume, which is consistent with the findings in Luu and Martens (2003). In the two foreign exchange futures markets, volume from trading between traders and customers does not Granger-cause price volatility, but volume from trading between customers Granger-causes price volatility. There is a Granger-causal relation from volume to volatility for trading between floor traders in Swiss franc futures market but not in German mark futures market. Overall, the results in Table 3 show that the dynamic relation between volatility and volume depends on trader types involved on both sides of the transactions.

Contemporaneous Volatility-Volume Relation--Total Volume

The volatility-volume relation at the aggregate level is established before testing the impacts of volume from three major types of trading. Panel A of Table 4 presents the results from Hausman's specification test. Except for frozen pork bellies futures market, Hausman's m-statistic is not significant and thus the null hypothesis that volume is exogenous to volatility is not rejected. As a result, GMM with lagged values as instrumental variables is used for frozen pork bellies contract and OLS is used for all other three futures contracts in the estimation of equation (5).

As shown in Panel B of Table 4, the coefficient for total volume is positive and significant in all markets, which is consistent with those in Foster (1995) and Wang and Yau (2000) as well as other studies surveyed in Karpoff (1987). Because a positive relation generally is found between intraday volatility and intraday aggregate volume, the results in the paper support both the mixture-of-distributions hypothesis and the informed trading models.

Notice that the coefficient for the dummy variable Monday is not significant in all markets except for the live cattle futures market, in which it is significantly positive. Therefore, there is no evidence that volatility decreases on Mondays, as found in Chang, Pinegar, and Schachter (1997) and Han, Kling, and Sell (1999).

Whose Trades Drive the Contemporaneous Volatility-Volume Relation--Individual Volume

Table 5 presents results of contemporaneous relation between volatility and the three major individual types of trading. Again, there is no evidence that volatility tends to decrease on Monday. For trading between floor traders, volatility is significantly negatively related to volume in the two foreign exchange markets and is not correlated to volume in the agricultural markets. Therefore, the results are consistent with conventional wisdom that professional floor traders are informed traders. Floor traders have at least some access to information circulating on the floor including the level of communication between brokers and the phone desks. Thus, trading between them probably is motivated by position adjustments and injects little noise into the price, which in return will not increase price volatility. The results are also consistent with those in Daigler and Wiley (1999).

For trading between floor traders and customers, volatility is significantly positively related to volume in all markets except for the frozen pork bellies futures market. In the frozen pork bellies futures market, volatility is not related to volume. For trading between customers, volatility is significantly related to volume in the agricultural markets but not in the foreign exchange markets. Therefore, evidence supports the empirical results in Daigler and Wiley (1999) that retail customers are noise traders only in the agricultural markets. It is clear that two types of trading, trading between customers and trading between floor traders and customers, drive the positive volatility-volume relation.

Notice the differences in both data and methodology between this study and that by Daigler and Wiley (1999), i.e., intraday data versus daily data and identifying a trade by both parties versus by only one of the two parties involved. Nevertheless, the results from this study are more consistent than contradicting with those in Daigler and Wiley (1999). For example, the results on volume from trading between floor traders are consistent with their results. On the other hand, this study also provides support that retail customers, at least in the agricultural futures markets, are noise traders.

Volatility and Expected and Unexpected Volume

Panel A of Table 6 shows the orders of the autoregressive and moving average parts in the ARIMA models, which is determined by the Schwarz Bayesian information criterion. Panel B of Table 6 reports the results of regressing volatility on expected and unexpected parts of volume from three major types of trading.

There is almost symmetric volatility-volume relation between expected and unexpected volume, which is consistent with the results in Bessembinder and Seguin (1993). The only exception is for trading between floor traders in the two foreign exchange markets. In this special case, expected volume is negatively related to price volatility and unexpected volume is not related to volatility.

By decomposing volume into expected and unexpected parts, it is more evident that trading between customers and floor traders becomes the main driving force behind the positive volatility-volume relation. For example, volatility is related to both expected and unexpected volume from trading between customers and floor traders in all markets. Overall, the results in Table 6 are consistent with the results in Table 5. Therefore, the conclusions in Table 5 are not affected by the decomposition of volume into expected and unexpected components.

Robustness Check--Different Sessions during the Day

It is well known that there is a U-shaped pattern in both intraday volume and volatility see, e.g., Chan, Chart, and Karolyi (1991) and Daigler (1997). It is of interest if the observed volatility-volume relation is present only during certain trading sessions. As a robustness check, each trading day is divided into four sessions: opening (the first 30 minutes), morning (after the opening but before 11:30am Chicago time), afternoon before closing, and closing session of last 30 minutes of trading. Table 7 presents estimation results during the four different sessions.

Overall, Table 7 shows that opening and closing sessions are not the only two periods that observe a positive volatility-volume pattern as in Table 6. For example, for trading between floor traders and customers, a significantly positive volatility-volume relation is present more during opening, morning, and afternoon sessions than during closing sessions in all markets. For frozen pork bellies contracts, it is only during the afternoon session that volatility is positively related to volume.

For trading between customers, the effects from both unexpected volume and expected volume are more or less present in all four sessions. For trading among floor traders, a negative volatility-volume relation is more likely to be observed during opening and closing sessions than during morning and afternoon sessions. There is no evidence that a single trading session drives the results in Table 6. In summary, U-shaped volatility and volume patterns do not affect the volatility-volume relation severely.

Summary and Conclusion

This study analyzes the impact of trading parties on the intraday volatility-volume relation in four futures markets. Three types of most active engagements are trader-customer, customer-customer, and trader-trader. These three types of trading account for an average of 80 percent of all volume in these markets. Conventional wisdom is that floor traders are the most informed due to their presence in the trading floor, whereas outside retail customers are the least informed.

The dynamic relation between volatility and volume is first examined in a VAR model. A bi-directional causal relation is found generally. On the other hand, the results also show that trading party does have an impact on market volatility. Volume from trading between floor traders--the most informed one--is associated with either no change or a reduction in volatility. Volatility is positively related to volume from trading between floor traders and customers. Only partially consistent with the noise trading models and common perception, however, trading between customers is significantly positively related to volatility only in two of the four markets. Using ARIMA models to decompose volume into expected and unexpected parts, it is more evident that trading between customers and floor traders drives the observed positive volatility-volume relation.

The results in the paper raise questions on two issues. First, are retail customers the least-informed traders in all markets? Second, will trading by customers, even if they are less informed, always increase market volatility? This study highlights that caution is needed against the perception that trading by retail customers increases market volatility in all markets.

References

(1.) Admati, A. and P. Pfleiderer, "A Theory of Intraday Patterns: Volume and Price Variability," Review of Financial Studies, 1, no. 1 (Spring 1988), pp. 3-40.

(2.) Andersen, T., and T. Bollerslev, "Answering the Skeptics: Yes, Standard Volatility Models Do Provide Accurate Forecasts," International Economic Review, 39, no. 4 (November 1998), pp. 885-905.

(3.) Bessembinder, H., and P. Seguin, "Price Volatility, Trading Volume and Market Depth: Evidence from Future Markets," Journal of Financial and Quantitative Analysis, 28, no. 1 (March 1993), pp. 21-39.

(4.) Campbell, J., and A. Kyle, "Smart Money, Noise Trading and Stock Price Behavior," Review of Economic Studies, 50, no. 202 (January 1993), pp. 1-34.

(5.) Chan, K., K.C. Chan, and A. Karolyi, "Intraday Volatility in the Stock Index and Stock Index Futures Markets," Review of Financial Studies, 4, no. 4 (Winter 1991), pp. 657-684.

(6.) Chang, E., M. Pinegar, and B. Schachter, "Interday Variations in Volume, Variance and Participation of Large Speculators," Journal of Banking and Finance, 21, no. 6 (June 1997), pp. 797-810.

(7.) Chang, E., R. Chou, and E. Nelling, "Market Volatility and The Demand for Hedging in Stock Index Futures," Journal of Futures Markets, 20, no. 2 (February 2000), pp. 105-125.

(8.) Copeland, T., "A Model of Asset Trading under the Assumption of Sequential Information Arrivals," Journal of Finance, 31, no. 4 (September 1976), pp. 1149-1168.

(9.) Corsi, F., G. Zumbach, U. Muller, and M. Dacorogna, "Consistent High-precision Volatility from High-frequency Data," Economic Notes, 30, no. 2 (July 2001), pp. 183-204.

(10.) Daigler, R., "Intraday Futures Volatility and Theories of Market Behavior," Journal of Futures Markets, 17, no. 1 (February 1997), pp. 45-74.

(11.) Daigler, R., and M. Wiley, "The Impact of Trader Type on the Futures Volatility-Volume Relation," Journal of Finance, 54, no. 6 (December 1999), pp. 2297-2316.

(12.) De Long, B., A. Shleifer, L. Summers, and R. Waldmann, "Noise Trader Risk in Financial Markets," Journal of Political Economy, 98, no. 4 (August 1990), pp. 703-738.

(13.) Foster, D., and S. Viswanathan, "A Theory of the Interday Variations in Volume, Variance, and Trading Costs in Securities Markets," Review of Financial Studies, 3, no. 4 (Winter 1990), pp. 593-624.

(14.) Foster, A., "Volume-Volatility Relationship for Crude Oil Futures Markets," Journal of Futures Markets, 15, no. 8 (December 1995), pp. 929-951.

(15.) Fujihara, R., and M. Mougoue, "An Examination of Linear and Nonlinear Causal Relationships between Price Variability and Volume in Petroleum Futures Markets," Journal of Futures Markets, 17, no. 4 (June 1997), pp. 385-416.

(16.) Garcia, P., R. Leuthold, and H. Zapata, "Lead-Lag Relationships between Trading Volume and Price Variability: New Evidence," Journal of Futures Markets, 6, no. 1 (Spring 1986), pp. 1-10.

(17.) Grammatikos, T., and A. Saunders, "Futures Price Variability: A Test of Maturity and Volume Effects," Journal of Business, 59, no. 2 (April 1986), pp. 319-330.

(18.) Han, L., J. Kling, and C. Sell, "Foreign Exchange Futures Volatility: Day-Of-The-Week, Intraday, and Maturity Patterns in the Presence of Macroeconomic Announcements," Journal of Futures Markets, 19, no. 6 (September 1999), pp. 665-693.

(19.) Harris, M., and A. Raviv, "Differences of Opinion Make a Horse Race, Review of Financial Studies, 6, no. 3 (Fall 1993), pp. 473-506.

(20.) Irwin, S. and S. Yoshimaru, "Managed Futures, Positive Feedback Trading, and Futures Price Volatility," Journal of Futures Markets, 19, no. 7 (October 1999), pp. 759-776.

(21.) Koch, P., "Reexamining Intraday Simultaneity in Stock Index Futures Markets," Journal of Banking and Finance, 17, no. 6 (December 1993), pp. 1191-1205.

(22.) Karpoff, J., "The Relation Between Price Changes and Trading Volume: A Survey," Journal of Financial and Quantitative Analysis, 22, no. 1 (March 1987), pp. 109-126.

(23.) Locke, P., and S. Mann, "Professional Trader Discipline and Trade Disposition," Journal of Financial Economics 76, no. 2 (May 2005), pp. 401-444.

(24.) Luu, J., and M. Martens, "Testing the Mixture-of-Distributions Hypothesis Using 'Realized' Volatility," Journal of Futures Markets, 23, no.7 (July 2003), pp. 661-679.

(25.) Maddala, G. S., Introduction to Econometrics (Englewood Cliffs, New Jersey: Prentice Hall, 1992).

(26.) Martell, T., and A. Wolf, "Determinants of Trading Volumes in Futures Markets," Journal of Futures Markets, 7, no. 3 (June 1987), pp. 233-244.

(27.) Manaster, S., and S. Mann, "Life in the Pits: Competitive Market Making and Inventory Control," Review of Financial Studies, 9, no. 3 (Fall 1996), 953-975.

(28.) Sahin, O., and P. Sarajoti, "The Impact of Trading Party on the Execution Spread: Evidence from Futures Markets," Quarterly Journal of Business and Economics, 44, no. 1 (Winter 2005), pp. 55-65.

(29.) Shalen, C., "Volume, Volatility, and the Dispersion of Beliefs," Review of Financial Studies, 6, no. 2 (Summer 1993), pp. 405-434.

(30.) Stiglitz, J., "Using Tax Policy to Curb Speculative Short-Term Trading," Journal of Financial Services Research, 3, no. 2 (December 1989), pp. 101-115.

(31.) Summers, L., and V. Summers, "When Financial Markets Work Too Well: A Cautious Case for s Securities Transaction Tax," Journal of Financial Services Research, 3, no. 2 (December 1989), pp. 261-286.

(32.) Tauchen, G., and M. Pitts, "The Price Variability-Volume Relationship on Speculative Markets," Econometrica, 51, no. 2 (March 1983), pp. 485-505.

(33.) Wang, C., "Information, Trading Demand, and Futures Price Volatility," Financial Review, 37, no. 2 (May 2002), pp. 295-316.

(34.) Wang, G., and J. Yau, "Trading Volume, Bid-Ask Spread, and Price Volatility in Futures Markets," Journal of Futures Markets, 20, no. 10 (November 2000), pp. 943-970.

(35.) Wiley, M., and R. Daigler, "Volume Relationships among Types of Traders in the Financial Futures Markets," Journal of Futures Markets, 18, no. 1 (February 1998), pp. 91-113.

(1) The informed trading models, such as Admati and Pfleiderer (1988) and Foster and Viswanathan (1990), show that both informed traders and liquidity traders trade during high liquidity periods. Liquidity traders are assumed to be noise traders in these models. Thus, trading by liquidity traders presumably increases price volatility during higher volume period.

(2) See, e.g., Stigleitz (1989), and Summers and Summers (1989).

(3) In comparison, a small discrepancy does exist if trades from Types 44, 43, 42, 41, 34, 32, 31, 22, 21, and 11 are used--probably due to errors in the data. Also, Types 11, 22, 33, and 44 trades are adjusted so that there is no double-counting.

(4) There are several contracts with different maturity months are traded simultaneously. Therefore, the total trading volumes are the sum of each contract month over the period of 30 minutes. Notice that the last interval does not always have 30 minutes depending the opening minute and closing time specific to the day and contracts. The main results remain unchanged when a 15-minute or a 60-minute sampling interval is used.

(5) Some researchers even suggest the use of one-hour interval; see for example, Corsi, Zumbach, Muller, and Dacorogna (2001).

(6) See Maddala (1992), pp. 506-508 for a discussion of the Hausaman m-statistic and its application.

Haiwei Chen *

California State University, San Bernardino

* I am indebted to Peter Locke for providing the data. Many helpful comments and suggestions are also received from anonymous referees. Editorial assistance is also received from Greg Richey. The usual disclaimer applies.

Table 1--Summary Statistics

Trading hour is Chicago time. Sample size is the number of 30-minute
intervals during which the total number of contracts traded (Volume)
is aggregated. Each trade involves two parties. The three most active
are trading between floor traders and customers (Type 14), trading
between customers (Type 44), and trading between floor traders (Type
11). Market volatility is measured by realized volatility, i.e.,
squared 5-minute returns. Similar to volume aggregation, volatility is
also aggregated over the 30-minute interval and is scaled by a factor
of 1,000. In parentheses are standard deviations

Panel A: Sample Size
                Live Cattle   Pork Bellies   Deutsch Mark   Swiss Franc
Trading Hours   9:10-13:00     9:10-13:00     7:20-14:00    7:20-14:00
Sample Size        2,246         1,996          3,481          3,481

Panel B: Composition of Volume by Three Major Types of Trading
Type 14           58.87%         60.79          40.37%        48.68%
Type 44           24.92%         24.88%         14.84%        15.28%
Type 11            6.75%          7.82%          9.07%         7.86%

Panel C: Volume by Three Major Types, Mean and Standard Deviations
Type 14             839           169            718            544
                   (735)         (149)          (643)          (463)
Type 44             355            69            264            171
                   (370)          (89)          (349)          (170)
Type 11             97             22            162            88
                   (85)           (21)          (142)          (79)

Panel D. Mean Realized Volatility
                  0.0343         0.0297         0.0217        0.0229

Table 2--Correlations between Volatility and Three Major Types of
Volume

The table presents Pearson correlation coefficients between
volatility and trading volume for each of the three major types of
trading, i.e., trading between floor traders and customers (Type 14),
trading between customers (Type 44), and trading between floor
traders (Type 11). The null hypothesis is that there is no
correlation between the two variables

             Volume 11   Volume 14   Volume 44

Panel A. Live Cattle

Volatility   0.42 ***    0.53 ***    0.51 ***
Volume 11                0.79 ***    0.63 ***
Volume 14                            0.89 ***

Panel B. Pork Bellies

Volatility   0.32 ***    0.54 ***    0.57 ***
Volume 11                0.65 ***    0.41 ***
Volume 14                            0.76 ***

Panel C. Deutsch Mark

Volatility   0.18 ***    0.29 ***    0.23 ***
Volume 11                0.78 ***    0.48 ***
Volume 14                            0.73 ***

Panel D. Swiss Franc

Volatility   0.22 ***    0.33 ***    0.27 ***
Volume 11                0.83 ***    0.53 ***
Volume 14                            0.72 ***

*** indicates statistical significance at 1 percent level.

Table 3--Granger Causality Test for the Volatility-Volume Relation

The [chi square] statistics tests the null hypothesis of no
Granger-causality from volume to ([right arrow]) volatility or from
volatility to volume. Schwarz Bayesian Information Criterion is used
to determine the optimal lag length (p) with up to 5 lags in the
model. In parentheses are the p-values

                                    Live Cattle    Pork Bellies

Optimal Lag Length                       2              2

Panel A. Trader-Trader Volume

Volume [right arrow] Volatility        14.97           8.58
                                       (0.00)         (0.01)
Volatility [right arrow] Volume         5.94          16.12
                                       (0.05)         (0.00)

Panel B. Trader-Customer Volume

Volume [right arrow] Volatility        32.75           6.80
                                       (0.00)         (0.03)
Volatility [right arrow] Volume        21.09          19.71
                                       (0.00)         (0.00)

Panel C. Customer-Customer Volume

Volume [right arrow] Volatility        29.68          15.48
                                       (0.00)         (0.00)
Volatility [right arrow] Volume        10.22          40.33
                                       (0.01)         (0.00)

                                    Deutsch Mark   Swiss Franc

Optimal Lag Length                       2              2

Panel A. Trader-Trader Volume

Volume [right arrow] Volatility         4.57           6.35
                                       (0.10)         (0.04)
Volatility [right arrow] Volume         2.28          18.63
                                       (0.32)         (0.00)

Panel B. Trader-Customer Volume

Volume [right arrow] Volatility         1.84           0.16
                                       (0.40)         (0.92)
Volatility [right arrow] Volume         1.49           7.01
                                       (0.47)         (0.03)

Panel C. Customer-Customer Volume

Volume [right arrow] Volatility        11.32          26.26
                                       (0.00)         (0.00)
Volatility [right arrow] Volume        18.87           7.58
                                       (0.00)         (0.02)

Table 4--Contemporaneous Volatility and Total Volume

The regression model is [V.sub.t] = [alpha] + [beta]TV + [psi]Monday
+ [e.sub.t]. The dependent variable is intraday volatility (V).
The independent variables are trading volume (TV) and dummy variable
Monday for the Monday effect. In Panel A, Hausman's m-statistic tests
the null hypothesis that volume is exogenous to volatility. If the
null hypothesis of no simultaneity is rejected, GMM with lagged values
as instrumental variables in the estimation. Panel B presents the
estimation results of the same model. Coefficients for volume and the
dummy variable are adjusted by a factor of 1,000. In parentheses are
t-statistics with autocorrelation and heteroskedasticity consistent
standard errors

                    Live         Pork        Deutsch        Swiss
                   Cattle      Bellies        Mark          Franc

Panel A. Hausman Specification Test for Simultaneity between
  Volatility and Volume

Hausman            8.67       34.96          7.06          2.02
  m-statistic
(p-value)         (0.12)      (0.00)        (0.07)        (0.57)

Panel B. Contemporaneous Volatility-Volume Relation--Three Tykes of
  Trading

[alpha]           -0.01        0.08         -0.03         -0.02
                 (-2.36 **)   -1.15        (-3.58 ***)   (-3.38 ***)
[beta]             0.03        0.74          0.03          0.04
                 (6.85 ***)   (3.19 ***)    (5.92 ***)    (5.68 ***)
[psi]             6.40        81.77         14.66          2.80
                 (2.19 **)    -1.68         -1.43         -0.49
N                 2,246        1,995         3,481         3,481
Adj. [R.sup.2]    0.29         0.25          0.08          0.10

*** and ** indicate statistical significance at 1 percent level and
5 percent level, respectively

Table 5--Contemporaneous Volatility-Volume Relation--Three Types of
Trading

The regression model is [V.sub.t] = [alpha] + [[beta].sub.1]
[TV.sub.11,t] + [[beta].sub.2] [TV.sub.14,t] + [[beta].sub.3]
[TV.sub.44,t] + [psi]Monday +
[e.sub.t]. The dependent variable is intraday volatility (V). The
types of trading volume (TV) are: trading between floor traders and
customers (Type 14), trading between customers (Type 44), and trading
between floor traders (Type 11). Monday is a dummy variable for the
Monday effect. GMM with lagged values as instrumental variables is
used in the estimation for frozen pork bellies futures market.
Coefficients for volume and the dummy variable are adjusted by a
factor of 1,000. In parentheses are t-statistics with autocorrelation
and heteroskedasticity consistent standard errors

                    Live         Pork        Deutsch        Swiss
                   Cattle       Bellies       Mark          Franc

A                 -0.01          0.12        -0.02         -0.02
                 (-2.32 ***)    -1.90       (-4.06 ***)   (-3.31 ***)
[[beta].sub.1]     0.03         -2.59        -0.11         -0.24
                  (1.02)       (-2.59)      (-2.50 ***)   (-2.60 ***)
[[beta].sub.2]     0.03          0.39         0.08          0.11
                  (5.05 ***)    -0.47        (5.39 ***)    (4.82 ***)
[[beta].sub.3]     0.04          2.27         0.01          0.04
                  (2.67 **)     (2.24 **)    (0.45)        (1.21)
[psi]              6.35         82.72        14.77          2.71
                  (2.14 **)     -1.74        (1.46)        (0.56)
N                  2,246         1,995        3,481         3,481
Adj. [R.sup.2]     0.29          0.27         0.09          0.12

*** and ** indicate statistical significance at 1 percent level and 5
percent level, respectively

Table 6--Contemporaneous Volatility and Volume--Expected versus
Unexpected Volume

The dependent variable is intraday volatility. The independent
variables are expected, unexpected parts of volume, and a Monday
dummy variable. Volume is decomposed into expected (E) and unexpected
(U) parts using ARIMA models. Schwarz Bayesian Information Criterion
is used to determine the orders of autoregressive and moving average
in the model, i.e., ARIMA(p,d,q). The types of trading volume are:
trading between floor traders and customers (Type 14), trading
between customers (Type 44), and trading between floor traders (Type
11). GMM is used in the estimation. Coefficients for volume and the
dummy variable are adjusted by a factor of 1,000. In parentheses are
t-statistics with autocorrelation and heteroskedasticity consistent
standard errors

                        Live          Pork
                       Cattle        Bellies

Panel A. ARIMA(p,d,q) Specifications

Volume 11              (2,0,1)       (2,0,1)
Volume 14              (3,0,2)       (5,0,5)
Volume 44              (3,0,3)       (2,0,2)

Panel B. Regression Results

Constant              -0.03          0.06
                     (-2.78 ***     (1.29)
E([volume.sub.11])    -0.11         -5.49
                     (-1.37)       (-1.88)
U([volume.sub.11])     0.03          0.19
                      (1.09)        (0.15)
E([volume.sub.14])     0.04          1.51
                      (4.19 ***)    (2.99 ***)
U([volume.sub.14])     0.04          0.94
                      (5.21 ***)    (2.29 **)
E([volume.sub.44])     0.10          1.27
                      (4.62 ***)    (2.34 **)
U([volume.sub.44])     0.03          3.11
                      (2.18 **)     (2.79 ***)
Monday                 6.11         80.55
                      (2.07 **)     (1.78)
N                      2,246         1,996
Adj. [R.sup.2]         0.29          0.36

                       Deutsch        Swiss
                        Mark          Franc

Panel A. ARIMA(p,d,q) Specifications

Volume 11              (5,0,5)       (5,0,5)
Volume 14              (5,0,5)       (5,0,5)
Volume 44              (1,0,5)       (5,0,5)

Panel B. Regression Results

Constant               0.002         0.021
                      (0.28)        (1.29)
E([volume.sub.11])    -0.12         -0.45
                     (-3.88 ***)   (-4.86 ***)
U([volume.sub.11])    -0.05         -0.09
                     (-0.84)        (0.09)
E([volume.sub.14])     0.04          0.10
                      (3.19 ***)    (5.01 ***)
U([volume.sub.14])     0.08          0.11
                      (5.01 ***)    (4.73 ***)
E([volume.sub.44])     0.02          0.02
                      (1.28)        (0.57)
U([volume.sub.44])     0.01          0.04
                      (0.56)        (1.08)
Monday                13.65          2.27
                      (1.35)        (0.41)
N                      3,481         3,481
Adj. [R.sup.2]         0.10          0.13

*** and ** indicate statistical significance at I percent level and 5
percent level, respectively

Table 7--Volatility, Expected Volume, and Unexpected Volume
--Robustness Check with Different Sessions During the Day

The dependent variable is intraday volatility and independent
variables are Monday dummy variable and expected and unexpected parts
of volume, which is decomposed into expected (E) and unexpected (U)
parts using ARIMA models. The types of trading volume are: trading
between floor traders and customers (Type 14), trading between
customers (Type 44), and trading between floor traders (Type 11). Each
trading day is divided into four sessions: opening session of the
first half hour of trading; morning session from open to 11:30 am;
closing for the last half hour; and afternoon session. Coefficients
for volume and the dummy variable are adjusted by a factor of 1,000.
In parentheses are t-statistics with autocorrelation and
heteroskedasticity consistent standard errors

Panel A: Opening and Morning Sessions

                           Live Cattle              Pork Bellies

                      Opening      Morning     Opening      Morning

Constant              -0.12       -0.01         0.18        0.02
                     (-0.84)     (-0.99)       (0.35)      (0.55)
E([Volume.sub.11])    -0.51       -0.10       -20.53        0.29
                     (-0.51)     (-1.41)      (-1.15)      (0.16)
U([Volume.sub.11])     0.23       -0.03        -0.44        -0.21
                      (2.10 **)  (-0.75)      (-0.13)     (-0.33)
E([Volume.sub.14])     0.07        0.03         3.46        0.27
                      (0.88)      (3.75 ***)   (1.25)      (0.79)
U([Volume.sub.14])     0.05        0.04         1.44        0.19
                      (2.10 **)   (4.28 ***)   (1.50)      (0.39)
E([Volume.sub.44])     0.02        0.06        -3.02        2.20
                      (1.57)      (3.44 ***)  (-1.14)      (2.84 ***)
U([Volume.sub.44])     0.06        0.02         4.27        2.57
                      (2.10 **)   (1.24)       (1.94 **)   (2.00 **)
Monday                27.49        4.52       405.01       62.15
                      (1.73)      (1.22)       (1.60)      (1.50)
N                       252         748          252         756
Adj. [R.sup.2]         0.22        0.32         0.26        0.31

Panel B: Afternoon and Closing Sessions

                          Live Cattle              Pork Bellies

                    Afternoon      Closing     Afternoon   Closing

Constant              -0.04        -0.02        0.42        -0.02
                     (-2.63 ***)  (-1.24)      (1.13)      (-0.27)
E([Volume.sub.11])     0.20         0.31       -1.45        -0.23
                      (1.19)       (2.22 **)  (-0.82)      (-0.08)
U([Volume.sub.11])     0.02         0.02        0.50         2.37
                      (1.89)       (0.21)      (1.03)       (1.98 **)
E([Volume.sub.14])     0.02        -0.02        0.24         0.67
                      (1.71)      (-0.89)      (1.34)       (1.58)
U([Volume.sub.14])     0.03        -0.02        0.53         0.36
                      (2.51 ***)  (-0.56)      (3.97 ***)   (1.66)
E([Volume.sub.44])     0.09         0.02        2.11         1.45
                      (3.29 ***)   (0.27)      (6.27 ***)   (3.06 ***)
U([Volume.sub.44])    -0.01         0.002       0.66         0.64
                     (-0.46)      (-0.04)      (2.91 ***)   (1.16)
Monday                 1.88         4.52       14.20       -43.34
                      (0.52)       (1.22)      (0.76)      (-1.78)
N                       978          252         730          252
Adj. [R.sup.2]         0.24         0.04        0.27         0.31

Panel A: Opening and Morning Sessions

                           Deutsch Mark             Swiss Franc

                     Opening      Morning      Opening      Morning

Constant              -0.10        -0.002       -0.05        -0.001
                     (-0.56)      (-0.71)      (-0.37)      (-0.49)
E([Volume.sub.11])     0.74        -0.09        -1.78        -0.05
                      (0.78)      (-5.05 ***)  (-0.89)      (-2.14 **)
U([Volume.sub.11])    -0.56        -0.06        -1.45        -0.01
                      (1.67)      (-4.63 ***)  (-2.55 **)    (-.48)
E([Volume.sub.14])    -0.20         0.03         0.48         0.02
                     (-1.00)       (4.97 ***)   (2.16 **)    (3.09 ***)
U([Volume.sub.14])     0.08         0.03         0.38         0.02
                      (1.57)       (5.57 ***)   (4.20 ***)   (3.83 ***)
E([Volume.sub.44])     0.93         0.02         0.12         0.03
                      (2.70 ***)   (2.10 **)    (0.20)       (1.21)
U([Volume.sub.44])     0.28        -0.01         0.15        -0.001
                      (1.88)       (1.59)       (0.82)      (-0.01)
Monday               165.30        -2.95        39.70        -4.59
                      (1.39)      (-1.91)       (0.54)      (-3.17 ***)
N                       252         1,764         252        1,784
Adj. [R.sup.2]         0.19         0.22         0.22         0.11

Panel B: Afternoon and Closing Sessions

                          Deutsch Mark             Swiss Franc

                     Afternoon   Closing   Opening     Closing

Constant             -0.01        -0.003   -0.004      -0.0003
                    (-1.46)      (-1.38)  (-1.61)      (-0.16)
E([Volume.sub.11])   -0.04         0.03     0.06        -0.06
                    (-0.77)       (0.84)   (0.67)      (-0.58)
U([Volume.sub.11])   -0.04         0.02     0.08        -0.02
                    (-1.24)       (0.59)   (1.04)      (-0.56)
E([Volume.sub.14])    0.04        -0.01     0.04         0.02
                     (2.03 **)   (-0.86)   (3.53 ***)   (0.76)
U([Volume.sub.14])    0.03        -0.03     0.03        -0.01
                     (4.37 ***)  (-0.65)   (3.04 ***)  (-0.97)
E([Volume.sub.44])    0.02         0.04    -0.01        -0.02
                     (1.12)       (1.74)  (-1.07)      (-0.48)
U([Volume.sub.44])   -0.01        -0.01    -0.03        -0.01
                    (-0.73)      (-0.21)  (-2.21 **)   (-1.69)
Monday               -2.05        -2.04    -0.18        -1.31
                    (-0.99)      (-1.46)  (-0.11)      (-1.02)
N                     1,213         252     1,213         252
Adj. [R.sup.2]        0.10         0.07      0.19        0.00

*** and ** indicate statistical significance at 1 percent level and 5
percent level, respectively