Traders’ preferences for electronic vs. floor trading

2. Traders’ preferences for electronic vs. floor trading

2.1. Trading systems Floor-based trading systems vary considerably in their design, the same is true of electronic screen-based trading systems. A discussion of traders’ preferences for one trading system versus another one requires a characterization of both systems. The characterization chosen here relates closely to the trading systems of the LIFFE and the DTB since data of these exchanges will be analyzed. 3 The floor system is a dealer-driven system with continuous trade through open outcry. Quotes are valid as long as Abreath is warmB. An official order book does not exist. Transaction prices are published immediately, volumes of transactions are published with short delays. Names of traders are not disseminated; this information is available on the floor. The electronic screen-based trading system is a continuous auction system with automatic order matching in which traders communicate only via computer screens without revealing their names. 4 If two orders can be matched, then orders with the best prices are matched. Information on transaction prices and volumes is published instantaneously in the electronic system. Although both trading systems appear to be very different, they share many features. Both systems are operating continuously. Execution risk is eliminated in both systems: the quotes from the floor tell the trader at which prices he can trade. Similarly, in the electronic system the trader knows the limit order book and, thus, the prices at which an order can be executed. In both systems, traders are dual capacity traders, i.e. they may trade on their own and on customers’ accounts. The following discussion is based on these trading systems. 2.2. The static impact of information Analyzing traders’ preferences for one or the other system, we assume that traders have access to both systems. Hence, the fixed costs are irrelevant for the short-term decision to trade in one or the other system. This decision depends also on the information available in each system. First, we discuss the information impact in a static, then in a dynamic framework. Suppose that arbitrage between the electronic and the floor system functions well. Hence, whether an order is executed at a better price in one or the other system, depends on bid–ask spreads and on price sensitivities to order volume. Usually, the bid–ask spread is split into three components, a storage cost 3 Ž . For a general discussion of different trading systems, see Pagano and Roell 1990, 1992 . ¨ 4 Trade is opened in many electronic exchanges through a batch auction. component, a premium for bearing price risk and an asymmetric information cost component. There is little reason to believe that the first two components differ among electronic and floor systems. Hence, differences in the bid–ask spread should be explained by differences in available information. Ž . Beneviste et al. 1992 argue that the bid–ask spread should be lower on the floor since observation of traders and sanctioning power of dealers allow them to distinguish information traders and liquidity traders. 5 Hence, the adverse selection problem should be weaker on the floor leading to lower bid–ask spreads and higher trading volume. 6 This argument needs to be qualified, however. Risk-averse traders can put very small orders into the electronic limit order book to protect themselves against adverse selection whereas on the floor quotes are valid for Ž . larger order sizes Glosten, 1994 . Hence, bid–ask spreads in electronic systems might be smaller since they relate to smaller order sizes. Large orders may face a higher spread in the electronic system. Also they are unlikely to be put into the electronic limit order book because of the free option problem. Thus, large orders 7 Ž . preferably go to the floor. But, as pointed out by Bernhardt and Hughson 1997 , traders are likely to benefit from splitting large orders between both markets to equate marginal costs. Therefore, large orders add to the liquidity of both markets although perhaps more to the floor. 8 Some empirical evidence concerning these issues is accumulating. The impact of competition among exchanges on the bid–ask spread has been demonstrated by Ž . McInish and Wood 1992 for the USA and for Europe by Pagano and Roell ¨ Ž . Ž . 1993 . Schmidt et al. 1993 show for Germany that regional floor exchanges with small trading volume compete through smaller bid–ask spreads against an interbank electronic trading system with much higher trading volume. de Jong et Ž . al. 1995 find, in contrast to theoretical predictions, that the effective bid–ask spread does not increase in trade size, neither in the Paris bourse nor in the SEAQ International. 5 Ž . Madhavan 1992 shows for a continuous dealer system and a continuous non-anonymous auction system that price competition between dealers eliminates the AwedgeB between the transaction price and the expected value of the asset whereas strategic behavior in auction markets distorts prices and thus induces inefficiency. 6 Higher trading volume, in turn, implies that prices are based on a larger set of information so that Ž . adverse selection is even more unlikely cf. Glosten and Milgrom, 1985; Stoll, 1989 . The inverse Ž . relationship between the bid–ask spread and trading volume is questioned by George et al. 1994 . They show that the impact of adverse selection on trading volume and the bid–ask spread depend on whether liquidity trading decreases in transaction costs at an increasing or a decreasing rate. 7 Ž . Similarly, Pagano and Roell 1992 argue that large investors get a better price in dealer than in ¨ auction markets. 8 Ž . This reasoning is also supported by Madhavan and Cheng 1997 who argue that the Adownstairs marketB provides significant liquidity to the Aupstairs marketB so that the execution costs for large trades are only slightly better in the Aupstairs marketB. Finally, there are some empirical studies about the Bund-Future trade at LIFFE Ž . and DTB, mostly based on short time intervals. Kofman et al. 1994 investigate data from 6 weeks. Using the Roll measure, they find that the DTB offers a tighter bid–ask spread. Correcting this measure for conditional expected returns as Ž . Ž . suggested by George et al. 1991 , they find the opposite result. Pirrong 1996 finds for the period July 1992 to June 1993 that spreads as given by the Roll measure were not higher at the DTB and sometimes lower than at the LIFFE. 2.3. The dynamic impact of information Now we analyze the impact of changes in information intensity on traders’ behaviour. The likelihood of adverse selection increases with information inten- sity. Therefore, in periods of high information intensity, non-informed traders preferably go to the system with faster information diffusion. Informed traders, however, prefer trading in the system with slower information diffusion to raise profits from insider trading. Since insiders represent only a small fraction of traders, their impact on trading volume is very likely to be smaller than that of non-informed traders. More precisely, in a period of intensive information arrival, non-informed traders are likely to concentrate their trading in the system with faster information diffusion. Therefore, the market share of this system should increase with information intensity. The market share of a trading system is defined as its trading volume per period, divided by the aggregate trading volume of both systems per period. Information has two components, the information about market transactions and Ž . behavior of traders endogenous information and the information about other Ž . events being relevant for pricing exogenous information . This information can be Ž . private or public Admati and Pfleiderer, 1988 . It is private information or the Ž . difference in opinion that drives trading Harris and Raviv, 1993 . Hence, it is likely to raise volume, frequency of transactions, and price volatility and, thereby, the intensity of endogenous information arrival. In periods of low information intensity, trade volume is low and transactions are infrequent; therefore information on the last trade is fairly old. Then the limit order book information of the electronic system is more updated and an important indicator of market developments. Also, in such a period, traders are relatively inactive so that observation of their behavior on the floor does not permit reliable predictions of their activities. Finally, in such a period there is not much to be gained from conversation among floor traders since new information to be evaluated is lacking. Hence, in a period of low information intensity, the order book information of the electronic system appears to offer more signals for predicting market developments than observation of traders on the floor. This picture changes significantly in periods of high information intensity. Then Ž . various effects reduce the importance of the electronic order book information: 1 Due to high price volatility, traders will reduce their limit orders in the electronic Ž . order book because the value of the free option increases with volatility. 2 Frequent trading with high trading volume 9 yields continuously brand-new infor- Ž . mation on prices and trading volume. 3 Observation of other traders on the floor becomes more informative. Also, discussing new information with other traders Ž . helps each trader to better understand and evaluate the new information. 4 In periods of high exogenous information intensity, the danger of adverse selection is high. This danger is stronger in an anonymous electronic system than in a floor system. All the preceding arguments support the hypothesis that floor trading gains attractiveness relative to electronic trading in periods of intensive information arrival. There are various ways to test this hypothesis. One way is to find out which Ž . system is leading price innovations. Shyy and Lee 1995 find for the period November 8–19, 1993 that the anonymous DTB-system is leading price innova- tions and that information asymmetries at the DTB are smaller. They argue that this may be explained by Germany being the Ahome marketB for German government bonds. Such a conclusion appears premature since we expect the relative speed of information diffusion at both exchanges to depend on the Ž . intensity of information arrival Franke and Hess, 1995 . This is confirmed by Ž . Martens 1997 . He investigates the information share of the DTB as defined by Ž . Hasbrouck 1995 . The information share of a market is the proportion of the efficient price innovation variance that can be attributed to that market. Martens finds for the period September 8 to December 20, 1995 that in times of high price volatility the information share of the DTB is lower than that of the LIFFE and vice versa in times of low price volatility. This finding is consistent with our conjecture that the speed of information diffusion in the electronic system relative to that in the floor system declines in terms of high information intensity. We test our conjecture by investigating the DTB’s market share across periods of varying information intensity. This leads to our first hypothesis. Hypothesis 1. The market share in trading volume of the floor vis-a-vis the ´ electronic system increases with the intensity of information arrival. Testing Hypothesis 1, we use volatility as a proxy for information. Given the positive relation between the intensity of exogenous information arrival and price volatility, Hypothesis 1 leads to Hypothesis 2. 9 High trading volume is likely to improve the reliability of information aggregated in transaction Ž . prices Blume et al., 1994 . Hypothesis 2. The market share of the electronic system declines when price volatility increases. Since high exogenous information intensity also raises trading volume, the market share of the electronic system should also be inversely related to the trading volume aggregated over both markets. 10 Hypothesis 3. The market share of the electronic system declines when aggregate trading volume increases. Similarly, an increase in endogenous information intensity renders the elec- tronic order book less informative. Hypothesis 4. The market share of the anonymous electronic trading system declines when trading frequency increases. Hypotheses 2 and 3 relate the market share to price volatility and aggregate trading volume. A high aggregate trading volume may be generated by high price volatility andror by a large average order size, independent of volatility. Since large orders allegedly also favor floor trading, we separate both effects in Hypothesis 5. Hypothesis 5. The market share of the electronic system declines when price volatility increases andror the average order size increases independently of price volatility.

3. Empirical results