Dynamics of Market Liquidity of Tunisian Stocks: An Analysis of Market Resiliency

Transcription

1 SPECIAL SECTION: FINANCIAL MARKET ENGINEERING Dynamics of Market Liquidity of Tunisian Stocks: An Analysis of Market Resiliency DORRA MEZZEZ HMAIED, ADEL GRAR AND OLFA BENOUDA SIOUD INTRODUCTION Advances in technology have led many emerging capital markets to implement market microstructure changes, in recent years. These changes concern trading rules and various institutional features of the market aiming at improving liquidity and controlling price volatility. The main microstructure change is the automation of the trading system (Israel in 1987, Singapore in 1989 and Morocco in 1998, etc.). 1 However, in spite of the rapid adoption of electronic limit order markets all over the world, many questions concerning the nature and the characteristics of liquidity in automated systems remain unanswered, especially for emerging markets. The extant microstructure literature has essentially focused on Asian stock markets and has paid little attention to MENA stock markets. The issue of market liquidity has long been under focus, given its great importance to academics and market participants. A liquid market is one in which participants can trade desired amounts quickly, cheaply and without greatly affecting prices. Kyle (1985) proposed a three-component classification of liquidity, covering tightness, depth and resiliency. Tightness is measured by the bid-ask spread which expresses the transaction cost for those who wish to execute a marginal trade in the market. Depth corresponds to the volume that can be traded at current price level and resiliency represents the period of time required to reach a new equilibrium after trade executions. Unfortunately, extant empirical studies of liquidity fail to fully explore Kyle s notions. The rich theoretical literature developed on bid-ask spreads and the variety of spread measures has incited researchers to focus especially on tightness. Liquidity is a major determinant of market viability and depends on the ability of the trading mechanism to match the trading desires of sellers and buyers. It arises from rules and market practices governing the trading process. The market microstructure literature typically distinguishes dealer markets and order-driven markets. Market makers are the only providers of liquidity in dealer markets and have the obligation to fill gaps arising from imperfect synchronization between the arrivals of buyers and sellers. In fact, they are the counterparty in all transactions at the quoted prices: the bid price, at which they are willing to buy securities and the ask price, at which they will sell. However, in an order driven market, A b s t r a c t In spite of the rapid adoption of electronic limit order markets all over the world, many questions concerning the nature and characteristics of liquidity in automated systems remain unanswered. This paper examines the dynamic behaviour of market liquidity on the Tunisian stock exchange (BVMT) using high frequency data from a reconstructed limit order book. The BVMT is an electronic pure order driven market that relies only on limit orders to supply liquidity, which may affect its viability and its resiliency. First, we apply a VAR model to stocks traded in continuous in order to examine if dynamic interactions exist between liquidity and volatility. Second, we study the resiliency of the BVMT through the impulse response function of the VAR model. Our findings show dynamic relationships between spread, depth and volatility. Some differences exist in the dynamics of liquidity when we take into account the trading intensity of the stock. Furthermore, we note that shocks are absorbed more quickly for frequently traded stocks than for infrequently traded ones. Keywords: resiliency, market liquidity, VAR model, emerging market A u t h o r s Dorra Mezzez Hmaied (dorra_hm@yahoo.fr) is an Assistant Professor of Finance at HEC Carthage. She is a graduate of the HEC Carthage and holds a PHD from Paris-Dauphine University. She is Head of the Finance Department HEC, Carthage. She is a researcher at the research centre ERECA. Adel Grar (dg@ameninvest.com.tn) is the manager of a brokerage house. He is a graduate of the HEC Carthage and holds a PHD from Paris-Dauphine University. Olfa Benouda Sioud (olfa.benouda@ihec.rnu.tn) is a Professor of Finance at HEC Carthage. She is a graduate of the HEC Carthage and holds a PHD from Paris-Dauphine University. She received the Tunisian Presidential Award offered by the President of the Republic of Tunisia for the best PhD in July She has been Head of Study and Training HEC, Carthage. She is a researcher at the research centre ERECA. Copyright ß 2006 Electronic Markets Volume 16 (2): DOI: /

2 Electronic Markets Vol. 16 No investors trade with the intervention of a broker acting as an agency trader only. In such markets, investors can place either market orders or limit orders. Market orders demand immediacy, requiring execution as soon as practicable, at the best available price. Limit orders allow a trader to set a limit price at which the order might fill, but there is a risk the order does not execute. An order driven market then relies only on limit orders to supply liquidity. This paper explores dynamic aspects of liquidity on an emerging stock market of the MENA region, the Tunisian stock exchange (BVMT). To the best of our knowledge, this study is the first in its kind on the BVMT, so certainly little is known about its microstructure. The BVMT is a pure order driven market without market makers to supply liquidity. Liquidity will vary then over time depending on the submission of limit orders and may even be absent at certain times. In 1996, the BVMT undertook a modernization reform replacing the manual trading system by a fully automated system. The new market microstructure offers advantages of speed and lower operating costs and has as a main objective the stimulation of market activity by attracting higher domestic and international order flows. This new trading system is the Super CAC, developed by Euronext and used in the Paris Bourse. The question thus arises as to whether the adoption of a sophisticated electronic trading system used in developed markets is a sufficient condition to improve market characteristics, especially market liquidity? Sioud and Hmaied (2003) document an important increase of the trading volume after the automation of the trading system on the Tunisian stock exchange. But, the dynamic aspects of liquidity were not studied because a period of adaptation is necessary before evaluating the benefits of this microstructure modernization. This paper analyses the dynamic behaviour of liquidity on the BVMT and its mechanism s resiliency. The evolution of liquidity is examined through information from a reconstructed limit order book by using a vector autoregressive model to explore causal and feedback effects among spread, depth and midquote return volatility. Using high frequency data, the model is applied to 22 stocks traded in continuous from September 2003 to November The stocks are divided into two groups according to the intensity with which they are traded. Impulse-response functions are used to explore the reaction of the system to a perturbation of its long-run equilibrium. We extend existing research by considering depth at each side of the limit order book and we examine whether resiliency depends on the trading intensity of the stock. Market liquidity dynamics is an important element for investors decision making, developing efficient trading strategies and managing portfolio risks. Moreover, this analysis enlightens market authorities on the ability and the rapidity of the market to absorb shocks, which can help them to define corrective institutional measures. The paper is organized as follows. The next section focuses on market microstructure research relevant for the analysis. We then present the BVMT and the dataset. The methodology used is then discussed. Results are then presented and discussed, and the paper ends with a concluding section. DYNAMIC LIQUIDITY BEHAVIOUR IN LIMIT ORDER MARKETS Extensive empirical work on market liquidity has focused on static indicators of liquidity, namely the bid-ask spread and depth. Battacharya and Spiegel (1998) highlight the dangers of using unidimensional static proxies for liquidity like the bid-ask spread of a stock. Muranaga and Shimizu (1999) suggest that market impact (price changes upon trade execution) and market resiliency should be considered as dynamic measures of liquidity to examine how market liquidity affects the price setting process. Theoretical and empirical literature on market microstructure has paid little attention to the issue of market impact and market resiliency. The main theoretical hypothesis underlying market impact suggests that information is conveyed by trades and the response of security prices to trading activity is a consequence of asymmetric information. Information-based microstructure models including Admati and Pfleiderer (1988), Easley and O Hara (1987), Foster and Viswanathan (1990), Glosten and Milgrom (1985) and Kyle (1985), predict that greater information asymmetry between informed and uninformed traders leads to wider spreads and lower depths and then affects liquidity, 2 as uninformed liquidity traders attempt to minimize losses from trading with informed traders. These theoretical models yield two important empirical predictions: the asymmetry is positively related to the bid-ask spread and to the price impact of a trade. Hasbrouck (1991a, 1991b) introduces the VAR analysis to model the price impact of trades by examining the interactions of security trades and quote revisions. He stipulates that in a security market with asymmetrically informed participants, trades are signals of private information and, therefore, cause temporary and persistent impacts on the security price. The permanent price impact is due to agent s beliefs about the private information content of the trade. Hasbrouck (1991a, 1991b) reports that the price impact of a trade is larger for wider bid-ask spreads and more significant for firms with smaller market capitalization. The recent availability of high frequency data has enhanced studying the dynamic behaviour of liquidity. Muranaga (2000) studies the dynamics of the Tokyo

3 142 Dorra Mezzez Hmaied, Adel Grar and Olfa Benouda Sioud & Market Resiliency of the Tunisian Stock Market stock exchange liquidity and finds a positive correlation between trade frequency and each of the following liquidity indicators: the bid-ask spread, market impact and market resiliency. Considering one week transaction data from an electronic FX broking system, Danielsson and Payne (2002) observe a clear predictable pattern in the evolution of liquidity and find that the liquidity supply process is self-regulated in this order-driven market. Furthermore, they show that high trading activity and volatility tend to inhibit the liquidity supply process by increasing spreads and decreasing depth. Coppejans et al. (2003) analyse the stochastic dynamic of liquidity and its relation to returns and volatility on an automated futures market. They find a significant variation in liquidity over time unrelated to calendar time effects. They show that this dynamic of liquidity affects trading strategy, in that, a concentration of volume is observed when liquidity is high. Empirical studies use two approaches to examine market resiliency either by observing market activity after a real event or by observing the way the system responds to a simulated shock. Battacharya and Spiegel (1998) study the resiliency of the NYSE by investigating trade suspensions in a cross-sectional analysis. They show that large cap stocks have lower bid-ask spreads but halt more often. Degryse et al. (2002) analyse the resiliency of the Paris Bourse after the submission of an aggressive order by observing the evolution of the bid-ask spread, the depth at the best quotes and the duration between best quote updates. They find that the Paris Bourse recovers quickly after an aggressive order since the variables return to their pre-order levels within few quotes updates. Large (2005) develops a dynamic model of limit order book activity which takes the form of multivariate point process in continuous time. The model is based on ten types of event related to the placement of orders with different degrees of aggressiveness and is estimated for the Barclays stock listed on the London Stock Exchange (LSE). The results show that the resilient replenishment of the limit order book is fast but it occurs quite infrequently. The second approach of studying resiliency is based on the impulse response analysis. 3 Spierdijk et al. (2004) examine the temporary and persistent price impact of trades for infrequently traded stocks listed on the NYSE, by applying a VAR model to returns, bid-ask spreads and trade sign. They find that prices overshoot before they revert to the full information price and that the degree of overshooting depends upon the bid-ask spread. In addition, both the temporary and persistent price impacts depend on the trading intensity. Coppejans et al. (2003) provide evidence about the high degree of resiliency of the Swedish stock index futures market since the reduction of liquidity after a volatility shock dissipates quickly (within the hour following the shock). They attribute this to the self-correcting ability of automated auctions. Manganelli (2002) jointly models trading volume, volatility and duration as autoregressive conditional models for ten stocks from NYSE. She finds that the dynamics of frequently traded stocks differ significantly from those of the infrequently traded ones. The results also show that the more the stock is traded, the faster the market returns to its full information equilibrium. DESCRIPTION OF THE MARKET AND THE DATASET Structure of the Tunis Stock Exchange The Tunis Stock Exchange, BVMT, is a private company owned by its member brokers. The trading system on the BVMT was fully automated in Orders are submitted by brokers on the behalf of investors and executed through an automated trading system, using a computerized limit-order book, known as SUPERCAC. Trading is carried out from 8.30 a.m. to a.m. from Monday to Friday. It starts by a pre-opening session (from 9:00 am to 10:00 am) during which investors can place, modify or cancel orders but no trades are permitted. A theoretical opening price is displayed in real time to show the market tendency. The BVMT applies two trading methods: call auction and continuous trading. The market opens by a call auction for all stocks at some point of time during the first fiveminute opening period. For liquid stocks, this is followed by a continuous market until 11:30 a.m. However, for illiquid stocks, a second call is set at 10:15 a.m. for securities not traded at the open call and a last call takes place at 11:00 a.m. 4 The BVMT is a pure order-driven market where investors can choose between market and limit orders, so as liquidity is only provided by limit order traders. Market orders have no limit on prices and look for immediate execution while limit orders specify a price either above the current ask or below the current bid and offer price improvement relative to market orders. A market order is matched with the best opposite quote of the order book. Limit orders are held in the limit order book until they are matched with incoming market orders to produce trades; otherwise, they are cancelled or modified. A limit order faces the risk of nonexecution whereas a market order executes with certainty. At the end of each month, all orders are purged from the limit order book. During the continuous session, limit orders are executed according to order price and time of submission priority basis. First, orders are matched in the order in which they are entered into the system, based on the best price: the higher bid for the buy side and the lower ask for the sell side. Moreover, the system applies a time priority: at the same price, the order first entered into the system must be fully executed before an order placed later.

4 Electronic Markets Vol. 16 No Investors can submit orders at any price taking into account the tick size representing the permissible price increments, at which the stock may be quoted. The tick size ranges from 0,01 Dinars to 0,1 Dinars depending on the price level of the security. Furthermore, during trading, orders for a stock must be priced within 3% of the day s reference price (previous day s closing price). Otherwise, a trading halt occurs for this stock for 30 minutes after which new limits ( 1.5 per cent) are applied. The trading system used on the BVMT offers good transparency, in that the five best quotes on the buy and sell sides of the book with the corresponding cumulated volumes are made available to market participants in real time through computer screens. However, the limit order book is fully visible to brokers and regulatory authorities. Data The data employed in this study are available for the first time. 5 They are obtained from a brokerage house that chose the real time feed to collect historical information on best quotes and trades. The data give information on quotes and trades on two separate datasets. The trade dataset includes transaction prices and volume with a time stamp recorded to the nearest second. The quotes dataset contains the five best bid and ask prices updated with the corresponding order depth, and the number of orders in each quote. Different filters are applied to the data in order to ensure accuracy of the calculated variables: the first filter eliminates the pre-opening session and the call auction as we are interested in the continuous session. The second filter includes relationships such as ask price being greater than the bid price and eliminates observations without quotes on one or two sides of the limit order book. We consider the 22 stocks traded in continuous on the BVMT from September 2003 to November Using information from the order flow and transactions datasets, the limit order book for each stock is reconstructed, at the end of each 15 minutes of the continuous session. METHODOLOGY We analyse, first, the dynamic interaction between liquidity and volatility for each stock of the sample using a standard vector autoregressive model. In a second step, we construct impulse response functions that show how quickly the system returns to its long run equilibrium after a shock. This can be helpful to understand market resiliency. The stocks are divided into two groups according to their trading intensity. 6 A first group includes ten stocks for which the trading intensity is very high (frequently traded stocks) and a second group is composed of 12 stocks presenting a low trading intensity (infrequently traded stocks). Variables We use jointly two measures of liquidity in the VAR model: market depth on each side of the book and bidask spreads. Market depth is measured by the logarithm of the cumulated volume in Tunisian Dinars at the five best quotes, for the bid side, D b and the ask side, D a. D q, t ~ X5 i~1 P qi, t Q qi,t with q: a (ask) or b (bid); P q,t : the limit price at one of the five best quotes on the ask or bid side q at t; and Q q,t : the number of shares at one of the five best quotes in the bid or the sell side at t. The relative spread, S t, is measured by the difference between the prevailing best ask and bid prices to the mid-quote at t. S t ~ ask t{bid t ðask t zbid t Þ=2 The absolute value of the relative change in the quote midpoint, V t, is used as proxy for volatility at t. V t ~ m t{m t{1 m t{1 with m t and m t-1 the quote midpoints at t and (t-1), respectively. This volatility measure is standard in market microstructure studies. Taking the average of the bid and ask quotes limits the bid-ask bounce problems. 7 The variables considered are measured at the end of each 15-min interval from the reconstructed limit order book. Model specification The relationships between the different measures of liquidity and volatility are predicted on the basis of the theoretical literature. Foucault (1999) states that in a volatile market, the probability of mispricing an asset is higher and limit order traders quote relatively wide bidask spreads. Handa and Schwartz (1996) show that large spreads increase short-run volatility, thus increasing the incentive to provide liquidity by limit orders as the gains from supplying liquidity exceed the potential loss from trading with informed traders. Accordingly, we predict an increase in depth after a period of high volatility.

5 144 Dorra Mezzez Hmaied, Adel Grar and Olfa Benouda Sioud & Market Resiliency of the Tunisian Stock Market Finally, Depth is a function of demand and offer. Therefore, buy and sell depth will affect subsequent volatility and spreads. Given these relationships, we expect bi-directional causalities and we adopt a four-equation vector autoregression in order to explore the dynamic relationship between spread, bid depth, ask depth and volatility. The VAR model is a system of equations with more than one dependent variable. It has some advantages relative to the single equation model since it allows dynamic interactions among variables and allows for impulse response analysis. 8 The vector auto-regressive model is specified for Y t 5(D at, D bt, S t, V t ) with lag s as follows: 9 Y t ~A 0 z XQ s~1 A s Y t{s zv t We, therefore, specify the different vectors as follows: Y t-s is the vector of the lagged variables at (t-s), A 0 is the vector of the constant terms, A s is the vector of lagged variables coefficients and v t the vector of mean-zero disturbances jointly and serially uncorrelated D a, t{s ada 0 D b, t{s a 0 Db Y t{s ~, A 0 ~, S t{s C B as 0 C A 0 A s ~ V t{s a Da Da, s a Da Db, s a 0 V a Da S, s a Db S, s a S S, s a V S, s a Da V, s 1 ada, Db s ada, S s ada, V 0 s adb, s adb, S s adb, V s, v t ~ C B av Db, s av S, s av V, s v Da,t v Db,t The dynamic responses of spread, depth and volatility to shocks of a particular intensity on each one of them are examined using impulse response functions. Shocks consist in an increase of the variable of one standard deviation. An impulse response is the change of Y over time at period t+p (p51, 2, 3, ) when there is shock in one of the variables. The impulse response function gives an idea on the speed with which the variables under study tend to converge to equilibrium values after a shock. It is computed on the basis of the estimated version of the model specified above. Since we consider four dependent variables, we have four different sets of impulse response functions. RESULTS We attempt to explain the determination of liquidity variables, i.e. bid-ask spreads and depth at each side of the limit order book in terms of midquote return v S,t v V,t 1 C A volatility. Tables 1a and 1b report results from the estimation of the VAR model for the frequently traded stocks and the infrequently traded ones, respectively. Overall, the results show significant interactions between liquidity measures and volatility. The Fisher statistic is significant for all equations and all stocks. We note that all variables are positively autocorrelated. Spread and depth appear to be very persistent processes for the infrequently traded stocks. The persistence is lower for the frequently traded stocks. If we look at the spread equations, both lagged volatility and lagged depth on the bid side affect spread for the two groups. As volatility rises, the subsequent bid-ask spread increases. The coefficient of the lagged volatility is positive for 9 out of 10 frequently traded stocks (significant only in 5 cases); and for 8 out of 12 infrequently traded stocks (5 are significant). This result is consistent with the theoretical predictions and the findings of Danielsson and Payne (2002) who find that increased volatility leads to significantly larger spreads for an electronic FX broking system. For the two groups, a negative correlation is noted between the depth on the bid side and subsequent spreads. However increased depth on the ask side tends to decrease the subsequent spread only for the frequently traded stocks. In fact, lower spreads can result from hitting the quotes in order to maximize the execution probability of submitted orders. The volatility equations show that the spread and the depth on the bid side positively affect the expected volatility. Little differences exist between the frequently and the infrequently traded stocks. In fact, large bid-ask spreads lead eager traders to place aggressive orders inside the spread inducing midquote variations. However, an increase in the sell depth leads to lower subsequent volatility. This result holds more for the infrequently traded stocks. The relationship between lagged volatility and buy depth is different for the two groups. The volatility coefficients for the infrequently traded stocks are almost not significant. However, volatility leads to increased subsequent buy depth for all frequently traded stocks and the relationship is significant for 7 out of 10 stocks. The incentive to provide liquidity by buy limit orders increases after an increase in transitory volatility (Handa and Schwartz, 1996). In an earlier paper, Hmaied et al. (2004) show that short-term volatility affects only buyers on the Tunisian stock market, in that buy limit orders are likely to be placed when volatility increases. A positive correlation is noted between volatility and sell depth for the two groups, but weakly significant. Furthermore, there is no significant relationship between spreads and subsequent depth whereas Danielsson and Payne (2002) find that increased spreads and volatility are associated with significantly lower subsequent depth, both buy and sell depth.

6 Table 1a. VAR estimation for frequently traded stocks SPDIT ELECTROSTAR SFBT SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * * * * ** * ** (76.128) (4.589) (21.711) (20.315) (47.744) (1.172) (20.209) (0.022) (70.826) (22.306) (23.438) (1.993) VOLAT(21) * ** * * ** * * (0.458) (6.413) (2.551) (0.673) (3.844) (10.068) (1.740) (1.452) (2.701) (10.760) (2.645) (1.289) DEPTH B (21) ** * * ** * ** ** 23.51E * ** (22.291) (3.352) (94.599) (21.248) (22.547) (20.775) (86.761) (2.170) (22.664) (20.340) (75.126) (1.957) DEPTH A (21) ** * * 0.917* E ** * (21.738) (22.717) (1.465) (98.929) (21.247) (0.718) (3.605) ( ) (1.491) (20.803) (2.155) (90.168) Constant (3.702) (0.984) (6.432) (11.422) (5.530) (1.553) (8.366) (6.546) (1.878) (1.931) (11.468) (9.532) R F statistic SOTETEL TUNISAIR SOTRAPIL SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * ** * * * * ** (45.583) (2.832) (1.607) (20.477) (32.167) (3.682) (20.849) (22.967) (64.923) (21.767) (21.536) (2.055) VOLAT(21) * * * * * * * ** (23.646) (10.819) (3.116) (3.427) (0.689) (12.899) (0.818) (1.239) (4.128) (12.324) (0.949) (2.833) DEPTH B (21) 23.78E * ** * * 23.62E E * (20.211) (21.282) (65.218) (20.507) (0.702) (2.318) (73.011) (3.382) (20.187) (20.133) (94.038) (0.803) DEPTH A (21) ** 21.51E * * 21.51E * * ** * (21.839) (20.098) (1.448) (78.323) (23.236) (20.091) (6.311) (76.494) (2.727) (21.524) (1.071) (103.93) Constant (2.898) (1.476) (10.788) (12.553) (4.453) (20.711) (8.625) (10.709) (20.652) (2.088) (9.841) (8.165) R F statistic Electronic Markets Vol. 16 No 2 145

7 Table 1a. (Continued) STIP SIAME SIPHA SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * * * * * (72.124) (1.205) (20.037) (0.711) (60.048) (3.752) (20.173) (20.400) (44.404) (3.225) (1.393) (0.932) VOLAT(21) * ** * * * * ** (1.439) (9.994) (2.519) (0.941) (0.887) (7.194) (3.187) (1.248) (7.706) (9.712) (3.257) (20.495) DEPTH B (21) ** * * * * * * (21.869) (0.763) ( ) (0.938) (21.690) (3.050) (97.393) (0.501) (21.199) (4.063) (93.233) (5.103) DEPTH A (21) * * * * * * * (20.382) (21.537) (0.784) ( ) (23.675) (23.934) (1.290) ( ) (23.577) (21.457) (4.492) (96.138) Constant (3.172) (1.894) (6.270) (7.408) (5.604) (2.134) (5.890) (9.365) (5.737) (21.486) (8.238) (5.342) R F statistic STEQ SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * ** (45.272) (1.941) (21.592) (21.063) VOLAT(21) * * ** (4.067) (11.050) (2.124) (1.034) DEPTH B (21) ** * * (22.722) (1.739) (90.113) (3.839) DEPTH A (21) * * ** * (23.260) (23.015) (2.598) (83.616) Constant (6.923) (2.626) (6.450) (11.214) R F statistic Notes: Market depth is measured by the logarithm of the volume in Tunisian Dinars at the five best quotes for bid and ask sides. The relative spread is measured by the difference between the prevailing best ask and bid prices to the mid-quote at t. The absolute value of the change in the quote midpoint is used as a proxy for volatility at t. t statistics in parentheses. * and ** represent 1 and 5% significance levels, respectively. 146 Dorra Mezzez Hmaied, Adel Grar and Olfa Benouda Sioud & Market Resiliency of the Tunisian Stock Market

8 Table 1b. VAR estimation for infrequently traded stocks BANQUE DU SUD BTEI MONOPRIX SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * * ** * (76.769) (20.130) (20.119) (1.155) (89.871) (0.211) (2.589) (0.451) (100.94) (20.430) (0.379) (21.131) VOLAT(21) * * * * * * (24.223) (9.986) (1.576) (1.096) (3.0351) (4.476) (0.534) (1.029) (0.791) (9.051) (4.102) (20.446) DEPTH B (21) * 28.10E * ** * * (25.636) (20.991) ( ) (1.198) (21.835) (1.082) (172.04) (20.125) (20.451) (0.532) (140.25) (1.415) DEPTH A (21) ** * * * ** * * (21.663) (21.876) (1.136) ( ) (20.497) (24.738) (20.789) (126.63) (22.702) (24.407) (1.379) (129.81) Constant (6.108) (3.169) (2.774) (5.327) (2.241) (4.154) (4.165) (8.974) (2.874) (3.092) (5.127) (3.862) R F statistic UIB UBCI TUNISIE LEASING SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * ** * * * ** * * ** (73.719) (2.438) (3.372) (1.283) (75.102) (5.125) (20.488) (2.295) (54.638) (4.652) (1.059) (2.797) VOLAT(21) * ** * ** * (0.464) (6.054) (1.347) (1.871) (3.049) (0.206) (0.565) (20.732) (22.910) (6.005) (21.231) (1.204) DEPTH B (21) * * * * (0.129) (4.036) (153.28) (20.555) (20.124) (20.069) ( ) (21.746) (1.114) (1.445) (99.310) (21.578) DEPTH A (21) * * * ** * (1.104) (23.076) (0.523) (104.75) (21.471) (21.414) (20.337) ( ) (1.340) (21.026) (22.532) (90.798) Constant (0.754) (1.642) (2.806) (10.262) (1.647) (1.003) (7.147) (4.905) (20.176) (20.099) (8.365) (6.733) R F statistic Electronic Markets Vol. 16 No 2 147

9 Table 1b. (Continued) BNA ATL BANQUE DE L HABITAT SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * ** * * ** (107.93) (20.203) (0.508) (2.483) (102.55) (0.802) (1.304) (0.951) (99.284) (1.661) (20.690) (2.178) VOLAT(21) * ** * * * ** * ** (1.764) (5.856) (1.808) (20.336) (6.592) (4.978) ( ) (23.672) (2.569) (9.307) (1.471) (2.658) DEPTH B (21) ** ** * E ** * ** ** * (21.951) (21.979) (169.64) (20.263) (20.267) (2.036) (171.40) (21.129) (22.072) (2.299) (121.31) (20.021) DEPTH A (21) ** * 3.42E * 7.80E ** * (1.992) (1.067) (20.159) (171.72) (0.226) (1.0126) (21.468) ( ) (0.578) (22.145) (21.661) (158.41) Constant (0.672) (0.994) (4.120) (6.151) (1.133) (21.131) (6.494) (6.825) (1.857) (0.713) (7.781) (5.783) R F statistic STB BIAT BANQUE DE TUNISIE SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD VOLAT DEPTH B DEPTH A SPREAD(21) * ** * * * ** (70.935) (0.316) (1.950) (0.761) (74.069) (4.737) (21.365) (20.799) (97.863) (1.915) (0.966) (1.642) VOLAT(21) ** * ** * ** ** * * (2.443) (7.529) (0.485) (1.003) (22.520) (4.783) (3.038) (22.533) (20.609) (5.568) (0.637) (3.758) DEPTH B (21) * ** ** * ** * (21.064) (126.49) (0.308) (21.892) (22.902) (20.647) (127.68) (22.297) (20.356) (0.257) (115.55) (20.659) DEPTH A (21) ** ** * ** * ** * * * (0.375) (21.870) (21.984) (174.75) (22.262) (25.484) (22.281) (81.220) (1.416) (3.491) (20.419) (154.77) Constant (2.268) (1.916) (8.561) (5.832) (4.396) (13.426) (7.301) (13.426) (0.803) (21.755) (7.308) (5.885) R F statistic Notes: Market depth is measured by the logarithm of the volume in Tunisian Dinars at the five best quotes for bid and ask sides. The relative spread is measured by the difference between the prevailing best ask and bid prices to the mid-quote at t. The absolute value of the change in the quote midpoint is used as a proxy for volatility at t. t statistics in parentheses. * and ** represent 1 and 5 per cent significance levels, respectively. 148 Dorra Mezzez Hmaied, Adel Grar and Olfa Benouda Sioud & Market Resiliency of the Tunisian Stock Market

10 Electronic Markets Vol. 16 No Finally, we find strong evidence that the quantities available at and around the best bid and ask are positively correlated for frequently traded stocks. Thus, providing liquidity on one side of the limit order book encourages providing it on the other side. This result does not hold for the less frequently traded stocks. To some extent, our findings show that the dynamics of the frequently traded stocks differ from those of the infrequently traded stocks. Furthermore, sellers and buyers do not behave similarly. The dynamic of liquidity on a limit order market is then a consequence of traders order placement strategies which are affected by liquidity needs and information arrivals. The dynamic responses of shocks to spread, volatility and depth on each side of the order book are given by the impulse response functions of the VAR model. For the two groups, the effect of a shock on spread is small in magnitude and dissipates relatively quickly compared to buy and sell depth responses. A shock on depth declines slowly as waiting limit orders are submitted. Especially for infrequently traded stocks, the order book is usually small so the book takes more time to refill and depth on the five best quotes takes longer time to return to its prior level. No differences are noted between the patterns of buy depth and sell depth after a shock of spread or volatility. This result is consistent with the findings of Degryse et al. (2002) on the Paris Bourse. These authors note that the evolution of the depth at the best ask does not differ much from the evolution at the best bid. Large (2005) find also little evidence that the bid and the ask sides of the limit order book react differently for Barclays stock on the LSE. When we look at the impulse response functions of the frequently and the infrequently traded stocks separately, we observe sharp differences in the time required to return to equilibrium between the two groups. In Table 2, we report some summary statistics on the time required for each variable to reach the equilibrium value. For example, a shock to volatility on spread takes a minimum of 8 (10) intervals and a maximum of 25 (45) intervals to dissipate for frequently traded stocks (infrequently traded stocks). Depth takes a minimum of 20 (25) intervals and a maximum of 50 (80) intervals to return to equilibrium for frequently traded stocks (infrequently traded stocks). The adjustment to the full information price can then take several days for the infrequently traded stocks on the BVMT. A similar result is noted by Spierdijk et al. (2004) for infrequently traded stocks listed on the NYSE. Our results show that the minimum time required on the BVMT to reach a new equilibrium after a perturbation is 2 hours (8 intervals) whereas Degryse et al. (2002) note that on the Paris Bourse, the bid-ask spread and the depth return to their pre-order levels within a few quote updates (a maximum of 30 minutes) after the submission of an aggressive order. Large (2005) find that limit order book replenishment follows a shock less than 40 per cent of the time and with a half life of under 20 seconds. Table 2. Summary statistics on time required to return to equilibrium Panel A: Frequently traded stocks (10 stocks) Min Max Median S. Error A shock to volatility on Spread Depth ask Depth bid A shock to spread on Depth ask Depth bid Volatility A shock to buy depth on Depth ask Spread Volatility A shock to sell depth on Spread Depth bid Volatility Panel B: Infrequently traded stocks (12 stocks) Min Max Median S. Error A shock to volatility on Spread Depth ask Depth bid A shock to spread on Depth ask Depth bid Volatility A shock to depth bid on Depth ask Spread Volatility A shock to sell depth on Spread Depth bid Volatility The period of time required to return to equilibrium is given in intervals of 15 minutes. Shocks consist in an increase of the variable of one standard deviation.

11 150 Dorra Mezzez Hmaied, Adel Grar and Olfa Benouda Sioud & Market Resiliency of the Tunisian Stock Market Figures 1 and 2 report impulse response functions for the most frequently traded stock on the BVMT, TUNISAIR and the less frequently traded one, BNA, 10 respectively. Panels A, B, C and D present responses to a shock to spread, volatility, buy depth and sell depth, respectively. Let s look at the effects of a shock to volatility on liquidity (spread and depth). A shock to volatility on spread is fully absorbed after 8 intervals for TUNISAIR which corresponds on average to a period of 2 hours. The effect is strong only in the first 2 intervals (30 minutes). Depth on the bid and ask sides increase after a positive shock of volatility during the following 2 intervals and takes 5 hours (20 intervals) to dissipate completely. For the less frequently traded stock, BNA, however, a volatility shock on spread is absorbed within 40 intervals (10 hours). The effect is largest over the first 4 intervals following the shock (60 minutes). The response of the buy and sell depth to a volatility shock takes about 80 intervals. The impulse response analysis shows that shocks are absorbed more quickly for frequently traded stocks than for infrequently traded ones on the BVMT. In fact, the more frequently traded the stock, the less time the shock takes to dissipate because more trading causes more information to be revealed earlier. However, the magnitude of the shock does not depend on the trading intensity of the stock. Our results are consistent with the findings of Manganelli (2002). Considering stocks listed on the NYSE divided into two groups according to the intensity with which they are traded, she documents that the more frequently traded the stock, the faster the market returns to its full information equilibrium after a perturbation. CONCLUSION The rapid adoption of electronic limit order book systems denote that they are gaining importance as trading mechanisms of financial markets. This paper addresses the issue of the relationship between microstructure design of an emerging limit order market (the BVMT) and its dynamic liquidity behaviour. The BVMT has modernized its trading system since 1996 and operates as a pure order-driven market. The exchange is totally based on a computerized and centralized limit order book. The new trading system is the Super CAC, developed by Euronext and used in the Paris Bourse. This study investigates the dynamic interaction between market liquidity and volatility using a vector autoregressive model that incorporates causal and feedback effects among four variables: spreads, bid depth, ask depth and volatility. This model uses high frequency data from a reconstructed limit order book for each of the 22 firms traded in continuous on the BVMT. In order to examine the reaction of the system to a shock and to estimate the period of time required to reach a new equilibrium, we construct impulse-response functions for each stock of our sample. Interestingly, dynamic relationships are documented between spreads, depth and volatility, consistent with major theoretical literature and empirical studies. Our results show also that buyers and sellers seem to behave differently. The explanation commonly suggested is that buyers are likely to be more information motivated than sellers. Little differences exist in the liquidity dynamics of frequently and infrequently traded stocks. However, the impulse response analysis shows that shocks are absorbed more quickly for frequently traded stocks than for infrequently traded ones. The less frequently traded the stock, the more time it takes to attain the equilibrium. In spite of the adoption of a sophisticated electronic trading system, the Tunisian stock market is weakly resilient compared to developed stock markets. On the BVMT, the time required to reach a new equilibrium after a shock is at least 2 hours while it is a maximum of 30 minutes on the Paris Bourse (Degryse et al. 2002). Degryse et al. (2002) and Coppejans et al. (2003) suggest that automated limit order markets are relatively resilient. Their self-correcting ability is an attractive feature and important element of their success. But Tunisian experience shows that the implementation of an electronic trading system trickled down from a developed stock exchange is not sufficient to reach international stock markets standards. Other aspects of market design should be then examined to enhance market activity and to resolve illiquidity problems such as the adoption of a system relying on limit orders and market makers to supply liquidity. Finally, it would be interesting to have benchmarks from comparable emerging stock markets. We think that the non-availability of high frequency data on these markets explains the rarity of studies on dynamic aspects of liquidity. Notes 1. Amihud et al. (1997), Derrabi (1998), Naidu and Rozeff (1994). 2. Lee et al. (1993) argue that wider (narrower) spread and smaller (greater) depths may be interpreted as a decrease (increase) in liquidity. 3. This analysis is based on response impulse functions calculated using a VAR model. 4. A committee designed by the BVMT revises at the end of each semester the lists of liquid and illiquid stocks. 5. No high frequency dataset on the Tunisian stock exchange is available for researchers. 6. The criterion applied is the average number of transactions on the study period. 7. See Manganelli (2002).

12 Electronic Markets Vol. 16 No Figure 1. Impulse response functions for TUNISAIR (the more frequently traded stock) to shocks to spread, volatility and depth. Dotted lines are 95 per cent confidence intervals

13 152 Dorra Mezzez Hmaied, Adel Grar and Olfa Benouda Sioud & Market Resiliency of the Tunisian Stock Market Figure 2. Impulse response functions for BNA (the less frequently traded stock) to shocks to spread, volatility and depth. Dotted lines are 95 per cent confidence intervals

14 Electronic Markets Vol. 16 No The VAR model has a forecasting ability and is easy to estimate requiring only OLS. However, it critically depends on the lag determination and using too many parameters leads to a loss of degree of freedom. 9. Lag length is truncated at s51 according to Schwartz information criterion (SIC) and Akaike Information criterion (AIC). 10. BNA is the abbreviation of Banque Nationale Agricole. References Admati, A. R. and Pfleiderer, P. (1988) A Theory of Intraday Patterns: Volume and Price Variability, The Review of Financial Studies 1: Amihud, Y., Mendelson, H. and Lauterbach, B. (1997) Market Microstructure and Securities Values: Evidence from the Tel Aviv Stock Exchange, Journal of Financial Economics 45: Battacharya, U. and Spiegel, M. (1998) Anatomy of Market Failure: NYSE trading Suspensions ( ), Journal of Business and Economic Statistics 16: Coppejans, M., Domowitz, I. and Madhavan, A. (2003) Resiliency in an Automated Auction, Unpublished work, Duke University. Danielsson, J. and Payne, R. (2002) Liquidity Determination in an Order Driven Market, Unpublished work, London School of Economics. Degryse, H., De Jong, F., van Ravenswaaij, M. and Wuyts, G. (2002) Aggressive Orders and the Resiliency of a Limit Order Market, Unpublished work, K.U. Leuven. Derrabi, M. (1998) Changement de Microstructure et Comportement des Prix des Actifs Financiers: Cas d un Marché Émergent, Unpublished work, AFFI Conference, Paris. Easley, D. and O Hara, M. (1987) Price, Trade Size and Information in Securities Markets, Journal of Financial Economics 19(1): Foster, D. F. and Viswanathan, S. (1990) A Theory of Interday Variations in Volumes, Variances and Trading Costs in Securities Markets, The Review of Financial Studies 3: Foucault, T. (1999) Order Flow Composition and Trading Costs in a Dynamic Limit Order Market, Journal of Financial Markets 2: Handa, P. and Schwartz, R. (1996) Limit Order Trading, Journal of Finance 51: Hasbrouck, J. (1991a) The Summary Informativeness of Stock Trades: An Econometric Ana1ysis, The Review of Financial Studies 4(3): Hasbrouck, J. (1991b) Measuring the Information Content of Trades, Journal of Finance 46(1): Hmaied, D., Sioud, O. and Grar, A. (2004) Order Placement Strategy on an Order Driven Market: Evidence from the Tunis Stock Market, Quarterly Journal of Finance India, April May, Glosten, L. and Milgrom, P. (1985) Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders, Journal of Financial Economics 14: Kyle, A. (1985) Continuous Auctions and Insider Trading, Econometrica 53(6): Large, J. (2005) Measuring the Resiliency of an Electronic Limit Order Book, Unpublished work, Nuffield College, University of Oxford. Lee, C., Mucklow, B. and Ready, M. (1993) Spreads, Depths and the Impact of Earnings Information: An Intraday Analysis, The Review of Financial Studies 6(2): Manganelli, S. (2002) Duration, Volume and Volatility Impact of Trades, Working paper number 125, European Central Bank. Muranaga, J. (2000) Dynamics of Market Liquidity of Japanese Stocks: An Analysis of Tick-by-tick Data of the Tokyo Stock Exchange, Unpublished work, Bank of Japan. Muranaga, J. and Shimizu, T. (1999) Market Microstructure and Market Liquidity, Unpublished work, Bank of Japan. Naidu, G. N. and Rozeff, M. S. (1994) Volume, Volatility, Liquidity and Efficiency of the Singapore Stock Exchange Before and After Automation, Pacific-Basin Finance Journal 2(1): Sioud, O. and Hmaied, D. (2003) The Effects of Automation on Liquidity, Volatility, Stock Returns and Efficiency: Evidence From the Tunisian Stock Market, Review of Middle East Economics and Finance 1(2): Spierdijk, L., Nijman, T. and van Soest, A. H. O. (2004) Temporary and Permanent Price Effects of Trades in Infrequently Traded Stocks, Unpublished work, University of Twente.