Technical trading strategies with market impact

Transcription

1 Alfred Ka Chun Ma (Hong Kong), Mandy Wai Man Chan (Hong Kong), Jonathan Chun Yu Poon (Hong Kong), Yan Yan (Hong Kong) Tehnial trading strategies with market impat Abstrat The paper proposes an empirial estimation of the market impat on the profitability of tehnial trading strategies for institutional inestors. A benhmark for the performane of large trades among institutional inestors, olume weighted aerage prie (VWAP) is used as a proxy for the exeution prie of large trades. By omparing to the performane using the losing prie as the exeution prie, the authors inestigate a representatie set of tehnial rules, inluding the moing aerage and trading range break-out rules, using the NYSE/AMEX seurities with the NYSE TAQ data set from 1993 to 005. Empirial results show that the market impat is estimated to be a spread of 1.4% per annum. The authors also find that tehnial trading strategies are not profitable after onsidering market impat. Keywords: olume weighted aerage prie, market impat, tehnial trading strategies. JEL Classifiation: G10, G0. Introdution It is ommon that institutional inestors implement trading strategies for their portfolios. While institutional inestors adopt different lasses of trading strategies subjet to their inestment objeties, they typially ealuate and selet optimal ones through bak-testing using historial data. The most frequently used data is the losing prie. Howeer, the assumption that institutional inestors an always exeute their large orders at the losing prie without prie impat an be iolated. In this study, we hypothesize that failing to exeute trading strategies at losing prie an result in a signifiant profit redution. Tehnial analysis is a disipline of seurity analysis to foreast the future prie trend by using historial finanial data mainly prie and olume. Many empirial studies, inluding Fama and Blume (1966), Jensen and Bennington (1970), Knez and Ready (1996), Allen and Karjalainen (1999), Marshall et al. (008), onlude that tehnial analysis is not useful for improing returns. Despite its many ritiisms, tehnial analysis has been popular among inestors and finanial analysts. Brok et al. (199) show the foreasting ability of 6 tehnial trading rules on the Dow Jones Industrial Aerage (DJIA) oer a period of 90 years. Furthermore, some other empirial studies, inluding Sweeney (1988), Allen and Taylor (1990, 199), Neely et al. (1997), Genay (1998), Sullian et al. (1999), Lo et al. (000), show the usefulness of tehnial trading rules. Blume et al. (1994), Friesen et al. (009) also adoate using theoretial models that tehnial analysis an be aluable. Park and Irwin (007) summarize main reent findings regarding the profitability of tehnial trading strategies. Reently, Shynkeih (01) shows that Alfred Ka Chun Ma, Mandy Wai Man Chan, Jonathan Chun Yu Poon, Yan Yan, 013. after adjusting for data snooping bias, tehnial trading strategies an not outperform buy-and-hold strategies. Menkhoff (010) finds that tehnial trading strategies are ommon in Germany, Switzerland, the United States, Italy and Thailand. All preious studies in the literature, howeer, assume impliitly that trades an be exeuted at the daily losing prie. We take a different approah and test the hypothesis of whether the performanes of tehnial trading strategies are affeted by market impat. In tehnial analysis, a majority of traders base their inestment plan on the losing prie of seurities. The losing prie represents the final ealuation of the stok made by the market on a gien trading day, whih is readily aailable and well-published. Hene, traders often trak errors relatie to the lose of stoks. This post-trade benhmark promotes trading at the losing prie, through market orders being plaed towards the end of the day or guaranteed market-on-lose orders. Trading at the losing prie inoles hidden osts whih an be signifiant. Cushing and Madhaan (000) show that there are greater market impats if trading at the lose beause pries are more sensitie to order flows at this point. Een ordinary retail inestors an hardly exeute their trade at around the lose without market impats. Chan and Lakonishok (1995) use a speial data set of 37 large inestment firms to illustrate the market impat of their trades. Therefore, we onlude that it is generally diffiult for institutional inestors to exeute all their trades at the losing prie as their trades are usually of large olume. The transation osts institutional inestors bear are not just expliit suh as broker ommissions and taxes but also impliit ones. Compared with expliit transation osts, impliit transation osts for institutional inestors an be muh higher. One signifiant impliit ost to institutional inestors is the market impat, whih is aused by unfaorable prie moements due to the exeution of large trades. 93

2 To aoid market impat, large trades are not exeuted at one. Instead, the order is typially split up for exeution oer the day to partiipate proportionately in the day s olume. When institutional inestors diide their trade into separate orders, the first order an affet the prie of subsequent trades. Suh market impat osts an be substantial. Although many researhers, inluding Dufour and Engle (000) and Lee and Ready (1991), attempt to measure the prie impat of a trade, there is still no aurate method to estimate impliit osts before the trade. The olume weighted aerage prie (VWAP) is a popular benhmark for measuring the performane of traders and omputing trading osts. Reealed by a surey onduted by the Bank of Ameria (007), VWAP exeution orders represent around 50% of all the trading atiities by institutional inestors. The popularity of using VWAP as a benhmark is mainly beause of its omputational simpliity. Although the omputation of VWAP may inole data-intensie alulations, it is proided by a number of endors suh as Reuters and Bloomberg. In addition, VWAP is better than any fixed time benhmarks as it improes both market transpareny and effiieny (Cushing and Madhaan, 000). Ting (006) also shows empirially that VWAP is loser to the effiient prie ompared with the losing prie. Market impat an be measured by omparing the exeution prie of a large order with the VWAP benhmark (Berkowitz et al., 1988). As a result, institutional inestors implement VWAP strategies (Madhaan, 000; Bialkowski et al., 008) to redue impliit transation osts. A VWAP strategy inoles buying or selling a fixed number of shares traking VWAP at an aerage prie. There are some examples of VWAP strategies: 1. Diret aess: orders are traded by inestors themseles, either through partiipation strategies or market timing strategies to beat VWAP.. Ageny trading: orders are gien to broker-dealers to trade on an ageny basis to trak VWAP. 3. Automated partiipation strategies: orders are split up oer the day to partiipate proportionately in the day s olume, trading as intelligently as possible and with minimal market impat. Manual trading is labor-intensie and ostly. For a large equity trade, in order to get the aerage exeution prie as lose to VWAP as possible to aoid prie moement risk, the orders are typially plaed in automated partiipation strategies, whih lower the expliit osts and minimize prie impat by spreading the liquidity demand of large orders aross the trading period. Both the losing prie and the VWAP are potentially informatie and onenient referene pries. Howeer, sine institutional inestors annot generally exeute large orders at the losing prie, it is unreasonable to ealuate their trading strategies based on the losing prie. Theoretially, the exeution ould be ompleted as if it was traded by any randomly seleted trader implementing VWAP strategies whih ould outperform or underperform the VWAP benhmark. We argue that on aerage they an exeute large orders at the daily VWAP. Therefore, ompared with the losing prie, VWAP should be a more realisti proxy to ealuate trading strategies for institutional inestors. Under the assumption that institutional inestors exeute orders at VWAP, it also reates an impetus to ompare the losing prie with VWAP in generating trading strategies through an empirial study of any trading strategy using historial prie as input. Aordingly, we test the following hypotheses: H1: Ealuating tehnial trading strategies using VWAP produes lower returns than using losing prie. H: For institutional inestors who exeute orders at VWAP, using of VWAP to generate tehnial trading strategies produes higher returns than using losing prie. If H1 is true, it supports the intuition that market impat affets the profitability of trading strategies. To test the two hypotheses, we follow the framework of Brok et al. (199) who test 6 tehnial trading rules under moing aerages and trading range breaks on the daily prie of DJIA oer the period from 1897 to They find that buy signals onsistently generate higher returns than sell signals, and proide eidene for the preditie power of the 6 tehnial rules. The findings has raised the interest of many researhers to inestigate whether similar results hold for other major stok markets using similar researh methods, suh as the London Stok Exhange FT30 index for the period from 1935 to 1994 (Mills, 1997), the Finanial Times Industrial Ordinary Index in the UK for the period from 1935 to 1994 (Hudson et al., 1996), 6 stok market indies in Asia, namely Japan, Hong Kong, Korea, Malaysia, Thailand, and Taiwan oer the period from 1975 to 1991 (Bessembinder and Chan, 1995), and the Chilean equity market index oer the period of 1987 to 1998 (Parisi and Vasquez, 000). All of them onlude that the trading rules are quite suessful in produing a return greater than a buy-and-hold strategy in their respetie sample periods. We fous on those tehnial rules studied in Brok et al. (199), namely the moing aerage rules and the trading range break rules, to test our hypotheses empirially. The moing aerage rules are espeially 94

3 hosen as they are widely used. Allen and Taylor (199) surey dealers in the London Foreign Exhange and find that oer 90% of the respondents use some forms of tehnial analysis to predit returns. The moing aerage rules are among the most popular trading rules being used. Early inestigations of the moing aerage rules by Van Home and Parker (1967) and James (1968) show that none of them were suessful when ompared with a buy-and-hold strategy. Howeer, some reent researhes support the use of moing aerage rules. El-Khodary (004) points out that moing aerage rules are the most profitable tehnial trading rules whih help traders gauge trend diretions. They rekon that moing aerages are exellent indiators to onfirm existing trends in spite of their lag. The main idea behind the moing aerage rule is to aerage a number of past referene pries. The alulated aerage represents the prie area nearest to a proportionately large number of trades in that time period. Suh area is where the fewest people hae extreme gains or losses for the period. Therefore, the pressure to trade out of fear or greed tends to be diminished. Aeraging an smooth the flutuation in the pries so that the underlying trends an be signaled out. Aording to El-Khodary (009), all types of moing aerage rules do not predit market moements lagging the urrent prie. In a bull market, the moing aerage is below the rising prie line and ie ersa for a bear market. This is the idea behind ariable length moing aerage rule. When the prie hanges its diretion, the moing aerage line rosses the prie line. Hene, depending on the diretion of the rossing, buy or sell signals an be identified. This is known as the fixed length moing aerage rule. Howeer, the simplest form of moing aerage rules is ritiized for its equal weighting in aeraging the referene pries in the speified period, assuming that old pries are equally releant to more reent ones. Conrad and Kaul (1998) suggest that momentum strategy, inluding the moing aerage rule, is useful and usually profitable at the medium horizon (3 to 1 months), while Lesmond et al. (004) find that stoks generating large momentum returns are preisely those with high trading ost and onlude that the magnitude of abnormal returns from trading strategies does not imply a profit opportunity. Thus, it is interesting to inestigate the momentum strategies, partiularly the ariable-length moing aerage rule, fixed-length moing aerage rule and trading range breakout rule using different exeution prie suh as VWAP. This empirial study examines the results of applying tehnial trading rules, inluding ten ariable-length moing aerage rules, ten fixedlength moing aerage rules and six trading range break rules, to the stoks in New York Stok Exhange (NYSE) and Amerian Stok Exhange (AMEX), oer a 13-year period from January 1993 to Deember 005, with losing prie and VWAP ating either as trading strategy generator, or exeution prie, or both. Existing similar studies inlude Kaajez and Odders-White (004) who test the liquidity effet of tehnial analysis using limit order book data. The main ontributions to existing literature are, first, to find out the profitability of tehnial analysis to institutional inestors with market impat as opposed to ordinary retail inestors, and seond, to ompare whether losing prie or VWAP is more faorable as trading strategy generator for institutional inestors with market impat. 1. Data and tehnial rules 1.1. Data. The data used in this study inlude all seurities listed on the New York Stok Exhange (NYSE) and Amerian Stok Exhange (AMEX) from the Center for Researh in Seurity Pries (CRSP) in Wharton Researh Data Series (WRDS). Only ommon stoks listed on NYSE are inluded. This is done by exluding all the stoks in the CRSP database with share ode (SHRCD) other than 10 or 11, and exluding those with exhange ode (EXCHCD) other than 1. The intraday transation data, inluding the prie and olume traded for the same seurities, is also retrieed from WRDS but through another dataset Trade and Quote (TAQ). Sine WRDS started proiding TAQ data from 1993, we ollet data from January 1993 to Deember 005. We lean the reords in TAQ aording to the standards in the literature (see for example Liu and Maheu, 008). Inalid trades are filtered out by using orretion indiator. The trade data is kept only if the orretion indiators equal 0 or 1, whih refer to regular trade and later-orreted trades respetiely. The daily VWAP is alulated based on the intraday transation data with the formula: VWAP N i1 Trasation Priei Volume i. Daily Share Volume Sine the losing prie and VWAP are obtained from two different datasets, inonsisteny is unaoidable. We inlude a partiular stok in our study only if both its losing prie and VWAP reords are aailable for the same period of time. 95

4 Sine small-ap stoks are usually not onsidered by institutional inestors, we exlude stoks with low market apitalization. In order not to exasperate the suriorship bias, we sort all the stoks with trading data oer the period from January 1993 to Deember 005 by their market apitalizations of their first trading days throughout this period. We impose a 80th perentile of market apitalization rule on all the stoks as the riterion for stok seletion, whih is roughly equialent to onsidering stoks with market apitalization greater than billion US dollars. As a result, all stoks with market apitalization on their first trading day greater than billion US dollars are inluded in our studies. This stok filtering riterion is better than the one whih just sorts the stoks aording to the market apitalization of one single trading day, in that we would not exlude stoks with large market apitalization but not being traded on the hosen trading day. The number of stoks satisfying all the aboe riteria is Tehnial trading rules. We ealuate the same set of twenty-six tehnial trading rules following Brok et al. (199), inluding ten Variable Length Moing Aerage (VMA) rules, ten Fixed Length Moing Aerage (FMA) rules and six Trading Range Break (TRB) rules, with the losing prie and VWAP ating as signal generators and exeution pries. The assumption behind the rules is that, upon a buy signal, an inestor is to borrow and double the inestment in the stok; upon a sell signal, an inestor is to sell shares and use the proeed to inest in a risk-free asset. The moing aerage rule aries with the length of time periods (short/long term), inlusion of band, type of MA models (ariable-length moing aerage/fixed-length moing aerage), and type of prie data (losing prie/vwap). The length of time periods inoles two moing aerages of stok prie leel, namely short-term moing aerage of order n and long-term moing aerage of order m (m > n). A trend is being identified if the relatie positions of short-term and long-term moing aerages an be loated. Generally speaking, when the short-term moing aerage penetrates the long-term moing aerage, buy and sell signals are generated. The rule an be aried by the length of time periods. Following Brok et al. (199) to use the most popular rules, we inestigate the 1-50 (short period is 1 day and long period is 50 days), 1-150, 5-150, 1-00 and -00 MA rules. Banding an eliminate false buy or sell signals, i.e. signals that result in losses, when the short and long term moing aerages are lose to eah other. Brok et al. (199) also introdue a one perent band so that a signal is generated only when the short-term moing aerage is aboe or below the long-term moing aerage by one perent. In this paper, the MA rules with and without the one perent band are tested. The VMA rules generate a buy (sell) signal when the short-term moing aerage is aboe (below) the long-term moing aerage. This rule simulates a strategy that traders go long as the short-term moing aerage moes aboe the long-term moing aerage and go short ie ersa. For a zero band, all days are lassified as either buy or sell. For a one perent band, signals are generated only if the short-term moing aerage is aboe (below) the long-term moing aerage by an amount larger than the band. The FMA rules generate a buy (sell) signal when the short-term moing aerage uts the long-term moing aerage from below (aboe). The holding period is 10 days as suggested by Brok et al. (199) 1. Returns are reorded at the end of eah holding period. In addition, signals ourring during this 10-day period are ignored. As the signal an only be based on histori data, a oneday lag is unaoidable. The TRB rules generate a buy (sell) signal when the prie penetrates the resistane (support) leel, whih is defined as a loal maximum (minimum) oer m trading days. The idea behind TRB rules is that many inestors are willing to buy at around the minimum prie, thus making it diffiult for the prie to further penetrate its support leel. In ase the prie drops and penetrates the support leel, it is expeted to further drift downwards. The rationale is ie ersa for that of the resistane leel. Therefore, tehnial analysts reommend buying when the prie rises aboe the last peak and selling when the prie sinks below its last trough. Following Brok et al. (199), the number of days being tested (m) are 50, 150 and 00. The rule is also tested with and without a one perent band to eliminate false signals. Similar to the FMA rules, returns are alulated after 10 trading days and any signal during the 10-day holding period is ignored.. Empirial results In the first hypothesis, we test the signifiane of differene between the returns ealuated by losing prie and VWAP respetiely. Sine the signals are generated from the same set of losing pries, the most suitable test statisti for this is the t-test for paired samples. 1 Aording to Brok et al. (199), the seletion of 10-day returns is arbitrary. For some rules they tried two-week return and obtained essentially the same results. 96

5 The t-statisti is D, / N D where D and are the mean and ariane of the D return differenes respetiely, and N is the number of signals. In the seond hypothesis, the signals generated from the losing prie and VWAP an differ. Therefore, we use the Welh t-test (Welh, 1938) where the t- statisti for the differenes is N N, where and are the means of the exess returns oer the unonditional mean return using the losing prie and VWAP as the input respetiely, and are the arianes of the exess returns using the losing prie and VWAP as the input respetiely, and N and N are the numbers of signals generated by the losing prie and VWAP respetiely. The degree of freedom assoiated with this t-distribution N N. N N N 1 N 1 To test the two hypotheses, we take a different approah ompared to Brok et al. (199) when we ompute the differene of the ombined returns between buy and sell strategies. The buy-sell returns here aggregate the returns from the buy signal and the negatie returns from the sell signal. Eah trading rule is applied to indiidual eligible stoks and eah return generated from the trading rule is used as a sample return..1. Is ealuating trading strategies using VWAP less profitable? For the institutional inestors with market impat, we are interested in whether exeuting orders with VWAP differs signifiantly ompared with losing prie. Table 1, Table and Table 3 (see Appendix) examine the differenes with VMA, FMA, and TRB rules respetiely. Table 1 shows that all the differenes of buy returns are positie. Under six out of ten sets of parameters, we rejet the null hypothesis that ealuating buying strategies with VWAP is at least as profitable as with losing prie at 5 perent signifiane. For the differenes of sell returns, the results also support the hypothesis. Similar results an be obtained from the tests for oerall return. Table shows results een more supportie to the hypothesis. All differenes of buy returns are positie and we rejet the null hypothesis at 5 perent signifiane leel for eah set of parameters. Similar onlusions are drawn for sell returns and oerall returns. Table 3 shows the results for TRB. We rejet the null hypothesis at 5 perent signifiane leel in all sets of parameters for buy, sell, and oerall returns. The empirial results for the three popular trading rules in tehnial analysis support the hypothesis that ealuating trading strategies using VWAP is less profitable than using the losing prie. Giing equal weights on eah of the 6 trading rules, the aerage estimated market impat is 1.379% per annum... Is using VWAP to generate trading strategies more profitable than using losing prie for institutional inestors? The results presented in the last setion suggest that there is a statistially signifiant differene between ealuating trading strategies using VWAP and losing prie. In this setion, we ask a further question as to whether generating signals using VWAP is more profitable ompared with using losing prie for institutional inestors, i.e., we assume that we ealuate all trading strategies using VWAP. Table 4, Table 5, and Table 6 show the results for the VMA, FMA, and TRB rules respetiely. In Table 4 (see Appendix), the results for the trading strategies generating signals with the losing prie are presented in olumn, 4, 6, 8, 10 with the number of buy signals, the number of sell signals, mean of the returns from buy signals, mean of the returns from sell signals, and mean of the oerall returns from the VMA rules. Similarly, the results for the trading strategies generating signals with VWAP are presented in olumn 3, 5, 7, 9, 11. The aerage of the differene is about perent, or 0.95 perent per annum. Howeer, most of the tests results are only signifiant at 0 perent leel and only one of them is signifiant at 10 perent leel. We fail to rejet the null hypothesis that the returns of generating signals with VWAP is better than those with the losing prie at 10 perent signifiane leel using a one-tailed test. For sell returns, the results are similar. Only one of the ten tests an rejet the null hypothesis at 10 perent signifiane leel. For the oerall returns presented in olumn 10 and 11, all exess returns from signal generator VWAP is around 0 while those from the losing prie is negatie. The aerage of these differene is perentage, whih is perent per annum. While it an be seen that using VWAP to generate trading strategies an improe performane, the results are not signifiant. 97

6 Table 5 shows the results for FMA. Generating signals with VWAP produes a higher return than generating with the losing prie. The aerage of the differenes for oerall returns is about 0.064, whih is an perent per annum. Howeer, results are not statistially signifiant. Table 6 shows the results for TRB. Generating signals with VWAP produes a higher return than generating with losing prie. The aerage of the differenes in oerall returns is about perent. Howeer, none of the ten one-tailed tests is signifiant for buy, sell, and oerall returns. While the empirial tests for the seond hypothesis is not onlusie ompared with the first hypothesis, they partially support that using VWAP as an input for generating trading strategies an improe the performane for institutional inestors. One explanation for the differene being less signifiant is that losing prie atually ontains similar if not the same information as VWAP and therefore using VWAP to generate trading strategies may not improe the performane signifiantly. The empirial results also show that een if VWAP is used as the signal generator, the tehnial trading rules are not profitable after onsidering market impat during the period from 1993 to 005. While most of the studies in the literature tend to support the usefulness of tehnial analysis, their studies mainly fous on market indies while in this paper we fous on a large lass of indiidual stoks together with market impat. As Referenes 98 a result, we onlude that market impat does affet the profitability of tehnial trading strategies. Conlusion By using VWAP as a proxy for ealuating tehnial trading strategies under market impat, we empirially show that the effet of market impat on the profitability is signifiant. Empirial results show that ealuating tehnial trading strategies using VWAP is less profitable ompared with using losing prie. A possible reason is that institutional inestors need to ompensate for the impliit trading osts resulted from large trades. The redution in returns ould also be due to the mismath between the tehnial trading strategy generator and the exeution prie. We then further test on whether using VWAP to generate trading strategies improes the performane of those tehnial rules. The empirial results generally support this hypothesis but with less statistial signifiane. This ould partially be explained by the fat that losing prie and VWAP ontain roughly the same market information. Therefore, using VWAP to generate trading strategies may not improe the performane substantially. Finally, empirial results also show that the tehnial trading strategies are not profitable if VWAP is the exeution prie. The data set oers a period of time with arious ondition of market olatility and therefore the results are alid for a range of market olatility ondition. Future researh may inlude testing more tehnial trading strategies to further inestigate the profitability under market impat. 1. Allen, F. and Karjalainen, R. (1999). Using geneti algorithms to find tehnial trading rules, Journal of Finanial Eonomis, 51 (), pp Allen, H. and Taylor, M. (1990). Charts, noise and fundamentals in the London Foreign Exhange market, Eonomi Journal, 100 (400), pp Allen, H. and Taylor, M. (199). The use of tehnial analysis in the foreign exhange market, Journal of International Money and Finane, 11 (3), pp Bank of Ameria (007). Marhing up the learning ure: The seond buy-side algorithmi trading surey, Tehnial report, Bank of Amerian. 5. Berkowitz, S., Logue, D. and Noser, E. (1988). The total ost of transations on the NYSE, Journal of Finane, 43 (1), pp Bessembinder, H. and Chan, K. (1995). The profitability of tehnial trading rules in the Asian stok markets, Paifi-Basin Finane Journal, 3 (-3), pp Bialkowski, J., Darolles, S., and Le Fol, G. (008). Improing VWAP strategies: A dynami olume approah, Journal of Banking and Finane, 3 (9), pp Blume, L., Easley, D., and O Hara, M. (1994). Market statistis and tehnial analysis: the role of olume, Journal of Finane, 49 (1), pp Brok, W., Lakonishok, J., and LeBaron, B. (199). Simple tehnial trading rules and the stohasti properties of stok returns, Journal of Finane, 47 (5), pp Chan, L. and Lakonishok, J. (1995). The behaior of stok pries around institutional trades, Journal of Finane, 50 (4), pp Conrad, J. and Kaul, G. (1998). An anatomy of trading strategies, Reiew of Finanial Studies, 11 (3), pp Cushing, D. and Madhaan, A. (000). Stok returns and institutional trading at the lose, Journal of Finanial Markets, 3 (l), pp Dufour, A. and Engle, R. (000). Time and the prie impat of a trade, Journal of Finane, 55 (6), pp El-Khodary, I.A. (004). Day trading with andlestiks and moing aerage, Futures: Newsk Analysis and Strategies for Futures, pp

7 15. El-Khodary, I.A. (009). A deision support system for tehnial analysis of finanial markets based on the moing aerage rossoer, World Applied Sienes Journal, 6 (11), pp Fama, E. and Blume, M. (1966). Filter rules and stok market trading profits, Journal of Business, 39 (1), pp Friesen, G., Weller, P., and Dunham, L. (009). Prie trends and patterns in tehnial analysis: a theoretial and empirial examination, Journal of Banking and Finane, 33 (6), pp Genay, R. (1998). The preditability of seurity returns with simple tehnial trading rules, Journal of Empirial Finane, 5 (4), pp Hudson, R., Dempsey, M., and Keasey, K. (1996). A note on the weak form effiieny of apital markets: The appliation of simple tehnial trading rules to UK stok pries 1935 to 1994, Journal of Banking and Finane, 0 (6), pp James, E. (1968). Monthly moing aerages an effetie inestment tool? Journal of Finanial and Quantitatie Analysis, 3 (3), pp Jensen, M. and Bennington, G. (1970). Random walks and tehnial theories: some additional eidene, Journal of Finane, 5 (), pp Kaajez, K. and Odders-White, E. (004). Tehnial analysis and liquidity proision, Reiew of Finanial Studies, 17 (4), pp Knez, P. and Ready, M. (1996). Estimating the profits from trading strategies, Reiew of Finanial Studies, 9 (4), pp Lee, C. and Ready, M. (1991). Inferring trade diretion from intraday data, Journal of Finane, 46 (), pp Lesmond, D., Shill, M.J., and Zhou, C. (004). The illusory nature ol momentum profits, Journal of Finanial Eonomis, 71 (), pp Liu, C. and Maheu, J. (008). Are there struture breaks in realized olatility? Journal of Finanial Eonometris, 6 (3), pp Lo, A., Mamaysky, H., and Wang, J. (000). Foundations of tehnial analysis: omputational algorithms, statistial inferene, and empirial implementation, Journal of Finane, 55 (4), pp Madhaan, A. (000). VWAP strategies, Inestment Guides, Transation Performane: The Changing Fae of Trading, Institutional Inestor In. 9. Marshall,., Cahan, R., and Cahan, J. (008). Does intraday tehnial analysis in the U.S. equity market hae alue? Journal of Empirial Finane, 15 (), pp Menkhoff, L. (010). The use of tehnial analysis by lund managers: international eidene, Journal of Banking and Finane, 34 (11), pp Mills, T. (1997). Tehnial analysis and the London Stok Exhange: testing trading rules using the FT30, International Journal of Finane and Eonomis, (4), pp Neely, C, Weller, P., and Dittmar, R. (1997). Is tehnial analysis in the foreign exhange market profitable? A geneti programming approah, Journal of Finanial and Quantitatie Analysis, 3 (4), pp Parisi, F. and Vasquez, A. (000). Simple tehnial trading rules of stok returns: eidene from 1987 to 1998 in Chile, Emerging Markets Reiew, 1 (), pp Park, C. and Irwin, S. (007). What do we know about the profitability of tehnial analysis? Journal of Eonomi Sureys, 1 (4), pp Shynkeih, A. (01). Performane of tehnial analysis in growth and small ap segments of the US equity market, Journal of Banking and Finane, 36 (l), pp Sullian, R., Timmermann, A., and White, H. (1999). Data-snooping, tehnial trading rule performane, and the bootstrap, Journal of Finane, 54 (5), pp Sweeney, R. (1988). Some new filter rule tests: methods and results, Journal of Finanial and Quantitatie Analysts, 3 (3), pp Ting, C. (006). Whih daily prie is less noisy? Finanial Management, 35 (3), pp Van Home, J. and Parker, G. (1967). The random walk theory: an empirial test, Finanial Analysts Journal, 3 (6), pp Welh, B. (1938). The signifiane of the differene between two means when the population arianes are unequal, Biometrika, 9 (3/4), pp Appendix Table 1. Differene between exeuting with VWAP and losing for VMA rules Test N(Buy) N(Sell) Buy Sell Buy-Sell (1, 50, 0) (1, 50, 0.01) (1, 150, 0) (1, 150, 0.01) (3.1856) (3.4066) (.5047) (.5485) (-.8767) ( ) (-.3304) (-.0183) (4.703) (4.554) (3.4100) (3.08) 99

8 100 Table 1 (ont.). Differene between exeuting with VWAP and losing for VMA rules Test N(Buy) N(Sell) Buy Sell Buy-Sell (5, 150, 0) (5, 150, 0.01) (1, 00, 0) (1, 00, 0.01) (, 00, 0) (, 00, 0.01) (0.740) (0.667) (1.7907) (1.9580) (1.167) (1.630) (-0.771) (-0.366) ( ) (-1.763) ( ) ( ) (1.0364) (0.7158) (.3647) (.5957) (1.50) 0.00 (1.561) Aerage Notes: The sample period is from January 1993 to Deember 005. Test is the VMA trading rule speified as the number of days to alulate the long-term moing aerage, the number of days to alulate the short term moing aerage, the band as a perentage of the short-term moing aerage. N(Buy) and N(Sell) are the numbers of buy and sell signals in the sample. Buy and Sell are the mean returns generated from VMA rules with exeuting at the losing prie less exeuting at the VWAP by buy signals and sell signals respetiely. The t-ratios below in the parentheses test the null hypothesis that exeuting at VWAP generates at least the same return as exeuting at the losing prie against exeuting at VWAP generates lower buy (higher sell) return than exeuting at the losing prie. Buy-Sell are the oerall returns generated from VMA rules with the losing prie as the exeution prie less those with the VWAP as the exeution prie. The t-ratios below are from the one-tailed test for the null hypothesis that exeuting at VWAP generates at least the same oerall return smaller as exeuting at the losing prie. All returns are reported in perentage leel. Table. Differene between exeuting with VWAP and losing for FMA rules Test N(Buy) N(Sell) Buy Sell Buy-Sell (1, 50, 0) (1, 50, 0.01) (1, 150, 0) (1, 150, 0.01) (5, 150, 0) (5, 150, 0.01) (1, 00, 0) (1, 00, 0.01) (, 00, 0) (, 00, 0.01) (.0018) (5.440) (4.9639) (3.5435) (3.9547) (1.75) (3.3930) (.3971) (.3300) (.7196) ( ) ( ) ( ) ( ) (-3.359) (-1.496) ( ) ( ) (-5.14) (-4.385) Aerage (7.0851) (9.406) (7.8346) (6.7057) (5.1086) (.56) (7.5859) (6.609) (5.5195) (5.110) Notes: The sample period is from January 1993 to Deember 005. Test is the FMA trading rule speified as the number of days to alulate the long-term moing aerage, the number of days to alulate the short-term moing aerage, the band as a perentage of the short-term moing aerage. N(Buy) and N(Sell) are the numbers of buy and sell signals in the sample. Buy and Sell are the mean returns generated from FMA rules with exeuting at losing prie less exeuting at the VWAP by buy signals and sell signals respetiely. The t-ratios below in the parentheses test the null hypothesis that exeuting at VWAP generates the same return as exeuting at the losing prie against exeuting at VWAP generates lower buy (higher sell) return than exeuting at the losing prie. Buy-Sell are the oerall returns generated from FMA rules with the losing prie as the exeution prie less those with the VWAP as the exeution prie. The t-ratios below are from the one-tailed test for a null hypothesis that exeuting at VWAP generates at least the same oerall return as exeuting at the losing prie. All returns are reported in perentage leel. Table 3. Differene between exeuting with VWAP and losing for TRB rules Test N(Buy) N(Sell) Buy Sell Buy-Sell (1, 50, 0) (1, 50, 0.01) (6.9343) (7.34) ( ) (-5.379) (9.097) (8.6187)

9 Table 3 (ont.). Differene between exeuting with VWAP and losing for TRB rules Test N(Buy) N(Sell) Buy Sell Buy-Sell (1, 150, 0) (1, 150, 0.01) (1, 00, 0) (1, 00, 0.01) (5.6878) (5.974) (5.4383) (6.1684) (-.790) (-.5385) (-.8489) (-.6084) (5.7316) (5.5994) (5.6859) (5.847) Aerage Notes: The sample period is from January 1993 to Deember 005. Test is the TRB trading rule speified as the number of days to alulate the long-term moing aerage, the number of days to alulate the short-term moing aerage, the band as a perentage of the short-term moing aerage. N(Buy) and N(Sell) are the numbers of buy and sell signals in the sample. Buy and Sell are the mean returns generated from TRB rules with exeuting at losing prie less exeuting at the VWAP by buy signals and sell signals respetiely. The t-ratios below in the parentheses test the null hypothesis that exeuting at VWAP generates the at least same return as exeuting at the losing prie against exeuting at VWAP generates lower buy (higher sell) return than exeuting at the losing prie. Buy-Sell are the oerall returns generated from TRB rules with the losing prie as the exeution prie less those with the VWAP as the exeution prie. The t-ratios below are from the one-tailed test for the null hypothesis that exeuting at VWAP generates at least the same oerall return as exeuting at the losing prie. All returns are reported in perentage leel. Table 4. Differene between generating with VWAP and losing for VMA rules Test N(Buy) N(Buy)* N(Sell) N(Sell)* Buy Buy* Sell Sell* Buy-Sell Buy-Sell* (1, 50, 0) (1, 50, 0.01) (1, 150, 0) (1, 150, 0.01) (5, 150, 0) (5, 150, 0.01) (1, 00, 0) (1, 00, 0.01) (, 00, 0) (, 00, 0.01) (1.081) (-0.933) (1.41) (1.479) (-1.80) (1.9304) (1.0100) (-0.849) (1.978) (1.01) ( ) (1.) (0.6000) ( ) (0.7755) (0.4596) (-0.167) (0.463) (0.6911) ( ) (0.8916) (0.808) ( ) (1.1419) (0.681) (-0.531) (0.816) (0.5483) ( ) (0.6531) Aerage Notes: The sample period is from January 1993 to Deember 005. Those olumns with (without) * are the results for VMA rules with VWAP (the losing prie) as the signal generator. N(Buy) and N(Sell) are the numbers of buy and sell signals reported in the sample. All returns are reported in perentage leel and in terms of exess return, i.e. daily return less the unonditional daily return of eah stok. The t-ratios below in the parentheses test the null hypothesis that generating signals with the losing prie produes at least the same returns as generating with the VWAP. Table 5. Differene between generating with VWAP and losing for FMA rules Test N(Buy) N(Buy)* N(Sell) N(Sell)* Buy Buy* Sell Sell* Buy-Sell Buy-Sell* (1, 50, 0) (1, 50, 0.01) (1, 150, 0) (1, 150, 0.01) (0.417) (0.637) (0.7481) (0.9411) ( ) (1.56) ( ) ( ) (0.1934) (0.1609) ( ) (0.9416) 101

10 Table 5 (ont.). Differene between generating with VWAP and losing for FMA rules Test N(Buy) N(Buy)* N(Sell) N(Sell)* Buy Buy* Sell Sell* Buy-Sell Buy-Sell* (5, 150, 0) (5, 150, 0.01) (1, 00, 0) (1, 00, 0.01) (, 00, 0) (, 00, 0.01) (0.3135) ( ) (0.517) (0.013) ( ) (0.671) (0.7844) ( ) (0.8746) (1.0055) (-0.938) (1.3719) (0.3548) (-0.811) (0.8479) (0.8146) (-0.745) (1.086) Aerage Notes: The sample period is from January 1993 to Deember 005. Those olumns with (without) * are the results for FMA rules with VWAP (the losing prie) as the signal generator. N(Buy) and N(Sell) are the numbers of buy and sell signals reported in the sample. All returns are reported in perentage leel and in terms of exess return, i.e. daily return less the unonditional daily return of eah stok. The t-ratios below in the parentheses test the null hypothesis that generating signals with VWAP produes better returns than generating with the losing prie. Table 6. Differene between generating with VWAP and losing for TRB rules Test N(Buy) N(Buy)* N(Sell) N(Sell)* Buy Buy* Sell Sell* Buy-Sell Buy-Sell* (1, 50, 0) (1, 50, 0.01) (1, 150, 0) (1, 150, 0.01) (1, 00, 0) (1, 00, 0.01) (0.871) (-1.154) (1.4655) (0.0363) ( ) (0.7603) (0.81) ( ) (1.1748) (0.3364) (-0.55) (0.6383) (1.0704) ( ) (1.0879) (0.159) ( ) (0.643) Aerage Notes: The sample period is from January 1993 to Deember 005. Those olumns with (without) * are the results for TRB rules with VWAP (the losing prie) as the signal generator. N(Buy) and N(Sell) are the numbers of buy and sell signals reported in the sample. All returns are reported in perentage leel and in terms of exess return, i.e. daily return less the unonditional daily return of eah stok. The t-ratios below in the parentheses test the null hypothesis that generating signals with VWAP produes better returns than generating with the losing prie. 10