Effects on Liquidity, Trading Activity and Volatility

Research and Product Development Department of Borsa İstanbul Working Paper Series No.20 July, 2014 Technology Upgrades in Emerging Equity Markets: Effects on Liquidity, Trading Activity and Volatility M. Kemal Yılmaz Orhan Erdem Veysel Eraslan Evren Arık Borsa İstanbul Borsa İstanbul Borsa İstanbul Borsa İstanbul 1

Technology Upgrades in Emerging Equity Markets: Effects on Liquidity, Trading Activity and Volatility July 2014 Mustafa Kemal Yılmaz i Orhan Erdem i Veysel Eraslan i,* Evren Arık i Abstract This study examines the effects of technological changes on selected stock market qualities such as liquidity, turnover and volatility. The data set includes daily data of 361 stocks from 10 emerging market exchanges, namely Colombia, Indonesia, Johannesburg, Korea, Malaysia, Mexico, Russia, Shanghai, Shenzhen and Thailand. The analysis is based mainly on the comparison between the pre- and post-launch of a new trading platform for equity markets. A panel data regression analysis shows that technological upgrade decreases the bidask spread and increases trading activity. In other words, launching a more sophisticated trading platform contributes to the overall efficiency of the market. Moreover, we find that, in some exchanges, an important upgrade in the technological infrastructure of the exchange decreases its level of volatility. Key Words: Market Microstructure, Technological Upgrade, High Frequency Trading, Emerging Markets, Liquidity i Borsa İstanbul,Reşitpaşa Mahallesi, Tuncay Artun Caddesi, Emirgan Sarıyer, 34467 İstanbul, Turkey. * Corresponding author, veysel.eraslan@borsaistanbul.com, +90 212 298 28 21. 1

1. Introduction For many years, developments in stock market technology have been a core topic on the agenda of all exchanges. Since technological upgrades modernize the way financial assets are traded, technological investments in exchange infrastructures improve the strength of the link between investment and savers. As the inclusion of technology in trading activities increases, both the friction and cost incurred by intermediaries decreases, thereby not only enabling more efficient risk sharing, but also improving hedging strategies, liquidity, and efficiency of prices (Hendershott et. al, 2011). Competition among the exchanges causes exchange-owners to recognize the importance of having an efficient market design and trading mechanism so as to receive a higher market share. With the advance of electronic platforms, institutional investors have switched their focus to fully electronic trading engines, namely algorithmic trading systems. Since all markets currently have electronic limit order books, the speed and quality of access to these markets encourages the use of algorithmic trading. Algorithmic trading may be defined as electronic trading whose parameters are determined by strict adherence to a predetermined set of rules aimed at delivering specific execution outcomes (Deutsche Bank, 2011). Currently the most popular type of algorithmic trading is high-frequency trading (HFT) where a large number of orders are sent into the market at high speed, with round-trip execution times measured in microseconds (Brogaard, 2010). In this paper we investigate the effects of technological changes on stock market qualities, namely liquidity, trading activity and volatility. Implementing technological changes reflects the market aim of attracting global competition by increasing and enhancing the technological facilities that a particular exchange can offer current and potential investors. What is meant by technological change or technological upgrade here is an upgrade in trading systems and/or the launching of a new trading platform which significantly decrease latency. Those technological changes which do not significantly contribute to reducing latency are not considered as technological upgrades. While previous studies primarily focus on liquidity and volatility, we give attention to trading activity metrics. The reason for doing so is that, different from advanced markets, emerging stock exchanges expect significant improvements not only in market quality, but also in the overall level of trading activity. 2

In the current study, we focus on 10 emerging market exchanges which have upgraded their trading technologies. The motivation behind these upgrades has been, for the most part, to achieve a substantial increase both in total trading activity and in foreign interest, which necessitates a long-term perspective. Furthermore, by using stock-level daily data and relevant stock-related dependent variables, a better comprehension of trading activity patterns may be obtained. As such, we document the effect of such upgrades on spreads, volume and volatility. To the best of our knowledge, we provide one of the first studies in this particular field of research focused on emerging markets. By both utilizing broader exchange-level monthly data and stock-level daily data, we determine that liquidity and trading activity are positively affected by technology upgrades. Moreover, we find that upgrades decrease the level of volatility. After having controlled for global trends and stock-related variables, our results were found to be robust with different dates of upgrades further strengthening our results. 2. Literature As algorithmic and high-frequency trading increases and as the IT capacity of trading venues continues its rapid development, so has the importance of conducting such studies increased. Using either the time of introduction of new technologies or algorithmic and/or HFT as a proxy for technological development, these studies reveal the relationship between several variables -mainly liquidity, efficiency and volatility- and technological changes. Focusing on the introduction of new technologies is also useful when the reverse side is considered. If algorithmic and/or HFT 1 finds itself as the main concern, using the introduction of new technologies as an instrumental variable reflecting the algorithmic trading/hft activity solves the endogeneity problem which arises from the fact that such trading activity is also affected by market variables (Hendershott et al. (2011), Boehmer et al. (2014)). Difficulty in tracking the algorithmic trading and HFT data is, of course, another reason behind the decision to use such a variable. 1 To clarify the descriptions, it is worth noting one more time that high-frequency trading (HFT) is a subset of algorithmic trading, which refers to all trading activity performed through computer algorithms. As Hasbrouck and Saar (2013) suggests, the more structural difference lies in the nature of trading: We can categorize algorithmic activity as proprietary or agency. We consider HFT a subcategory of proprietary algorithms for which low latency is essential... Agency algorithms are used by buy-side institutions (and the brokers who serve them) to minimize the cost of executing trades in the process of implementing changes in their investment portfolios... Such proprietary algorithms try to make profit from the trading environment itself (as opposed to investing in stocks), employed by hedge funds, proprietary trading desks of large financial firms, and independent specialty firms. 3

There are a number of studies focusing on spreads as liquidity measure. Hendershott and Moulton (2011) document that, for the New York Stock Exchange, by increasing automation and by reducing the execution time for market orders from 10 seconds to less than one second, not only does the cost of immediacy (bid-ask spread) increase due to increased adverse selection, but so is the noise in prices reduced, thereby making prices more efficient. Hendershott et al. (2011), utilizing daily average data, also report that for large stocks in particular, algorithmic trading narrows spreads, reduces adverse selection, and reduces traderelated price discovery, which then improves liquidity and enhances the informativeness of quotes. Their sample includes 943 common stocks traded on the NYSE. Hasbrouck and Saar (2013) report that rise in the volume of low-latency trades lowers spreads and short-term volatility while also increasing the depth of the limit order book. The work of Riordan and Storkenmaier (2012) provides similar evidence from Europe. They examine the upgrade of the Deutsche Börse trading system with the release of Xetra 8.0, which decreases latency in the market. The observation period covers the intra-daily data; that is, 40 trading days prior to and after the release of the new trading system. Their study shows that decreasing latency leads to increased liquidity by decreasing bid-ask spreads while also improving the efficiency of prices. On the other hand, Gai et al. (2013) find that an increase in speed does not decrease the bidask spread. In their study, they analyze the results of increases in the speed of trading by utilizing the NASDAQ TotalView-ITCH and daily TAQ data. They find that although an increase in trading speed also increases message flow, the rise in message flow is due to an increase in order cancellations without any real increase in actual trading volume. They also find that as a result of more cancellations, market depth decreases and short-term volatility increases. Some papers provide evidence that volatility increases following upgrades or HFT activity. In his study, Zhang (2010) analyze the effect of high frequency trading on stock price volatility and price discovery in the US capital market. His study uncovers a positive correlation between HFT and stock price volatility after controlling for the volatility of a stock s fundamentals and other volatility drivers. It is also emphasized that HFT and price discovery are negatively correlated. Boehmer et al. (2014) reach a similar result for algorithmic trading. By concentrating on the effect of algorithmic trading intensity on equity market liquidity, short-term volatility and informational efficiency in 42 equity markets around the world, they 4

find that more intense algorithmic trading is associated with improved liquidity, improved efficiency and increased volatility. With this being said however, findings of other studies present a different picture for volatility in opposition to the above-mentioned findings. Hagströmer and Nordén (2013) distinguish between market making HFTs and opportunistic HFTs on NASDAQ OMX Stockholm, finding that market makers share in order and trade traffic is higher and that market making HFTs decrease intraday volatility. Focusing on the role of algorithmic trading on market efficiency in the Korean stock market, Seo and Chai (2013) find that algorithmic trading reduces the asymmetric volatility causing inefficiency of information. Moreover, algorithmic trading also increases the operational efficiency of the stock market. They also find that algorithmic trading provides liquidity for market participants contributing to friction-free transactions. Similarly, Easley et al. (2014) examine the impact on stock prices of a major upgrade to the New York Stock Exchange, reporting that the upgrade generated a relatively greater turnover and relatively lower transaction costs. There are two studies focusing on a single emerging country. In their study, Krishnamurti et al. (2003) argue that level of technology in India s two exchanges (NSE and BSE) positively affects the attention of small and medium investors. Dicle and Levendis (2013) study the effect of a technological upgrade on the Johannesburg Stock Exchange in 2002. Utilizing daily data for ten years, they find that while liquidity improves following the upgrade, the level of efficiency remains unchanged. They also argue that more independent market in terms of returns after the upgrade provides international investors more opportunities for diversification. 2 3. Data In order to study the effects of significant technological upgrades on market qualities, such as liquidity, trading activity and volatility, we use two different data sets for two different analyses. First of all, we explore the technological upgrades in equity markets of exchanges in 2 Our study also touches on another strand of literature which deals with a broader scope and which gives more attention to the volume of trading activity and cost of equity. Jain (2004, 2005), Torre et al. (2007), Li (2007), and Lagoarde-Segot (2009) are among those examining the country-level factors affecting the equity trading activity. Such studies utilize data aggregated at the exchange level (monthly in general) and search for the effects of several institutional factors. These cross-country studies mostly document the effects of introducing electronic trading into stock exchanges and are especially appropriate for the emerging markets. The introduction of electronic platforms was a structural shift in the history of stock exchanges, which has been widely researched in the literature. 5

emerging countries searching for from the beginning of 2005. If we are able to obtain the initial implementation date of the new technology, we add that market to our sample. Our final sample consists of 10 markets. The dates that technology upgrades were applied, along with other related information, are gleaned from the World Federation of Exchange Focus s monthly newsletter (Table 1). Table 1: Date of Technology Upgrades Exchange Date of Upgrade Technology Provider Bursa Malaysia 01.12.2008 NYSE Shenzhen SE 12.01.2009 STS (in-house) Colombian SE 09.02.2009 NASDAQ X-Stream Indonesia SE 09.03.2009 JATS-NEXT (by NASDAQ) Korea Exchange 23.03.2009 EXTURE (in-house) Shanghai SE 04.12.2009 NGTS (in-house) Johannesburg SE 02.07.2012 Millenium IT Mexican SE 03.09.2012 MoNet (in-house) Thailand SE 03.09.2012 SET CONNECT (in-house) Moscow Exchange 08.12.2012 Spectra The geographical dispersion of the sampled exchanges is thought to have a high level of representative power: There are two exchanges from the Americas, six exchanges from the Asia-Pacific, and one exchange from Africa and Eastern Europe, each. The exchanges in our sample are also diversified in terms of level of development and fields of specialization. Our study consists of two main parts. While the first part of the study utilizes monthly aggregate data of the sampled exchanges, the second part exploits these exchanges daily stock data. In the first part of our analysis, using the World Federation of Exchanges monthly database, we focus on the monthly traded value share (TVS) 3, turnover and volatility of exchanges themselves. For each of the exchanges mentioned above, which we call main exchange or main exchanges hereafter, we create a different control group consisting of 3 Traded value share (TVS) is defined as (trading value of exchange/ total trading value of WFE members). We calculate this measure in order to better capture the change in trading activity due to the technological upgrade by controlling the global change in trading activity. 6

five or six exchanges whose average level of trading values in 2005 are close to the exchange with which the main exchange in question is being tested (see Appendix A for main exchanges and control groups). Prior to performing our analysis, we obtain two monthly series, one of which belonging to main exchanges and the other consisting of the average of their corresponding control group, meaning that we had acquired 10 pairs before starting our analysis. The time series of these pairs are symmetric around the date of upgrade. For each of the pairs, we divide the time line into two separate periods of either 1, 2, or 3 years both before and after the main exchange upgraded its trading platform (hence, the total lengths of the time series are 2, 4, or 6 years). In the event that we have insufficient data for any case, we limit ourselves with the available data; e.g. if the date that the update was implemented is January 2012, then we would not have a sufficient time frame to construct a series of 6 years in length and would use only 2 or 4-year worth of data. To deepen the analysis, we focus on stock based data in the second part of our study. The second part of the analysis is based on selected individual stocks listed on the aforementioned main exchanges. We conduct a panel data analysis by using daily data sets which included 361 stocks from the 10 main exchanges listed above. Stocks listed on the main exchanges blue-chip index are included in the data set. If a blue-chip index contains more than 50 stockswhich is the case for two of ten exchanges-we sort the stocks according to their weights on the index and select the first 50 of them (see Appendix B for the number of stocks selected for each main exchange). The change on liquidity, trading activity and volatility after implementing the upgrade is tested by using stock based panel data. The Bloomberg terminal is used as the main data source in this part. The variables obtained are price, return, traded value, trading volume, bid-ask spread, market capitalization and market to book ratio. Each exchange s data set includes the daily close values of the variables both three years before and after the upgrade. If the post-upgrade period is less than 3 years for an exchange, we adjust the estimation windows accordingly, keeping the length of pre- and post-upgrade periods the same. We also obtain daily high and low values of prices again from the Bloomberg database. The daily closing values of the CBOE Volatility Index (VIX) and MSCI Emerging Index (MXEF) are also included. 7

4. Methodology 4.1. Exchange Based Analysis For the first part of the analysis, we follow two steps. In the first step, we check whether the average traded value share (TVS) and the average turnover of any exchange (for both the main exchanges and control group exchanges) during the post-upgrade period (1, 2, or 3 years after the upgrade) is greater than its value during the pre-upgrade period (1, 2, or 3 years before the upgrade). By doing this, we want to observe the trend in tested market qualities for each main exchange and its control group. Our first hypothesis is related to the change in average traded value share (TVS, average ratio of traded value of exchange to total traded value of WFE members) before controlling against the control group: Hypothesis 1a where represents average traded value share of exchange j at year t+n, t represents the date of upgrade, n is the preceding or succeeding year (n = 1, 2, or 3 years and j = 1,,10). In the second step, we test whether the difference between the main exchange and its control group, in terms of average TVS, during post-upgrade period was greater than its value during pre-upgrade period. This approach shows how the technological upgrade affects the respective values of the stated market qualities in the main exchanges. Hypothesis 1b where is the average TVS of the corresponding control group exchanges (n = 1, 2, or 3 years and j = 1,,10). Hypotheses 2a and 2b are constructed for our second variable turnover, which is calculated as dividing traded value to market capitalization, and analogous to 1a and 1b. 8

Hypothesis 2a where represents average turnover of exchange j at year t+n, t represents the date of upgrade, n is the preceding or succeeding year (n = 1, 2, or 3 years and j = 1,,10). Hypothesis 2b where is the average turnover of the corresponding control group exchanges (n = 1, 2, or 3 years and j = 1,,10). Lastly, we test for changes in volatility. In order to calculate volatility, we construct a model using the GARCH (1, 1) specification: (Eqn.1) where represents the benchmark index return of exchange j at day t, represents the return of MSCI Emerging Markets Index at day t, h t is the conditional variance, γ and φ are the ARCH and GARCH coefficients, is the dummy variable for the technology upgrade which takes value of 1 for the days succeeding the implementation of the new technology. Then, we form Hypothesis 3 accordingly: Hypothesis 3 where is the dummy variable for the technology upgrade which takes value of 1 for the days succeeding the implementation of the new technology. In order to ascertain the effects of global conditions on volatility, we add, as a control variable, the return series of the MSCI Emerging Markets Index in the mean equation. We add 9

the dummy variable of technology, dt, in the variance equation which takes the value of 1 for the post-upgrade periods. The dummy variable is added to the standard GARCH (1,1) variance equation in order to test the structural change in the unconditional variance (Ryoo and Smith (2004), Gökbulut et al. (2009)). In the event that the coefficient of this dummy is significant and negative, then we argue that volatility is lower in the post-upgrade period. 4.2. Stock Based Analysis In the second part of our study, we explore the nature of change in the market following the technological upgrade via a stock-by-stock analysis using Hendershott et al. (2011) as the main guide. We focus on the effect of technological upgrades on three market qualities; namely liquidity, trading activity and volatility by using daily individual stock data set. Firstly, in order to test the effect of technology on stock liquidity, we adopt the following panel regression model: (Eqn.2) where denotes the liquidity of stock i at day t. is the dummy variable, taking the value of 0 before the technology upgrade implementation date by the exchange i, and 1 afterward. X consists of the matrix of control variables, including price change of the stock on day t (P it - P it-1 ); daily volatility, calculated as ; and the daily close value of the VIX 4. The reason we choose to use this index as a control variable is that the VIX is known to be an appropriate indicator for global stress and reflects the volatility effect coming from outside of the market. While firm fixed effects are included in, day fixed effects are included in. We use relative quoted spread as the proxy for liquidity. For a given day t and stock i, the relative quoted spread, standardized by the quote midpoint, may be defined as: (Eqn.3) where refers to ask price, which is the minimum price that sellers are willing to receive for stock i at time t, refers to bid price, which is the maximum price that buyers are 4 VIX is the abbreviation of Chicago Board Options Exchange Volatility Index. It is a measure of implied volatility of S&P 500 index options. It is a representation of one measure of the market s expectations of stock market volatility over the next 30 day period. It is sometimes referred as fear index. 10

willing to pay for stock i at time t. We, in equation 2, expect the coefficient of T, β, to be negative and significant. Secondly, since, from among the various market qualities, it is on the trading activity of stocks that technological upgrades exert an effect, we include three different measures to gauge trading activity: trading volume, traded value and turnover. Trading volume is the total number of stocks traded. Traded value is the total number of shares traded multiplied by their respective matching prices. Turnover is the ratio of traded value to the market capitalization. To analyze the effects on these measures, we adopt the following panel data regression model, which we repeat for all three trading activity measures: (Eqn.4) where denotes the trading activity of stock i at day t and is the same variable included in Eqn.3. Y represents the matrix of control variables, including lagged returns of the stock for up to five days (following Chodra et al., 2007), the daily closing values of the VIX, and the market to book ratio. Thirdly, we focus on the effects of technology on volatility. In order to achieve this goal, we adopt the following model, which is similar to (Eqn.1) in terms of methodology: (Eqn.5) where is the daily returns of the stock i; is one-day lagged return of the VIX, which we add to capture the global volatility; and T is the dummy variable, taking the value of 0 before the exchange implemented the technology upgrade, and 1 afterward; h t is the conditional variance; and γ and φ are the ARCH and GARCH coefficients, respectively. For each of the variables referring liquidity, trading activity, and volatility, we conduct both a stock exchange based panel data analysis and an overall panel data analysis. A stock exchange based panel data means that we perform the three analysis denoted by Equations 2, 4, and 5 for each of the main exchanges. 5 We then perform an overall panel data analysis, meaning that, for each market quality, we combine all of exchanges data sets into a single data set then test the same equations on that new data set. All series are checked for stationarity and, when necessary, appropriate adjustments are made. 5 Recall that we have 10 main exchanges. However, the liquidity analysis is tested for 9 exchanges, with the exclusion of the Colombian Exchange, whose bid and ask values are not available for the specified time interval. 11

5. Results 5.1. Exchange Based Analysis As shown in Table 2, we find that, following a technology upgrade, an exchange s average traded value share (TVS) increased for 8 of the 10 exchanges over a two-year period (oneyear pre-upgrade and one-year post-upgrade). For the four- and six- year windows, we are able to calculate the t-statistics for 6 exchanges, observing that the average TVS increased for 5 exchanges over a four-year period, and 6 exchanges over a six-year period. When we repeat this analysis to measure differences between the main exchanges and their corresponding control groups, the number of exchanges in which the traded value share increased relatively is 5 over a two-year period, 4 over a four-year period, and 4 over a six-year period (See Appendix C for figures). Table 2: Difference Test Results for TVS Panel A: Results for Hypothesis 1a Exchange Traded Value Share 2-year 4-year 6-year Indonesian SE *** *** *** Control Group *** *** *** Colombia SE *** *** *** Control Group *** *** *** Johannesburg SE ** Control Group ** Korea Exchange *** *** *** Control Group (-) (-) (-) Bursa Malaysia ** (-) *** Control Group ** *** *** Mexican Exchange *** Control Group (-) Moscow Exchange (-) Control Group (-) Shanghai SE (-) ** ** Control Group (-) *** *** 12

Shenzhen SE *** *** *** Control Group *** *** *** Thailand SE *** Control Group (-) (8/10) (5/6) (6/6) *, ** and *** denote that null hypothesis is rejected at the 10%, 5% and 1% significance levels, respectively while (-) denotes that null hypothesis is not rejected at these levels. Panel B: Results for Hypothesis 1b Traded Value Share 2-year 4-year 6-year Indonesia SE -Control Group * (-) (-) Colombia SE -Control Group (-) ** *** Johannesburg SE -Control Group (-) Korea Ex. -Control Group *** *** *** Bursa Maleysia -Control Group (-) (-) (-) Mexican Ex. -Control Group *** Moscow Ex. -Control Group (-) Shanghai SE -Control Group (-) * ** Shenzhen SE -Control Group *** *** *** Thailand SE-Control Group *** (5/10) (4/6) (4/6) *, ** and *** denote that null hypothesis is rejected at the 10%, 5% and 1% significance levels, respectively while (-) denotes that null hypothesis is not rejected at these levels. The results for turnover, a trading measure adjusted for market capitalization, show weaker evidence (Table 3). Three of the 10 exchanges exhibit an increase in turnover following the upgrades over a two-year period, whereas, out of the 6 exchanges whose upgrade effects are able to be measured over a period of six years, only 1 over a four-year period and 2 over a sixyear period exhibit an increase in turnover. These results are robust to changes in control groups. Table 3: Difference Test Results for Turnover Panel A: Results for Hypothesis 2a Exchange Turnover 2-year 4-year 6-year Indonesian SE (-) (-) (-) Control Group ** ** (-) Colombia SE (-) (-) (-) Control Group ** (-) (-) 13

Johannesburg SE (-) Control Group (-) Korea Exchange (-) (-) * Control Group (-) (-) (-) Bursa Malaysia (-) (-) (-) Control Group ** ** ** Mexican Exchange *** Control Group (-) Moscow Exchange (-) Control Group (-) Shanghai SE (-) (-) (-) Control Group (-) (-) (-) Shenzhen SE *** *** * Control Group *** *** *** Thailand SE *** Control Group (-) (3/10) (1/6) (2/6) Panel B: Results for Hypothesis 2b Turnover 2-year 4-year 6-year Indonesia SE -Control Group (-) (-) (-) Colombia SE -Control Group (-) (-) (-) Johannesburg SE -Control Group (-) Korea Ex. -Control Group *** ** *** Bursa Maleysia -Control Group (-) (-) (-) Mexican Ex. -Control Group ** Moscow Ex. -Control Group (-) Shanghai SE -Control Group (-) (-) (-) Shenzhen SE -Control Group *** ** * Thailand SE -Control Group *** (4/10) (2/6) (2/6) *, ** and *** denote that null hypothesis is rejected at the 10%, 5% and 1% significance levels, respectively while (-) denotes that null hypothesis is not rejected at these levels. The final analysis for the first part of the study is on volatility. Here, utilizing daily return data, we show that the dummy variable for technological change is significantly negative, 14

implying that the upgrade results in a decrease in volatility for 7 of the 10 exchanges. Table 4 provides the results of GARCH estimations for each of the exchange. Table 4: Results of GARCH (1,1) Estimations (Eqn.1) Exchange Mean equation Variance equation Constant MSCI Emerging Index return Constant ARCH coefficient GARCH coefficient Technology dummy Indonesian SE 0.0907 *** 0.6233 *** -2.5949 *** 0.1745 *** 0.7924 *** -0.5573 *** Johannesburg SE 0.0399 *** 0.6303 *** -4.1241 *** 0.0963 *** 0.8867 *** -0.5865 ** Colombian SE 0.0502 ** 0.3935 *** -0.2764 *** 0.3070 *** 0.2997 *** -0.9549 *** Bursa Malaysia 0.0513 *** 0.2732 *** -3.5459 *** 0.1725 *** 0.7867 *** -0.5945 *** Mexican Exchange Moscow Exchange 0.0323 0.4866 *** -4.6550 *** 0.0680 *** 0.9245 *** -0.1914 0.0361 * 0.8519 *** -2.7112 *** 0.1558 *** 0.8262 *** -0.5326 ** Shanghai SE 0.0343 0.4710 *** -3.1945 *** 0.0486 *** 0.9376 *** -0.6670 *** Shenzhen SE 0.0730 ** 0.4297 *** -2.7158 *** 0.0632 *** 0.9180 *** -0.2431 Thailand SE 0.0833 *** 0.5254 *** -1.8483 *** 0.2736 *** 0.6510 *** -0.4885 *** Dependent variable is the daily return of the benchmark index for each of the exchange. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels, respectively. 5.2. Stock Based Analysis In order to explore the effects of the technological changes more precisely, we next perform a stock-level analysis with the daily data. Here, we again perform three analyses on liquidity, trading activity and volatility. We perform both a pooled analysis which included 361 stocks from the10 exchanges as well as separate analyses for each of the exchanges. Table 5: Results of the Pooled Panel Regressions Dependent Variable Effect of Technological Change Relative Quoted Spread (Eq.2) -0.0027*** Traded Value (Eq.4) 3.0002*** Trading Volume (Eq.4) 1.9349*** Turnover (Eq.4) 0.9519*** Volatility (Eq.5) -1.9484*** 361 stocks from 10 exchanges are included in the estimations. *, ** and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. 15

As shown in Table 5, we find that while the upgrades have a positive effect on trading activity measures in agreement with Easley et al. s (2014) findings, they have a negative effect on spreads. Our findings on spreads are in line with Riordan and Storkenmaier (2012) and Boehmer et al. (2013), whereas they contradict Hendershott and Moulton s (2011) findings. Riordan and Storkenmaier (2012) explain the reason behind their results differing with those of Hendershott and Moulton (2011) by pointing out that their study (2012) contains a microstructure change which may unintentionally encompass other effective factors, such as an increase in anonymous trading. We may argue that, as our sample consists of emerging markets, marginal effect of improvements in technical capacity on liquidity may be more significant with respect to the advanced markets. Our results also indicate that volatility decreases following the upgrades, which is in line with Hasbrouck and Saar (2013) whereas they contradict those of Hendershott & Moulton (2011) and of Boehmer et al. (2014). Hendershott and Moulton (2011) argue that increasing volatility reflects information being incorporated into prices more rapidly. On the other hand, Hasbrouck and Saar (2013) argue that high-frequency traders contribute to liquidity provision and that their activity appears to lower volatility. When we look at the stock-level results across the exchanges, our results show that spreads narrowed in 6 of the 9 exchanges 6 (Panel A of Table 6), the spreads widened in 2 exchanges, and remained unchanged in 1 exchange. Regarding trading activity, we find that although the traded value increased in all of the sampled exchanges, trading volume and turnover increased in 9 and 7 out of the 10 exchanges, respectively (Panel B of Table 6). Lastly, our results show that volatility decreased following upgrades in 5 of the 10 exchanges, increased in 1 exchange, and remained unchanged in 4 exchanges (Panel C of Table 6). Table 6: Results of the Panel Regression for Technology Upgrades across Exchanges Panel A: Dependent Variable: Relative Quoted Spread (Eqn.2) Exchange Beta-Coefficient Indonesian SE (-0.0058)*** Johannesburg SE 0.0001** 6 We were not able to find the spread data for the Colombian exchange. 16

Colombian SE - Korea Exchange (-0.0054)*** Bursa Malaysia (-0.0027)*** Mexican Exchange (-0.0015)*** Moscow Exchange 0.0002** Shanghai SE (-1.8)x10^-6 Shenzhen SE (-0.0007)*** Thailand SE (-0.0002)*** Pooled (-0.0027)*** *, ** and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. Panel B: Dependent Variable: Trading Activity Measures (Eqn.4) Exchange Beta-Coefficient Beta-Coefficient Beta-Coefficient (Dep.Var: Traded Value) (Dep.Var: Trading Volume) (Dep.Var: Turnover) Indonesian SE 2.5534*** 1.5131*** 0.7690*** Johannesburg 3.2364*** 0.0723*** 1.5943*** Colombian SE 0.5163*** 0.0615-0.0051 Korea Exchange 2.7622*** 1.2379*** 0.7722*** Bursa Malaysia 0.6936*** 0.3881*** 0.0350 Mexican Exc. 2.8385*** 0.2847*** 0.2575*** Moscow Exc. 2.7796*** 2.1370*** 0.0001 Shanghai SE 3.9460*** 3.2862*** 1.3729*** Shenzhen SE 3.5756*** 2.7381*** 1.4553*** Thailand SE 4.3712*** 3.3368*** 0.1128*** Pooled 3.0002*** 1.9349*** 0.9519*** *, ** and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. 17

Panel C: Dependent Variable: Volatility (Eqn.5) Exchange Coefficient Indonesian SE -0.6864*** Johannesburg 0.2293 Colombian SE -0.5633*** Bursa Malaysia -0.9106*** Mexican Exchange -0.3042 Shanghai SE -1.2003*** Shenzhen SE -0.9134*** Thailand SE 11.07105*** Pooled -1.9484*** *, ** and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. 6. Conclusion Utilizing the monthly and daily data series of 10 exchanges from 9 countries, we explore the effects of sophisticated technological upgrades on market structures. We detect 10 important upgrades in emerging markets over the last decade. These changes are expected to be game changers for these exchanges, reflecting their strategy of reaching higher levels of trading volumes together with more efficient markets. The effects of technological changes on stock markets have recently developed and there are a limited number of studies analyzing emerging markets. Our study is one of the first studies focusing on emerging markets and gives new evidence for ongoing debates. We find that spreads narrow following upgrades. We then find that, for both the exchange and stock-levels, trading activity increases after implementing upgrades. Lastly, we find evidence that volatility decreases following upgrades in some exchanges. Our findings on spreads confirm, for the most part, previous studies findings. On the other hand, our findings on volatility provide new evidence regarding the findings of previous literature, itself being a mixed issue. All results support the idea that technological capacity is worth the investment regarding the business environment of the stock exchange industry since it is shaped by competitive pressures exerted by other exchanges and new trading schemes such as algorithmic trading and HFT. 18

For the emerging markets, as they continue to develop their systems, it seems that the effects of technological changes will remain prominent in the near future. Hence, the different characteristics of these changes as well as how they compare with advanced markets may be valuable questions for future research. 19