Liquidity Measurement in Frontier Markets

Transcription

1 Liquidity Measurement in Frontier Markets Ben R. Marshall* Massey University Nhut H. Nguyen University of Auckland Nuttawat Visaltanachoti Massey University Abstract Frontier markets, which are countries that have not yet reached emerging market status, have been shown to provide diversification benefits for international investors. However, many stocks in these markets are thinly traded so liquidity is an important consideration. We investigate which liquidity proxies best measure the actual cost of trading in 19 frontier markets that can be accessed by foreign investors. We find that the Gibbs, Amihud, and Amivest proxies have the largest correlation with liquidity benchmarks, while the FHT measure provides the best measure of the magnitude of actual transaction costs. JEL Classification: G12, G15 Keywords: Frontier market, liquidity proxy, transaction cost First Version: October 17, 2012 This Version: October 30, 2012 Corresponding author: Ben Marshall, School of Economics and Finance, Massey University, Private Bag , Palmerston North, New Zealand. Tel: ext. 5402; Fax: Electronic copy available at:

2 Liquidity Measurement in Frontier Markets Abstract Frontier markets, which are countries that have not yet reached emerging market status, have been shown to provide diversification benefits for international investors. However, many stocks in these markets are thinly traded so liquidity is an important consideration. We investigate which liquidity proxies best measure the actual cost of trading in 19 frontier markets that can be accessed by foreign investors. We find that the Gibbs, Amihud, and Amivest proxies have the largest correlation with liquidity benchmarks, while the FHT measure provides the best measure of the magnitude of actual transaction costs. JEL Classification: G12, G15 Keywords: Frontier market, liquidity proxy, transaction cost 2 Electronic copy available at:

3 1. Introduction Frontier markets are attracting increased attention from investors and researchers. These markets, which are less developed than emerging markets, have low integration with the world market and thereby offer significant diversification benefits. (Berger, Pukthuanthong, and Yang (2011, p. 227). However, frontier markets are relatively illiquid. Marshall, Nguyen, and Visaltanachoti (2012) show spreads are, on average, over two and a half times larger in frontier markets than the U.S, while Minovic and Z ivcovic (2010, p. 33) find illiquidity and liquidity risk significantly impact price formation in the frontier market of Serbia. Correctly measuring and accounting for liquidity is clearly an important issue in frontier market research. We adopt a similar approach to Goyenko, Holden, and Trzcinka (2009) and Fong, Holden, and Trzcinka (2011) and run horse races between popular liquidity proxies in frontier markets to determine which measures have the largest correlations and lowest root mean squared errors with high frequency transaction cost benchmarks. As Goyenko, Holden, and Trzcinka (2009) note, studies often consider issues like the link between liquidity and returns using a particular liquidity proxy, without first verifying that the proxy in question is an accurate measure of liquidity. These authors note that tick data availability is a contributing factor. They state (p. 153) in many countries transaction data are not available at all. Frontier market research is in its infancy compared to developed and emerging market work. Most research to date has concentrated on the integration of frontier markets, on their correlations with other markets, and on the diversification these markets provide. Cheng, Jahan- Parvar, and Rothman (2010) show investment in nine North African and Middle Eastern markets result in diversification benefits to investors with global market exposure, Jayasuriya and Shambora (2009) find a U.S. investor would have earned higher risk-adjusted returns if they had 3

4 invested in five frontier markets, while Speidell and Krohne (2007) show frontier markets have lower correlations with the S&P 500 than emerging markets. Frontier market research in other areas includes de Groot, Pang, and Swinkels (2010) who show value and momentum effects prevail in frontier markets. There is one frontier market liquidity paper that we are aware of. Minovic and Z ivcovic (2010) measure liquidity using a proxy based around the prevalence of zero return days 1 and show liquidity and liquidity risk play an important role in price formation in Serbia. Our analysis is based on tick and low-frequency data for 19 frontier market countries for the period We use two metrics to determine the extent to which the liquidity proxies represent the liquidity benchmarks. The first is correlation analysis and the second is root mean squared errors. As Goyenko, Holden, and Trzcinka (2009) point out, the correlation results are likely to be of most interest to researchers in asset pricing who are seeking to answer questions such as the impact of liquidity on stock returns, while the root mean square error results should be most relevant to researchers in areas like market efficiency who require a proxy with an accurate scale. The transaction cost benchmarks are effective spread, quoted spread, and price impact. Each of these is calculated using tick data for all stocks within each frontier market. These data are sourced from the Thomson Reuters Tick History (TRTH) database. The liquidity proxies we test include Roll from Roll (1984), Gibbs from Hasbrouck (2004, 2009), Zeros and Zeros2 from Lesmond, Ogden, and Trzcinka (1999), FHT from Fong, Holden, and Trzcinka (2011), Amihud from Amihud (2002), Amivest from Amihud, Mendelson, and Lauterback (1997), and Pa st Stam from Pa stor and Stambaugh (2003). We also follow Goyenko, Holden, and 1 This measure has its origins in Lesmond, Ogden, and Trzcinka (1999) and Bekaert, Harvey, and Lundblad (2007). 4

5 Trzcinka (2009) and convert the spread proxies, such as Roll to price impact proxies by dividing by average daily dollar volume. We find the Amihud and Gibbs measures have the largest average correlations across the 19 frontier markets and the largest number of statistically significant country correlations with the effective spread and quoted spread benchmarks. Amivest, Roll, FHT, and Zeros also perform well. However, Zeros2 does not appear to be an effective proxy for either of the spread benchmarks. Amihud is the best performing proxy for price impact. However, the Roll Impact, FHT Impact, Gibbs Impact, Amivest, and Zeros Impact proxies also perform adequately. Neither Pa st Stam nor Zeros2 Impact correlates well with the price impact benchmark. The FHT proxy performs best in the root mean square error analysis. Its scale is, on average, the closest to the scale of both the effective and quoted spread benchmarks. However, the null hypothesis of a root mean square error of zero can be rejected for FHT as well as the other proxies. Therefore none of the proxies are consistently of a similar magnitude to the spread benchmarks. We hope the results from this paper will be useful for subsequent work that considers different aspects of liquidity. S&P (2007, p. 7) suggest frontier markets have become increasingly accessible to foreign investors, to the point where dedicating a significant investment to this space is feasible. Frontier markets are clearly of increased relevance to investors so subsequent research might like to consider important questions such as the role liquidity and liquidity risk plays in asset pricing in frontier markets, and whether the returns to momentum and value strategies in frontier markets withstand transaction costs. The remainder of this paper is organized as follows: The data, liquidity benchmarks and proxies, and method are described in Section 2. The results are presented in Section 3, and Section 4 concludes the paper. 5

6 2. Data and Methodology 2.1. Data The dataset is the same as that used in Marshall, Nguyen, and Visaltanachoti (2012). The 19 countries include: Argentina, Bahrain, Bulgaria, Croatia, Estonia, Jordan, Kuwait, Lebanon, Lithuania, Oman, Pakistan, Qatar, Romania, Serbia, Slovenia, Sri Lanka, the United Arab Emirates (UAE), Ukraine, and Vietnam. 2 These countries were all part of the MSCI Frontier Markets Index in March Each liquidity benchmark is calculated using high-frequency data from the Thomson Reuters Tick History (TRTH). These data are provided to us by the Securities Industry Research Centre of Asia Pacific (SIRCA). As noted in Fong, Holden, and Trzcinka (2011), TRTH data users include central banks, regulators, hedge funds, and investment banks Transaction Cost Benchmarks The transaction cost benchmarks we use are effective spread, quoted spread, and price impact. 3 Effective spread is calculated as per equation 1: Effective Spread = 2 ln - ln, (1) 2 MSCI previously classified three of these countries (Argentina, Jordan, and Pakistan) as emerging markets. 3 We only calculate each benchmark in the market session in each country. Some exchanges have a pre-market and post-market session but these are not included to ensure consistency. 6

7 where P k (M k ) is the price (midpoint of the bid and ask quotes) when kth trade occurs. We follow Goyenko, Holden, and Trzcinka (2009) and calculate monthly average effective spreads by weighting intraday spreads by dollar volume. Quoted spread is calculated as per equation 2: Quoted Spread = ( )/M k, (2) where,, and M k is the ask price, bid price, and midpoint of these two prices respectively. The monthly average quoted spread is, following Fong, Holden, and Trzcinka (2011), calculated by time weighting the intraday spreads. Five-minute price impact is calculated as per equation 3: Price Impact 2 ln ln when the k th trade is a buy, 2 ln ln when the k th trade is a sell, (3) where M k+5mins (M k ) are the midpoints five minutes after (at the time of the kth trade). We use the Lee and Ready (1991) algorithm to classify trades, and monthly averages are calculated using the same approach as for effective spreads. 7

8 2.3. Liquidity Proxies The first four measures are estimates of spread, while the remaining proxies are price impact measures. Following, Goyenko, Holden, and Trzcinka (2009), we test the ability of the Amihud and Amivest measures to proxy for both spread and impact benchmarks Roll Consistent with Goyenko, Holden, and Trzcinka (2009), the modified Roll (1984) effective spread estimator we use is in equation 4: Roll = 2 Cov P t, P t-1 when Cov P t, P t-1 < 0 0 when Cov P t, P t-1 0 (4) This measure assumes the price change serial covariances can be used to estimate the spread. Roll is calculated for each stock each month Gibbs The second proxy we consider is the Hasbrouck (2004, 2009) Bayesian Gibbs sampling Roll (1984) model which is generalized with the inclusion of a market factor. 5 4 The list of liquidity proxies we consider is not exhaustive. For instance, we do not test the related Holden and Effective Tick measures (see Holden (2009) and Goyenko, Holden, and Trzcinka (2009)) because the lack of information regarding several changes in tick rule in many exchanges. The Corwin and Schultz (2012) spread estimator is calculated due to incomplete high and low data in many countries. 5 We appreciate Joel Hasbrouck providing his code on his website: 8

9 p t =c q t +β m r m,t +u t (5) where is the log trade price, is a trade indicator, and c is the first autocovariance of price changes (half spread), and r m,t is the market return factor. Similar to Hasbrouck (2009), the priors are the normal distribution with zero mean and variance equal to for c and unit mean and unit variance for β m. The prior of residual variance follows the inverted gamma distribution with shape and scale parameters equal to We estimate each of the model parameters using Markov Chain Monte Carlo simulation in accordance with Hasbrouck (2009, p. 1451) Zeros Lesmond, Ogden, and Trzcinka (1999) propose that the frequency of days with zero returns gives an indication of liquidity. Less liquid stocks are said to be more likely to have no volume traded and a zero return. The zeros measure is in equation 6. Zeros # of days with zero returns / (6) where T represents the total number of trading days in a month. A modified version of the Zeros proxy, which is termed Zeros2 by Goyenko, Holden, and Trzcinka (2009) determines frequency of days with some volume having a return of zero. They suggest that stocks that are costly to trade provide less incentive to obtain private information, which suggests information is less likely to be reflected in the price of stocks with larger transaction costs. The Zeros2 measure is in equation 7: Zeros2 # of positive-volume days with zero returns / (7) 9

10 FHT Proxy equation 8: The FHT proxy comes from Fong, Holden, and Trzcinka (2011). This measure is given in FHT = 2σN -1 1+z (8) 2 where σ, z, and N -1 ( ) are the standard deviation of the commodity s daily returns, the proportion of zero returns, and the inverse of the cumulative normal function respectively Amihud The Amihud (2002) proxy is given in equation 12. Amihud = r t Volume t (12) where r t and Volume t is the is the return and dollar volume on day t respectively Amivest 13. Amihud, Mendelson, and Lauterback (1997) use the Amivest measure which is in equation Amivest = Volume t r t (13) 10

11 P stor and Stambaugh The Pa stor Stambaugh (2003) or gamma liquidity measure is in equation 14: 1 = θ + φr t + sign( )Volume t + ε t (14) where r t ( ) is the stocks return (excess return, r t - r mt ) on day t, sign( ) is one (zero) if is positive (negative), and Volume t is the dollar volume on day t Other Impact Proxies Consistent with Goyenko, Holden, and Trzcinka (2009) we convert the spread proxies to price impact proxies by dividing by average daily dollar volume Method We follow Goyenko, Holden, and Trzcinka (2009) and assess each liquidity proxy using two distinct metrics. The first is correlation analysis which involves determining the correlation between a liquidity benchmark and a liquidity proxy. We first calculate the average monthly liquidity benchmark and proxy for each stock. We then calculate the equal-weighted average market liquidity benchmark and proxy each month. The final step involves calculating the timeseries correlation of these monthly observations. The second aspect of the analysis is calculating root-mean squared errors between the liquidity benchmark and proxy. As noted by Goyenko, Holden, and Trzcinka (2009), the correlation results should be of particular interest to asset pricing researchers, while the root-mean square error results are likely to be seen as the most important by market efficiency researchers for whom the scale of each proxy is particularly important. 11

12 3. Results 3.1. Benchmark and Proxy Means The mean effective and quoted spread benchmarks and liquidity proxies are given in Table 1. Most countries have a reasonable number of stock month observations. These range from 406 in Lebanon to 18,657 in Sri Lanka. The average effective spread across the 19 frontier markets is 4.84%. This compares to an average effective spread across 41 developed and emerging exchanges of 2.3% reported by Fong, Holden, and Trzcinka (2011). 6 The average frontier market quoted spread is 7.04% compared to 2.9% in Fong, Holden, and Trzcinka (2011). There is considerable variation in the mean effective and quoted spreads across the various frontier markets. The average effective (quoted) spread in Qatar is just 1.33% (2.00%), yet in Ukraine the averaged effective (quoted) spread is 12.50% (20.22%). The frontier market sample includes seven Gulf countries and 11 non-gulf countries. The Gulf countries have certain similarities, such as a focus on oil exports, so we calculate the measures separately for these two groups. This should give researchers focusing on Gulf countries insight into the most appropriate liquidity metrics for these countries. Gulf markets tend to be more liquid. The average effective and quoted Gulf spreads are 2.87% and 4.34% respectively compared 5.99% and 8.61% respectively for non-gulf countries. Figure 1 shows the mean effective spreads through time. Both the Gulf and non-gulf spreads are volatile. This was particularly evident during the global financial crisis. The average Roll, Gibbs, and FHT measure across all markets is 1.58%, 1.94%, and 2.53%. All three proxies have a larger average for non-gulf than Gulf countries, which is consistent with the spread benchmarks. There is evidence that each of the measures tend to be 6 Argentina is the only country that is in both the Fong, Holden, and Trzcinka s (2011) sample and our own. Argentina was previous classified as an emerging market by MSCI. Fong, Holden, and Trzcinka s (2011) report an average effective spread for Argentina of 2.3% which is very similar to the 2.36% we report. 12

13 larger in countries with higher benchmarks. However, the difference in the size of these measures in countries with low and high transaction costs is not as pronounced as it is in the spread benchmarks. For instance, the Gibbs proxy ranges from 0.81% in Vietnam to 5.03% in the Ukraine. Table 2 shows that the average price impact across all countries is 1.66%. The Gulf country average is 1.30% and the non-gulf country average is 1.87%. The variation in price impact across countries is not as pronounced as the variation in both effective spread and quoted spread. Bahrain has the smallest average price impact (0.93%) and the Ukraine has the largest (2.96%). [Insert Table 1 About Here] [Insert Table 2 About Here] [Insert Figure 1 About Here] 3.2. Correlations The correlations of each proxy with the effective spread benchmark are presented in Table 3. We follow Goyenko, Holden, and Trzcinka (2009) and also include the Amihud and Amivest proxies in this analysis. Correlations that are statistically significantly different to zero are in bold. Correlations that are statistically significant and have the correct sign are in bold with a box around them. The Table 3 results show that Gibbs and Amihud have the largest and most consistent correlation with the effective spread benchmark across the 19 frontier markets. The Gibbs proxy has the largest average correlation (0.515) and has a statistically significant positive correlation in 15 of the 17 countries it can be calculated for, while Amihud has the second largest average 13

14 correlation (0.432) and a positive statistically significant correlation in 15 of the 19 countries. Gibbs and Amihud are the best performers in both Gulf and non-gulf markets, although their relative rankings differ. Gibbs has the larger average correlation in non-gulf countries, while Amihud has the largest average correlation in Gulf countries. Many of the individual country correlations are large. The largest Gibbs correlation is in Pakistan and the largest Amihud correlation is in Kuwait. Amivest, FHT, Roll, and Zeros also do a good job of measuring liquidity. They each have average correlations that have the correct sign and are statistically significant different to zero. These average correlations range from (Zeros) to (Amivest). Note that a positive relation between liquidity in a benchmark and the Amivest proxy results in a negative correlation due to the construction of the Amivest variable. Each of these four proxies has a statistically significant correlation with the benchmark which is of the correct sign in 11 or 12 of the countries. Zeros2 is the only proxy that does not accurately capture liquidity. This proxy frequently has a negative correlation. The correlations are of a similar magnitude to those reported by Fong, Holden, and Trzcinka (2011) for developed and emerging markets. The global average they report for the effective spread benchmark ranges from to This suggests that liquidity proxies correlate with transaction cost benchmarks in frontier markets as well as they do in developed and emerging markets. [Insert Table 3 About Here] Both the Gibbs and Amihud proxies have the largest correlations with the quoted spread benchmark. These correlations are larger than the effective spread equivalents in Table 3. Gibbs has an average correlation of and is statistically significant and positive in all 17 of the 14

15 countries it can be calculated in, while Amihud has an average correlation of and is statistically significant in 17 of the 19 countries. The results are also similar to Table 3 in that Roll, FHT, Zeros, and Amivest have average correlations that are statistically significantly different to zero. These range from for Zeros to for Amivest. Roll, FHT, and Zeros have statistically significant correlations with the correct sign in 11 or 12 of the 19 countries while Amivest has correlations that have the correct sign and are statistically significant in 14 of the 19 countries. Zeros2 is again the poorest performer. It has just one positive statistically significant correlation, seven negative statistically significant correlations. The average correlation across all 19 countries is also negative. [Insert Table 4 About Here] The price impact correlations in Table 5 are similar to those in Table 3 and 4 in that Gibbs and Amihud have the largest average correlations. There are however a number of differences. Firstly, the correlations are lower than those for the quoted and effective spread benchmark. Secondly, the Gibbs measure is Gibbs Impact which is Gibbs from Tables 3 and 4 scaled by dollar volume. Thirdly, unlike the Table 3 and 4 results, Amihud has a slightly larger correlation. The average Amihud correlation is compared to for Gibbs Impact. Amihud has a positive and statistically significant correlation in 14 of the 19 countries, while Gibbs Impact correlation is statistically significant and positive in 12 of the 17 countries it can be calculated for. Roll Impact, FHT Impact, Amviest and Zeros Impact all have an average correlation with the correct sign that is statistically significant. The number of instances of these proxies having a statistically significant correlation at the country level ranges from 10 to 12. Neither the Pa stor and 15

16 Stambaugh (2003) nor Zeros2 Impact measure has an average correlation coefficient that is statistically significantly different to zero. [Insert Table 5 About Here] Researchers in areas such as market efficiency are likely to be interested in the magnitude of each proxy relative to the benchmark. Not all proxies (e.g. Amihud) are intended to be of the same size as the benchmark, but for those that are (Roll, Gibbs, and FHT) we calculate the rootmean square errors (RMSEs) and measure whether the null hypothesis that the root mean square error is zero can be rejected. The RMSEs, which are presented in Table 6, show that FHT has the lowest average RMSE across all countries, Gulf countries, and non-gulf countries versus both the quoted spread and effective spread benchmarks. Gibbs has the next lowest, while Roll has the largest RMSE. The average FHT RMSE is 4.81% versus the effective spread and 6.33% versus the quoted spread benchmark. These RMSEs are larger than the global developed and emerging market average reported by Fong, Holden, and Trzcinka (2011) of 2 3%, but they are similar in magnitude to the size of the benchmarks themselves, which is broadly consistent with Fong, Holden, and Trzcinka (2011). We conclude that a researcher who needs an estimate of the level of transaction costs should use FHT, however it is important to note that even the FHT RMSE is statistically significantly different from the benchmarks. [Insert Table 6 About Here] 16

17 4. Conclusions Frontier markets, which are countries that have not been given emerging markets status, have been shown to offer investors with diversification benefits. This has resulted in increased attention for these markets by researchers and their accessibility to investors has led to an increase in the funds allocated to these markets by the investment community. However, relatively little is known about many important aspects of these markets. Our contribution is documenting which liquidity proxies best measure the actual cost of transacting in frontier markets. These markets are relatively illiquid so it is important that researchers accurately measure liquidity in their work. We form effective spread, quoted spread, and price impact transaction cost benchmarks from high frequency data for 19 frontier markets and then conduct horse races of various liquidity proxies including Roll from Roll (1984), Gibbs from Hasbrouck (2004, 2009), Zeros and Zeros2 from Lesmond, Ogden, and Trzcinka (1999), FHT from Fong, Holden, and Trzcinka (2011), Amihud from Amihud (2002), Amivest from Amihud, Mendelson, and Lauterback (1997), and Past Stam from Pa śtor and Stambaugh (2003). We also convert the spread proxies, such as Roll to price impact proxies by dividing by average daily dollar volume following Goyenko, Holden, and Trzcinka (2009). We find Gibbs and Amihud have the largest correlations. Researchers who need to focus on movements in liquidity, such as those in asset pricing, should focus on these two measures. However, they can also rely on Roll, Zeros, Amivest, FHT, Roll Impact, FHT Impact, Gibbs Impact, and Zeros Impact as these all perform well. The FHT proxy consistently is consistently closest to the spread benchmarks in size. However, even this proxy has a scale that is statistically significantly different to the benchmarks. 17

18 References Amihud, Yakov. 2002, Illiquidity and stock returns: Cross-section and time-series effects. Journal of Financial Markets 5, Amihud, Yakov, Mendelson, Haim., and Beni Lauterbach, 1997, Market microstructure and securities values: Evidence from the Tel Aviv Stock Exchange, Journal of Financial Economics, 45(3), Bekaert, Geert, Harvey, Campbell R., and Christian Lundblad, 2007, Liquidity and expected returns: Lessons from emerging markets. Review of Financial Studies, 20(5), Berger, Dave, Kuntara Pukthuanthong, and J. Jimmy Yang, 2011, International diversification with frontier markets, Journal of Financial Economics, 101, Cheng, Ai-Ru, Mohammad R. Jahan-Parvar, and Philip Rothman, 2009, An empirical investigation of stock market behaviour in the Middle East and North Africa, Journal of Empirical Finance 17, Corwin, Shane, and Paul Schultz, 2012, A simple way to estimate bid ask spreads from daily high and low prices, Journal of Finance, 67(2), de Groot, Willma., Pang, Juan, and Laurens Swinkels, 2010, Value and momentum in frontier emerging markets. SSRN working paper, Fong, Kingsley, Craig Holden, and Charles Trzcinka, 2011, What are the best liquidity proxies for global research? SSRN working paper, Goyenko, Ruslan, Craig Holden, and Charles Trzcinka, 2009, Do liquidity measures measure liquidity? Journal of Financial Economics, 92, Hasbrouck, Joel., 2004, Liquidity in the futures pits: Inferring market dynamics from incomplete data. Journal of Financial and Quantitative Analysis, 39, Hasbrouck, Joel., 2009, Trading costs and returns for U.S. equities: Estimating effective costs from daily data. Journal of Finance, 65(3), Holden, Craig., 2009, New low-frequency spread measures. Journal of Financial Markets, 12, Jayasuriya, Shamila A., and William Shambora, 2009, Oops, we should have diversified! Applied Financial Economics, 19, Lee, Charles, and Mark Ready, 1991, Inferring trade direction from intraday data, Journal of Finance, 46,

19 Lesmond, David A., James P. Ogden, and Charles A. Trzcinka, A new estimate of transaction costs. Review of Financial Studies, 12, Marshall, Ben R. Nhut H. Nguyen, and Nuttawat Visaltanachoti, 2012, Frontier Market Transaction Costs and Diversification. SSRN working paper, Minovic, Jelena Z, and Bos ko R. Z ivcovic, Open issues in testing liquidity in frontier financial markets: The case of Serbia. Economic Annals, 55(185), Pa stor, L ubos, and Robert F. Stambaugh., 2003, Liquidity risk and expected stock returns, Journal of Political Economy, 111, Roll, Richard., 1984, A simple implicit measure of the effective bid-ask spread in an efficient market. Journal of Finance, 39, Speidell, Lawrence, and Axel Krohne, 2007, The case for frontier equity markets, Journal of Investing, Fall, Standard & Poor s, 2007, Frontier markets: investment rationale, accessibility and risks (Standard and Poor s, New York). 19

20 Table 1 Spread Benchmark and Proxy Means Start Date No. Stock Months Effective Spread Quoted Spread Roll Gibbs Zeros Zeros2 FHT Argentina , Bahrain , Bulgaria , Croatia , Estonia , Jordan , Kuwait , Lebanon Lithuania , Oman , Pakistan , Qatar , Romania , Serbia , Slovenia , Sri Lanka , UAE , Ukraine , Vietnam , Overall Average Gulf Average Non Gulf Average Table 1 contains mean spread benchmarks and proxies. Roll is from Roll (1984), Gibbs is from Hasbrouck (2004, 2009), Zeros and Zeros2 are from Lesmond, Ogden, and Trzcinka (1999), and FHT was developed by Fong, Holden, and Trzcinka (2011). 20

21 Table 2 Price Impact Benchmark and Proxy Means Price Impact Amihud Amivest Past Stam Roll Impact FHT Impact Zeros Impact Zeros2 Impact Gibbs Impact Argentina 1.41% 3.4E E E E E E E E+03 Bahrain 0.93% 8.4E E E E E E E E+03 Bulgaria 2.65% 1.6E E E E E E E E+04 Croatia 1.66% 2.0E E E E E E E E+03 Estonia 0.99% 1.5E E E E E E E E+03 Jordan 1.49% 1.9E E E E E E E E+03 Kuwait 1.25% 2.0E E E E E E E E+02 Lebanon 1.19% 2.3E E E E E E E E+04 Lithuania 1.81% 6.9E E E E E E E E+00 Oman 1.58% 3.8E E E E E E E E+03 Pakistan 1.90% 1.8E E E E E E E E+03 Qatar 1.09% 1.2E E E E E E E E+01 Romania 1.92% 1.0E E E E E E E E+04 Serbia 2.03% 4.0E E E E E E E E+00 Slovenia 1.51% 7.7E E E E E E E E+05 Sri Lanka 2.31% 2.7E E E E E E E E+02 UAE 1.53% 2.3E E E E E E E E+02 Ukraine 2.96% 4.1E E E E E E E E+05 Vietnam 1.26% 3.6E E E E E E E E-02 Overall Average 1.66% 8.6E E E E E E E E+04 Gulf Average 1.30% 1.3E E E E E E E E+03 Non Gulf Average 1.87% 1.3E E E E E E E E+04 Table 2 contains the median of the price impact benchmark and proxies. Amihud is from Amihud (2002), Amivest is from Amihud, Mendelson, and Lauterback (1997), Past Stam is from Pa stor Stambaugh (2003), and the other proxies are as per Table 1 except they are scaled by dollar volume. 21

22 Table 3 Effective Spread Correlations Roll FHT Gibbs Zeros Zeros 2 Amihud Amivest Argentina Bahrain Bulgaria Croatia Estonia Jordan Kuwait Lebanon Lithuania Oman Pakistan Qatar Romania Serbia Slovenia Sri Lanka UAE Ukraine Vietnam Overall Average Gulf Average Non Gulf Average Table 3 contains spearman correlations for the effective spread benchmark and liquidity proxies. Monthly averages are calculated for each stock, cross-sectional averages are then generated for each country, and the correlations of these observations through time is generated. Correlations that are statistically significantly different to zero at the 10% level or more are in bold. Correlations that are statistically significant and have the correct sign are in bold with a box around them. 22

23 Table 4 Quoted Spread Correlations Roll FHT Gibbs Zeros Zeros 2 Amihud Amivest Argentina Bahrain Bulgaria Croatia Estonia Jordan Kuwait Lebanon Lithuania Oman Pakistan Qatar Romania Serbia Slovenia Sri Lanka UAE Ukraine Vietnam Overall Average Gulf Average Non Gulf Average Table 4 contains spearman correlations for the quoted spread benchmark and liquidity proxies. Monthly averages are calculated for each stock, cross-sectional averages are then generated for each country, and the correlations of these observations through time is generated. Correlations that are statistically significantly different to zero at the 10% level or more are in bold. Correlations that are statistically significant and have the correct sign are in bold with a box around them. 23

24 Table 5 Price Impact Correlations Roll Impact FHT Impact Gibbs Impact Amihud Amivest Past Stam Zeros Impact Zeros2 Impact Argentina Bahrain Bulgaria Croatia Estonia Jordan Kuwait Lebanon Lithuania Oman Pakistan Qatar Romania Serbia Slovenia Sri Lanka UAE Ukraine Vietnam Overall Average Gulf Average Non Gulf Average Table 5 contains spearman correlations for the price impact benchmark and liquidity proxies. Monthly averages are calculated for each stock, cross-sectional averages are then generated for each country, and the correlations of these observations through time is generated. Correlations that are statistically significantly different to zero at the 10% level or more are in bold. Correlations that are statistically significant and have the correct sign are in bold with a box around them. 24

25 Table 6 Root-Mean Square Errors Versus Quoted Spread Versus Effective Spread Roll Gibbs FHT Roll Gibbs FHT Argentina Bahrain Bulgaria Croatia Estonia Jordan Kuwait Lebanon Lithuania Oman Pakistan Qatar Romania Serbia Slovenia Sri Lanka UAE Ukraine Vietnam Overall Average Gulf Average Non Gulf Average Table 6 contains root mean squared errors. Those that are statistically significantly different from zero at the 10% level or more are in bold. 25

26 Figure 1 Mean Effective Spreads (%) All Gulf Non Gulf