Market Frictions and the Geographical Location of Global Stock Exchanges. Evidence from the S&P Global Index

We examine the impact of trading costs on investor average holding periods for the S&P global 1200 index. We report overwhelming evidence that global equity indices cannot be pooled. When we differentiate between stock indices based on their geographical location, we discover that for companies listed in the USA, Europe, Canada and Australia, there are no market frictions enabling continuous high frequency trading. However, companies listed in Latin America and Asia face significant barriers to trading in the form of transaction costs. We ascertain that the geographical location of stock markets plays a vital role in achieving international portfolio diversification. JEL Classifications: G10, C33.


Introduction
Investment theory is based upon the principle of minimizing risk through portfolio diversification. This encourages investors to spread their risk by investing in multiple assets. Therefore, from a theoretical perspective, investment in a global index can achieve diversification leading to higher risk adjusted returns. In reality this is not achieved because trading costs provide a stumbling block in trading a large number of equities. This is why there are numerous market microstructure studies that focus on the importance of trading costs in financial markets (see among others Stoll (1978), Atkins and Dyl (1997), Boinet et al (2008), and Florackis at al (2011)).
The academic literature computes the 'optimal holding period', for investors in the presence of new information by balancing the costs of portfolio matching with the gains from trading. Amihud and Mendelson (1986) and Wilcox (1993) provide the solution for the optimal holding period by an explicit proportional relationship between investors' holding periods, in the presence of transaction costs, and the arrival of new information. Atkins and Dyl (1997) and Gregoriou and Ioannidis (2006) undertake extensive econometric analysis that relate the bid-ask spread (used as a proxy for trading costs) to investors' holding periods. Their analysis involves all the stocks traded in the New York and the London Stock Exchange over the periods 1975-1989 and 1990-1999 respectively. In the context of static linear models, they find strong support for the postulated positive relationship between transaction costs as measured by the bid-ask spread and the holding period. Boinet et al (2008) discover that the speed of adjustment increases the greater the deviation of the holding period of an investor from its optimal value in the London Stock Exchange. They also find that for heavily traded stocks, even small misalignments of the holding period from its 'optimal' value, trigger trading. However, trading costs prevent such rapid adjustment in less liquid securities.
We contribute to the literature on holding periods and trading costs in a number of ways.
First, we re-examine the empirical association between investor average holding periods and trading costs in a global context by analyzing stocks listed on the S&P Global 1200 Index, which covers approximately 70% of global market trading. This enables us to test if trading costs affect international portfolio diversification in a global framework rather than in just a US or a UK context. In our opinion this is a vital piece of research as we are the first study to witness if trading costs have an impact on international portfolio diversification, which is the fundamental focus of modern portfolio financial theory.
Second, given that the S&P Global 1200 Index consists of seven regional indices we are able to assess the pooling assumption of average holding periods and trading costs for indices across various geographical locations. In particular, we can determine whether you can cluster stocks listed in the USA, Canada, Latin America, Europe, Japan, Asia and Australia into the same empirical specification. This is a very important element to test because the pooled effects may not even provide consistent estimates of the average (Pesaran and Smith, 1995). This enables us to witness if the geographical location of stocks is a fundamental factor in achieving international portfolio diversification.
Third, unlike prior research we overcome contemporaneous correlation, endogeneity and jointly determination of investor average holding periods and trading costs by employing the Generalized Method of Moments (GMM) system panel estimator established by Blundell and Bond (1998) on our data. This makes our empirical estimates robust and therefore more reliable than the previous empirical research in this area conducted by Atkins and Dyl (1997) and Gregoriou and Ioannidis (2006).
Fourth, we undertake our empirical analysis on a very large comprehensive dataset consisting of 1200 stocks over a fifteen year period. Previous research used an extensively shorter time horizon and completed their research in 1999. We believe that the empirical literature requires updating in light of the evolution of high frequency trading (see among others Johnson and Jain (2006)) and the global financial crises that occurred in 2008. This is because high frequency trading could imply that trading costs are not a significant factor in deterring trading volume, which is contrary to earlier research. On the other hand, the financial crises might have caused a loss in confidence in equity markets thus decreasing trading volume, which may possibly lead to greater market frictions. Given these two significant issues occurring in financial markets over the last fifteen years, we believe that research in this field should provide more recent econometric analysis.
We provide empirical evidence that trading costs are a significant determinant in increasing investor average holding periods, when we look at the companies listed on the S&P Global 1200 Index as a whole. This supports the previous literature conducted on the US and UK equity markets in isolation.
When we proceed to test the validity of the pooling assumption of trading costs in global financial markets we discover some very interesting results. The objective is to determine if we can group all global data together into one testable model to compute the empirical relationship between trading costs and investor average holding periods of the 1200 firms listed on the S&P Global Index. We find overwhelming evidence that firms listed on the S&P 1200 Global Index cannot be pooled into a single regression model. Given this finding, we then assess whether the seven regional indices that make up the Global Index can be pooled.
The test results suggest that the pooling hypothesis cannot be rejected for companies listed on the S&P 500 index (US), TSX 60 Index (Canada), ASX 50 (Australia) and the S&P 350 Europe Index (Eurozone markets including Denmark, Norway, Sweden, and Switzerland; and the S&P United Kingdom). This result suggests that we can run a single econometric specification of each regional index for investor average holding periods and trading costs for companies listed in the USA, Canada, Australia and Europe. We therefore proceed to compute a regression where we attempt to explain investor average holding periods with trading costs for investment of companies in the USA, Europe, Canada and Australia. Our results imply that trading costs no longer provide a market friction to trading suggesting that high frequency trading can be utilized in order to achieve international portfolio diversification. From a practical point of view, fund managers can trade continuously to rebalance the portfolios of clients when they receive information to ensure that investors are full diversified against financial risk.
In addition, we find evidence that for companies listed on the S&P Latin America 40 Index (Mexico, Brazil, Peru, Chile, Colombia), S&P/TOPIX 150 Index (Japan) and the S&P Asia 50 Index (Hong Kong, Korea, Singapore, Taiwan) the data can also be pooled into one estimation for each regional index. Further econometric analysis reveals a positive and significant relationship between investor average holding periods and trading costs for these companies suggesting that transaction costs do provide significant obstacle in accomplishing international portfolio diversification.
Our results are robust to the global financial crises of 2008 and to the econometric problems of contemporaneous correlation, endogeneity and joint determination which are present in the data. Furthermore, our findings hold when we look at institutional trades, made up prominently of insurance, pension, mutual and investment funds which dominate the stock market.
Our findings indicate that the geographical location of exchanges could be a leading factor in preventing international portfolio diversification in global stock markets. This could indicate that stock indices located in Latin America and Asia have a liquidity problem. This could lead to the requirement of specialist market makers to provide liquidity for these equity markets, by allowing trading to occur outside the market makers' ask and bid quotes, in order to achieve stock market liquidity. 1 The rest of the paper is organised as follows. The following section discusses the econometric specification; Section 3 discusses data and the tests of poolability; Section 4 presents the empirical results; and Section 5 summarises and concludes.

Econometric Specification
In order to conduct our empirical analysis we follow the mainstream literature on average holding periods and liquidity by estimating a similar linear specification to Atkins and Dyl (1997) and Gregoriou and Ioannidis (2006), which models the average holding period ( ) H holding period and a set of stock characteristics, of the form: In order to formally test the explanatory variables for endogeneity, a Hausman (1978) test for the hypothesis that the explanatory variables are strictly exogenous is performed. If the null hypothesis is rejected, it leads to the conclusion that the explanatory variables in equation (1)  In order to accommodate endogeneity and the possibility of joint determination we employ a GMM system of equations in first differences and levels to estimate equation (1). 3 The estimation of the systems of equations simultaneously using the GMM system should be Where it X is a vector of the explanatory variables, and z represents the lag structure in the econometric model. 5

Data
In this study we collect monthly data from Datastream Advance for all the firms that are listed on the S&P Global 1200 Index over the time period of 2000-2014, resulting in 216,000 firm-year observations. This is a near comprehensive dataset given that the S&P Global 1200 Index was launched on 25 th October 1999. We will now show the procedure used to obtain all the variables displayed in equation (1).

The average holding period, ( ) H
We calculate the average holding period for firm I at time period T as

Shares outstanding for firm I in month T Average Holding Period
Trading volume for firm I in month T IT = Thus, the average holding period of each firm's investors for each month is computed by dividing the number of outstanding shares in the firm by the firm's monthly trading volume. 4 The time-varying matrix of instruments for the first difference GMM estimator can be observed in Blundell and Bond (1998). 5 The Three Stage Least Squares panel estimator also estimates a system of equations simultaneously and is regarded as an alternative to the GMM system estimator. However, we implement the GMM system estimator, given that it accommodates for the possibility of joint determination of an equation system with different instruments for different equations (Cornwell et al (1992)).
This average holding period, observed ex post, is a proxy for the average investors' ex ante investment horizon. The computation of investors' average holding period is only a crude approximation of investors' time horizons, because a particular firm's investors are unlikely to hold the firm's shares for the same length of time.
The bid-ask spread, S Datastream Advanced provides the bid and ask quotes originally used to compute the bid-ask spread for our research. The average monthly bid-ask spread for each stock in the data set is computed with the use of the formula proposed by Atkins and Dyl (1997) That is, the average spread for each stock I for each month T is computed as follows: Once all the data has been collected for all the 1200 companies listed on the S&P Global 1200 Index over the time period of 2000-2014, we place each of the firms in our sample in one of the following seven stock indices in order to test the possible heterogeneity in stock market liquidity.
Consists of the 500 leading companies with respect to market capitalization in the US economy. The index is widely used as a proxy for the US stock market as represents approximately 75% coverage of U.S. equities.

S&P Europe 350.
Consists of the 350 leading companies with respect to market capitalization in the European region. The index represents 70% of the region's market capitalization spanning seventeen exchanges.

S&P/TSX 60.
Consists of the 60 leading companies with respect to market capitalization in for Canada.

S&P/TOPIX 150.
Consists of the 150 leading companies with respect to market capitalization in the Tokyo market.

S&P/ASX Australian 50.
Consists of the 50 largest index-eligible Australian securities listed on the ASX.

S&P Asia 50.
Combining coverage of Hong Kong, Korea, Singapore and Taiwan, S&P Asia 50 measures four major markets in Asia. This index provides coverage of the 50 largest cap, liquid constituents of each of these key countries in Asia.

S&P Latin America 40.
The S&P Latin America 40 index includes the 40 largest market value securities from major sectors of the Brazil, Chile, Mexico and Peru equity markets. 6

Liquidity Heterogeneity
We investigate data poolability through the tests of parameter homogeneity. We estimate Equation (1) and test the null of parameter ( 1 π ) for equality of bid-ask spreads (our liquidity proxy) for all the firms listed on the S&P 1200 Global Index and between the firms listed on the seven regional indices based on geographical location that make up the S&P 1200 Global Index. We explicitly test poolability across these categories because we want to empirically examine the impact of the geographical location of indices on the liquidity of financial markets in a global context. If the null hypothesis is not rejected across the sample of categories, then this forms a basis for pooling the seven regional indices that form the S&P 1200 Global Index, because this essentially implies homogeneity in the average holding period of stocks and liquidity within each regional index that belongs to the S&P 1200 Global Index. We then test for the null of group-wise error homocsedasticity treating the liquidity of each regional index as a separate entity. A rejection of group-wise homoscedasticity indicates that the liquidity of each regional index heterogeneity is dynamic.
Chow-type F tests under the null of parameter equality across liquidity between the S & P 1200 Global Index and the seven regional indices based on their geographical location explained in Section 3.1 are reported in Tables 1 and 2. In Table 1 we report the results for the S&P 1200 Global Index as a whole. There is overwhelming econometric evidence that firms listed on the S&P 1200 Global Index cannot be pooled due to their significant differences in liquidity. The LM tests of group-wise homoscedasticity are also reported in Table 1, which confirm that error variances across all the firms in terms of liquidity are significantly different from each other (i.e., heteroscedastic). This implies that a single regression examining the empirical association between the liquidity and average holding periods is not applicable for the firms listed on the S&P 1200 Global Index.

[INSERT TABLE 1 HERE]
Motivated by the construction of the S&P 1200 Global Index, we investigate the poolability of firms that are listed on the index based on their geographical location. The regional indices results can be viewed in Table 2. We witness some very interesting and innovative econometric results. We find strong evidence which accepts the notion that firms listed within the seven regional indices (

Main Results
Eventhough the parameter equality results displayed in Table 1 suggest that we cannot pool the firms listed on the S&P 1200 Global Index, for completeness we initially report the results for all the 1200 firms listed on the S&P Global Index. We do this to detect the impact of our findings if we did not undertake the extensive econometric analysis concerning parameter heterogeneity of liquidity. Table 3 Tables 4 and 5 respectively.

[INSERT TABLES 3, 4 AND 5 HERE]
From Tables 3-5  for first order serial correlation is insignificant, which suggests that the panels do not suffer from serial correlation. The Jarque-Bera normality test indicates that the residuals of the models are normally distributed, implying that the empirical estimates obtained are not due to any outliers in the data. The Sargan tests confirm the validity of the instruments in all GMM system models. Finally, we witness that for all panel estimations the control variables for firm size and volatility have the hypothesized signs and are significant at all conventional levels.
We find that investors have longer average holding periods for stocks associated with larger firms that are less risky. This agrees with the previous research conducted by Atkins and Dyl (1997) and Gregoriou and Ioannidis (2006).
We observe some very interesting findings regarding the role of market frictions. Table 3 reports the results assuming a homogenous panel where we pool all the S&P 1200 Global Index into one signal panel estimation. We detect that the bid-ask spread is positive and significant when we analyse the S&P 1200 Global Index as a whole. This reaffirms the previous literature which states that market frictions cause a reduction in trading, resulting in longer average holding periods of stocks for investors.
However, once we allow for a heterogeneous panel based on our findings from Section 3.2, our results become very striking indeed. Table 4 reports the results of the companies listed on the S&P 500 index (US), TSX 60 Index (Canada), ASX 50 (Australia) and the S&P 350 Europe Index. We find that the bid-ask spread is insignificant suggesting that trading costs do not provide a stumbling block to international portfolio diversification, as they do not possess a significant impact on the frequency of trading.

This is an extremely prominent result as it suggests that investors should always trade when
there is new information, which supports the notion of the recent developments in financial markets along the lines of high frequency trading. Furthermore, our results do not agree with the previous research conducted on the US and UK equity markets, where bid-ask spreads were found to be positive and significant. We believe that the differences can be explained through the evolution of high frequency trading demonstrated by the dramatic increases in the turnover of stocks listed on highly traded stock exchanges as reported in Florackis at al (2011). Our findings indicate that the geographical location of stock exchanges is a important element in where individuals choose to distribute their wealth. This is because if they invest in Latin America or Asia, they cannot rebalance their portfolios as easily then if they trade in North America, Europe or Australia. This possible prevention of international portfolio diversification in equity markets depending on their location can have significant long term economic growth problems in global stock markets. This could be because stock exchanges located in Latin America and Asia have a liquidity problem. As a consequence of this, they may require specialist market makers to provide liquidity for these equity markets, by allowing trading to occur outside the market makers' ask and bid quotes, in order to achieve stock market liquidity.

Robustness Tests
In order to validate our findings further we undertake the econometric analysis displayed above on large (block) transactions defined in the US (Madhavan and Cheng, 1997) and UK (Gemmill, 1996) stock exchanges as transactions of 10,000 shares or more in a single trade.
We believe this is a good robustness test as block trades involve mostly institutional trades that essentially drive the stock exchange. In order to save space will place all the empirical results on block trades into Table 6. We can see from Table 6 that the results displayed in Tables 3-5 remain intact, suggesting that the geographical location of stock exchanges plays a vital role in portfolio diversification for insurance, pension, mutual and investment funds. 9

[INSERT TABLE 6 HERE]
Our results have implications for operations research. This is because the influence of liquidity is becoming an emerging research area in the operational research field. For example Mercurio (2001) derives the bid-ask spread for a portfolio of assets and bonds. Albanese and Tompaidis (2008) show that trading costs provide a significant market friction to investor hedging strategies. Castellano and Cerqueti (2014) compute optimal portfolios for investors faced with thinly traded stocks characterized by their lack of liquidity. One of the shortcomings of the research conducted on liquidity in the operational research discipline is that is based purely on a theoretical framework. Therefore, we feel that an empirical piece of research showing the importance of trading costs in a global framework complements and extends the previous literature in the finance stream of operational research.

Conclusion
In this paper we initially examine the influence of trading costs on investor average holding periods in a global framework, by conducting our analysis on all the companies listed on the S&P Global 1200 Index as a whole. The results provide evidence that trading costs increase investor average holding periods of common stocks which agrees with financial theory and with the previous research undertaken on the US and UK equity markets in isolation.
When we proceed to test the validity of the pooling assumption of trading costs in global financial markets we discover some very interesting results. Our evidence suggests that we are able to compute a regression where we attempt to explain investor average holding periods with trading costs for investment of companies in the USA, Europe, Canada and Australia. Our results imply that trading costs no longer provide a market friction to trading suggesting that high frequency trading can be utilized in order to achieve international portfolio diversification. From a practical point of view, fund managers can trade continuously to rebalance the portfolios of clients when they receive information to ensure that investors are fully diversified against financial risk.
In addition, we find overwhelming evidence that for companies listed on the S&P Latin America 40 Index (Mexico, Brazil, Peru, Chile, Colombia), S&P/TOPIX 150 Index (Japan) and the S&P Asia 50 Index (Hong Kong, Korea, Singapore, Taiwan) the data can also be pooled. Further econometric analysis reveals a positive and significant relationship between investor average holding periods and trading costs for these companies suggesting that transaction costs do provide significant residence in accomplishing international portfolio diversification.
Our results are robust to the global financial crises of 2008 and to the econometric problems of contemporaneous correlation, endogeneity and joint determination, which are present in our data. Furthermore, our findings hold when we look at institutional trades, made up prominently of insurance, pension, mutual and investment funds which dominate the stock market.
Our findings specify that the geographical location of exchanges could be a leading factor in preventing international portfolio diversification in global stock markets. This could indicate that stock indices located in Latin America and Asia have liquidity problems. This could lead to the requirement of specialist market makers to provide liquidity for these equity markets, by allowing trading to occur outside the market makers' ask and bid quotes, in order to achieve stock market liquidity. Finally, avenues for future research could be directed upon alterative liquidity measures to the bid-ask spread. In particular, the price impact ratios of Amihud (2002) and Florackis et al (2011) could be implemented to provide further robustness to our empirical findings.

TABLES TABLE 1: Heterogeneous Liquidity Effects of Trading Costs on Average Holding Periods of Common Stocks listed on the S&P 1200 Global Index
The specification is where i represents the companies within the index and t denotes the monthly time period; i α captures the time-invariant unobserved average holding period firm-specific fixed effects (e.g., differences in the initial level of investor average holding periods), and the t b captures the unobservable individual-invariant time effects (e.g., stock market shocks that affect all investors