CHAPTER 6 DETERMINANTS OF LIQUIDITY COMMONALITY ON NATIONAL STOCK EXCHANGE OF INDIA

Transcription

1 CHAPTER 6 DETERMINANTS OF LIQUIDITY COMMONALITY ON NATIONAL STOCK EXCHANGE OF INDIA 6.1 Introduction In the previous chapter, we established that liquidity commonality exists in the context of an order-driven emerging market, India. Although there is an extensive research to document the presence of liquidity commonality, the literature on what drives liquidity commonality is still at a nascent stage. The sources of liquidity commonality can be due to the microstructure effects or due to the market conditions. Sources due to market microstructure effects are due to the sources that commonly affect inventory costs and asymmetric information of stocks simultaneously (CRS, 2000). Alternatively, the sources due to market conditions effect liquidity commonality due to the co-movement in market states. The co-movement in market states is due to the common variation in supply or/and demand for liquidity in the market (Coughenour and Saad, 2004; Amihud, Mendelson, and Perdersen, 2006; Karolyi, Lee, and van Dijk, 2012). Although the above two strands are developed for the quote-driven markets, they can be extended to the order-driven markets (Brockman and Chung, 2002). We fill this important gap in the literature by examining the sources of liquidity commonality for an order-driven emerging market. In the present chapter, we examine various sources of liquidity commonality. First among them related to microstructure effects is asymmetric information. Asymmetric information can exist both at the market level and/or the industry level. When a market participant comes across a buyer or seller who has private information about a company, then he tends to lose by trading with informed trader as his knowledge regarding the stock is inferior.

2 Similarly, if there is new information in the market that affects all the stocks in the market, covariation of liquidity across stocks tends to occur. For the quote driven markets, the tests of correlation between trading frequency and commonality in liquidity show that the trading frequency of an individual stock positively impacts liquidity proxies (CRS, 2000; Huberman and Halka, 2001). In line with quote-driven markets, asymmetric information can be a driver of liquidity commonality. Trading volume contains relevant information regarding informed trading and informed traders split their orders into small to medium size trades to hide their existence. This process of tactfully breaking up orders containing information results in empirically positive relation between the trading frequency and their degree of information. This results in both a market-wide and industry-wide trading frequencies having common components contributing to liquidity commonality (Brockman and Chung, 2002). However, we do not consider inventory costs as a fundamental source of liquidity commonality as an order-driven market is devoid of market makers who incur inventory costs. After examining asymmetric information as a source of liquidity commonality, we examine market conditions related sources of liquidity commonality. Following Karolyi, Lee, and van Dijk (2012), we group them into supply-side and demand-side factors of liquidity commonality. When there is uncertainty in the market about the fundamentals, the providers of liquidity are forced to liquidate their positions across many assets to recover from losses. This results in declining of market liquidity and leads to further losses for the intermediaries creating an illiquidity spiral. This decrease in market liquidity or increase in volatility results in commonality in liquidity. This supply-side hypothesis predicts that commonality is higher during high market volatility, higher interest rates in the economy, and poor financial market conditions such as low liquidity, negative market returns, etc. affecting the availability of capital to the

3 financial intermediaries (Brunnermeier and Pederson, 2009; Karolyi, Lee, and van Dijk, 2012). The demand-side explanation for sources of commonality in liquidity for an emerging country like India mainly lies in the intense trading by institutional investors. The trading by various institutional investors such as foreign institutional investors, mutual fund institutions, banking and insurance companies is correlated to a large extent. The reason being that rise in institutional trading may result in an increase in correlated trading across many stocks leading to a common selling or buying pressure and hence increased variation in commonality in liquidity (Koch, Ruenzi, and Starks, 2010). When market participants are constrained by source of capital to trade, the market experiences a large negative return which in turn reduces the amount of funds tied up with tradable securities resulting in a decrease of liquidity supply in the market (Brunnermeier and Pederson, 2009). Hence, we examine the behavior of commonality in liquidity due to change in overall market returns, especially due to large negative market returns. We regress monthly liquidity commonality on the market returns, controlling for return volatility, and bank return volatility which is a proxy of supply constraint. We examine this relation between liquidity commonality and market returns both at the market level and size-based portfolio level and in particular focusing on large market declines leading to financial crisis.

4 6.2 Literature Review Existing literature views liquidity as an individual firm phenomenon and hence each firm has its own liquidity and also, the factors affecting liquidity are distinct for each individual stock. The main determinants of liquidity of individual stocks are order flow, trading frequency, stock returns, return volatility, and volume traded (Stoll, 1978). CRS (2000) shift the research from single asset focus to a market-wide context. CRS (2000) is the first paper to suggest that asymmetric information and inventory costs are two main sources of liquidity commonality related to microstructure effects. These can impact liquidity commonality as inventory holding cost, information asymmetry cost, and order processing cost are three main determinants of costs to attract market makers to provide liquidity to the market (Stoll, 1978; Copeland and Galai, 1983). Hence, if there is a common component to this liquidity cost, changes in this will cause market-wide effects and impact individual stocks causing liquidity commonality. Even though the common components of any of these costs may impact commonality, CRS (2000) examine inventory and asymmetric information costs and find asymmetric information has common underlying determinants. Similar to CRS (2000), Harford and Kaul (2005) examine the commonality in order-flow to explore the determinants and implications on trading. Their main focus is on the impact of index-inclusion as well as market-wide and industry-wide order flow, returns, and trading costs. They find strong commonality in order flow and returns for the index stocks. They also find commonality in order flow for non-indexed stocks which are due to industry-wide and marketwide effects, though, they are statistically insignificant. Similar to Hasbrouck and Seppi (2001) and contradicting CRS (2000) their results suggest that commonality in order flow or returns are stronger than commonality in trading costs. Similarly, Hughen and McDonald (2006) find that

5 the trading by retail investors is a significant determinant of commonality across stocks. This may be due to the fact that retail investors are distinctly sensitive to market factors resulting in a commonality in order flow and trading. Their results confirm the earlier findings that commonality in order flow significantly affects liquidity. Karolyi, Lee, and van Dijk (2012) provide a comprehensive understanding of supply-side and demand-side sources of liquidity commonality at the global level. Their hypothesis is to test how and why level of liquidity commonality variation among stocks within a country is different from other countries and over time. Their sample is taken from 40 countries out of which 21 are developed and 19 are emerging market nations. Their sample consists of 27,447 stocks for a period of 15 years. They use Amihud s price impact liquidity measure at daily and monthly interval as a liquidity proxy. They find only one supply-side factor market volatility having a significant impact on commonality and commonality is higher in countries where market volatility is higher. There are several demand-side factors significant in explaining cross-country variation in commonality. Commonality in liquidity is lower for countries where there is high mutual fund ownership, and international investors. It is higher in countries where there is a less transparent environment and weak investor protection laws. The results show that commonality increases during periods of high fluctuations in returns. There is weaker evidence for funding constraints with respect to the supply-side factors. For the demand-side changes in co-variation of trading activity, globalization, presence of foreign investors, and investor sentiment play a significant role.

6 From the above literature review and the gaps identified, we specifically examine the determinants of liquidity commonality such as asymmetric information, supply-side, demandside, impact of index-inclusion on liquidity commonality. Apart from the time-series determinants, we examine the cross-sectional determinants of liquidity commonality. Finally, we examine the relationship between liquidity commonality and market returns empirically. 6.3 Data and Methodology We use high frequency intraday transactions and order-book snapshot data for equity market for a period of two years from April, 2010 to March, 2012 to examine the index inclusion hypothesis. The transactions data has record of all transactions that took place for the period under study. For the stocks, the trade data comes in a single file with information regarding each and every transaction (with time stamp) happened on that day. NSE collects the snapshots data of the limit order book at four different time periods of the day at 11 A.M., 12 noon, 1 P.M., and 2 P.M. which gives detailed information regarding the quotes (with time stamp) placed by various market participants on that particular day. Similarly for the options, the trade & snapshot data is obtained for all the option series except that limit order book snapshot data is collected at five different time periods of the day as provided by NSE at 11 A.M., 12 Noon, 1 P.M., 2 P.M., and 3 P.M. The operating time of stock and options market is synchronized from 9.15 A.M. to 3.30 P.M. for our sample period. 1 For our sample period, the number of stocks traded on NSE is Following prior literature, we apply certain data filters for our equity dataset. Our first filter is to delete all those stocks with a price less than Rs. 10 which results in a sample size of We apply this filter 1 The stock market follows a dual auction mechanism of call and continuous auction with call auction at the opening from AM and continuous auction throughout the day from PM.

7 following CRS (2000) to avoid any contaminating effect of tick size. CRS (2000) and Fabre and Frino (2004) argue that stocks with infrequent trades do not provide reliable information for estimating commonality coefficients. To remove less frequently traded stocks, we delete all those stocks with less than 40 per cent active trading days over our sample period resulting in a sample size of 1404 firms. Using the criteria followed by NSE to identify less liquid stocks, we delete all those stocks with an average daily trading volume less than 10,000 shares and number of trades less than 50 in a quarter which reduces our final sample to 981 firms. 2 For testing the supply and demand-side sources of liquidity commonality we use daily data for a period of 12 years from to construct quarterly measures of liquidity commonality. As the time period is very long and we don t have the intra-day data for this long period we depend on Amihud s liquidity measure (LIQ) to capture liquidity commonality. We use the R 2 of regressions of the individual stock liquidity on market liquidity to compute the liquidity commonality measure. First we perform the following filtering regression for each stock J based on observations on each day d within each month t: Here Dum k is the weekly dummy for controlling seasonality. We have lagged liquidity measure as an explanatory variable and take the estimated residuals of daily liquidity as our interest lies in examining if the changes in individual liquidity of firms co-move. We use the innovations from EQ1 to obtain quarterly measures of liquidity commonality denoted by R 2 liq for each firm by making use of R 2 from the following regressions, using daily observations within a quarter: 2 NSE uses this criterion to separate out less liquid stocks and they are traded under a different window.

8 Where is the sum total of estimated market residuals from EQ1 computed as market value weighted mean of the estimated residuals for all the firms in the sample excluding the firm in question. We also include one lead and one lag market residuals. This measure capturing commonality is not appropriate to use as a dependent variable in the regressions to follow because its value ranges between 0 and 1. So, we apply the logistic transformation of the commonality measure (Morck, Yeung, and Yu, 2000) as shown below: Here, is the monthly liquidity commonality for all the stocks in the sample. The measure is constructed in a similar fashion for different size-based portfolios. 6.4 Results and Discussion Summary Statistics Table 6.1 shows the descriptive statistics of the monthly time-series variables. The monthly mean market liquidity commonality is 0.23 and the maximum is 0.6. The mean monthly market return for our sample period is 1.16% with a standard deviation of 7.58%. The minimum monthly market returns reported is % and the maximum is 24.74%. The mean market capitalization of the firms is Rs Million and the maximum is Rs Million. The mean monthly market turnover is and the maximum is a high of The mean BankReturns are double that of the mean market returns. The BrokerReturns reported are higher those of the BankReturns. The mean net FII flow for the sample period is 6.58%, whereas the MF flow is a negative This shows that mutual funds are net sellers and FIIs show interest in Indian markets over our sample period.

9 Table 6.1: Descriptive Statistics of Monthly Time-series Variables Variable Mean Std Min Max LiqCom Market Return (%) Market Volatility (%) Market Cap (Rs. Million) Market Turnover ExchangeRate Exports (Rs. Million) Net % FII Flow Net % MF Flow CP Spread Short-term Interest Rate (%) BrokerReturns (%) BankReturns (%) Year-wise Commonality in Liquidity To examine market-wide commonality in liquidity for equity market each year between 2001 and 2012, we follow CRS (2000) and run market model time series regressions for each stock in each year. We regress the percentage change in individual Amihud stock liquidity measure on the percentage change in market liquidity measure. The market liquidity measure is an equally weighted average liquidity of all stocks in the market excluding the stock under examination. We exclude the stock liquidity from market liquidity measure to eliminate any cross-sectional dependence in the estimated coefficients. The market returns are estimated in the same excluding the return of the stock in question. The market model time series regression is = ( denotes the percentage change in Amihud liquidity measure used in the study on a given day t for a firm j. is the concurrent change in the

10 corresponding average market liquidity measure. We also include a lag and lead market liquidity variables in EQ3 to capture any nonsynchronous change in liquidity due to thin trading. Crosssectional means of time series slope coefficients are reported with the t-statistics to test the null hypothesis that there is no market-wide commonality in liquidity for stocks listed on NSE in line with Fama-Macbeth (1973). The concurrent, lag and lead market return along with idiosyncratic firm volatility act as control variables for the model. These control variables help to segregate the impact of changes in market-wide liquidity on an individual firm's liquidity after taking into account market-wide price changes and idiosyncratic volatility. Year Table 6.2: Year-wise Commonality in Liquidity for NSE Listed Stocks No. of Firms Concurrent beta t-stat for concurrent beta Percentage Positive Percentage Positive and significant Sum t-stat for Sum Note: Market-wide commonality in liquidity for actively traded stocks in each year is estimated by regressing percentage change in the Amihud liquidity measure on the percentage change in equallyweighted market liquidity measure on a daily basis. The equally-weighted market average measure excludes the liquidity of the dependent variable stock. Cross-sectional mean of the time-series slope coefficients are reported in the Fama-Macbeth fashion with the corresponding t-statistics in the parentheses. Concurrent, lag, and lead refer to same day, previous and the next trading day of the market liquidity measure. 'Percentage Positive' reports the percentage of positive slope coefficients. 'Percentage Positive and Significant' reports the percentage of positive coefficients significant at 5% level. Sum reports the sum of concurrent, lag, lead coefficients.

11 The results are shown in Table 6.2. There is an increase in the number of actively traded stocks on NSE during our sample period. The number of firms in the sample in 2001 is 318 and the same for 2012 is The concurrent beta coefficient is around 0.9 for the entire sample period. The percentage of firms having positive beta coefficients increased from 76% to 87% over the sample period. The number of firms with a positive and significant beta coefficient is around 50% for the sample period. These results prove that commonality in liquidity exists on the equity market of NSE Asymmetric Information and Liquidity Commonality To test for the impact of asymmetric information on liquidity commonality, we make use of the following regression model: Where measures the percentage change in the transaction frequency which is the overall trades for the firm on a given day. ( the equally-weighted transaction frequency of all the firms in the sample for the market (industry) except firm except the industry (except the firm). 3 For the options market, the specification is similar to EQ4 except that the transaction frequency is calculated for the options. We run firm by firm time-series regressions and the average coefficients for the stock and options markets are reported separately in Table 6.3 and Table 6.4 respectively. We test hypotheses 2a and 2b by examining the significance of and for stock and options markets separately. 3 We include the trading frequency at the industry level too in EQ4 because CRS (2000) argue that information asymmetry may be present at the industry or market level in the form of technological advancements.

12 Table 6.3: Commonality and Asymmetric Information for Equity Market Ntrades (Market) Mean Coefficient Ntrades (Market and Industry) Market Mean Industry Mean Coefficient Coefficient Concurrent (20.39) (7.22) (3.37) % Positive % +ve Significant % Negative % -ve Significant Lag (-2.06) (-0.95) (0.47) % Positive % +ve Significant % Negative % -ve Significant Lead (2.94) (-1.72) (2.49) % Positive % +ve Significant % Negative % -ve Significant Sum (19.85) (5.86) (3.60) Adj. R-Squared Mean (%) Note: The table presents results for asymmetric information as a determinant of liquidity commonality for the NSE equity market. Percentage change in daily trading frequency of each of the stock is regressed in time series on the percentage change in equally-weighted average of trading frequency for all the stocks in the market (as well as market and industry). The equally-weighted average of market excludes the industry and industry excludes the firm in question. The concurrent, lag, and lead coefficients are estimated and the mean cross-sectional time-series slope coefficients are reported similar to Fama-Macbeth methodology with the associated t-statistics in the parenthesis.

13 The mean concurrent coefficient for the market-wide transaction frequency is and is statistically significant with a t-statistic of The percentage of firms with a positive coefficient is 84.96% and 55.93% of the firms have a significant and positive concurrent coefficient. This is 25% higher than that reported by Brockman and Chung (2002) for the Hong Kong market. The sum of concurrent, lag, and lead coefficients is and highly significant with a t-statistic of When the analysis is performed for market and industry, market-wide concurrent coefficient is and the industry concurrent coefficient is which suggests that asymmetric information at the market level is stronger than that of the industry-wide asymmetric information. Also, the percentage of firms with a positive and significant concurrent coefficient for the market is 42.37% and that of industry is 37.71%. It can be inferred from these results that, if transaction frequency is a reliable measure of asymmetric information, there exist a common underlying source in transaction frequency both at the market and industry level indicating that asymmetric information is a possible source effecting liquidity commonality. Similarly, for the options markets, we observe that the market-wide commonality in liquidity denoted by the concurrent mean coefficient is 1.72 which is higher than that of 1.05 reported for the equity market. The number of firms with positive and significant commonality coefficient is 86.71% and 55.94% respectively. However, there are no firms with a negative and significant coefficient. This indicates that asymmetric information is a significant source of liquidity commonality for the options market. When we examine the asymmetric information as a source of liquidity commonality for call and put options separately, the mean estimated coefficient is for call options which is higher than that of reported for the put options. The number of firms with a positive and significant coefficient is 55.24% for the call options which is higher than 34.42% that of put options. Overall, these results show that

14 asymmetric information is a significant factor contributing to liquidity commonality for the options market. Table 6.4: Commonality and Asymmetric Information for Options Market All Options Put Options Call Options Mean Estimated Coefficient Mean Estimated Coefficient Mean Estimated Coefficient Concurrent (5.49) (1.98) (5.31) % Positive % +ve Significant % Negative % -ve Significant Lag (-0.99) (1.31) (-0.68) % Positive % +ve Significant % Negative % -ve Significant Lead (-3.36) (-1.75) (-2.78) % Positive % +ve Significant % Negative % -ve Significant Sum (9.02) (2.89) (8.03) Adj. R-Squared Mean (%) Note: The table presents results for asymmetric information as a determinant of liquidity commonality for the NSE options market. Percentage change in daily trading frequency of each option is regressed in time series on the percentage change in equally-weighted average of trading frequency for all the options in the market (as well as calls and put separately). The equallyweighted market average excludes the firm in question from the market average. The concurrent, lag, and lead coefficients are estimated and the mean cross-sectional time-series slope coefficients are reported similar to Fama- Macbeth methodology with the associated t-statistics in the parenthesis.

15 6.4.4 Supply-side Determinants of Liquidity Commonality Following Karolyi, Lee, and van Dick (2012), we examine the time-series behavior of supplyside sources of liquidity commonality. Monthly liquidity commonality ( ) for all stocks in the sample is estimated as described in section 6.3. We use the following time-series for the analysis: SInt is the short-term interest rate (%) which is the 91-day treasury-bill rate. CPSpread is the commercial paper spread, is the equally-weighted average returns of the brokerage industry. BankReturns is the equally-weighted average returns of the banking stocks listed on NSE. The above four variables serve as proxies of supply-side sources of liquidity commonality. Along with the supply-side sources, we include four other market conditions factors; Market return (, market liquidity, market volatility (, and market turnover ( as additional regressors, since these factors impact liquidity of a stock (Karolyi, Lee, and van Dick, 2012). We test hypotheses 3a.1, 3a.2, 3a.3, and 3a.4 by examining the significance of respectively of EQ5. Table 6.5 provides the regression results for EQ5. Model (1) shows that out of the four capital market conditions variables, market volatility and market turnover impact liquidity commonality positively with significance at 5% level. However, market volatility has a higher impact on commonality compared to market turnover. These results support the argument that an increase in volatility and turnover result in an increase in liquidity commonality. Market returns has a negative and market liquidity has a positive impact on commonality, though they are

16 insignificant. The directions of the coefficients reported support prior literature in explaining determinants of liquidity. Table 6.5: Supply-Side Determinants of Liquidity Commonality Model (1) (2) (3) (4) (5) (6) (7) Capital Market Conditions Market Return (-0.90) (-1.31) (-1.36) * (-1.96)) *** (-2.83) ** (-2.41) *** (-3.05) Market Liquidity (0.61) (0.30) (-0.14) (0.59) 1.31 (0.93) (0.06) (0.29) Market Volatility 0.267*** (4.76) 0.252*** (4.48) 0.250*** (4.45) 0.263*** (4.74) 0.277*** (5.07) 0.242*** (4.32) 0.256*** (4.61) Market Turnover 0.001** (2.08) 0.001** (2.23) 0.002*** (2.73) 0.001** (2.05) 0.001* (1.70) 0.001** (2.23) 0.001** (2.06) Supply-Side Factors SInt * (-1.72) * (-1.70) (-1.52) CPSpread * (-1.81) (-0.73) (-0.94) Broker Returns 0.003* (1.89) 0.002* (1.81) Bank Returns 0.012*** (2.79) 0.011** (2.58) Constant (-0.58) (-0.27) (-0.76) (-0.64) (-0.35) (-0.40) (-0.18) Adj. R-squared F-Value 11.83*** 10.19*** 10.29*** 10.36*** 11.48*** 8.13*** 8.80*** N Note: This table reports the monthly time-series regressions of supply-side determinants of liquidity commonality for the period Jan-2001 to Dec-2012 (144 months). Equally-weighted liquidity commonality for all stocks in each month is estimated by using Amihud liquidity measure and the methodology of Karolyi, Lee, and van Dick (2012). The capital market condition factors effecting liquidity commonality are used as the additional regressors in the model. The supply-side factors used in the study are Short-term interest rate, commercial paper spread (CP spread), broker returns, and returns of the banking industry. The regression coefficients are reported in the table along with the associated t-statistics in the parenthesis. *, **, *** indicate significance at 10%, 5%, and 1% respectively. We use models (6) and (7) separately as Broker returns and Bank returns are correlated.

17 Models (2) to (5) provide the evidence for impact of each of the source on liquidity commonality. The results suggest that short-term interest rate and CP Spread negatively, though have weak significant impact on commonality as an increase in interest rates decreases supply of limit orders and hence less trading activity leading to a decrease in commonality. However, an increase in broker returns or bank returns impact liquidity commonality positively. In models (6) and (7), we test the joint effect of the supply-side sources of liquidity commonality. As we find multicollinearity between broker returns and bank returns, we do not include them together in the models. Overall, the results show that bank returns, broker returns, and short-term interest rate significantly impact commonality. Also, in the presence of the supply-side sources, the coefficient on market returns is negative providing evidence that during the periods of negative market returns liquidity commonality increases (Hameed, Kang, and Vishwanathan, 2010). Table 6.6 shows the analysis portfolio-wise for three size-based portfolios; small, medium, and large. The results show that none of the sources explain the liquidity commonality of small firms which is quite possible as small firms are less affected by liquidity commonality as shown in Chapter 5. The commonality of medium-size firms is explained by short-term interest rate, CP Spread, and bank returns, but the significance is weak. However, the commonality of large firms is explained significantly by broker returns and bank returns and the signs of the coefficients are positive. The coefficient for broker returns is and that for bank returns is Also, in the presence of the supply-side factors, market liquidity negatively impacts liquidity commonality, though the significance level is low. This implies that as market liquidity decreases, commonality increases for large firms.

18 Table 6.6: Supply-Side Determinants of Liquidity Commonality for Size Portfolios Model Small Small Medium Medium Large Large Capital Market Conditions Market Return (-1.58) (-0.42) (-1.33) (0.90) (-1.23) (0.55) Market Liquidity (1.52) (1.40) (0.99) (0.77) * (-1.67) * (-1.92) Market Volatility (-0.07) (-0.34) 1.213** (2.78) 1.131** (2.55) 3.526*** (5.46) 3.345*** (5.06) Market Turnover (0.09) (0.28) (0.35) (0.54) (1.54) (1.59) Supply-Side Factors Short-term Interest Rate (-0.48) (-0.59) (-1.44) * (1.68) (-0.99) (-1.34) CP Spread (0.58) (0.58) 0.003* (1.80) 0.003** (1.94) (0.32) (0.59) Broker Returns (-1.0) (1.54) *** (3.03) Bank Returns (1.02) ** (2.77) 0.003*** (4.06) Constant 0.207*** (15.92) 0.208*** (16.05) 0.185*** (12.59) 0.187*** (12.54) 0.135*** (5.99) 0.140*** (6.06) Adj. R-squared F-Value *** 3.77*** 10.82*** 9.35*** N Note: This table reports the monthly time-series regressions of supply-side determinants of liquidity commonality for the period Jan-2001 to Dec-2012 (144 months) is analyzed for size-based portfolios. Equally-weighted liquidity commonality for all stocks in each month is estimated by using Amihud liquidity measure and by the methodology of Karolyi, Lee, and van Dick (2012). The capital market condition factors effecting liquidity commonality are used as the additional regressors in the model. The supply-side factors used in the study are Short-term interest rate, commercial paper spread (CP Spread), Broker returns, and returns of the banking industry. The regression coefficients are reported in the table along with the associated t-statistics in the parenthesis. *, **, *** indicate significance at 10%, 5%, and 1% respectively.

19 6.4.5 Demand-side Determinants of Liquidity Commonality In this section, we examine the time-series behavior of demand-side determinants of liquidity commonality. Monthly liquidity commonality ( ) for all stocks in the sample is estimated as described in section 6.3. We use the following time-series for the analysis: NetFII is the net FII flow in a month in percentage terms which is calculated as (Net buy/ (buy+sell/2)). NetMF is net mutual fund flow in a month calculated to NetFII. ExchangeRate is the monthly percentage change in exchange rate of Indian rupee vis-à-vis dollar. LnExports is the natural logarithm of exports each month. The above four variables serve as proxies of demandside sources of liquidity commonality. Along with the demand-side sources, we include four other market conditions factors; Market return (, market liquidity, market volatility (, and market turnover ( as additional regressors, since these factors impact liquidity of a stock (Karolyi, Lee, and van Dick, 2012). We test hypotheses 3b.1, 3b.2, 3b.3, and 3b.4 by examining the significance of respectively in EQ6. Table 6.7 provides the regression results for EQ6. Model (1) shows that out of the four capital market conditions variables, market volatility and market turnover impact liquidity commonality positively at 5% significance. However, market volatility has a higher impact on commonality compared to market turnover. These results support the argument that an increase in volatility and turnover result in an increase in liquidity commonality. Market returns has a negative and market liquidity has a positive impact on commonality, though they are

20 insignificant. The directions of the coefficients reported support prior literature in explaining determinants of liquidity. Table 6.7: Demand-Side Determinants of Liquidity Commonality Model (1) (2) (3) (4) (5) (6) (7) Capital Market Conditions Market Return Market Liquidity Market Volatility Market Turnover (-0.90) (0.61) 0.267*** (4.76) 0.001** (2.08) (-1.35) (0.572) 0.283*** (4.91) 0.001** (2.32) (-0.87) (0.42) 0.261*** (4.34) 0.001** (2.08) (-0.90) (0.71) 0.271*** (4.74) 0.001** (2.09) * (-1.82) 1.46 (1.05) 0.225*** (4.05) 0.003*** (3.52) ** (-2.36) 3.244** (2.14) 0.238*** (4.35) 0.003*** (4.37) ** (-2.19) 2.898* (1.67) 0.232*** (3.93) 0.003*** (4.32) Demand-Side Factors Net % FII Flow 0.374* (1.75) (1.02) Net % MF Flow (0.23) (0.41) Exchange Rate (0.43) 0.023** (2.76) 0.023*** (2.73) LnExports *** (-3.15) *** (-4.22) *** (-3.95) Constant (-0.58) (-0.41) (-0.62) (-0.71) 1.144** (2.30) 1.001** (2.05) 0.923* (1.75) Adj. R-squared F-Value 11.83*** 11.92*** 9.41*** 9.44*** 11.06*** 11.81*** 8.76*** N Note: This table reports the monthly time-series regressions of demand-side determinants of liquidity commonality for the period Jan-2001 to Dec-2012 (144 months). Equally-weighted liquidity commonality for all stocks in each month is estimated by using Amihud liquidity measure and the methodology of Karolyi, Lee, and van Dick (2012). The capital market condition factors effecting liquidity commonality are used as the additional regressors in the model. The demand-side factors used in the study are Net% FII Flow, Net% MF Flow, Exchange Rate, and Exchange Rate. The regression coefficients are reported in the table along with the associated t-statistics in the parenthesis. *, **, *** indicate significance at 10%, 5%, and 1% respectively.

21 Models (2) to (5) show the impact of each of the demand-side determinant on liquidity commonality. We observe that out of the four measures, FII flow is positive and significant with a coefficient of and Ln_Exports is negative and significant with a coefficient of Table 6.8: Demand-Side Determinants of Liquidity Commonality Model Small Medium Large Capital Market Conditions Market Return Market Liquidity Market Volatility * (-1.84) (1.24) (1.11) (0.29) (1.13) 1.183** (2.48) 0.127* (1.75) (-1.01) 3.509*** (4.91) Market Turnover (1.03) ** (2.37) *** (2.97) Demand-Side Factors Net % FII Flow (0.53) (0.61) (1.21) Net % MF Flow (0.13) (0.13) (0.67) Exchange Rate (0.88) 0.002** (2.22) 0.004** (2.41) Ln_Exports (-1.05) (-1.46) ** (1.99) Constant 0.258*** (4.61) 0.163** (2.42) (0.97) Adj. R-squared F-Value ** 7.72*** N Note: This table reports the monthly time-series regressions of demand-side determinants of liquidity commonality for the period Jan-2001 to Dec-2012 (144 months) for size-based portfolios. Equally-weighted liquidity commonality for all stocks in each month is estimated by using Amihud liquidity measure and the methodolo gy of Karolyi, Lee, and van Dick (2012). The capital market condition factors effecting liquidity commonality are used as the additional regressors in the model. The demand-side factors used in the study are Net% FII Flow, Net% MF Flow, Exchange Rate, and Exchange Rate. The regression coefficients are reported in the table along with the associated t-statistics in the parenthesis. *, **, *** indicate significance at 10%, 5%, and 1% respectively.

22 The reason for a positive impact of FII flow on commonality may be due to the correlated trading activity of FIIs. In model (6) where we include ExchangeRate as well as Ln_Exports, we find both these variables significantly impacting liquidity commonality; ExchangeRate positively and Ln_Exports negatively. The liquidity commonality of small-firms as shown in Table 6.8 is not explained by any of the demand-side factors of interest. This may be due to the less degree of liquidity commonality for small stocks. For the medium-size portfolio, ExchangeRate significantly explains liquidity commonality with a coefficient of For the large-firms portfolio, both ExchangeRate and Ln_Exports significantly explain liquidity commonality Cross-sectional Determinants of Liquidity Commonality In this section, we examine the cross-sectional determinants of liquidity commonality for our sample stocks listed on NSE. First, we estimate liquidity commonality for each firm each month by using EQ1 and EQ2. We run Fama-Macbeth monthly cross-sectional regressions for 144 months to examine the possible cross-sectional determinants of liquidity commonality. We report the coefficients and Newey-West corrected t-statistics in the parenthesis in Table 6.9. We consider five explanatory variables for our analysis; Ln_Price, Ln_Mcap (proxy for firm size), Monthly_Return which is the monthly stock return, volatility is the monthly stock return volatility, and Stk_Liq is the Amihud liquidity measure. We test hypotheses 5a, 5b, 5c, 5d, and 5e in this section. It can be observed from Table 6.9 that Ln_price, Ln_mcap, and Stk_Liq significantly impact liquidity commonality in the cross-section. The coefficient of market returns is negative, but insignificant. The results show that stock price impacts liquidity commonality positively.

23 Liquidity commonality is high for high priced stocks as the coefficient is positive and significant. This can be explained by the fact that there exists a size effect in commonality as shown in Chapter 5. Also, liquidity commonality increases with an increase in liquidity of the stock. However, unlike time-series regressions, there is a significant negative impact of stock return volatility on commonality. Table 6.9: Cross-sectional Determinants of Liquidity Commonality Model (1) (2) (3) (4) (5) (6) Ln_Price 0.003** (2.27) Ln_Mcap (0.34) Monthly_Return Volatility Stk_Liq Constant 0.216*** (40.13) 0.198*** (20.43) (-0.31) 0.201*** (60.12) * (-1.77) 0.202*** (71.99) 0.01* (1.71) 0.210*** (60.59) 0.008*** (5.09) 0.004** (2.5) (-1.12) (0.31) 0.001*** (2.64) 0.202*** (22.26) Adj. R-squared Industry FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Note: The above table shows the cross-sectional determinants of liquidity for stocks listed on NSE. Monthly liquidity commonality of each stock is regressed on Ln_Price, Ln_Mcap (proxy for firm size), Monthly_Return which is the monthly stock return, volatility is the monthly stock return volatility, and Stk_Liq is the Amihud liquidity of each stock in the sample. We control for the Industry as well as time fixed effects. We estimate monthly Fama-Macbeth regressions and the time-series means of cross-sectional slope coefficients are reported with Newey-West corrected t- statistics in the parenthesis. *, **, *** represents the significance of each coefficient at 10%, 5%, and 1% level respectively Index Inclusion as a Determinant of Liquidity Commonality To examine market-wide commonality in liquidity for the subsamples, we follow CRS (2000) and run market model time series regressions and estimate the coefficients of interest. We regress the percentage change in individual stock liquidity measures on the percentage change in

24 market liquidity measure. The market liquidity measure is an equally weighted average liquidity of all stocks in the market excluding the stock under examination. We exclude the stock liquidity from market liquidity measure to eliminate any cross-sectional dependence in the estimated coefficients. The market model time series regression is = ( denotes each of the six liquidity measures used in the study on a given day t for a firm j. is the concurrent change in the corresponding average market liquidity measure. We also include a lag and lead market liquidity variables in EQ7 to capture any nonsynchronous change in liquidity due to thin trading. Cross-sectional means of time series slope coefficients are reported with the t-statistics to test the null hypothesis that there is no market-wide commonality in liquidity for stocks listed on NSE in line with Fama-Macbeth (1973). The concurrent, lag and lead market return along with idiosyncratic firm volatility act as control variables for the model. These control variables help to segregate the impact of changes in market-wide liquidity on an individual firm's liquidity after taking into account market-wide price changes and idiosyncratic volatility. In Table 6.10, we report the results of EQ5 separately for two portfolios of stocks; 124 index based stocks and 857 non-index based stocks. We compare the concurrent mean coefficient of the two portfolios of firms. For each portfolio, we report concurrent mean coefficient, the percentage of firms which are positive, percentage of firms which are positive and significant at 5% level, percentage of firms which are negative, percentage of firms which are negative and significant, and also the sum of concurrent, lag, and lead coefficient.

25 Table 6.10: Relation between Liquidity Commonality and Index Arbitrage Potential Subsample No. of Firms Concurrent Mean Coefficient % Pos. % Pos. Sig. % Neg % Neg. Sig. Sum Spread Index Constituents Non-Index Constituents Pspread Index Constituents Non-Index Constituents Depth Index Constituents Non-Index Constituents Roll Spread Index Constituents Non-Index Constituents Spread_HL Index Constituents Non-Index Constituents Amihud Index Constituents (19.23) (12.34) (25.67) (14.18) (6.12) (4.21) (32.13) (18.56) (25.23) (16.29) (6.44) (14.46) (15.68) (22.32) (7.89) (4.12) (3.56) (21.87) (10.12) (12.34) (9.18) (4.22) Non-Index Constituents (3.88) (4.65) Note: The total number of 981 firms in the sample are divided into two subsamples; one subsample having a total of 124 firms which are included in one of the five equity indices (Bank Index, CNX IT, CNX Infra, CNX Midcap 50, and NIFTY) that trade on the derivative segment of NSE and the remaining 857 firms which are not part of any of the five equity-index products that trade on the derivatives segment. For each subsample, Market-wide commonality in liquidity is estimated by regressing percentage change in the individual stock liquidity measure on the percentage change in equally-weighted market liquidity measure on a daily basis. The equally-weighted market average measure excludes the liquidity of the dependent variable stock. Cross-sectional mean of the time-series slope coefficients are reported in the Fama-Macbeth fashion with the corresponding t-statistics in the parentheses. Concurrent, lag, and lead refer to same day, previous and the next trading day of the market liquidity measure. % Positive (% Negative) reports the percentage of firms with positive (negative) slope coefficients. % Positive Significant (% Negative Significant) reports the percentage of firms with positive (negative) coefficients significant at 5% level. Sum reports the sum of concurrent, lag, and lead coefficients.

26 For the Spread measure 54% of the firms are positive and significant at 5% for the indexbased firms and it is 29% for the non-index based firms. For the Pspread measure, the percentages are 43% and 27% respectively. Similarly, for other liquidity measures, we observe a higher percentage of firms having positive and significant coefficients for the index-based firms compared to the non-index based firms. The Sum values for all the liquidity coefficients are also highly significant. The results in Table 6.10 prove that index inclusion hypothesis of Brockman and Chung (2006) that firms included in any equity product trading in the corresponding derivatives market holds good for the NSE equity market. Next, we estimate EQ7 by disaggregating the index portfolio into five separate indices to examine the commonality of individual indices and compare the coefficients of each of the index with that of the non-index firm portfolio. We do this analysis to provide robust evidence that the results reported in Table 6.8 are not driven by any one or two indices and generalize the above mentioned hypothesis. Hence, according to the index inclusion hypothesis, all underlying indices with active derivatives market should exhibit higher levels of liquidity commonality since arbitragers trade in blocks of the underlying index. The results for the commonality coefficients for the five constituent indices and non-index portfolio of 857 stocks are reported in Table 6.9. To save space, we report the coefficients of Spread, Pspread, and Depth only and the coefficients for other measures of liquidity are provided in the appendix. For the Spread measure, 33.3%, 60%, 40%, 76%, 86% of the firms of Bank Index, IT Index, Infra Index, Midcap 50 Index, and Nifty Index respectively have a positive and significant commonality coefficient. This is 29.31% for the non-index firm portfolio.

27 Table 6.11: Relation between Liquidity Commonality and Index Arbitrage Potential for Individual Indices Spread Subsample No. of Firms Concurrent Mean Coefficient % Pos. % Pos. Sig. % Neg. % Neg Sig. Bank Index (9.12) (6.34) CNX IT Index (18.26) (8.99) CNX Infra Index (13.23) (7.22) CNX Midcap 50 Index (21.33) (8.11) Nifty Index (18.71) (9.01) Non-Index Firms (12.34) (15.6) Pspread Bank Index (12.11) (7.33) CNX IT Index (14.22) (5.99) CNX Infra Index (18.11) (8.11) CNX Midcap 50 Index (9.11) (7.38) Nifty Index (17.45) (10.1) Non-Index Firms (14.18) (7.89) Depth Bank Index (6.36) (3.45) CNX IT Index (5.12) (4.33) CNX Infra Index (4.28) (3.84) CNX Midcap 50 Index (6.66) (5.10) Nifty Index (9.15) (6.52) Non-Index Firms (4.21) (3.56) Note: The total number of 981 firms in the sample are divided into six subsamples; five subsamples having firms which are included in one of the five equity indices that trade on the derivative segment of NSE and the remaining 857 firms which are not part of any of the five equity-index products that trade on the derivatives segment. For each subsample, Market-wide commonality in liquidity is estimated by regressing percentage change in the individual stock liquidity measure on the percentage change in equally-weighted market liquidity measure on a daily basis. The equally-weighted market average measure excludes the liquidity of the dependent variable stock. Cross-sectional mean of the time-series slope coefficients are reported in the Fama-Macbeth fashion with the corresponding t-statistics in the parentheses. Concurrent, lag, and lead refer to same day, previous and the next trading day of the market liquidity measure. % Positive (% Negative) reports the percentage of firms with positive (negative) slope coefficients. % Positive Significant (% Negative Significant) reports the percentage of firms with positive (negative) coefficients significant at 5% level. Sum reports the sum of concurrent, lag, and lead coefficients. Sum