Liquidity as risk factor A research at the influence of liquidity on stock returns Bachelor Thesis Finance R.H.T. Verschuren 134477 Supervisor: M. Nie
Liquidity as risk factor A research at the influence of liquidity on stock returns Supervisor: M. Nie Department: Finance Author: R.H.T. Verschuren 134477 Study programm: Bachelor Bedrijfseconomie Faculty: Faculty of Economics and Business Administration Number of words: 7273 Date: 27-5-2011 2
Table of content 1.Introduction 4 2.Literature review 6 2.1 The relationship between stock return and liquidity 6 2.2 Measuring liquidity 7 3.Methodology 10 4.Empirical study 13 4.1 Descriptive statistics 13 4.2 Preliminary regression 15 4.3 Individual regression 17 4.4 Portfolio regression 20 4.5 Cross-sectional regression 21 4.6 Hypothesis testing 21 5.Conclusion 23 5.1 Conclusions 23 5.2 Economic implications 24 5.3 Limitations of the thesis 24 5.4 Recommendations for further research 24 References 25 3
1. Introduction A well known phenomenon in the financial world are illiquid assets. These assets are hard to be sold on the financial market, due to the illiquidity of the market. The financial market is illiquid when there is a gap between the bid and the offer on a the asset or when potential buyers can t be found by sellers. The risk that derives from this phenomenon is called liquidity risk. Liquidity risk is a measure for the volatility of liquidity. In times of crisis there is a lack of liquidity on the financial market. So for instance during the last financial crisis it was hard to sell your stocks. Liquidity risk compounds other risks such as market risk and credit risk. In fact there are two types of liquidity risk. The one I already introduced is called market liquidity risk. The other one is called funding liquidity risk. This is the risk that arrives when a firm has to meet their obligations. This funding liquidity risk depends on future cash flows and outstanding liabilities. In this thesis I will focus on the market liquidity risk. In earlier research by Acharya and Pedersen (2005) a liquidity adjusted capital asset pricing model was build. This model helps understanding the various ways liquidity influences asset prices. In another paper about this subject Amihud and Mendelson (1986) studied on the effect of bid-ask spread on asset prices. They tested their hypotheses and found out that the expected return is an increasing and concave function of the spread. Chordia et al. (2001*) studied the effects of liquidity on the trading activity of the market. They state that daily changes in liquidity and trading activity are highly volatile and have a negative relationship. In the paper of Chordia et al. (2009) they approached liquidity estimation from a theoretical perspective. Their empirical research provided evidence that this theorybased estimates of illiquidity are priced in the cross-section of expected stock returns. The paper of Amihud (2002) shows that expected market illiquidity positively affects ex-ante stock excess return. According to Amihud this excess return consists of a illiquidity premium. He also stated that the effect is larger with small firm stocks. Brennan and Subrahmanyam (1996) also studied on the compensation for liquidity risk in stock return. From their empirical research they found significant evidence that there s a relation between required rates of return and these measures after adjusting for risk factors. Chordia et al. (2001) analyzed the relation between expected equity returns and the level and volatility of trading activity. Trading activity is used as a proxy for liquidity. They show the importance of measures of trading activity in the cross-section of expected stock returns. 4
In the study of Pástor and Stambaugh (2003) they tried to find out the relationship between marketwide liquidity and asset pricing. They conclude that marketwide liquidity is indeed an important variable for stock prices and that stocks that are sensitive for liquidity fluctuations have higher expected returns. Hölmstrom and Tirole (1993) studied the value of stock as an managerial performance monitor. They say that the amount of information that is generated from stock value depends on the market liquidity. The aim of the study will be the influence of market liquidity on stock returns. In a later stage of the research I would like to investigate the effect of market liquidity risk too. I think this effect of liquidity on stock returns is an interesting study object because of the recent financial crisis and thus the growing importance of understanding all types of risks. The research question of this thesis is: How does market liquidity affect stock returns? I expect that the liquidity has a negative and small influence on stock returns. This means that illiquid stocks have higher returns than liquid stocks. To measure the illiquidity of a stock, the bid-ask spread of the stocks are used. The bigger this spread is the more illiquid the stock is. The following model explains the relationship between stock return and bid-ask spread:. In this model is the excess return of the stock, is the bid-ask spread of the same stock, is the effect of the bid-ask spread on the return and is a constant. The other variables in the model are the Fama-French stock-market factors. I expect to be positive. This means that illiquidity has a positive effect on stock returns. This hypothesis will be tested in chapter 4. The stocks that are used for the regression analysis are the 25 stocks from the Dutch AEX Index. The returns of this stock will be corrected for the risk-free interest rate. For each stock the betas will be calculated as well as to significance of this effect. Also the stated hypothesis will be tested for each individual stock. There also will be regression analysis s of the equal weighted portfolio instead of the individual stocks. In chapter 2 I m going to give you a full review of the literature that is relevant for this thesis. This chapter will consists of two parts. The first part is about the relationship between stock returns and the bid-ask spread and the second part is about measuring liquidity. Chapter 3 will be about that data and the methodology of the empirical part. The fourth chapter will provide descriptive statistics, regression analysis s and hypothesis testing. The last chapter includes the conclusion of this thesis. In this chapter the conclusions, economic implications, limitations of the thesis and the recommendations for further research can be found. At the end of the thesis, the used references can be found. 5
2. Literature review In this chapter I will explain all the relevant literature for this thesis. The academic research at this topic began to shape more after the work of Amihud and Mendelson in 1986. Before that research the effects of liquidity were virtually ignored. This chapter will consist of two main parts. The first paragraph will be quite general about the relationship between stock return and liquidity. The second paragraph will be about measuring liquidity and the bid-ask spread. 2.1 The relationship between stock return and liquidity Pastór and Stambaugh (2003) describe liquidity as follows: liquidity is a broad and elusive concept that generally denotes the ability to trade large quantities quickly, at low cost, and without moving the price. 1 According to Brennan and Subrahmanyam (1996) illiquidity is primary caused by the adverse selection that arises from the presence of privately informed traders. Amihud (2002) says that positive relationship between illiquidity and return has been examined across stocks in a number of studies. Therefore he researched this relationship over time. According to him stock excess return reflects compensation for expected market illiquidity, and is thus an increasing function of expected market illiquidity. (Amihud, 2002) 2 The results of his study proved that his hypotheses was right. The excess returns were not constant but varied over time with the changes in market illiquidity. In addition he found that there s a negative relationship between unexpected market illiquidity and stock prices. He also found that the effects of liquidity are stronger for small firms stocks. In this thesis 25 large firm stocks will be used, so according to Amihud the effect will be smaller. Pastór and Stambaugh (2003) and Brennan and Subrahmanyam (1996) agree with Amihud (2002) about the fact that there is a compensation for liquidity. Pastór and Stambaugh (2003) think that it s reasonable that investors might require higher expected returns on assets that are more illiquid. They state that stocks have liquidity betas, these betas provide a higher expected return, which is the compensation for illiquidity. They found that marketwide liquidity is a state variable that is important for pricing common stocks. (Pastór and Stambaugh, 2003) 3 They also conclude that stocks that are 1 P.644 2 P.32 3 P.683 6
sensitive for liquidity fluctuations have higher expected returns. In this thesis the findings of Pastór and Stambaugh (2003), Brennan and Subrahmanyam (1996) and Amihud (2002) will be tested. So this thesis will try to confirm that there is a compensation for liquidity in stock returns. Acharya and Pedersen (2005) gives a liquidity-adjusted capital asset pricing model. This model is a framework that can explain the empirical findings of Pastór and Stambaugh (2003) and Amihud and Mendelson (1986). Acharya and Pedersen (2005) found that positive shocks to illiquidity, if persistent, are associated with a low contemporaneous returns and high predicted future returns. 4 They also found that liquidity risk explains about 1,1% of cross-sectional returns. Another result of their exercises was that illiquid securities also have high liquidity risk. They state that illiquid assets tend to have high return sensitivity to market liquidity and high liquidity sensitivity to markets returns. Chordia et al. (2001*) studied the relationship between liquidity and trading activity. They found that liquidity and trading activity are influenced by several factors such as default spreads, market volatility and recent market movement. Their results provided evidence for a negative relationship between daily changes in liquidity and trading activity. In another research Chordia et al. (2001) used liquidity as a proxy for trading activity. They studied the relation between expected returns and the level and volatility of trading activity. Their results provided significant evidence for a negative relationship. This results were consistent with the findings of Amihud and Mendelson (1986) and Brennan and Subrahmanyam (1996). 2.2 Measuring liquidity Chordia et al. (2009) argue that the issue in relating illiquidity and asset prices is the measurement of illiquidity. Because this thesis wants to focus on the relationship between illiquidity and stock return, it s is important to know how to measure illiquidity. When this method is known, a model can be build and tested. Amihud and Mendelson (1986) state that the cost of immediate execution can be a measure for illiquidity. Investors have the choice to wait and trade assets at a favorable price or to trade immediately with probably a less favorable price. The quoted ask (offer) price includes a premium for immediate buying, and the bid price similarly reflects a concession required for immediate sale. Thus, a natural measure of illiquidity is the spread between the bid and ask prices, which is the sum of the buying premium and the selling concession. (Amihud and Mendelson, 1986) 5 They also state that they think that this spread isn t a indication for market efficiency, but it s a representation for a 4 P.405 5 P.223 7
rational response to an efficient market. The conclusion of their study was that expected return is an increasing and concave function of the bid-ask spread. Chordia et al. (2001*) found that spreads increase dramatically in down market, but decrease only marginally in up markets. Eleswarapu (1997) also used the bid-ask spread to examine the liquidity premium predicted by Amihud and Mendelson (1986). In his research he used daily returns, bids and asks. He calculated every monthly spread by averaging every daily relative spread. Eleswarapu (1997) and Amihud and Mendelson (1986) both state the average spread as the actual bid-ask spread divided by the average of bid and ask prices. Eleswarapu (1997) divided the researched Nasdaq stocks into portfolios based on estimated beta and average spreads. The results of his research supported the model of Amihud and Mendelson (1986). It s curious to see that two other tests of this model, Chen and Kan (1988) and Eleswarapu and Reinganum (1993), used NYSE stocks and provided weak evidence. Eleswarapu (1997) conjectures that this difference is caused by what quoted spreads represent in each market. In particular, the dealers inside spreads on the Nasdaq are likely to better proxy the actual cost of transacting as compared to the specialist s representatives quotes on the NYSE. (Eleswarapu, 1997) 6 Brennan and Subrahmanyam (1996) used other techniques to measure liquidity. They measured illiquidity with intraday transactions data. They used portfolios with stocks sorted on market depth and firm size. Where Amihud and Mendelson (1986) used the capital asset pricing model to adjust for risk, they use the Fama and French three-factor model. They also made a distinction between the variable and fixed components of trading cost. They found that there is a significant return premium associated with both fixed and variable elements of the cost of transacting. (Brennan and Subrahmanyam, 1996) 7 Chordia et al. (2001) used two measures to proxy for liquidity: dollar share volume and share turnover. Breen et al. (2002) developed price impact as a measure for liquidity. This price impact is a measurement for the change of the stock associated with the net trading volume. Price impact captures the extent to which trade execution influences the stock price: A perfectly liquid asset trades without any price impact while a perfectly illiquid asset cannot be traded at any price. Therefore, our measure encompasses important aspects of liquidity which are not captured by existing measures like the bid-ask spread and quoted depth. (Breen et al. 2002) 8 6 P.2127 7 P.463 8 P.470 8
Chordia et al. (2009) say that these various methods helped understanding illiquidity, but because of the mixed results it s hard to interpret the significance of the findings. They also give two more reasons to doubt about the significance of the findings. The first one is that there are no theory-based arguments to justify the liquidity proxies. The second one is that illiquidity also depends on variables that could be related to asset prices, for instance volatility. 9
3. Methodology As can be seen from the literature review a measure for market liquidity is the bid-ask spread. With this measure, the market liquidity of different assets can be compared with each other. If the bid-ask spread is large the stock is illiquid and if the bid-ask spread is small the stock is liquid. So to be precisely: the bid-ask spread of an asset is its measure for illiquidity. Amihud and Mendelson (1986) used the bid-ask spread as a measure for market liquidity. In their research they use the relative spread, which is the dollar spread divided by the average of the bid and ask. To find out the effect of market liquidity on stock returns, I will use the real bid-ask spreads of stocks as well as their returns. The data of the stocks of the 25 companies that are listed at the Dutch AEX index are used in this thesis. With DataStream all monthly prices, ask prices and bid prices of the 25 stocks were collected. The data consists of all prices from base date till today. The ask and bid prices are not available till around 1996. So the months before 1996 could not be used. Therefore this thesis will use the data from the stocks from the month the ask and bid price are collected. In the next chapter the number of researched months for each stock can be found. The returns of the stocks are calculated from the prices and the bid-ask spreads are calculated from the bid and ask price. To improve the model, the returns were corrected for the risk-free interest rate. This monthly excess return is calculated by subtracting the risk free rate from the individual stock returns. With this subtraction the estimation bias was eliminated. Using regression analysis I want to examine the direction and size of the effect of market liquidity as an independent variable on the dependent variable, the stock return. I expect that this effect will be negative but relatively small. In that case illiquid stocks will have higher returns than liquid stocks. So I expect that the effect of the bid-ask spread on the returns is positive. The next chapter will start with providing descriptive statistics about the monthly excess returns and bid-ask spreads of the 25 AEX stocks and an equal weighted portfolio including all 25 stocks. There also will be descriptive statistics of the Fama-French factors and the risk-free interest rate. The empirical study will start with a preliminary part. In this part of next chapter a simple model will be tested. The following model explains the relationship between excess stock return and bid-ask spread: 10
In this model the return of each stock is corrected for the risk-free interest rate ( ). In this model is the return of the stock, is the bid-ask spread of the same stock, is the effect of the bid-ask spread on the return and is a constant. In the preliminary part of next chapter there will be a two-sided significance test of the models, due to test if 0. If 0 the model is useful. Also the degree of usefulness of the model can be found in this section. In this thesis this model is also tested for the equal weighted portfolio of the 25 stocks. After the preliminary part the model will be extended with the Fama-French factors. Fama and French (1993) studied the common risk factors in stock and bond returns. According to them there are at least five common risk factors in returns. Three of this five factors are so called stock-market factors. These factors produce common variation in stock returns. The three stock-market factors are RMO, SMB and HML. Fama and French (1993) state that the slopes on RM-RF are, however, the same as the slopes on RMO 9. To calculate the factors Fama and French used 6 value weighted portfolios formed on size and book-to-market. SMB, which stands for small minus big, is the so-called size-related factor. SML is the average return on the three small portfolios minus the average return on the three big portfolios. HML, which stands for high minus low, is the book-to-market-related factor. HML is the average return on the two portfolios with high market-to-book ratios minus the average return on the two portfolios with low market-to-book ratios. With adding the three stock-market factors, the tested model will have a higher r 2. The r 2 gives the percentage of the variation in excess return that can be explained by the model. So with including this factors the degree of usefulness of the model will rise. First a model with only the Fama-French factors as independent variables will be tested: In this model is the return of the market minus the risk-free interest rate and is the effect of this variable on the excess return. is the effect of SMB on the excess return. The last variable of the model is HML. is the effect of the HML variable on the excess return. After the analysis of the model above, the model with the bid-ask spread as fourth independent variable will be tested: In this model is the bid-ask spread of each individual stock and is the effect of this bid-ask spread on the excess return. For both models all betas and corresponding P-values can be found. Both models are also used to test the regression of the equal weighted portfolio of the 25 stocks. The data of the months from July 2005 till March 2011 will be used in all analysis s concerning the 9 P.52 11
portfolio. Next to all the time-series analysis s that will be done, there also will be a cross-sectional regression analysis. The average excess return and bid-ask spread of each stock will be used for this part. This will be followed by the testing of the hypothesis. As can be read, I expect (which is the effect of the bid-ask spread on excess return) to be positive. The hypothesis testing will be done by a onesided test. The following hypotheses will be tested: H 0 : β 4 0 vs H 1 : β 4 > 0 This method will be used for the model with the three Fama-French factors and the bid-ask spread. 12
4. Empirical study 4.1 Descriptive statistics Table 1: Descriptive statistics of the excess returns over the of the 25 AEX stocks Stock µ (%) min (%) max (%) σ (%) N AEGON 0,38-48,48 34,67 12,50 180 Ahold 0,48-68,76 61,13 10,74 180 Air France-KLM 0,50-60,55 117,79 18,33 268 Akzo Nobel 0,64-23,35 27,76 9,10 180 Arcelor Mittal 2,37-44,79 54,47 17,41 127 ASML Holding 2,21-40,10 62,40 17,06 169 BAM Groep 1,02-29,14 25,35 10,29 127 Boskalis Westminster 1,79-35,76 37,09 10,74 129 Corio 0,50-19,30 15,74 6,10 119 DSM 0,83-35,33 23,89 8,30 180 Fugro 1,33-28,36 16,72 9,08 129 Heineken 0,48-23,96 23,41 6,52 180 ING Groep 0,66-52,07 57,11 12,52 180 KPN 0,76-45,78 80,93 13,25 180 Philips 0,96-30,63 36,10 11,33 180 Randstad Holding 1,27-34,49 64,61 14,21 180 Reed Elsevier -0,10-23,46 22,86 6,66 180 Royal Dutch Shell 0,35-23,36 23,68 6,59 180 SBM Offshore 0,61-36,28 29,93 9,40 150 TNT 0,01-22,32 34,36 7,85 151 Tom Tom 0,43-48,13 57,93 19,24 69 Unibail-Rodamco 0,48-19,20 22,70 6,49 273 Unilever Certs. 0,51-16,46 24,53 6,83 180 Wereldhave 0,35-13,63 15,11 5,80 129 Wolters Kluwer 0,03-36,65 28,28 8,56 180 Porfolio 0,58-16,78 13,61 6,49 69 As can be seen in table 1 most of the average monthly returns corrected for the risk free rate are positive. Also the highest and lowest return of each stock can be found above. Al numbers of the first four columns are rates. At the last column the number of the researched months can be found. For instance, when the number is 180 the data used is 180 months back from March 2011. So in that case the period April the 1 st 1996 till March the 1 st 2011 is researched. In the last row of table 1 the descriptive statistics of the excess return of the equal weighted portfolio can be found. 13
In the table below the descriptive statistics of the Fama-French risk factors and the risk-free interest rate is given. The 338 months are from the period February the 1 st 1983 till March the 1 st 2011. The average monthly return of the Mkt-rf can t be compared with the average excess returns of the stocks from table 1 because each stock has its own number of researched months. Table 2: Descriptive statistics of the Fama- French factors and the rf µ (%) min (%) max (%) σ (%) N Mkt-rf 0,60-23,14 12,43 4,56 338 SMB 0,12-16,67 22,19 3,22 338 HML 0,36-12,78 13,84 3,14 338 rf 0,37 0,00 1,00 0,21 338 Table 3: Descriptive statistics of the bid-ask spreads of the 25 AEX stocks Stock µ min max σ N AEGON 0,0436-0,2181 1,0577 0,1142 180 Ahold 0,0337-0,2682 0,5746 0,0911 180 Air France-KLM 0,4892-0,1677 13,7205 1,8086 268 Akzo Nobel 0,0746-0,0226 0,5446 0,1135 180 Arcelor Mittal 0,0979 0,0000 1,0545 0,1403 127 ASML Holding 0,0365 0,0000 0,3489 0,0649 169 BAM Groep 0,0192 0,0000 0,1121 0,0193 127 Boskalis Westminster 0,0588 0,0000 0,3500 0,0582 129 Corio 0,0803 0,0000 0,5000 0,0781 119 DSM 0,0475 0,0000 0,3781 0,0674 180 Fugro 0,0583 0,0000 0,3000 0,0604 129 Heineken 0,0753 0,0000 1,3600 0,1412 180 ING Groep 0,0391 0,0000 1,1863 0,1132 180 KPN 0,0279-0,0349 0,4529 0,0612 180 Philips 0,0428-0,0604 0,5676 0,0903 180 Randstad Holding 0,1146-0,1000 1,3613 0,2205 180 Reed Elsevier 0,0360-0,0456 0,7736 0,0846 180 Royal Dutch Shell 0,0376-0,0170 0,5332 0,0940 180 SBM Offshore 0,0400-0,0608 0,4616 0,0569 150 TNT 0,0411 0,0000 0,6635 0,0735 151 Tom Tom 0,0248 0,0000 0,1405 0,0273 69 Unibail-Rodamco 0,1858 0,0000 1,2288 0,1944 273 Unilever Certs. 0,0333-0,0077 0,4580 0,0691 180 Wereldhave 0,1412 0,0000 0,7500 0,1323 129 Wolters Kluwer 0,0579 0,0000 0,6557 0,1125 180 Portfolio 0,0312 0,0000 0,0754 0,0168 69 14
In table 3 the descriptive statistics of the bid-ask spreads of the 25 AEX stocks can be found. The average spread of each stock is positive, which is quite obvious. Because when the spread is negative, it means that people bid more than what is asked. This happened only a few times in the total dataset, as can be seen at the min-column in the table. The number of researched months for each stock are the same as in table 1, due to testing for regression. In the last row the descriptive statistics of the bid-ask spread of the equal weighted portfolio can be found. 4.2 Preliminary regression Table 4: Regression analysis of the 25 AEX stocks with the bid-ask spread as independent variable and the excess return as dependent variable Stock β 0 β 1 r² AEGON -0,0007 0,1025 0,0087 0,9440 0,2117 Ahold 0,0039 0,0263 0,0005 0,6514 0,7661 Air France-KLM -0,0039 0,0182 0,0321 0,7343 0,0032 Akzo Nobel 0,0027 0,0508 0,0040 0,7444 0,3980 Arcelor Mittal 0,0311-0,0750 0,0037 0,1026 0,4997 ASML Holding 0,0214 0,0190 0,0001 0,1585 0,9257 BAM Groep 0,0252-0,7838 0,0216 0,0512 0,0991 Boskalis Westminster 0,0325-0,2482 0,0181 0,0168 0,1290 Corio 0,0009 0,0516 0,0044 0,9101 0,4752 DSM 0,0081 0,0036 0,0000 0,2860 0,9692 Fugro 0,0198-0,1119 0,0055 0,0772 0,4014 Heineken 0,0047 0,0006 0,0000 0,3944 0,9868 ING Groep 0,0036 0,0759 0,0047 0,7137 0,3598 KPN 0,0030 0,1649 0,0058 0,7824 0,3094 Philips 0,0024 0,1689 0,0181 0,8002 0,0716 15
Randstad Holding 0,0196-0,0603 0,0087 0,1017 0,2119 Reed Elsevier -0,0041 0,0856 0,0118 0,4497 0,1463 Royal Dutch Shell 0,0034 0,0017 0,0000 0,5201 0,9741 SBM Offshore 0,0122-0,1533 0,0086 0,1948 0,2588 TNT 0,0053-0,1265 0,0140 0,4655 0,1472 Tom Tom -0,0141 0,7420 0,0111 0,6553 0,3888 Unibail-Rodamco -0,0002 0,0270 0,0066 0,9664 0,1820 Unilever Certs. 0,0000 0,1536 0,0241 0,9945 0,0374 Wereldhave 0,0000 0,0245 0,0031 0,9956 0,5302 Wolters Kluwer -0,0013 0,0285 0,0014 0,8519 0,6179 In table 4 the results of the preliminary regression analysis of the individual stocks can be found. In this preliminary part of the empirical study, a simple model is used. The relationship that is tested is between the bid-ask spread as independent variable and the excess return over the risk-free interest rate as dependent variable. β 1 stands for the influence of the bid-ask spread on the excess return. β 0 is the constant of this relationship. Below the β 0 and β 1 the corresponding P-value can be found. When the P-value is smaller than 0,05 there can be concluded with a certainty of 95% that 0 and that the tested model is useful. The degree of usefulness can be found at the r 2 column of table 4. If r 2 is 1, 100% of the variation of the returns is explained by the model. So in the case of AEGON 0,87% of the variation of the returns is explained by the bid-ask spread. As Table 4 makes clear the r 2 isn t large for this regressions. The highest value is only 3,21% for the Air France-KLM stock. Table 5: Significance of the effects of the bid-ask spread on excess stock return Interval Significant effect Positive significant Negative significant 95% 2 2 0 90% 4 3 1 85% 7 4 3 80% 8 5 3 75% 10 6 4 70% 11 6 5 16
Table 5 shows the significance of the tested model and. At a confidence level of 95% only 2 stocks have a significant effect. This are the stocks of Air-France KLM and Unilever, which both have a positive. With decreasing the confidence level, the number of significant models rises. At a lower confidence level there also seems to be significant effects with a negative. Table 6: Regression analysis of the portfolio with bid-ask spread as independent variable and excess return as dependent variable β 0 β 1 r² Porfolio 0,0143-0,2718 0,0049 0,3955 0,5663 In table 6 you can find the preliminary model tested for the equal weighted portfolio of the 25 stocks. As can be seen, the effect of the bid-ask spread on the excess return is negative and not significant. The r 2 states that only 0,49% of the variation of the excess return of the portfolio can be explained by the bid-ask spread. 4.3 Individual regression Table 7: Regression analysis of the 25 AEX stocks with the Fama-French factors as independent variables and excess return as dependent variable Stock β 0 β 1 β 2 β 3 r² AEGON -0,0054 1,7326-0,2540 0,5432 0,4278 0,4546 0,0000 0,2111 0,0115 Ahold 0,0004 0,7958-0,0397 0,2218 0,1239 0,9590 0,0000 0,8540 0,3280 Air France-KLM -0,0033 1,1556 0,0632 0,6111 0,0759 0,7633 0,0000 0,8526 0,0917 Akzo Nobel -0,0009 1,1687 0,0656 0,4913 0,3841 0,8664 0,0000 0,6680 0,0026 Arcelor Mittal 0,0190 1,6394 0,5843-0,0840 0,2629 0,1728 0,0000 0,2667 0,8387 ASML Holding 0,0167 1,7907 0,4529-0,9907 0,4297 0,1026 0,0000 0,1097 0,0009 BAM Groep 0,0036 0,8767 0,3552 0,5758 0,2390 0,6658 0,0000 0,2610 0,0216 Boskalis Westminster 0,0140 1,1020-0,5055 0,8068 0,2821 0,0985 0,0000 0,0893 0,0010 Corio 0,0036 0,7296-0,3866 0,2640 0,3288 0,4507 0,0000 0,0357 0,1318 DSM 0,0014 1,0960 0,1452 0,3727 0,4156 0,7683 0,0000 0,2866 0,0098 17
Fugro 0,0086 1,0946-0,0091 0,4958 0,3723 0,1997 0,0000 0,9689 0,0098 Heineken 0,0017 0,4040 0,0568 0,2959 0,0982 0,7224 0,0001 0,6687 0,0348 ING Groep -0,0064 1,8737 0,2352 1,0199 0,5412 0,3253 0,0000 0,1965 0,0000 KPN 0,0097 0,6301-0,0324-1,4720 0,2635 0,2672 0,0007 0,8941 0,0000 Philips 0,0028 1,2789 0,5557-0,3084 0,4478 0,6589 0,0000 0,0024 0,1049 Randstad Holding 0,0032 1,4579 0,3296 0,4748 0,2624 0,7309 0,0000 0,2086 0,0853 Reed Elsevier -0,0032 0,5614-0,0554-0,0881 0,1782 0,4930 0,0000 0,6684 0,5171 Royal Dutch Shell -0,0005 0,7928-0,1811 0,2338 0,3211 0,9112 0,0000 0,1212 0,0572 SBM Offshore 0,0010 0,6329 0,1643 0,6387 0,1479 0,8896 0,0000 0,4163 0,0020 TNT -0,0033 0,7452 0,1802 0,1709 0,2521 0,5583 0,0000 0,2541 0,2820 Tom Tom -0,0083 2,1012 1,2414-0,1548 0,3937 0,6578 0,0000 0,1528 0,8343 Unibail-Rodamco 0,0006 0,4792 0,1073 0,4709 0,1247 0,8800 0,0000 0,3597 0,0002 Unilever Certs. 0,0015 0,5407-0,1440 0,4296 0,1663 0,7468 0,0000 0,2820 0,0025 Wereldhave 0,0007 0,5696 0,1340 0,1813 0,2666 0,8760 0,0000 0,4073 0,1676 Wolters Kluwer -0,0034 0,4880 0,1911 0,2349 0,0881 0,5884 0,0003 0,2759 0,2026 In table 7 the regression analysis with the Fama-French factors as independent variables and the excess return as dependent variable can be found for all 25 stocks. In this table β 0 is a constant. β 1 is the influence of the market return minus the risk-free interest rate on the excess return. β 2 is the influence of SMB on the excess return and β 3 is the influence of HML on the excess return. As can be seen in this table all β 1 s are positive and significant. The r 2 of each stock is high, which means that the Fama-French factors explain a big part of the variation in excess return. 18
Table 8: Regression analysis of the 25 AEX stocks with the Fama-French factors and the bid-ask spread as independent variables and excess return as dependent variable Stock β 0 β 1 β 2 β 3 β 4 r² AEGON -0,0063 1,7269-0,2531 0,5523 0,0190 0,4280 0,4226 0,0000 0,2140 0,0112 0,7678 Ahold -0,0002 0,7949-0,0378 0,2212 0,0178 0,1241 0,9801 0,0000 0,8614 0,3305 0,8314 Air France-KLM -0,0122 1,1615 0,0382 0,6078 0,0183 0,1084 0,2759 0,0000 0,9093 0,0883 0,0022 Akzo Nobel -0,0027 1,1646 0,0701 0,4981 0,0238 0,3849 0,6796 0,0000 0,6483 0,0024 0,6201 Arcelor Mittal 0,0174 1,6486 0,5768-0,0874 0,0177 0,2631 0,3036 0,0000 0,2762 0,8331 0,8570 ASML Holding 0,0133 1,8037 0,4185-1,0056 0,0968 0,4310 0,2552 0,0000 0,1477 0,0008 0,5420 BAM Groep 0,0066 0,8612 0,3415 0,5910-0,1532 0,2397 0,5893 0,0000 0,2856 0,0208 0,7372 Boskalis Westminster 0,0128 1,1102-0,5081 0,8064 0,0206 0,2823 0,2927 0,0000 0,0896 0,0011 0,8900 Corio -0,0044 0,7456-0,3968 0,2737 0,0989 0,3446 0,5159 0,0000 0,0301 0,1158 0,0993 DSM 0,0035 1,1000 0,1434 0,3722-0,0436 0,4168 0,5557 0,0000 0,2937 0,0100 0,5416 Fugro 0,0143 1,0887 0,0058 0,5057-0,1008 0,3768 0,1149 0,0000 0,9801 0,0086 0,3474 Heineken 0,0025 0,4070 0,0553 0,2953-0,0104 0,0987 0,6461 0,0001 0,6776 0,0356 0,7540 ING Groep -0,0066 1,8725 0,2345 1,0219 0,0053 0,5412 0,3385 0,0000 0,1993 0,0000 0,9272 KPN 0,0097 0,6300-0,0325-1,4718 0,0009 0,2635 0,3142 0,0007 0,8943 0,0000 0,9951 Philips -0,0026 1,2504 0,5869-0,3132 0,1295 0,4583 0,7080 0,0000 0,0013 0,0975 0,0679 Randstad Holding 0,0094 1,4561 0,3135 0,4634-0,0530 0,2691 0,3732 0,0000 0,2313 0,0926 0,2058 Reed Elsevier -0,0051 0,5510-0,0478-0,0888 0,0544 0,1829 0,3090 0,0000 0,7120 0,5135 0,3165 Royal Dutch Shell 0,0002 0,7945-0,1783 0,2324-0,0172 0,3217 0,9700 0,0000 0,1286 0,0594 0,6967 SBM Offshore 0,0047 0,6193 0,1514 0,6422-0,0883 0,1506 0,6072 0,0001 0,4566 0,0019 0,4946 TNT 0,0034 0,7552 0,2020 0,1618-0,1664 0,2761 0,5944 0,0000 0,1964 0,3022 0,0293 Tom Tom -0,0312 2,1370 1,2307-0,2121 0,9196 0,4107 19
0,2177 0,0000 0,1538 0,7734 0,1794 Unibail-Rodamco -0,0040 0,4777 0,0775 0,4756 0,0248 0,1300 0,4421 0,0000 0,5160 0,0002 0,2043 Unilever Certs. -0,0019 0,5220-0,1253 0,4244 0,1050 0,1773 0,7170 0,0000 0,3495 0,0028 0,1274 Wereldhave -0,0035 0,5694 0,1361 0,1987 0,0292 0,2710 0,6027 0,0000 0,4004 0,1359 0,3920 Wolters Kluwer -0,0040 0,4859 0,1911 0,2353 0,0112 0,0883 0,5667 0,0003 0,2770 0,2030 0,8397 In table 8 the bid-ask spread is added to the model. β 4 is the influence of the bid-ask spread on excess return. As can be seen in table 9 below, at a confidence level of 95% only 2 stocks have a significant effect: Air-France KLM and TNT. But TNT s β 4 is negative. Table 9: Significance of the effects of the bid-ask spread on stock return Interval Significant effect Positive significant Negative significant 95% 2 1 1 90% 4 3 1 85% 5 4 1 80% 6 5 1 75% 8 6 2 70% 8 6 2 4.4 Portfolio regression Table 10: Regression analysis of the portfolio with the Fama-French factors as independent variables and excess return as dependent variable β 0 β 1 β 2 β 3 r² Porfolio 0,0012 1,1221 0,0202-0,0474 0,7796 0,7521 0,0000 0,9079 0,7528 In table 10 the regression analysis of the model with the Fama-French factors as independent variables on the portfolio excess return can be found. Only Mkt-rf has a significant effect on the excess return of the portfolio. The r 2 of this model is very high. 20
Table 11: Regression analysis of the portfolio with the Fama-French factors and the bid-ask spread as independent variables and excess return as dependent variable β 0 β 1 β 2 β 3 β 4 r² Porfolio -0,0009 1,1251 0,0207-0,0519 0,0665 0,7799 0,9139 0,0000 0,9064 0,7333 0,7728 In table 11 the bid-ask spread of the portfolio is added to the model. The effect of the bid-ask spread (β 4 ) on the excess return of the portfolio is positive but not significant. Also can be seen that, with adding the bid-ask spread to the model, the r 2 only increased with 0,0003. 4.5 Cross-sectional regression Table 12: Cross-sectional regression analysis of the 25 AEX stocks with the bid-ask spread as independent variables and excess return as dependent variable β 0 β 1 r² 0,0078-0,0032 0,0023 0,0001 0,8191 In the table 12 the results of the cross-sectional regression analysis can be found. For this analysis the 25 average excess returns and bid-ask spread of the stocks are used. The effect of the bid-ask spread (β 1 ) is negative and not significant. The r 2 states that only a very small part of the variation in the excess returns can be explained by the bid-ask spread. 4.6 Hypothesis testing As can be read earlier in this thesis I expect β to be positive. This hypothesis can be tested with the following one-sided test: H 0 : β 4 0 vs H 1 : β 4 > 0 The P-value given by the regression analysis is derived from a two-sided test. So to get the one-sided test P-value the two-sided value should be divided by 2. When this value is < 0,05 and β 4 > 0, H 0 can be rejected. And with a confidence level of 95% can be concluded that β 4 > 0. Table 13: Hypothesis testing with excess return as dependent variable, confidence level of 95% Stock β 4 P-Value P-Value/2 Reject H0: β4 0 AEGON 0,0190 0,7678 0,3839 No Ahold 0,0178 0,8314 0,4157 No Air France-KLM 0,0183 0,0022 0,0011 Yes Akzo Nobel 0,0238 0,6201 0,3100 No Arcelor Mittal 0,0177 0,8570 0,4285 No ASML Holding 0,0968 0,5420 0,2710 No 21
BAM Groep -0,1532 0,7372 0,3686 No Boskalis Westminster 0,0206 0,8900 0,4450 No Corio 0,0989 0,0993 0,0496 Yes DSM -0,0436 0,5416 0,2708 No Fugro -0,1008 0,3474 0,1737 No Heineken -0,0104 0,7540 0,3770 No ING Groep 0,0053 0,9272 0,4636 No KPN 0,0009 0,9951 0,4975 No Philips 0,1295 0,0679 0,0339 Yes Randstad Holding -0,0530 0,2058 0,1029 No Reed Elsevier 0,0544 0,3165 0,1583 No Royal Dutch Shell -0,0172 0,6967 0,3483 No SBM Offshore -0,0883 0,4946 0,2473 No TNT -0,1664 0,0293 0,0146 No Tom Tom 0,9196 0,1794 0,0897 No Unibail-Rodamco 0,0248 0,2043 0,1022 No Unilever Certs. 0,1050 0,1274 0,0637 No Wereldhave 0,0292 0,3920 0,1960 No Wolters Kluwer 0,0112 0,8397 0,4199 No As can be seen in table 13, there are 3 of the 25 stocks which have a significant and positive β 4. These stocks are Air-France KLM, Corio and Philips. 22
5. Conclusion 5.1 Conclusions The results of the preliminary regression analysis of the model, that can be found in table 4 and 5, show that this model is only useful for the Air France-KLM, Unilever, Bam and Philips stocks. Only the Bam stock has a negative the other stocks have a positive. For all other stocks it can t be proven that 0 at a confidence level of 90%. In table 5 can be seen that with decreasing the confidence level the number of significant effects rises. With a certainty of 90% can be concluded that for 4 of the 25 stocks the bid-ask spread has a significant effect on the excess return. This means that for 4 of the 25 stocks liquidity and illiquidity have a significant influence on the excess return of that stock. After testing this model on the portfolio of the 25 stocks (table 6) can be concluded that the bid-ask spread of the portfolio has no significant effect on its excess return. After testing the model for each of the 25 stocks can be concluded that, which is the effect of excess market return on excess stock return, is significant and positive for all 25 stocks. There are 2 significant 's and 14 significant 's at a confidence level of 90%. This information can be found in table 7. From the regression analysis of the effects of these three stock-market factors on the portfolio excess return (table 10) can be concluded that is the only significant beta. So HML and SMB don t have a significant effect on the excess return of the portfolio. The next model that is tested is:. In this model the bid-ask spread is added. From the regression analysis of this model (table 8 and 9) can be concluded, with a certainty of 90%, that for 4 of the 25 stocks the bid-ask spread has a significant effect on the excess return. These 4 stocks are Air- France KLM, Corio, Philips and TNT. In case of TNT there is a negative. From the regression analysis of this model for the portfolio (table 11) can be concluded that all betas except are not significant. So HML, SMB and the bid-ask spread don t have a significant effect on the excess return of the portfolio. From the cross-sectional regression analysis, as can be found in table 12, can be concluded that the effect of the average bid-ask spread on the average excess return is not significant. Acharya and Pedersen (2005) found that liquidity risk explains about 1,1% of cross-sectional returns. But is this situation only 0,23% is explained by liquidity-risk. The hypothesis of this thesis is that the bid-ask spread has a positive influence on the excess stock return. From the hypothesis testing of the model with excess return as dependent variable, as can be 23
found in table 13, can be concluded with a certainty of 95% that the stocks of Air France-KLM, Philips and Corio have a positive. So for 3 of the 25 stocks illiquidity has a significant and positive influence on the return of the stock. So it can be concluded that 3 of the 25 stocks that are used confirm the hypothesis of this thesis. Analyzing this conclusion it can be stated that the results of this thesis are not in line with the existing theory. 5.2 Economic implications According to Amihud (2002) there is a compensation in excess return for expected market illiquidity. This suggests that there is a positive relationship between excess return and market illiquidity. The results of this thesis are not in line with this compensation rule. Stocks or other securities that are illiquid are harder to trade and therefore this compensations seems logical. So in case that this compensation doesn t exist, it would be unattractive to invest in illiquid securities. 5.3 Limitations of the thesis The fact that only 25 stocks from the Dutch market are used to test the relationship can be seen as a limitation of this thesis. The results are also influenced by the fact that the tested Dutch firms, are relatively big firms. So these stocks are not representative for the whole market. Another limitation is the availability of the bid-ask spread data of the 25 stocks. For most stocks this data was only collected since 1995. Therefore most regressions are tested for a maximum of 180 months, which is 15 years. This number of months is a small time horizon, compared with other researches in the same field of study. 5.4 Recommendations for further research There are some opportunities for further research on this topic. Extending the model is seems an option. Some macro factors such as the economic state of the country or the currency exchange could be included to the model. The thesis could also focus on another market, a specific industry or another country. Another idea is to raise the number of stocks that are researched. Further research could also use other securities than stocks, such as options, funds and bonds. 24
References Acharya, V. V. and Pedersen, H. J. (2005). Asset Pricing with liquidity risk. Journal of Financial Economics, Vol. 77, p. 375-410 Amihud, Y. (2002). Illiquidity and stock returns: cross-section en time-series effects. Journal of Financial Markets, Vol. 5, p. 31-56 Amihud, Y. and Mendelson, H. (1986). Asset Pricing and the bid-ask spread. Journal of Financial Economics, Vol. 17, p. 223-249 Breen, W. J., Hodrick, L. S. and Korajczyk, R. A. (2002). Predicting Equity Liquidity. Management Science, Vol. 48, No. 4, p. 470-483 Brennan, M. J. and Subrahmanyam, A. (1996). Market microstructure and asset pricing: On the compensation for illiquidity in stock returns. Journal of Financial Economics, Vol. 41, p. 441-464 Chen, N. and Kan, R. (1989). Expected return and the bid-ask spread, University of Chicago. Chordia, T., Huh, S. and Subrahmanyam, A. (2009). Theory-based illiquidity and asset pricing. The Review of Financial Studies, Vol. 22, No. 9, p. 3629-3668 Chordia, T., Roll, R. and Subrahmanyam, A. (2001*). Market Liquidity and Trading Activity. The Journal of Finance, Vol. 56, No. 2, p. 501-530 Chordia, T., Subrahmanyam, A. and Anshuman, V. R. (2001). Trading activity and expected stock returns. Journal of Financial Economics, Vol. 59, p. 3-32 Eleswarapu, V. R. (1997). Cost of transacting and expected returns in the Nasdaq market. Journal of Finance, Vol. 52, p. 2113-2127 Eleswarapu, V. R. and Reinganum, M. R. (1993). The seasonal behavior of the liquidity premium in asset pricing. Journal of Financial Economics, Vol. 34, p. 373-386. 25
Fama, E. F. and French, K. R. (1993). Common risk factors in the returnson stocks and bonds. Journal of Financial Economics, Vol. 33, p. 3-56 Hölmstrom, B. and Tirole, J. (1993). Market Liquidity and Performance Monitoring. The Journal of Political Economy, Vol. 101, No. 4, p. 678-709 Pástor, L. and Stambaugh, R. F. (2003). Liquidity Risk and Expected Stock Returns. Journal of Political Economy, Vol. 111, No. 3, p. 642-685 26