STOCK LIQUIDITY AND VOLATILITY IN EMERGED MARKETS DURING THE FINANCIAL CRISIS

Master Thesis STOCK LIQUIDITY AND VOLATILITY IN EMERGED MARKETS DURING THE FINANCIAL CRISIS Student: Maurits Gaudesaboos Student number/anr: 1261147/233679 Master Thesis Supervisor: Dr. J. C. Rodriguez Second reader: Dr. R.C.P. Frehen Submission Date: 21-08-2014 Date of Defense: 28-08-2014 Tilburg University Master of Finance School of Economics and Management 1

Abstract This paper uses models to test the change of behavior of volatility and illiquidity in times of large shocks. The relationship between volatility and illiquidity is tested daily between 1991 and 2013 in order to predict how the relationship changes in times of a crisis. The crisis defined is the financial crisis, starting at early 2008 and ending at late 2012. The results have strong implications for how illiquidity and volatility affect each other. Illiquidity is found to affect volatility less during the crisis event. Volatility affects illiquidity more during the crisis event. A negative shock increases volatility in returns, and this affects the liquidity of stocks. The models control for firm market capitalization and market illiquidity. When controlling for these control variables the results still hold. Explanations for the strengthened effect of volatility on illiquidity can be found in the flight-to-safety and flight-to-liquidity theories. This paper finds evidence that volatility is more important for investors and illiquidity less important during crisis compared to common economic times. 2

Luctor et Emergo -Zeeuws proverb For this paper I would like to thank first of all my parents and my sister, without their support I would never be able to stand where I am standing now, dr. Rodriguez for the helpful comments on my work, my fellow students for supporting me during the lectures, Daan for the thesis talks and last but certainly not least Marjon for the mental support and encouragement. 3

Table of Contents 1. Introduction 5 2. Theory Development 8 2.1 Stock volatility and returns 8 2.2 Stock Liquidity and Returns 9 2.3 Stock volatility and liquidity 11 3. Methodology 13 3.1 Data 13 3.1.1 Data Sources 13 3.1.2 Data Cleaning 13 3.1.3 Data Constructing 14 3.2 Returns, volatility and illiquidity 15 3.3 Models 16 4. Results 19 4.1 Returns and the financial crisis 19 4.2 Volatility and the financial crisis 20 4.3 Illiquidity and the financial crisis 22 4.4 Change in standardized betas 23 5. Discussion 26 References 28 Tabulations 31 4

1. Introduction Liquidity and common stock returns are a popular topic of discussion in economic theory. From the aggregated level towards the single stock-based level, the relationship between the two has been widely discussed since the relationship was discussed as by Roll (1984). Sharpe (1964), Lintner (1965) and more argue that volatility and returns are even a bigger aspect of financial theory. Combining stock volatility with the market exposure of that stock has given us the idiosyncratic risk exposure. However, the underlying relationship between volatility and liquidity has not received full attention on the level of individual stocks. Shocks affect investor behavior (Barberis, Shleifer & Vishny, 1998; Kim & Wei, 2002) and investors tend to flee towards safe and liquid securities in times of shocks. Flight-to-quality (FTQ) and flight-toliquidity (FTL) are common theories providing explanations for these phenomena. FTQ and FTL are both commonly observed in times of crisis or shocks. FTQ can be observed as the phenomenon in which investors are willing to sell their high-risk assets and flee towards more safe assets in times of shocks, i.e. investors sell their volatile securities and replace them with less volatile securities, such as T-bills (Bernanke, Gertler & Gilchrist, 1999). FTL often occurs with FTQ as they are both common during shocks. FTL is the observed phenomenon in which investors flee towards more liquid assets and thus sell their illiquid assets (Longstaff, 2002). Both flights improve each other s effect, making the capital flow from high-volatile and highilliquid assets towards low-volatile and liquid assets. This could have implications for the relationship between liquidity and volatility (Vayanos, 2004). One would expect that, in common economic times, illiquidity and volatility interact positively. Single investors have more price influencing power over illiquid stocks than over liquid stocks as the bid-ask spread is larger for illiquid stocks (Amihud, 2002). The other way around it holds that investors that are more riskaverse are more likely to avoid high volatile securities, increasing the illiquidity of that security. More illiquid assets could be more volatile (Domowitz, Glen & Madhavan, 2001). When a shock kicks in, the effects could become even stronger due to the FTL and the FTQ. The change of this relationship is not tested in economic literature so far. The goal of this paper is to provide a contribution to the literature regarding the change in the relationship between illiquidity and volatility in times of shocks. The main testable hypothesis for this paper is that stock illiquidity and stock volatility become more dependent in times of shocks. This will be tested using separate hypotheses for the effect of volatility on illiquidity and of illiquidity on volatility. The formal hypotheses for both relationships are that lagged volatility has a significantly positive and increased effect on illiquidity during the crisis and that lagged illiquidity has a significantly positive and increased effect on illiquidity during the crisis. This paper focuses on the financial crisis starting at early 5

2008 and ending at late 2012. The main hypotheses will be discussed structured using supporting hypotheses. These hypotheses will contribute to structured models used to discuss the main hypothesis as stated. The results on this hypothesis are stated in Chapter 4.2 and 4.3. The relationship between return and volatility is discussed in Chapter 2.1. As literature does not comes to an agreement regarding the influence of volatility on return, this relationship will be tested in this paper. Also, testing this relationship gives insight in the importance for investors of the results regarding the change in relationship between illiquidity and volatility. The hypothesis developed here is that lagged daily common stock volatility has a positive and significant impact on common stock return. This hypothesis follows the main implications of Baillie and DeGennaro (1990). Regarding the relationship of stock illiquidity and stock return literature is more united. As discussed in Chapter 2.2, higher stock illiquidity results in higher illiquidity risk premium and thus higher stock return (f.e. Amihud, 2002). As stated in literature, the hypothesis developed here is that lagged daily illiquidity has a positive and significant influence on stock returns. The hypothesis on the effect of stock illiquidity and stock volatility on returns will be tested in Chapter 4.1. Regarding the impact of illiquidity on volatility it is difficult to determine in which direction effects will move. As explained in Chapter 2.3 both could affect each other for different reasons. In order to test if illiquidity affects volatility and/or volatility affects illiquidity different hypotheses will be tested. Regarding the hypothesis that illiquidity affects volatility, it is expected that lagged illiquidity has a positive and significant impact on volatility. This hypothesis is tested in Chapter 4.2 and is found to be supported in the data. Next the influence of volatility on illiquidity will be tested. The hypothesis stated regarding this relationship is that lagged volatility has a positive and significant influence on illiquidity. This relationship is tested in Chapter 4.3 and is found to be supported by the data. Both hypotheses support the fact that illiquidity and volatility affect each other. As said, the main testable hypothesis is whether stock illiquidity and stock volatility became more dependent during the crisis event. From Chapter 4.2 and 4.3 it can be concluded that the impact of lagged volatility on illiquidity has significantly increased during the crisis event. However, the impact of illiquidity on volatility seems to decrease significantly during the crisis event. This implies that volatility became more important for illiquidity, but illiquidity became less important for volatility during. The results come with implications for investors. Results show that the impact of illiquidity on volatility has decreased in the crisis event. The opposite if found for the impact of volatility on illiquidity. Here results show that volatility has a larger impact and significant impact on common stock illiquidity. This implies that investors tend to attach more value towards less volatile stocks than they do towards more liquid common stocks. This could imply that investors are more prone towards the FTQ phenomenon than the FTL during the shock event. 6

This paper proceeds as follows. Chapter 2 introduces the literature. The relation between volatility and returns is a widely discussed topic in economic theory. Also the effects of stock liquidity and stock returns are broadly discussed. However, the relationship between volatility and liquidity in times of shocks on the stock-based level is not broadly discussed. The FTQ and FTL could provide some insights in how this relationship might change. Chapter 3 discusses the empirical structure and the models used. Because of the ambiguity of the relationship and the question of whether the financial crisis has had any influence on the relationship between volatility and illiquidity, several models are developed. Chapter 4 presents the results of the tested models as given and discussed. Theoretical explanations are discussed regarding the results. Chapter 5 concludes and offers remarks regarding the performed study. 7

2. Theory Development Now that the research question has been discussed the theoretical framework will be elaborated upon. First, the relationship between stock volatility and return will be briefly discussed. Next, the current state of literature between illiquidity and returns will be discussed. This Chapter will conclude with possible theoretical relationships between illiquidity and volatility. 2.1 Stock volatility and returns The relationship between stock returns and volatility is a widely discussed topic in general finance. Theoretical asset pricing models (e.g. Sharpe, 1964) links the excess returns of stocks to their volatility and the correlation between its return and the return on the market portfolio using a simple form of the Capital Asset Pricing Model (CAPM). However, the magnitude and the direction of this relationship is a center of debate for decades. Baillie and DeGennaro (1990) summarize that most models show a positive relationship between stocks expected returns and volatility. However, there is not very much evidence of the statistical significance of that relationship when they control for kurtosis. Since this paper seems somewhat dated, it is interesting to review more recent literature. Recent studies provide no agreement on the relationship between returns and volatility. Chua, Goh and Zhang (2010) do find that there is a positive and significant relationship between stock returns and idiosyncratic volatility. They collected a wide data sample on a long horizon for stocks traded at the NYSE, AMEX and NASDAQ. This gives their results some robustness; however this is only evident on stocks traded on U.S. exchanges. Li, Yang, Hsiao and Chang (2005) examined the relationship between expected stock returns and volatility across 12 large markets. The paper uses a variety of measures for volatility. Doing this allows them to reform the question on the relationship between returns and volatility, towards how to measure volatility. Li et al. uses both parametric and semiparametric methods to model volatility. Building their parametric and semiparametric models on a GARCH framework, they conclude that their semiparametric model is more robust than a parametric conditional variance model. The results support the claim that stock return volatility is negatively correlated with stock returns (Bekaert & Wu, 2000). A theoretical explanation given is that this negative relationship is based on the amount of leverage. Negative returns increase a firm s leverage, which increases the stock s risk and accordingly increases its volatility. Another explanation is based on volatility feedback. When volatility is priced and it increases as anticipated, the required rate of return on equity increases. This leads to a stock price decline. This paper has some important implications for the relationship between stock returns and volatility and, thus, also for stock volatility and liquidity. It seems that the current state of literature still does not agree on the effects of stock volatility on stock returns. 8

2.2 Stock Liquidity and Returns Now that the ambiguity in the relationship between stock volatility and returns has been discussed, it is necessary to take a closer look at the relationship between liquidity and return. Stock liquidity and return are a special topic in the financial theory. Intuitively, it seems clear that liquidity could influence returns. This could be both on an aggregated level and on a stockbased level regarding liquidity. The liquidity (or illiquidity) of a security influences the return on that security. This follows from the fact that investors are only willing to hold more illiquid securities when they are compensated for the additional illiquidity (Amihud & Mendelson, 1986). Thus, a lack of liquidity is seen as more risky and according to the basic financial principles, this should be rewarded with a higher return. Amihud and Mendelson studied the effect on the bid-ask spread on asset pricing for NYSE stocks. When defining illiquidity as the cost of selling a stock immediately, they found that a low bid-ask spread indicates a high level of liquidity. Therefore, an asset with a large difference in the bid-ask spread is likely to have higher expected returns. The main testable hypothesis of Amihud and Mendelson is that expected return is an increasing and concave function of the spread. They performed empirical tests on the stocks on monthly based returns and on the average of the beginning and the end-of-year relative spreads. Because market capitalization effects could influence the relationship between volatility and liquidity Amihud and Mendelson control for market capitalization. Small firms have had significantly larger risk-adjusted returns than large firms (Banz, 1981). The intuition is that larger firms are more likely to have larger trade volumes, which is supported by empirical data (Amihud & Mendelson 1986). Therefore, in the models developed in this paper, controls for market capitalization are implemented. Amihud and Mendelson were unable to provide an explanation on the role of illiquidity in asset pricing. However, they were able to conclude that there is a relationship between liquidity and returns. Amihud (2002) expands upon the findings of Amihud and Mendelson (1986). The hypothesis on the relationship between stock return and stock liquidity is that return increases with illiquidity, suggesting that there is some kind of illiquidity risk premium. Amihud (2002) examines that relationship over time. Proposing that, over time, the ex-ante stock excess return is increasing in the expected illiquidity of the stock market. Amihud developed the illiquidity measure that is used as foundation for the illiquidity measure in this paper, ILLIQ. It is defined as the daily ratio of the absolute stock return to its dollar trade volume, averaged over some period. ILLIQ can be interpreted as the daily price response associated with one dollar of trading volume for that stock. As with Amihud and Mendelson (1984), Amihud (2002) focuses on NYSE stocks. The results of Amihud (2002) holds that over time, expected returns of stocks seem to increase with expected illiquidity, implying evidence for the existence of an illiquidity risk premium. ILLIQ has a positive and highly significant effect on expected returns. This is supported by the results shown in Chapter 4.1. 9

Amihud also provides insights into the role of market illiquidity. His paper proposes that expected excess stock returns reflect a compensation for expected market illiquidity. Excess stock returns are an increasing function of aggregated market illiquidity. Because of the significant role of market illiquidity in stock returns, market illiquidity is considered as a control variable into further modeling for this paper. This will be further elaborated in Chapter 3. Amihud (2002) also provides useful insights into the relevance of firm market capitalization on illiquidity, concluding that illiquidity effects are stronger for small firms. This supports the view that, besides that there needs to be a control for market illiquidity; models should also implement a control for market capitalization. Acharya and Pedersen (2005) continued on the framework developed by Amihud (2002). Acharya and Pedersen start with the statement that investors do not challenge the level of financial liquidity, but that they challenge the variability and uncertainty of the liquidity. The same measurement of illiquidity is used as in Amihud (2002) to proxy for illiquidity. Acharya and Pedersen develop a liquidity-adjusted CAPM and find that this model fits the data better than the models without liquidity adjustments. The implications of their model are that investors should care about stock tradability in both market downturns and when liquidity dries up and that the liquidity-adjusted CAPM shows that positive shocks in illiquidity are associated with low contemporaneous returns. This leaves room for finding unexpected results. Furthermore, Acharya and Pedersen find weak evidence that liquidity risk is important over and above the effects of market risk and the level of aggregated liquidity. Pastor and Stambaugh (2003) continued on the relationship between aggregated liquidity and returns. However, they focused more on the aggregated level of liquidity and stock sensitivities towards innovations in the aggregate level of liquidity. Aggregated liquidity is seen as a state variable describing the current market situation. This gives the aggregated liquidity, or market liquidity, applicable characteristics to test in the relationship between liquidity and volatility. A relevant conclusion made by Pastor and Stambaugh is that stocks that are more sensitive to aggregated liquidity have higher expected returns and market capitalization. Emerging markets have some interesting characteristics regarding liquidity. Emerging markets often face large shocks in liquidity and the role of liquidity is larger in emerging markets since there is less (Bekaert, Harvey & Lundblad, 2007). Bekaert et al. notice that literature written on liquidity is primarily based on U.S. securities. Observing that emerging markets offer unique characteristics, they collect data on the Standard and Poor s Emerging Markets Database. Also, for emerging markets, it holds that unexpected liquidity shocks are positively correlated with returns. Furthermore, they find that these liquidity shocks are negatively correlated with dividend yields, which support the previous views of, for example, Acharya and Pedersen (2005) and Pastor and Stambaugh (2003). Now that the relation between returns and volatility and returns and liquidity has been discussed it is now time to associate the 10

two relations. Above all, the purpose of this paper is to test the relation between liquidity and volatility and what happens to this kind of relation in times of shocks? To provide insights on this question, it is necessary to take a closer look at investor behavior in times of shocks and to connect this behavior to liquidity and volatility. 2.3 Stock volatility and liquidity The attribution of this paper is to test whether the relationship between stock liquidity and stock volatility changed in the past large financial shock starting at early 2008 and ending at late 2012. When following classical financial theorem, it is not necessary to expect large changes. As the volatility and/or beta increases, higher returns are expected. Liquidity is also affected by large shocks. Liquidity could dry up in markets; the classical financial theorem states that this will be corrected for higher returns. However, taking into account more recent literature, this does not always hold for all types of assets. Effects, such as the FTL and FTQ, increase the demand for low-volatile, liquid securities and reduce the demand for highly illiquid and volatile securities, demanding an additional premium for the latter type of securities. Focusing on bonds, it reveals that liquidity premiums vary over time and widen dramatically during extreme market changes (Vayanos, 2004). Assuming that most investors are fund managers, withdrawals are more likely to occur in highly volatile times. This is, because managers that are more likely to underperform, or even make losses, are more likely to face withdrawals. When managers are more likely to face withdrawals, they are less interested in illiquid assets as they cannot be liquidated quickly and cheaply. This increases the liquidity premium demanded by investors. During more volatile times, investors become more risk averse. This could improve the preference of fund managers regarding liquid assets in times of shocks. Making a distinction between the degree of both FTL and FTQ is difficult (Ericsson & Renault, 2006). Both have different motives, but the effects tend to pile up. However, both are usually positively correlated for Ericsson and Renault s test on U.S. markets. The model used by Beber, Brandt and Kavajecz (2009) captures the different levels of effects of the FTL and FTQ. They found that, regarding bonds, the largest part of the spread can be explained by the differences in credit quality and thus the FTQ. However, they found evidence that, in times of large cash in- or outflows, liquidity explains a greater proportion of the yield spreads than the quality. This implies that, in times of large shocks in cash flows the FTL is more important for the yield spread than the FTQ. A remark that needs to be made here is that the FTQ still provides a large fraction of the spread. The main finding of Beber et al. (2009) is that investors demand both quality and liquidity, yet they do so at different times and for different reasons. 11

Brunnermeier and Pedersen (2009) connect market liquidity to funding liquidity. As assumed by Vayanos (2004), most investors are fund managers who need to borrow capital in order to trade. The model used shows that the funding of traders affects market liquidity and also that market liquidity affects the funding of traders. Furthermore, they (Brunnermeier & Pedersen, 2009) show that market liquidity can suddenly dry up. In addition, market liquidity has commonality across securities, is related to volatility, is subject to FTQ and co-moves with the market. The finding that market liquidity is related to volatility can show some support to the hypothesis that stock volatility and stock liquidity are related. As market liquidity affects both volatility and liquidity, investors margin requirements are increasing with volatility. When combining the discussed insights and theory, it seems plausible to suggest that there is a direct link between stock liquidity and stock volatility and that this relation may change in times of large shocks. Market liquidity seems to affect both, which suggest that there is a combined factor. This is particularly true considering that stock liquidity also affects market liquidity (Brunnermeier & Pedersen, 2009). FTQ and FTL also seem to affect both (Spyrou, Kassimatis & Galariotis, 2007). As more volatile securities are traded for the more secure securities and liquid securities are replacing illiquid securities, it follows intuitively that both are affecting one other. 12

3. Methodology In this Chapter, the empirical design, models, data and variables will be explained. First, the data and variables will be discussed. Next, the models used will be elaborated upon. As discussed before, separate models are used in order to test whether the data fits the current state of literature and to test whether the financial crisis has had any effect on the relationship between volatility and liquidity in regard to the research question. 3.1 Data 3.1.1 Data Sources Data is retrieved from Thomson Reuters DataStream. The markets focused on are globally diverse. These markets include the S&P-500, DAX-30, CAC-525, FTSE-625 and the S&P/TSX-370. This selection is a sample of the FTSE developed market group list and supports the representativeness of the data as sample for emerged markets. This also excludes any large shocks in liquidity or volatility due to emerging economies or transitions, thus, controlling for large changes in the aggregated market liquidity and/or volatility. The total number of firms collected in this dataset is 2,050. The data is retrieved daily starting from 01/01/1991 until 31/12/2013. The daily collection is done because this paper is interested into fast changes between common stock volatility and liquidity. Therefore yearly or monthly data are not as suitable as daily data as volatility shocks can be harder to observe when returns are collapsed monthly. 3.1.2 Data Cleaning The observed variables retrieved from Thomson Reuters DataStream are stock prices (p), trade volume (vo), common shareholders equity (cse) and returns adjusted for dividends, or return index (ri). These variables are used to construct the daily dollar trade volume (dvt), illiquidity (ILLIQ), firm market capitalization (MarketCap), market illiquidity (MarketILLIQ) and volatility (volatility). The dataset is screened and filtered using static files. The datasets are collected per market and are merged into one dataset. The purpose of the filter is to obtain a clear dataset without data errors and includes only values or missing values per firm. The markets and variables chosen for the data are convenient are accessible. This avoids a lot of problems regarding the availability and correctness of the data. Still some unknowns are flagged as $$"ER", 2311, NO DATA AVAILABLE. These unknowns are cleared for the whole dataset across all variables. Also firms that have data only five years or less back from 2013 available are cleared for the whole data set. Firms that did not experience the crisis event are not present in the dataset, so there is no need to clear them. 13

3.1.3 Data Constructing The daily dollar trade volume is constructed by multiplying trade volume with the stock price. Firm market capitalization is constructed by multiplying common shareholders equity with stock prices. Returns are based on the return index. The return index corrects for dividend payments and is a more accurate measure of returns than deriving returns from stock prices. Market illiquidity is constructed as the aggregated level of illiquidity on any given date. Market capitalization is normalized for the complete data set. The crisis dummy is created for the crisis event which is assumed to start on 01-01-2008 to 28-12-2012. This dummy will have a value equal to 1 for this time event and 0 for all other time events. Using this crisis dummy allows the model to test if the crisis had a significant impact on the dependent variable. The period is chosen because the most crisis-related activities happened or their effects occurred within this time sequence, bank defaults, house price declines and unemployment increases. The assumption made here is that this period captures the effects of the past financial crisis. Remarked can be the time period tends to be somewhat long for U.S. and Canadian stocks. Including a high number of European stocks justifies the longer time period. Using the crisis dummy, interaction terms are constructed; this is done for lagged volatility and lagged illiquidity. These interaction terms allows the models to test how lagged volatility affects illiquidity and how lagged illiquidity affects volatility. The interaction terms, l1illiqcrisisdum and l1volatilitycrisisdum, are constructed by multiplying the crisis dummy by the lagged volatility and lagged illiquidity. Illiquidity and volatility are determinative components of the research question. Proxies for these variables need to be made, because both variables are not widely observable. The proxy selected for illiquidity is the Amihud (2002) illiquidity measure. This proxy is commonly used in literature (e.g., Acharya & Pedersen, 2005). This paper adjusts the measure towards a daily measurement of illiquidity. This allows the models developed to test for daily changes in illiquidity and the effect on/off volatility. The assumption made is that ILLIQ is an estimator for illiquidity. The Amihud illiquidity measure is constructed using the following formula: (1) ILLIQ is calculated as the average ratio of the daily absolute return to the dollar trading volume on that day. Intuitively, this indicates that when a security has a high value for ILLIQ, the security s price moves substantially in response to little trading volume (Acharya & Pedersen, 2005). This variable is estimated on a daily basis as this paper is interested in fast changes of volatility and illiquidity. 14

Volatility is estimated using a GARCH model to find the variance of returns. The GARCH model is estimated across the whole time series. GARCH models are widely used in financial literature to capture and estimate volatility. The GARCH variant estimated is a GARCH (1,1) model, as suggested by the findings of Hansen and Lunde (2005). Hansen and Lunde do not find any evidence that a GARCH (1,1) model is outperformed by more sophisticated models. Also, when testing if a GARCH (1,2) model or (1,3) model fits the data better, no concave relationship is found for the data. Therefore, this paper will adhere to the GARCH (1,1) model. Variance, used to construct the volatility, is therefore estimated as follows:. (2) The variance of the model is used as an estimate for the variance of the return. Next, volatility can be constructed by taking the square root of the variance. Controls are implemented in the models for market capitalization and market illiquidity. The models are estimated as normal ordinary least squares models. As discussed in Chapter 2, market capitalization and market illiquidity has historically had an influence on stock volatility and stock liquidity (Banz, 1981; Brunnermeier & Pedersen, 2009; Pastor & Stambaugh, 2003). Implementing control variables for both models adds theoretical power to the model. 3.2 Returns, volatility and illiquidity Returns and volatility and returns and liquidity are widely tested in literature. Following the theories described in Chapter 2, one would approximately expect increasing returns with increased volatility and likewise with increased illiquidity, as investors face more risk or higher costs and want to be compensated for these additional risks. The research question of this paper is whether that relationship changes during the financial crisis. Plotting the returns, it is observable that returns fluctuate more during the crisis period than during the non-crisis period. The median return is used for plotting, because there are some large outliers observed in the data when using the mean return. As graph 1 illustrates, the largest shifts in the median returns are observable between the start and the end of the crisis. Also, large shocks in returns can be found around the year 2000. This can be connected with the collapse of the internet bubble. The large shifts in daily returns indicate shocks in volatility. Graph 2 shows us that, regarding innovations in the mean liquidity, it is not that clearly observable that abnormal shocks happened in the crisis event, although it is noteworthy that two major shocks occurred in the illiquidity innovation during that period. However, the largest shocks are observed in 2007 and around the year 2000. This can be explained by the internet bubble in 2000 and the collapse of the trust amongst banks in late 2007. Note that these spikes are increases in illiquidity and thus indicate that illiquidity decreases as the graph tops. Large peaks indicate large shifts in 15

liquidity. Regarding the plotted graphs, shocks are observed in illiquidity and in returns. This gives some evidence towards the research question that the relationship might have changed during the crisis. 3.3 Models Now that shocks have been observed in stock return volatility and illiquidity innovation between 2008 and 2012, the formal models can be derived. Formal models will be distinctly derived for returns, volatility and illiquidity to test the hypothesis as stated in Chapter 1. However, when doing this, several problems arise. Because of reverse causality it is difficult to argue in which direction the effect holds, it cannot be addressed whether illiquidity affects volatility or vice versa. Because of this endogeneity, models cannot predict, for example, how today s illiquidity affects today s volatility. Therefore the models will use lagged explanatory variables (Reed, 2013). This allows testing how, for example, yesterday s illiquidity affects today s volatility. Testing with lagged explanatory variables allows to make claims on how yesterday s variables affects today s dependent. This also deals with the reverse causality issue, as lagged variables are intuitively not dependent on today s values. Using lags allows drawing conclusion on the relationship of the dependent and the lagged independent variables. When doing this with common independent variables, as illiquidity and volatility, endogeneity problems will arise. As discussed in Chapter 2, the relationship of volatility and liquidity on return is a commonly discussed topic. Consistency does not exist among the underlying relationships. Therefore a model is tested to support claims that volatility and illiquidity have an influence on returns. In Chapter 3.1, the control variables are discussed. Market capitalization and market illiquidity will be added to the model to control for any of the effects they could possibly have. The model of volatility and illiquidity also include a lag for market illiquidity as market illiquidity is structured as the total amount of stock illiquidity on any given day. This could blur any results regarding the effect of illiquidity and volatility when only taking into account the common market illiquidity as this is also dependent on the common stock illiquidity. Also the dummy variable for the crisis starting January 2008 and ending December 2012 are implemented in the model. This is done to test whether the crisis has had any significant effect on returns at all. Finally, lags are implemented into the model. Because of the clustering of volatility, the temporal persistency of illiquidity and the short-term momentum lags are included for returns, volatility and illiquidity. An assumption made here is that the information is incorporated into the prices within one day, so only level 1 lags are included. The formal testable expression of daily stock return then becomes: 16

. (3) This formal expression should capture the effects of illiquidity and volatility on returns while controlling for market capitalization effects and market illiquidity. Lagged values are used for illiquidity, volatility and market illiquidity because of endogeneity issues. The model should also capture whether the financial crisis, as defined, has had any significant impact on returns. Finally, it tests whether the impact of the lagged volatility and lagged illiquidity differs during the crisis using the interaction terms. Finding significant and supports the claim that effects of illiquidity and volatility on returns has changed during the crisis event. Results as estimated by equation (3) are shown in table 1 as model 5. After returns are related to illiquidity and volatility, it is appropriate to test the relationship of illiquidity and volatility. Here, a lag 2 is incorporated in the model for illiquidity. Volatility is tested as follows:. (4) Volatility is a function of lagged illiquidity with lag 1 and lag 2, lagged volatility, market illiquidity, lagged market illiquidity, market capitalization, crisis dummy and an interaction term of the crisis dummy and lag 1 illiquidity. Besides a lag 1 for illiquidity, an additional lag 2 is incorporated to test whether this could still has its effect on volatility. Finding a significant will provide support for the claim that the effect of lagged illiquidity on volatility changed during the crisis event. The results of this regression are shown in table 2 as model 10. Now that a formal model of volatility has been derived it is appropriate to describe the illiquidity model. Illiquidity is specified as a function (5) of lag 1 and lag 2 volatility, lagged illiquidity, market illiquidity, lag 1 market illiquidity, firm market capitalization, a crisis dummy and an interaction term of the crisis dummy and lag 1 illiquidity:. (5) The results of regression (5) are shown in table 4 as model 15. Equation 4 and 5 will be estimated structured on the whole dataset, the crisis event and the whole dataset excluding the crisis. Results regarding the coefficients are described in order to control whether the relationship of illiquidity and volatility has changed. Conclusions are drawn based on the tables. 17

The volatility and illiquidity model include a lag 2 for the independent illiquidity and volatility. This is done so, because the market may need to have additional time to incorporate full information on events regarding the crisis as a lot of economic related news is pumped into markets during such events. The beta coefficients are discussed in Chapter 4.4 to check if the relationship between the dependent and independent standard deviations has changed. This could provide further support for the claim that the effect of volatility on illiquidity and vice versa has changed during the crisis period. This will be discussed alongside any changes in correlations for the crisis event. 18

4. Results The results of the empirical tests performed as discussed in Chapter 3, will be discussed in this Chapter. This is done in five separate steps: (i) First stock return is tested against the dependent variables as described in equation (3). (ii) The model of illiquidity is tested as specified in equation (4). (iii) The model of volatility is tested as specified in equation (5). (iv) The betas of both models are tested for the different time periods. 4.1 Returns and the financial crisis Classical financial theory has widely discussed the relationship between stock volatility and excess returns. For this model, the normal returns are used as it is not necessary to derive a formal CAPM. The main purpose of this part is to test whether the financial crisis has affected stock returns. Finding a significant effect on the crisis variable will support the theoretical background that the crisis has had its influence and can give some direction for the change in effect in the relationship between liquidity and volatility. Stock returns are regressed as formal derivation (3). The results are displayed in table 1 with daily stock returns as the dependent variable. First, returns are regressed against lagged volatility. This test whether the lagged volatility does influence stock returns. Volatility is found to have a negative and insignificant influence on stock returns. The negative sign implies that an increase in volatility leads to a decrease in returns. However, this effect is found to be insignificant, so the hypothesis that there is a positive and significant relationship between lagged volatility and returns is not supported. Next, the lagged illiquidity factor is tested against returns. Illiquidity is found to have a positive and significant impact on returns. This implies that, as the lagged illiquidity rises, stocks require higher returns. The results corresponds with the current state of literature on the liquidity risk premium (Amihud, 2002; Acharya & Pedersen, 2005) and supports the hypothesis that lagged illiquidity has a positive and significant influence on stock returns. As illiquidity rises, the returns rise as well because of the higher illiquidity premium demanded by investors. Model 3 combines illiquidity and volatility. Here, it is observed that volatility is still negative and insignificant. In model 4, market capitalization and lagged market illiquidity are added to the model. Also, the dummy variable for the financial crisis is added. Model 4 yields the same results regarding volatility and liquidity. Market capitalization is found to be significantly negative but only at the 5% level for stock return. Market illiquidity is found to be significantly positive for the 10% level. The crisis dummy is found to be significantly negative. This implies, that returns were significantly lower during the crisis event. Model 5 introduces interaction terms for the lagged volatility and the crisis dummy and the lagged 19

illiquidity and the crisis dummy. These interaction terms indicate whether there is a significant difference of volatility and illiquidity during the crisis event. Results show that this claim may only be supported at the 10% significant level. The interaction terms are both found to be negative. This would imply that the financial crisis did not yield any differences between the effect of the lagged volatility and lagged illiquidity on the common stock return. However, this claim is not strong as the interaction terms are found to be significant at the 10% level. The crisis dummy is found to be significantly negative in test 4 and 5. Hence, claims that the financial crisis, as defined in Chapter 3, has had a negative impact on stock returns is supported. This supports the hypothesis that the ongoing relationship between volatility and illiquidity might have changed during that period. It is noteworthy that the goodness of fit remains low, implying that the model used does not fit the data very well and that some nuances need to be made regarding the claims made. Stock returns are significantly influenced by volatility, illiquidity, the crisis dummy, and lags of volatility, liquidity and returns as estimated. However, the predictive power of the models is relatively low. 4.2 Volatility and the financial crisis The relationship between the financial crisis and volatility and the financial crisis influence on volatility are at the core of the main testable hypotheses that are tested in this Chapter. First, the relation between volatility and lag 1 and lag 2 illiquidity is tested in model 6. The results are found to be significantly positive, implying that stock illiquidity significantly influences volatility. This supports the claim that as stock illiquidity increases, volatility will increase as well due to, for example, the higher demanded illiquidity risk premium. In model 7 market capitalization and market illiquidity are incorporated in the regression. Both are found to be significant at the 1% level. Market capitalization is found to be negative, implying that as firm market capitalization increases, volatility decreases. Market illiquidity is found positive. This implies that as market illiquidity increases, volatility increases. Theoretical reasoning behind this could be that as illiquidity dries up, which often occurs in shocks, stock prices tend to jump more (Brunnermeier & Pedersen, 2009). Model 8 adds the crisis dummy variable. The crisis dummy is found significantly and positively influencing volatility. This implies that during the crisis event stock volatility was higher than it was during the non-crisis period. Returns have jumped more during the period of 2008 to 2012 than they did for the whole dataset. Besides the crisis dummy, model 8 also adds a lag for volatility. This lag is found to be positive and significant and indicates some evidence for clustering of volatility. Clustering of volatility is a common phenomenon found in literature (Andersen and Bollerslev (1997); Lux & Marchesi, 2000; Cont, 2007), although this relationship is not often as clear as Poterba & Summers (1987) state. Model 9 adds the interaction term as discussed in Chapter 3.3 for lag 1 illiquidity. Besides that, model 9 excludes 20

the crisis dummy. The interaction term is found to be significantly negative. Combining the coefficient of the interaction term of the lagged illiquidity and the common lagged illiquidity it yields that the crisis event affects the relationship in a less positive way. Concluded can be that the combined coefficient of the lagged illiquidity on volatility decreased during the crisis. This does not support the hypothesis that the effect of illiquidity on volatility became larger during the crisis as stated in Chapter 1. Moreover, it is likely that the effect of illiquidity on volatility decreased during the crisis event. However, some remarks need to be made regarding the strong claims made. The goodness of fit of the model is found to be low. The estimated model as described by equation (3) only yields a goodness of fit of 0.009. This is found to be very low and imply that the model does not fit the data that well. Adding the lagged volatility and the crisis dummy does give some increase in goodness of fit, from 0.004 to 0.009. The positive lagged values of illiquidity imply that stock volatility decreases with its liquidity. This implies, that for a lower level of illiquidity, and thus higher liquidity, stock volatility tends to decrease. This can also be explained by a combination of effects of the flight to liquidity and flight to safety. Investors tend to flee towards more safe and liquid stocks in times of shocks as they are more that volatile and liquid. The interaction term indicates that the effect of the lagged illiquidity on volatility decreased. Combining the coefficients of the lagged illiquidity it yields that during the crisis period the beta dropped from 0.0394 to 0.0182 implying that the magnitude of the effect decreased. Comparing the relationship of illiquidity on volatility for different time periods, model (10) is tested for different time scopes. The model is run for (i) the whole data set, (ii) part of the dataset that excludes the crisis and (iii) the dataset that focuses on the crisis. The results are reported in table 3. The lag 1 of illiquidity is found to be significant and positive at the 5% level for the period that excludes the crisis. However, focusing on the crisis, the lag 1 illiquidity is no longer found to be significant. This supports the result of table 2 that the role of illiquidity for volatility decreased during the crisis. Dropping from a significant effect at the 5% level to an insignificant effect during the crisis period supports the view that the role of lagged illiquidity decreased. Regarding the effect of lag 2 illiquidity, no large changes are observed. The magnitude of the lagged volatility, being significant for all time periods, does increase much. Excluding the crisis, the effect of the lag 1 volatility equals 0.0457, focusing on the crisis, this value increases to 0.0832. The role of market illiquidity and market capitalization seems to not be different during the crisis compared to that of the other time scopes, although a slight increase in the effect of market illiquidity can be registered during the crisis. This can be explained by the market liquidity drying up, increasing the effect of market illiquidity. Regarding the predictive power of the models, it is observable that, during the crisis, the goodness of fit increases. Concluding, it can be stated that the influence of lagged illiquidity on volatility decreased during the crisis. 21

4.3 Illiquidity and the financial crisis From Chapter 4.2, especially model 10, it can be concluded that illiquidity does have a significant impact on volatility. This Chapter takes another perspective, following equation (5) as discussed in Chapter 3 and testing whether volatility could influence illiquidity. The results are displayed in table 4. Model 11 regresses lag 1 and lag 2 volatility on illiquidity; this mirrors the regression run in model 6. The results state that the lagged volatilities have a significant impact on illiquidity. The coefficient for the lag 1 volatility is found to be positive opposing the negative sign for the lag 2 volatility. Model 12 includes controls for market illiquidity and market capitalization. Both are found to be significant at the 1% level. The magnitude of market illiquidity is found to be positive. This implies that as market illiquidity increases, common stock illiquidity increases. Economic motives for this to happen are markets that dry up. As markets dry up, it is harder to provide liquidity for stocks and therefore their illiquidity increases with the increased market illiquidity. Market capitalization is found to be negative. This implies that as firm market capitalization increases, illiquidity decreases. This could also be supported by the phenomenon that larger firms are more visible and therefore more traded which decreases the illiquidity. Model 13 includes a lag for illiquidity and the crisis dummy. The lag for illiquidity is found to be significantly positive. This implies that the as the lagged illiquidity decreases, the common illiquidity will decrease as well. The crisis dummy is found to be significantly negative. This would imply that the crisis has a negative effect on illiquidity and thus increases liquidity. However, when concluding this, it needs to be taken into account that worldwide market has developed and increased. As the crisis event is near the end date of the dataset, it needs to be taken into caution that the increase in liquidity might also be explained by an increase in globalization of which its effect is larger than markets that dry up. Model 14 adds the interaction term regarding the lagged volatility and the crisis event. This variable is found to be positive and significant and this would imply that the effect of the lagged volatility on illiquidity has increased during the crisis event. This supports the hypothesis that the effect of lagged volatility on illiquidity increased during the crisis event. Combining the coefficients of lagged volatility and the interaction term of volatility it can be concluded that the effect of lagged volatility on illiquidity significantly increased during the crisis period from 0.00837 to 0.02247. Model 15 expands the model with lagged market illiquidity. This is found to be insignificant. An insignificant lagged market illiquidity could imply that illiquidity is not dependent on lagged market illiquidity. However, illiquidity is found to be significantly affected by common market illiquidity. Comparing tables 2 and 4, it can be concluded that common market illiquidity seems to be an important determinant of both stock illiquidity and stock volatility. However, it is hard to 22

determine in which direction this relationship goes as market illiquidity can also be affected by stock illiquidity and stock volatility. From tables 2 and table 4 it can be concluded that the effects of common market illiquidity are highly significant. Higher market illiquidity increases the volatility of stocks and increases the illiquidity. This also holds for market capitalization. An increase in market capitalization is associated with a decrease in volatility and a decrease in illiquidity. Larger firms tend to be less volatile and more liquid. The crisis dummy is also found to be significant in both models. This implies that liquidity did not dry up as much as would be expected for stocks as stated by the hypothesis. However, this can also be explained by the way the illiquidity measure is constructed as large shifts in prices are more common in crisis and these large shifts increase the illiquidity measure. The lags of illiquidity and volatility play an important role in both models. For volatility, an increase in the lagged illiquidity increases the next day s volatility. The same holds for the lagged volatility. As the lagged volatility increases, the next day s volatility will increase. Although this effect does not seem to be that large, it is found to be significant. For illiquidity, the role of the lagged variables is different. The lagged volatility implies that a decrease in the lagged volatility is followed by an increase in the illiquidity of the stock on the next day. For the lagged illiquidity it implies that an increase in the previous day s illiquidity does have a positive impact on illiquidity the next day. As with table 2 and 3, it holds that the goodness of fit of the model is found to be very low. Reaching a maximum of 0.003, the R-squared indicate that the model does not fit the data that well. To further support the claim that the effect of lagged volatility on illiquidity increased, model 15 is run for different time periods. First the model is run for the whole sample, next, the model is run for only the period that excludes the crisis, and finally, it is tested on the crisis. Supporting results are found and are reported in table 5. The first thing to notice is that the lagged volatility is significantly related to illiquidity only during the crisis and only at the 5% level. Lag 2 volatility is significantly related to illiquidity during the crisis at the 10% level. However, both lags are not found to be significantly affecting illiquidity during the time period when the crisis is excluded. The lagged value of illiquidity is found to be significant during all periods. However, it loses some of its statistical power and magnitude during the crisis event. Testing the effect of the lagged market illiquidity for different time events still no significant results are found. The effect of market capitalization is found to be similar for all periods. Concluding, table 5 supports the claims made regarding table 4 that the effect of volatility on illiquidity increased during the crisis event. 4.4 Change in standardized betas Now that the changes in coefficients have been discussed between time events, it is convenient to inspect the changes in the beta coefficients. Using the beta coefficients allows the model to 23

support claims made in Chapters 4.2 and 4.3. The beta coefficients, as estimated, indicate that a one standard deviation increase/decrease of the independent variable leads to a beta standard deviation increase/decrease of the dependent variable. This allows the inclusion of relative weight to the changes in the coefficients as estimated in Chapter 4.2 and 4.3 as it shows the direction of changes in the standard deviation of the independent variable. The models tested are as in equation (4) and (5). The results of the beta coefficients are displayed in table 6 and 7. Table 6 indicates the standardized beta coefficients of the independent variables as regressed against volatility. The standardized beta coefficient of lagged illiquidity has increased in the past crisis when comparing the beta with other time events. However, this increase is only found to be relatively small. The beta increased for the dataset that excludes the crisis and the crisis event from 0.0611 and 0.0662, respectively. This implies a slight increase in the beta impact. Regarding the lag 2 illiquidity, it is observed that the beta coefficient increases from 0.00791 to 0.0188 during the crisis. The increases of the lag 1 and lag 2 beta coefficients imply that the standard deviations of the lags influence the standard deviation of volatility more. The beta coefficient of the lagged volatility increases relatively much, from 0.0665 to 0.0913 during the crisis. The beta coefficient of market illiquidity and the lagged market illiquidity increased dramatically. For the period that excludes the crisis period, it is observed to be 0.00296 and 0.00369. The beta coefficients increase during the crisis to 0.0333 and 0.0109. This expresses the importance of market liquidity and lagged market illiquidity for volatility. The beta of market capitalization also increased dramatically. Having a negative magnitude for all time events, it is observed that, excluding the crisis, the beta is equal to -0.00132 and increases during the crisis towards -0.0398. This result supports the role of market capitalization as found in table 3. Concluding, it can be stated that all beta coefficients increased during the crisis event. This implies that the relationship between the standard deviation of the explanatory variables and the standard deviation of volatility increases during the crisis period. In table 7, an observation is made that the relationship between the standard deviation of explanatory variables and illiquidity over time. Regarding the beta coefficient of lag 1 volatility it can be observed that the beta coefficient increased during the crisis period from 0.00833 to 0.023. This increase seems to be relatively large and shows some support for the claim that the effect of lagged volatility on illiquidity increased during the crisis period. The magnitude of the beta of lag 2 volatility tends to increase from -0.00591 to -0.0101 during the crisis. This gives also support for the hypothesis on the increased role of volatility on illiquidity. The beta of lagged illiquidity and market illiquidity decreases slightly. However, the magnitude of the beta coefficient of the lagged market illiquidity and market capitalization seems to increase relatively much. Lagged market illiquidity is not found to be significant in table 4 and 5, 24

so no claims can be made regarding the influence of the lagged market illiquidity. Concluding, it can be noticed that the beta coefficients of the lag 1 and lag 2 volatility seems to increase during the crisis event. As these explanatory variables are found to be significant, it supports the claim that the influence of the lagged volatilities on illiquidity has increased during the crisis event. 25

5. Discussion This paper tests whether the relationship between stock illiquidity and stock volatility changes in times of economic shocks. The study performed focuses on the past financial crisis of 2008 to 2012. Using daily data on emerged markets for the time period of 1991 to 2013, different models are tested for stock illiquidity and stock volatility. Results show that the underlying relationship between volatility and illiquidity increased during the observation period. I found that illiquidity is significant for volatility when testing the whole data set and when excluding the crisis event. However, when measuring the relationship during crisis, illiquidity does not significantly influence volatility any longer. For volatility the opposite holds. Volatility is not found to be significant for illiquidity during the whole time period and the period excluding the crisis. However, when measuring the influence of volatility on illiquidity during the crisis it is found to be significant. Combining the effect of the interaction terms, it can be concluded that the effect of illiquidity on volatility decreased during the crisis event and that the effect of volatility on illiquidity increased during the crisis event. Investors tend to be more aware of volatility effects and are more prone to the FTQ and so less towards the FTL during the crisis event. For single investors it holds that low volatile stocks are more interesting than more liquid stocks when comparing common stocks on these characteristics. The results could help investors in times of a crisis event as they can conclude that investor behavior regarding volatility and illiquidity changes during such an event. When observing the beta coefficients, as estimated, it is observed that the relationship between the standard deviations of illiquidity and volatility increased during the crisis event. This implies that the inner relationship between the two increased as fluctuations of both became more aligned. This implies that the variance of the returns increased and that the spread of illiquidity increased. It can be deducted that these effects occur due to the crisis event because in crisis events there is more volatility (Beber et al., 2009). Regarding the model of volatility and the economic significance it is observed that the role of lagged common stock illiquidity seems to diminish when comparing the crisis event with the non-crisis results. The coefficient of market illiquidity has increased during the crisis event. This implies that the effect of market illiquidity on volatility increased. Regarding the model of illiquidity it can be observed that the effect of market illiquidity on common stock illiquidity decreased during the crisis event. Dropping from 0.000759 to 0.00048 during the crisis the coefficient is no longer found to be significant. The effect of lagged volatility on illiquidity increased from 0.00795 when excluding the crisis to 0.0279, becoming significant at the 5% level. For the lag 2 value of volatility, negative results are found. However, these results seems to be insignificant for this data. In both models, the role of market illiquidity and market capitalization is found to be important. Also the role of the lagged dependent variables, volatility or illiquidity, are found to 26

be significant across both models. The l1volatiltiycrisisdum and l1illiqcrisisdum are found to be significant. This implies that the effect of both have changed during the crisis. For l1volatiltiycrisisdum it implies that the effect of lagged volatility on illiquidity has increased and for the l1illiqcrisisdum it holds that the role of lagged illiquidity on volatility has decreased during the crisis period. However, some remarks need to be made regarding the results. Results and significance levels seems to be very strong. This is partly due to the fact that a large dataset is used. Significant trends can be observed, although the trend can be not that relevant. The goodness of fit of the estimated model for illiquidity is found to be very low. This implies that the model does not fit the data well. Thus, modeling illiquidity is more difficult than formal Derivation (5) states. Modeling volatility does not give a high goodness of fit either. Although the goodness of fit is slightly higher than when testing illiquidity, it is still found to be low. This implies that the model of volatility does not fit the data that well. Another remark that can be made is that other large shocks occurred during the whole dataset. This may have blurred the results, but it is assumed that shocks do occur in normal economic cycle. The financial shock between 2008 and 2012 was an abnormally large case. Regarding illiquidity, the illiquidity measure in this paper is modified towards a daily measurement instead of a monthly measurement as supposed by Amihud (2002). This could also blur results because information regarding liquidity may take time to be incorporated by markets. However, measuring daily illiquidity does give some beneficial characteristics compared it to daily volatility. Shocks are more observable and more common to associate with illiquidity shocks. Also, the illiquidity measure itself is a topic of discussion as Goyenko, Holden & Trzcinka (2009) state. There has not been a perfect measure of illiquidity developed yet and, therefore, this paper needs to assume that ILLIQ is a sufficient measurement for illiquidity. Another flaw is that this paper only tests one large shock. The shock as defined could be distinguished into several shocks as the credit crunch and the euro currency crisis. Although the credit crunch affected the latter it would be interesting to compare how the relationship could change amongst different countries. Regarding endogeneity issues, models used lagged explanatory variables to control for this issue. However, it could still be possible that some form of endogeneity, like reverse causality, appears between the dependent and the independent variables. Regarding the FTQ and FTL, it would be interesting to know how each explains the change in relationship between illiquidity and volatility. This could give insights regarding the level of effect of FTQ and FTL. Another interesting suggestion would be to perform this study on emerging economies. That data would provide larger shifts in illiquidity and it would be interesting to see how volatility responds during a large crisis for emerging economies. Though, data issues may be a problem, it would be interesting to see how the relationship might change. Further it could be interesting to test how the relationship changes for single countries. 27

References Acharya, V. V., & Pedersen, L. H. (2005). Asset pricing with liquidity risk. Journal of Financial Economics, 77(2), 375-410. Amihud, Y. (2002). Illiquidity and stock returns: cross-section and time-series effects. Journal of financial markets, 5(1), 31-56. Amihud, Y., & Mendelson, H. (1986). Asset pricing and the bid-ask spread. Journal of financial Economics, 17(2), 223-249. Andersen, T. G., & Bollerslev, T. (1997). Intraday periodicity and volatility persistence in financial markets. Journal of empirical finance, 4(2), 115-158. Baillie, R. T., & DeGennaro, R. P. (1990). Stock returns and volatility. Journal of financial and Quantitative Analysis, 25(02), 203-214. Banz, R. W. (1981). The relationship between return and market value of common stocks. Journal of financial economics, 9(1), 3-18. Barberis, N., Shleifer, A., & Vishny, R. (1998). A model of investor sentiment. Journal of financial economics, 49(3), 307-343. Beber, A., Brandt, M. W., & Kavajecz, K. A. (2009). Flight-to-quality or flight-to-liquidity? Evidence from the euro-area bond market. Review of Financial Studies, 22(3), 925-957. Bekaert, G., Harvey, C. R., & Lundblad, C. (2007). Liquidity and expected returns: Lessons from emerging markets. Review of Financial Studies, 20(6), 1783-1831. Bekaert, G., Wu, G. (2000). Asymmetric volatility and risk in equity markets. Review of Financial Studies, 13(1), 1-42. Bernanke, B. S., Gertler, M., & Gilchrist, S. (1999). The financial accelerator in a quantitative business cycle framework. Handbook of macroeconomics, 1, 1341-1393. Brunnermeier, M. K., & Pedersen, L. H. (2009). Market liquidity and funding liquidity. Review of Financial studies, 22(6), 2201-2238. Chua, C. T., Goh, J., & Zhang, Z. (2010). Expected volatility, unexpected volatility, and the cross section of stock returns. Journal of Financial Research, 33(2), 103-123. Cont, R. (2007). Volatility clustering in financial markets: empirical facts and agent-based models. In Long memory in economics (pp. 289-309). Springer Berlin Heidelberg. 28

Chordia, T., Sarkar, A., & Subrahmanyam, A. (2005). An empirical analysis of stock and bond market liquidity. Review of Financial Studies, 18(1), 85-129. Domowitz, I., Glen, J., & Madhavan, A. (2001). Liquidity, volatility and equity trading costs across countries and over time. International Finance, 4(2), 221-255. Ericsson, J., & Renault, O. (2006). Liquidity and credit risk. The Journal of Finance, 61(5), 2219-2250. Goyenko, R., Holden, C. & Trzcinka, C., (2009), Do liquidity measures measure liquidity? Journal of Financial Economics, 92(2), 153 181. Hansen, P. R., & Lunde, A. (2005). A forecast comparison of volatility models: does anything beat a GARCH (1, 1)?. Journal of applied econometrics, 20(7), 873-889. Kim, W., & Wei, S. J. (2002). Foreign portfolio investors before and during a crisis. Journal of international economics, 56(1), 77-96. Li, Q., Yang, J., Hsiao, C., & Chang, Y. J. (2005). The relationship between stock returns and volatility in international stock markets. Journal of Empirical Finance, 12(5), 650-665. Lintner, J. (1965). The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. The review of economics and statistics, 13-37. Longstaff, F. A. (2002). The flight-to-liquidity premium in US Treasury bond prices (No. w9312). National bureau of economic research. Lux, T., & Marchesi, M. (2000). Volatility clustering in financial markets: A microsimulation of interacting agents. International Journal of Theoretical and Applied Finance, 3(04), 675-702. Pastor, L., Stambaugh, R.F. (2003). Liquidity risk and expected stock returns. Journal of Political Economy, 111, 642 685. Poterba, J. M., & Summers, L. H. (1987). The persistence of volatility and stock market fluctuations. Reed, W. R. (2013). ON THE PRACTICE OF LAGGING VARIABLES TO AVOID SIMULTANEITY. Roll, R. (1984). A simple implicit measure of the effective bid ask spread in an efficient market. The Journal of Finance, 39(4), 1127-1139. Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk*. The journal of finance, 19(3), 425-442. 29

Spyrou, S., Kassimatis, K., & Galariotis, E. (2007). Short-term overreaction, underreaction and efficient reaction: evidence from the London Stock Exchange. Applied Financial Economics, 17(3), 221-235. Vayanos, D. (2004). Flight to quality, flight to liquidity, and the pricing of risk (No. w10327). National Bureau of Economic Research. 30

Tabulations Graph 1 Median returns for the complete time period starting at 01/01/1991 and ending at 31/12/2013. Start of the crisis event as indicated is registered at 01/01/2008 and ends at 31/12/2012. The median return is taken daily and is computed as the daily percentage changes of the return index. 31

Graph 2 Innovations in mean illiquidity for the complete time period starting at 01/01/1991 and ending at 31/12/2013. Start of the crisis event as indicated is registered at 01/01/2008 and ends at 31/12/2012. Innovations in mean illiquidity are measured as the percentage change of the daily mean illiquidity. 32