Does monetary policy determine stock market liquidity? New evidence from the euro zone

Does monetary policy determine stock market liquidity? New evidence from a b s t r a c t The recent financial crisis has been characterized by unprecedentedmonetary policy interventions of central banks with the intention to stabilize financial markets and the real economy. This paper sheds light on the actual impact ofmonetary policy on stock liquidity and thereby addresses its role as a determinant of commonality in liquidity. Our results suggest that an expansionary monetary Euro area Journal of Empirical Finance 21 (2013) 54–68 Contents lists available at SciVerse ScienceDirect Journal of Empirical Finance j ourna l homepage: www.e lsev ie r .com/ locate / jempf in 1. Introduction The liquidity of financial markets, defined as “the ease of trading” (Amihud et al., 2005), has recently attracted a lot of attention, as the financial crisis highlighted its role as a precondition for efficient markets. Although central banks all over the world tried to ease financialmarkets during the recent crisis period bymeans ofmassivemonetary policy interventions, we knowsurprisingly little so far about the actual relationship of monetary policy on stock liquidity. Stock liquidity Monetary policy ☆ The opinions are strictly those of the authors a Authority (FMA). ☆☆ We thank the co-editor, Prof. Franz C. Palm, as we Cuaresma for their very helpful feedback on our pap Campus for Finance Research Conference at the WHU - Midwest Finance Association in Chicago (March, 02–0 ⁎ Corresponding author. E-mail address: [email protected] (G. Peter). 0927-5398/$ – see front matter © 2012 Elsevier B.V. A http://dx.doi.org/10.1016/j.jempfin.2012.12.008 E51 E52 G12 Keywords: policy of the European Central Bank leads to an increase of aggregate stock market liquidity in the German, French and Italian markets. Furthermore, the effect of monetary policy is significantly stronger for smaller stocks, suggesting a non-linear impact of monetary policy on stock liquidity. © 2012 Elsevier B.V. All rights reserved. a r t i c l e i n f o Article history: Received 8 September 2011 Received in revised form 8 December 2012 Accepted 10 December 2012 Available online 20 December 2012 JEL classification: E44 the euro zone☆,☆☆ Octavio Fernández-Amador a, Martin Gächter b,c, Martin Larch d, Georg Peter e,⁎ a Johannes Kepler University of Linz, Department of Economics, Austria b Oesterreichische Nationalbank (OeNB), Foreign Research Division, Austria c University of Innsbruck, Department of Economics and Statistics, Austria d Austrian Financial Market Authority (FMA), Department of Integrated Financial Markets, Austria e University of Liechtenstein, Institute for Financial Services; Fürst-Franz-Josef-Strasse, 9490 Vaduz, Liechtenstein nd do in no way commit the Oesterreichische Nationalbank (OeNB), or the Austrian Financial Market ll as the three anonymous referees, as well as Michael Pfaffermayr, Doris A. Oberdabernig and Jesús Crespo er. Furthermore, we are very grateful for the earlier comments on our study from the participants of the Otto Beisheim School of Management in Vallendar (January 12–13, 2011), the 2011 Annual Meeting of the 5, 2011) and the 2011 Annual Meeting of the Eastern Finance Association in Savannah (April, 13–16, 2011). ll rights reserved. Since Amihud and Mendelson (1986) suggested that stock returns are an increasing function of illiquidity, numerous successive studies investigated this relationship. Indeed, the empirical literature generally confirms the theoretical proposition that investors demand higher gross returns as compensation for holding less liquid stocks.1 Another well-established strand of the literature on asset liquidity documents that the liquidity of individual stocks exhibits significant co-movement,which is usually referred to as commonality in liquidity.2 Covariation in the liquidity of stocks implies that the illiquidity risk cannot be diversified and therefore illiquidity should be regarded as a systematic risk factor.3 Furthermore, the observed commonality suggests the assumption that there needs to be at least 4 55O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 one common factor that simultaneously determines the liquidity of all stocks in a market, which might be monetary policy. The hypothesis we test in this paper is that the monetary policy of central banks is a common determinant of stock liquidity. In particular, we examine the relationship between the European Central Bank's (ECB) monetary policy interventions and the liquidity of German, French and Italian stocks. Interestingly, there are only few relevant theoretical approaches. The inventory paradigm of the market microstructure literature suggests that inventory turnover and inventory risk affect stockmarket liquidity.5 In a nutshell, this paradigmproposes that stocks are expected to be more liquid if market participants can cheaply finance their holdings and perceive low risk of holding assets. Since monetary policy influences both the costs of financing and the perceived risk of holding securities, it follows that monetary policy should also affect stockmarket liquidity. Similarly, Brunnermeier and Pedersen (2009) develop amodel that addresses the interaction between funding liquidity and asset liquidity. Their model suggests that traderswho face capital constraints experience difficulties to meet margin requirements and therefore fail to provide liquidity to the market. The other way around, a deterioration of market liquidity reduces traders' funding liquidity through higher margin requirements. This may lead to a loss spiral and a lower liquidity, higher margin equilibrium. Following this reasoning, an expansionary (restrictive) monetary policy eases (exacerbates) constraints for margin borrowing and thus, facilitates (impedes) the funding liquidity of market participants. Another argument could be that both monetary policy as well as stock market liquidity are closely linked to business cycle movements. Thus, we could expect a considerable impact of monetary policy on stock liquidity, where the real economy might serve as the transmission channel. Few academic studies empirically examine the relationship between monetary policy and aggregate stock liquidity, and their results are to some extent ambiguous. Goyenko and Ukhov (2009) document strong evidence for the U.S. market (NYSE and AMEX) that monetary policy predicts liquidity for the period from 1962 to 2003. A tightening of monetary policy, as indicated by positive shocks to the federal funds rate and negative shocks to non-borrowed reserves, is shown to decrease stockmarket liquidity. Moreover, the bondmarket seems to serve as a transmitter that forwardsmonetary policy shocks to the stockmarket. On the contrary, Chordia et al. (2005) report only modest predictive power of monetary policy for stockmarket liquidity. For a sample of NYSE traded stocks they find that an expansionary monetary policy is associated with a contemporaneous increase in aggregated liquidity only during periods of crisis. The authors measure monetary policy by means of net-borrowed reserves and the federal funds rate. Soederberg (2008) studies the influence of 14macroeconomic variables on themarket liquidity of three Scandinavian stock exchanges between 1993 and 2005 and also provides mixed evidence. He finds that the policy rate is able to predict market liquidity on the Copenhagen stock exchange, whereas broadmoney growth plays amajor role on the Oslo stock exchange and short-term interest rates andmutual fund flows predict liquidity on the Stockholm stock exchange. However, no variable is able to forecast liquidity for all three exchanges. Similarly, Fujimoto (2003) studies the relationship betweenmacroeconomic variables and liquidity for NYSE and AMEX stocks. For the period ranging from 1965 to 1982, a positive shock to non-borrowed reserves increases liquidity, whereas an increase in the federal funds rate decreases liquidity. However, for the period from 1983 to 2001, neither shocks to non-borrowed reserves nor to the federal funds rate are able to predict stock market liquidity. We find that an expansionary (contractionary) monetary policy of the ECB leads to an increase (decrease) in the liquidity of stocks, which is in line with the main findings of Goyenko and Ukhov (2009). However, we observe this relationship not only at the macroeconomic level for aggregate liquidity by using vector autoregressive (VAR) models, but also at the microeconomic level for individual stocks by applying panel estimations. Contrary to earlier studies, we are able to report non-linear effects on the individual stock level, i.e. that the effect ofmonetary policy becomesweaker the higher themarket capitalization of the traded stock. Noteworthy, our findings are robust for three different markets (Germany, France, and Italy), seven measures of (il)liquidity (capturing trading activity, price impact and transaction costs) and two variables of monetary policy (base money growth and the Euro Overnight Index Average (EONIA) interest rate). We contribute to the existing literature in threeways. First of all, while previous research focuses primarily on the U.S. stockmarket and offers to some extent ambiguous results, this study investigates European data. The effect of monetary policy on stock market liquidity might differ between currency areas and across countries, particularly when taking the differences in the statutes and policy aims between central banks into account. We are not aware of any study analyzing in depth the impact of ECB monetary policy. Secondly, we extend the analysis of monetary policy and liquidity to the individual stock level. From a methodological point of view, the application of panel-fixed-effects givesmuch stronger evidence as some effects could be canceled out at an aggregated level due to (unobserved) heterogeneity among assets. Our panel approach controls implicitly even for unobserved time-invariant characteristics at the individual stock level. To our knowledge, this is the first study applying both panel and VAR models to this specific research 1 For a comprehensive overview of the literature about asset pricing and liquidity see Amihud et al. (2005). 2 See Chordia et al. (2005), Hasbrouck and Seppi (2001) or Huberman and Halka (2001) for the U.S. and Kempf and Mayston (2008) for the German market. 3 See for example Pastor and Stambaugh (2003) and Acharya and Pedersen (2005). 4 Chordia et al. (2000) propose in their outline for future work on commonality in liquidity: “A sensible next step would attempt to identify specific macroeconomic influences that correlate with time series variation in liquidity.” 5 Market microstructure theory deals with the determinants of the liquidity of individual stocks by focusing on stock characteristics and trade mechanisms. For an overview see O'Hara (1998) and Hasbrouck (2007). question. As shown below, the results in our panel models are not only much more robust than the VAR analysis of the aggregated market, but also offer additional insights regarding individual interaction effects, i.e. the linkage between the effect of monetary policy and themarket capitalization of stocks. This leads to the conclusion that unobserved heterogeneity across individual stocksmight play an important role, which cannot be captured by an analysis of aggregated liquidity. Finally, we add additional insights by employing in this respect untested, but generally well-acknowledged measures for both monetary policy and asset (il)liquidity. The paper is structured as follows. Section 2 describes the data set and the applied variables, including the measures of monetary (1998)we assume that a higher traded volume implies increased liquidity. It should be noted that the two trading activity variables can prices of stocks. We use the relative Roll (1984) estimate (R_REL), as it can sensibly be interpreted as a proxy for the bid-ask-spread. 56 O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 Secondly, in line with Amihud and Mendelson (1986), we use the relative bid-ask-spread (S_REL). For the estimations we compute monthly averages of the individual (il)liquidity measures of each stock for our panel regressions, as well as equally weighted cross-sectional averages of the (il)liquidity measures for our VAR analysis. 6 The monthly observations of the Roll impact ratio and the relative Roll measure areWinsorized only by the most illiquid 1%, since the estimation procedure of the Roll measure exhibits an implicit trimming of the most liquid securities. The Roll estimate is set to zero whenever the autocovariance of daily stock price changes is positive. 7 See Appendix A for details of the definition and computation of the (il)liquidity measures. be interpreted as liquidity proxies, since higher values are associatedwithmore liquid stocks. The othermeasures following below can be considered as illiquidity proxies, since an increase in these variables is associated with less liquid stocks. Price impact proxies indicate the responsiveness of prices to order flow. Firstly, the Amihud's (2002) illiquidity ratio (ILLIQ) quantifies the response of returns to one euro of trading volume. This illiquidity measure is very well established, particularly since Hasbrouck (2009) and Goyenko et al. (2009) report its adequacy as ameasure of price impact.We also apply the turnover price impact (TPI) which was proposed by Florackis et al. (2010) and can be interpreted as the return impact of a one percent stock turnover. By using the stock turnover rate instead of the traded volume in euro, TPI should be, by construction, less related to market capitalization or inflation than Amihud's (2002) illiquidity ratio. In this sense, it is expected that in an environment where nominal shocks dominate (inflationary environment) the two measures could offer a different conclusion, since TPI is more isolated to nominal effects than Amihud's (2002) illiquidity ratio. The third price impact proxy is based on thework of Goyenko et al. (2009), who propose a new form of price impact measures by dividing proxies for the bid-ask-spread by the traded volume in euro.We include the Roll impact variable (R_IMP), since Goyenko et al. (2009) conclude that it is a qualified measure for price impact (even when estimated from daily data). Finally, in order tomeasure transaction costs, we employ two variables that proxy the relative difference between the bid and ask policy and (il)liquidity. Empirical results at themacro andmicro levelwith respect to theGerman stockmarket are illustrated in detail in Sections 3 and 4, while results from the French and Italian markets are presented as a robustness analysis. Finally, Section 5 summarizes the results and draws some conclusions. 2. Data and hypotheses 2.1. Data set For our analysis we consider data of three major markets of the euro area, namely Germany, France and Italy. The sample period spans from the introduction of the euro in January 1999 to December 2009 (132 months). The considered stock universe includes all German stocks traded at the Xetra trading system, all French stocks traded at the Euronext Paris and all Italian stocks traded at the Milan stock exchange. The source of the capitalmarket data is Thomson Reuters Datastream. In our analysiswe only include a stock if it hasmore than 100 trading days per year, a share price greater than one euro and at least 15 observations of the (il)liquiditymeasures described in Section 2.2 in the respective month. In order to eliminate outliers and erroneous data, we also exclude the highest and lowest 1% of the computed returns and of the monthly (il)liquidity measures.6 All macroeconomic variables are from the ECB Statistical Data Warehouse. 2.2. Dependent variables: (Il)liquidity measures Since stock market liquidity is a broad concept with various facets, we employ seven different measures that capture the aspects trading activity (turnover rate and traded volume in euro), price impact (Amihud's 2002, illiquidity ratio, turnover price impact and Roll impact) and transaction costs (relative Roll, 1984, estimate and relative bid-ask-spread).7 Trading activity is considered as an indirect measure of a stock's liquidity. According to Amihud and Mendelson (1986), in equilibrium, liquid stocks should be held by investors with short investment horizons and, therefore, exhibit a higher trading activity. Similarly, Constantinides (1986) predicts that investors reduce their trading frequency for illiquid assets. The first proxy of trading activity is the turnover rate (TO) as was proposed, for example, by Datar et al. (1998). The stock turnover can be interpreted as the reciprocal of the average holding period, implying that stocks with higher turnover are on average held for shorter time periods and thus, exhibit an increased trading activity. The second variable employed is the traded volume in euro (TV). Following Brennan et al. 2.3. Explanatory variables 2.3.1. Central bank policy measures In line with previous literature we approximate monetary policy by using the monetary aggregate and the interest rate. Firstly, we use the rolling twelve-month growth rate of base money (BM).8 Base money is defined as currency (banknotes and coins) in circulation plus the reserves credit institutions hold with the Eurosystem. We choose base money because it represents the monetary aggregate that is most easily influenced by the central bank. An expansionary monetary policy is characterized by a higher growth rate of the monetary base. Secondly, we measure the monetary stance of the ECB by the EONIA (Euro Overnight 57O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 Index Average), as proposed, for instance, by Sauer and Sturm (2007).9 Thereby, a higher EONIA interest rate (ceteris paribus) indicates a tighter monetary policy. We also carried out the analysis by using the policy rate, as measured by the interest rate of the main refinancing operations implemented by the ECB, and by using the deviation of the policy rate from a simple Taylor (1993) rule. Notably, all results are robust irrespectively to the applied measure of the monetary stance.10 Table 1 depicts the expected influence of monetary policy on each liquidity variable. We expect base money growth to affect the liquidity variables (TO and TV) positively, sincewe hypothesize that an expansionarymonetary policy (i.e. higher basemoney growth) will imply more liquid stock markets. Contrary, we expect that base money growth has a negative impact on the illiquidity measures (ILLIQ, TPI, R_IMP, R_REL and S_REL). The EONIA is expected to have the opposite signs as a loose monetary policy is indicated by low values of the EONIA. 2.3.2. Individual effects: Does the impact of monetary policy depend on the market capitalization? Amihud (2002) finds that small (illiquid) stocks aremore responsive to illiquidity shocks, while large (liquid) stocks becomemore attractivewhen aggregate liquidity declines. Following this reasoning,wewould expect that a loosening ofmonetary policy (i.e. higher base money growth or a lower EONIA interest rate) decreases illiquidity, but this effect should decrease with increasing firm size (i.e. market capitalization). Similarly, one could argue that the spread measures have a natural zero lower bound, as the spread cannot become negative. Consequently, an expansionary monetary policy should decrease the illiquidity of large firms to a lower absolute degree, as the spread of large firms is already close to zero. To capture individual effects at the stock level we therefore include an interaction term in our panel estimations in Section 4.. The interaction term is defined as Int= lnMV∗MP, whereMV is a stock's market capitalization in a specific month, andMP the corresponding monetary policy measure (either base money growth or EONIA). 2.3.3. Control variables In the time series analysis of Section 3 we follow Chordia et al. (2001, 2005) and Goyenko and Ukhov (2009) and account for monthly market returns and the monthly volatility of daily market returns. We compute the monthly market return as the equally weighted average of individual monthly stock returns. Similarly, themarket's monthly volatility of returns is computed as themonthly standard deviation of the equally weighted average of daily stock returns. The relationship between liquidity and macroeconomic factors has been theoretically demonstrated, for instance, by Eisfeldt (2004), and empirically investigated, among others, by Fujimoto (2003), Soederberg (2008) and Næs et al. (2011). In addition to this, it is clear that somemacroeconomic factors, in particular, business cycle and inflation developments, constitute an important piece of information for the decision rule of the monetary authority. We follow Goyenko and Ukhov (2009) and include the twelve-month growth rate of the euro area industrial production and the twelve-month inflation rate of the euro area. In the panel regressions in Section 4 we control for individual stock characteristics that are known to determine stock liquidity, and formacroeconomic variables thatmaybe related to themonetary policy or to stockmarket liquidity. The individual stock characteristics are one-month lagged and include the monthly return, the monthly standard deviation of daily returns and the natural logarithm of market capitalization. We include the return of the previous month (RET), since, among others, Brunnermeier and Pedersen (2009) have shown theoretically that past returnsmay influence stock liquidity and Hameed et al. (2010) have provided confirming empirical evidence. The inclusion of the monthly standard deviation of daily stock returns (STDV) is motivated by the findings of Copeland and Galai (1983)who showed theoretically that the volatility of stock returns should be negatively related to liquidity. To take into account the argument of Amihud (2002) that liquidity is negatively related to a stock's market value, we include the (log of the) market capitalization of stocks (lnMV). We also control for the potential effect of macroeconomic variables on stock market liquidity. Following Goyenko and Ukhov (2009), we include the rolling twelve-month growth rate of the euro area industrial production (IP) and the twelve-month inflation rate in the euro area (IR) in the panel regression. Finally, to account for an interdependence of liquidity and cyclical movements in the stock market we include the MSCI stock market index (IDX) for each market under consideration. 8 See Appendix A for details of the definition of base money growth. 9 As the EONIA is commonly referred to as the equivalent of the Federal Funds rate for the US, we focus on this measure of monetary policy. While the EONIA is significantly more volatile than the Federal Funds rate due to short-term liquidity needs, this is not relevant for our study, as we use monthly averages for our estimations. 10 We prefer to use the EONIA instead of the policy rate, because during the recent financial crisis the non-standard liquidity-providing measures implemented by the ECB have obscured the signaling role of the policy rate. In this sense, the EONIA remains as a better indicator of monetary conditions faced by the financial markets in the last part of our sample period. We also prefer the EONIA to the residual from a Taylor rule, since the latter is already implicitly considered by including both inflation and industrial production in our estimations, and the former is a simpler measure not conditioned by the selection of the model for the Taylor rule. matrix Th altern Table 1 Expected impact of base-money growth and EONIA on each liquidity variable. Expected signs Liquidity variable Base money growth EONIA TO + − TV + − ILLIQ − + TPI − + R_IMP − + R_REL − + S_REL − + 58 O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 production IP and the twelve-month inflation rate IR, the monthly standard deviation of equally weighted daily stock returns STDV, and the equallyweightedmonthly stock return RET.12 Since the ordering of the variables is relevant for the impulse response analysis, we follow the usual approach, such as Chordia et al. (2005) and Goyenko and Ukhov (2009), by placing variables according to the order in which they may (contemporaneously) influence other variables. This is the order of the above defined zt vector, where we place the macroeconomic variables IP, IR and MP first, followed by STDV, RET and LIQ. A lag length of one was set according to the Schwarz (1978) information criterion.13 3.2. Empirical results In order to interpret the estimated VAR models for the German stock market we report the Granger-causality tests (see Granger, 1969 and Sims, 1980) in Table 2 and the corresponding impulse response functions.14 We test the null hypothesis that the lagged endogenous variable of interest (either monetary policy or stockmarket liquidity) does not Granger-cause the dependent variable of interest (again, either stock market liquidity or monetary policy). Th marke and m 11 This 12 The impact of the s 13 We differen retain t 14 We total of the scop zt is the vector of endogenous variables (IP, IR, MP, STDV, RET, LIQ), c is the vector of intercepts, A is a 6×6 coefficients , and ut labels the vector of residuals. e variables of main interest are LIQ, which represents alternatively the sevenmarket (il)liquidity proxies, andMP, which labels atively the two monetary policy measures. The control variables include the twelve-month relative growth rate of industrial 3. The macro level: Aggregated stock market liquidity 3.1. Empirical method In a first step, we examine the influence of central bank policy on the aggregated liquidity of stock markets. Even though the main objective of the ECB is to maintain price stability, the objective of financial market stability has clearly gained importance in recent years. In this respect, Garcia (1989) outlines that central banks may ease market liquidity during periods of crisis by means of monetary policy, as a certain level of liquidity is essential for the functioning of financial markets. If this is the case, we would expect an endogenous relationship between the liquidity of stock markets, central bank interventions and other macroeconomic factors. Thus, on the one hand, stock market liquidity may be a function of central bank policy and macro variables while, on the other hand, central bank actions and macroeconomic variables may be influenced by stock market liquidity as well. In order to take that potential endogeneity into account we investigate the relationship between stock market liquidity and monetary policy by specifying the following VAR model:11 zt ¼ cþ Azt−1 þ ut ; ð1Þ where e results of the Granger-causality tests depicted in panel (a) indicate some evidence that monetary policy Granger-causes stock t liquidity. In particular, the basemoney growth and the EONIA significantly Granger-cause some of the price impactmeasures ost of the transaction cost variables. However, the two trading activity proxies as well as the first differences of the turnover approach is also employed by Chordia et al. (2005) and Goyenko and Ukhov (2009). Augmented Dickey and Fuller (1979) test was used to check for non-stationarity of the variables. Only for the illiquidity ratio ILLIQ, the turnover price TPI and the relative spread S_REL we could not reject the null of a unit root. Thus, to ensure that the (il)liquidity variables of the German stock market are ame order of integration we employ the first differences of those three measures. also considered the lag length resulting from the Akaike (1973) criterion in our estimations. When using the Akaike criterion, the number of lags in the t specifications considered ranges from 1 to 5. The results do not change and can be considered robust. We thus preferred a simpler structure in order to he maximum number of degrees of freedom. More details on those and other estimations can be obtained from the authors upon request. estimated such a VAR model for each of the seven (il)liquidity measures and the two monetary policy variables considered in our analysis. This entails a 14 different VAR estimates, each of which allows for 30 Granger-causality tests. Since reporting the results of all the Granger-causality tests would exceed e of this paper, Table 2 only presents the causality-tests between each of the two monetary stance measures and the seven different (il)liquidity proxies. Table 2 VAR Granger-causality tests between liquidity and monetary policy for the Xetra trading system. (il)Liquidity measure 59O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 price impact TPI are not significantly Granger-caused by monetary policy. Interestingly, the results of the Granger-causality test in panel (b) show only little evidence of a bidirectional relationship between stockmarket liquidity and central bank policy. Overall, the results favor the hypothesis that the ECBmonetary policy causes the aggregate stockmarket (il)liquidity, but offer only little evidence for the reverse is found. To get a deeper understanding of the interactions between the variables in the VAR system we also report impulse response functions. Thereby, we are able to investigate the dynamic reaction of the stock market (il)liquidity measures due to a unit standard deviation innovation in themonetary policy variable.15 Sincewe are primarily interested in the influence of the central bank policy on stock market liquidity, we only report the accumulated twelve-month responses of the seven different (il)liquidity measures to a one-time shock in base money growth (Fig. 1) and in the EONIA (Fig. 2). Fig. 1 illustrates the twelve-month responses of the seven (il)liquidity measures to a unit standard deviation innovation in base money growth. All the signs of the responses of aggregated market (il)liquidity to a one-time shock in base money growth are in line with the hypotheses outlined in Section 2 and significant. Thus, we conclude that stock market liquidity (illiquidity) tends to rise (decline) as base money growth increases. The impulse responses in Fig. 2 (EONIA) suggest again that the aggregated stock market liquidity (illiquidity) decreases (increases) in response to a tightening of the monetary policy. However, though the signs of the illustrated responses are all well in line with our hypotheses, at least in the mid-run, the two-standard-error bands indicate that the response of the (il)liquidity measures is in general not statistically significant for all specifications. Additionally, we considered the variance decomposition of the liquidity measures (not shown) in order to disentangle the amount of information contributed by monetary policy measures.16 Several conclusions can be extracted from this analysis. First of all, the percentage explained by monetary policy measures increases and stabilizes after a year, confirming that monetary policy realizes its effects with some lag, in the mid-run. Secondly, the effect of monetary policy is larger for base money growth than for the EONIA, and the magnitude of the effect depends on the liquidity measure. Thirdly, the effect of monetary policy is comparable to or even larger than the impact of other macroeconomic variables (i.e. IP and IR), and larger than the influence of market volatility and returns in some cases. Finally, even though monetary policy seems to substantially contribute to explain liquidity dynamics, we are not able to infer that it is the only driver of market liquidity. Indeed, stock-market variables still explain Monetary policy measure TO TV d(ILLIQ) d(TPI) R_IMP R_REL d(S_REL) Panel (a): monetary policy (row)→(il)liquidity (column) H0: The central bank policy (row) does not Granger-cause the (il)liquidity (column) Base money growth 2.040 2.627 3.687* 1.254 3.619* 5.248** 1.696 p-value (0.153) (0.105) (0.055) (0.263) (0.057) (0.022) (0.193) EONIA 2.567 0.337 3.637* 1.336 7.474*** 20.147*** 7.407*** p-value (0.109) (0.562) (0.057) (0.248) (0.006) (0.000) (0.007) Panel (b): liquidity (column)→monetary policy (row) H0: The (il)liquidity (column) does not Granger-cause the central bank policy (row) Base money growth 2.091 0.755 2.037 2.156 1.307 1.681 7.580*** p-value (0.148) (0.385) (0.154) (0.142) (0.253) (0.195) (0.006) EONIA 0.316 0.514 1.916 1.710 1.502 1.162 4.653* p-value (0.574) (0.474) (0.166) (0.191) (0.220) (0.281) (0.031) Note: p-values in parenthesis. *, ** and *** denote 10%, 5% and 1% significance levels. the major part of market liquidity. For robustness purposes, we ran similar VARs for the French and Italian markets to test the influence of the ECB on the aggregated liquidity of the Euronext Paris and the Milan stock exchange. Instead of reporting the graphs of the 14 impulse response functions for each market we qualitatively summarize the results in Table 3. The impulse response functions for the Euronext Paris stock exchange, summarized in panel (a) of Table 3, show that all but one of the responses confirmour expectations concerning the sign of the response, though only four of them are significant at the 5% level. For the Milan stock exchange we find weaker evidence for an influence of the ECB policy interventions on the aggregated stock market liquidity. While most signs of the impulse response functions are in line with our hypotheses, only the response of the relative Roll measure is statistically significant. The somehow weak significance might be due to (i) the heterogeneity of different stocks in the market, canceling out certain effects at an aggregated level, and (ii) the short time-series since the start of the euro, resulting in a small number of observations.17 All in all, the estimated VARmodels suggest that the monetary policy of the ECB, as measured by basemoney growth and the EONIA, indeed influences aggregatedmarket (il)liquidity formajormarkets of the euro area, which is in linewith the findings of Goyenko and Ukhov (2009) for the US. The next section will investigate this relationship at the individual stock level by taking into account potential interaction effects (i.e. market capitalization of individual stocks) in a panel setting. 15 In order to orthogonalize innovations we use the Cholesky decomposition. 16 Further details can be obtained from the authors upon request. 17 Unfortunately, an extension of the data set before the introduction of the euro would not necessarily improve our results, as structural breaks due to the introduction of the common currency are likely. -0.4 0.0 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.6 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of TO to Cholesky One S.D. BASE MONEY GROWTH Innovation -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of TV to Cholesky One S.D. BASE MONEY GROWTH Innovation -.06 -.05 -.04 -.03 -.02 -.01 .00 .01 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of d(ILLIQ) to Cholesky One S.D. BASE MONEY GROWTH Innovation -2,000 -1,600 -1,200 -800 -400 0 400 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of d(TPI) to Cholesky One S.D. BASE MONEY GROWTH Innovation -.0016 -.0012 -.0008 -.0004 .0000 .0004 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of R_IMP to Cholesky One S.D. BASE MONEY GROWTH Innovation -.014 -.012 -.010 -.008 -.006 -.004 -.002 .000 .002 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of R_REL to Cholesky One S.D. BASE MONEY GROWTH Innovation -.0024 -.0020 -.0016 -.0012 -.0008 -.0004 .0000 .0004 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of d(S_REL) to Cholesky One S.D. BASE MONEY GROWTH Innovation Fig. 1. Response of the Xetra trading system to a unit standard deviation innovation in the base money growth. 60 O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 -1.6 -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of TO to Cholesky One S.D. EONIA Innovation -.4 -.3 -.2 -.1 .0 .1 .2 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of TV to Cholesky One S.D. EONIA Innovation -.005 .000 .005 .010 .015 .020 .025 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of d(ILLIQ) to Cholesky One S.D. EONIA Innovation -400 -200 0 200 400 600 800 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of d(TPI) to Cholesky One S.D. EONIA Innovation -.0004 -.0002 .0000 .0002 .0004 .0006 .0008 .0010 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of R_IMP to Cholesky One S.D. EONIA Innovation -.002 -.001 .000 .001 .002 .003 .004 .005 .006 .007 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of AM_REL_ROLL_EW to Cholesky One S.D. EONIA Innovation -.0006 -.0004 -.0002 .0000 .0002 .0004 .0006 .0008 .0010 1 2 3 4 5 6 7 8 9 10 11 12 Accumulated Response of d(S_REL) to Cholesky One S.D. EONIA Innovation Fig. 2. Response of the Xetra trading system to a unit standard deviation innovation in the EONIA. 61O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 monet other c and th Yi,t−1) Germa Table 3 Summary of impulse response functions: Euronext Paris & Milan stock exchange. (il)Liquidity measure 62 O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 within (fixed-effects) estimator, whereas ci stands for fixed-effects in the cross-section.18 4.2. Empirical results Descriptive statistics and the correlation matrix for the variables employed in the panel estimations are presented in Tables 4 and 5. Of particular interest are the average monthly bivariate correlations between the seven (il)liquidity measures. As one would expect, the cross-sectional correlations between the trading activity measures (i.e. TO and TV) and the measures of price impact (i.e. ILLIQ, TPI and R_IMP) or transaction costs (i.e. R_REL and S_REL) are negative. This observation is intuitive, since higher tradin assets used i that o constr Moreo larger for all We estima marke to acc conve 18 We evidenc 19 We ary policy depends on firm size asmeasured by (logged)market capitalization (MPt−1⁎lnMVi,t−1). The vector Xi,t−1 represents ontrol variables on the stock level, such asmonthly return (RETi,t−1),monthly standarddeviation of daily stock returns (STDVi,t−1) e natural logarithmofmarket capitalization (lnMVi,t−1). The employedmacroeconomic variables (as represented by the vector include the twelve-month growth rates of industrial production (IPt−1), the twelve-month inflation rate (IRt−1) and theMSCI ny stock market index (IDXt−1). In order to account for time-invariant stock specific determinants of liquidity we use the 4. The micro level: Individual stock liquidity 4.1. Empirical method In a second step, we investigate in-depth whether and how monetary policy as exercised by the ECB determines the liquidity of individual stocks. For that purpose, we estimate panel regressions in which the liquidity (LIQi,t) of stock i in month t is modeled as a function of the (one-month lagged) ECB's monetary policy, the interaction term and other lagged control variables: LIQi;t ¼ cþ b1LIQi;t−1 þ b2MPt−1 þ b3Inti;t−1 þ b4Xi;t−1 þ b5Yi;t−1 þ ci þ ui;t ; ð2Þ where the dependent variable LIQi,t represents the seven above-described (il)liquidity measures. To account for autocorrelation induced by a dynamic relationship in stock liquidity, we include the one-month lagged (il)liquidity measures LIQi,t−1 as a regressor. MPt−1 stands for the monetary policy as exercised by the ECB and the interaction term Inti,t−1 indicates whether the influence of Monetary policy measure TO d(TV) d(ILLIQ) d(TPI) d(R_IMP) R_REL d(S_REL) (a) Euronext Paris Base money growtht−1 ++ + − − − − − EONIAt−1 − − + + − ++ + (il)Liquidity measure Monetary policy measure TO d(TV) d(ILLIQ) TPI d(R_IMP) R_REL S_REL (b) Milan stock exchange Base money growtht−1 + + − − − − + EONIAt−1 − − − + + ++ − Note: — (±) indicates a negative (positive) response of the seven aggregate (il)liquidity measures to a unit standard deviation innovation in the monetary policy variables. — (++) marks responses of which both corresponding bands representing plus/minus two standard errors are less than (exceed) zero. g activity translates into more liquid stocks, whereas higher levels of price impact or transaction costs indicate less liquid . Furthermore, the relatively small magnitude of the correlations highlights the fact that the several (il)liquidity measures n the analysis are not representing the same information, but different aspects of the broad concept of liquidity. It suggests ne should be careful when selecting a measure for the analysis and that, unless there is a priori information that allows us to ain our research to some of them, in principle, all measures should be considered to analyze the robustness of the results. ver, the positive (negative) correlation between the market value of firms and liquidity (illiquidity) suggests that stocks of firms tend to be more liquid. Besides that, the monthly standard deviation of daily returns is negatively related to liquidity variables, except for the stock turnover rate which implies that turnover increases during more volatile periods. estimate (2) for each of the seven (il)liquidity measures and the two monetary policy variables. This entails a total of 14 tions for every market under consideration.While we focus on the German stockmarket, evidence from the French and Italian ts is presented for robustness purposes. Empirical results are shown in Tables 6 (base money growth) and 7 (EONIA). In order ount for heteroscedasticity, all p-values are based on robust standard errors.19 Due to the use of the interaction term, it is nient to evaluate the effects of themonetary policy at different percentiles of the sample distribution for the interaction variable keep the number of lags equal to one, the lag length selected in the VAR model. In any case, there is no need of including more lags since there is no e of autocorrelation in the residuals. test for stationarity applying the panel unit root test developed by Levin et al. (2002) and Pesaran (2007). Table 4 Descriptive statistics for the Xetra trading system. Panel Mean of Mean of Mean of monthly Median of Min. monthly Max. monthly 63O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 variables monthly means monthly σ skewness monthly means mean mean TO 4.482 4.963 1.911 4.503 1.578 8.542 TV 8.406 2.466 0.890 8.371 7.644 9.522 ILLIQ 0.178 0.338 3.213 0.159 0.029 0.535 TPI 8716.097 13,590.580 3.465 8530.118 2980.804 16,940.220 R_IMP 0.002 0.004 2.852 0.002 0.001 0.005 R_REL 0.017 0.021 1.452 0.016 0.010 0.036 S_REL 0.018 0.012 0.740 0.017 0.012 0.035 RET −0.043 12.911 0.466 0.805 −20.351 20.184 MV 2159.542 8193.870 0.006 2022.399 1272.181 4599.777 STDV 2.862 1.242 0.429 2.669 1.765 6.103 Time variables Mean σ Skewness Median Min. Max. IP −0.004 6.186 −1.933 1.203 −22.047 7.773 IR 2.013 0.815 −0.611 2.100 −0.700 4.000 IDX 109.867 28.722 0.148 106.902 50.311 170.717 Table 5 Correlation matrix of time-series means of the monthly bivariate cross-sectional correlations for the Xetra trading system. TO TV ILLIQ TPI R_IMP R_REL S_REL RET lnMV STDV TO 1 0.582 −0.226 −0.402 −0.285 −0.026 −0.340 0.064 0.240 0.244 TV 1 −0.526 −0.430 −0.437 −0.193 −0.788 0.084 0.890 −0.142 ILLIQ 1 0.561 0.394 0.228 0.672 −0.058 −0.489 0.188 TPI 1 0.340 0.132 0.425 −0.038 −0.171 0.019 (market capitalization). The results presented here represent themarginal effects at themedian of the distribution, whereas they are robust across the cross-section of market capitalization.20 Table 6 depicts the estimation results when measuring monetary policy by the rolling twelve-month growth rate of base money. As hypothesized, an increase in the twelve-month growth rate of basemoney leads to a rise in trading activity, while price impact and transaction costs decline. This implies that an expansionary monetary policy triggers an increase (decrease) in individual stocks' liquidity (illiquidity). Particularly interesting are the results for the interaction term, i.e. whether the impact of monetary policy depends on themarket capitalization of stocks. The estimations for the two liquidity measures (capturing trading activity) show that the liquidity-providing effect of a loosemonetary policy is even stronger for larger firms. However, the estimations based on illiquidity variables show that while higher base money growth reduces illiquidity, this effect weakens with increasing market capitalization of stocks. In Table 7 we present estimation results for the models in which the central bank policy is approximated by the EONIA. From the second row, labeled EONIAt−1, it can be inferred that a higher interest rate leads to a decline in the two liquidity variables (stock turnover and trading volume).Moreover, such a restrictivemonetary policy tends to be followed by an increase in the other illiquidity measures. These results are well in linewith our hypotheses and all coefficients appear very significant. While these results generally confirm the empirical pattern observed for base money growth, the effect of the interaction term is slightly different. The impact of the EONIA tends to decreasewith increasing firm size (market capitalization) not only for the five illiquiditymeasures, but also for the two trading activity variables. Remarkably, in terms of economic significance, the influence of the two monetary policy variables (base money growth and EONIA) on stock liquidity is higher than the impact of the remainingmacroeconomic variables, i.e. inflation and industrial production, when comparing standardized coefficients (not shown). For instance, a change of the EONIA by one standard deviation (which amounts to 1.11 percentage points) increases the relative spread by 0.142 standard deviations (the coefficients for the remaining illiquidity proxies are of similar magnitude). The economic significance for the liquidity proxis (turnover and trading volume) is slightly lower, but still considerable. 20 The effect of monetary policy was evaluated at the minimum, 10%, 25%, 50%, 75%, 90%, and at the maximum. The sign, the magnitude and significance of the monetary policy effect did not change across the distribution of the market capitalization variable. More details can be provided by the authors upon request. R_IMP 1 0.536 0.422 −0.059 −0.331 0.060 R_REL 1 0.272 −0.081 −0.233 0.291 S_REL 1 −0.049 −0.771 0.344 RET 1 0.088 0.053 lnMV 1 −0.343 STDV 1 Table 6 Panel estimations for the Xetra trading system, growth rate of the monetary base. Dependent variable ((il)liquidity measure) Trading activity Price impact of order flow Transaction costs TO TV ILLIQ TPI R_IMP R_REL S_REL Dependent variable i,t−1 0.608*** 0.707*** 0.463*** 0.478*** 0.164*** 0.022*** 0.649*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.004) (0.000) Base money growtht−1 0.017*** 0.004*** −0.003*** −85.833*** −0.000*** −0.000*** −0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Interaction termt−1 0.006*** 0.001*** 0.001*** 4.792 0.000** 0.000*** 0.000* (0.000) (0.010) (0.000) (0.115) (0.020) (0.008) (0.067) Returni,t-1 −0.012*** −0.001*** −0.001*** −10.152*** 0.000* −0.000* −0.000*** (0.000) (0.000) (0.000) (0.002) (0.053) (0.096) (0.000) Standard deviationi,t−1 −0.235*** −0.053*** 0.003** 83.413* 0.000*** 0.002*** 0.000 (0.000) (0.000) (0.013) (0.064) (0.000) (0.000) (0.549) In(Market value)i,t−1 0.021 0.256*** −0.082*** −871.780*** −0.000*** −0.002*** −0.002*** (0.748) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Industrial productiont−1 −0.025*** 0.003*** −0.004*** −129.228*** −0.000*** −0.000*** −0.000*** (0.000) (0.005) (0.000) (0.000) (0.000) (0.000) (0.000) Inflationt−1 0.205*** −0.018*** 0.019*** 534.427*** 0.000*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Stock market indext−1 0.007*** 0.000 −0.000 −4.149 −0.000 0.000*** 0.000** (0.000) (0.463) (0.173) (0.159) (0.289) (0.000) (0.028) N 42,084 42,176 41,424 41,306 37,948 37,845 41,843 R2 0.380 0.686 0.319 0.256 0.038 0.044 0.584 Note: P-values are given in parentheses and *, **, *** denote 10%, 5% and 1% significance levels. 64 O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 To sum up, the liquidity-providing (-depriving) effect of expansive (restrictive)monetary policy is generally confirmed. However, some interesting relationships have been revealed that highlight the importance of using various variables covering different aspects of both liquidity andmonetary policy, especially if applied in a panel setting at the individual stock level.While a higher growth rate of the monetary base seems to directly influence trading volume and turnover in the markets, especially for larger stocks, a monetary loosening according to the EONIA indeed has a similar effect, but the impact decreaseswith increasing firm size. Theweakermonetary policy effect for larger firms is confirmed for the illiquiditymeasures for both basemoney growth and EONIA. A possible reason for the greater liquidity effects of monetary policy in smaller stocks may be its impact on funding costs of firms. In particular with respect to interest rate measures, the greater responsiveness of smaller firms would be consistent with a scenario where monetary policy increases the cost of bank financing. Particularly in bank-based economies such as Germany, smaller firms are more dependent on bank lending than larger firms. In this context, a restrictive monetary policy may increase the cost-of-funds for smaller firms to a greater degree. This may alter the expected return of small stocks and exacerbate the (adverse) effect on its liquidity. Table 7 Panel estimations for the Xetra trading system, EONIA. Dependent variable ((il)liquidity measure) Trading activity Price impact of order flow Transaction costs TO TV ILLIQ TPI R_IMP R_REL S_REL Dependent variable i,t−1 0.626*** 0.722*** 0.473*** 0.488*** 0.159*** 0.008 0.641*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.487) (0.000) EONIAt−1 −0.219*** −0.040*** 0.025*** 1405.263*** 0.000*** 0.003*** 0.002*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Interaction termt−1 0.036*** 0.004*** −0.002*** −25.971 −0.000** −0.000** −0.000*** (0.000) (0.010) (0.002) (0.188) (0.017) (0.014) (0.001) Returni,t−1 −0.011*** −0.001*** −0.001*** −2.073 0.000*** −0.000 −0.000*** (0.000) (0.004) (0.000) (0.529) (0.002) (0.659) (0.000) Standard deviationi,t−1 −0.218*** −0.049*** 0.001 -132.163*** 0.000** 0.002*** -0.000*** (0.000) (0.000) (0.594) (0.004) (0.014) (0.000) (0.000) ln(Market value)i,t−1 −0.027 0.238*** −0.065*** −698.892*** 0.000 −0.002*** −0.002*** (0.663) (0.000) (0.000) (0.000) (0.866) (0.000) (0.000) Industrial productiont−1 −0.030*** 0.002* −0.004*** −135.002*** −0.000*** −0.000*** −0.000*** (0.000) (0.062) (0.000) (0.000) (0.000) (0.000) (0.000) Inflationt−1 0.351*** 0.010 0.008*** −313.757*** −0.000*** −0.000 −0.000*** (0.000) (0.174) (0.002) (0.008) (0.000) (0.110) (0.002) Stock market indext−1 0.010*** 0.001*** −0.000*** −26.787*** −0.000*** −0.000 −0.000*** (0.000) (0.001) (0.000) (0.000) (0.000) (0.106) (0.000) N 45,306 45,412 44,619 44,482 40,834 40,723 44,234 R2 0.401 0.692 0.326 0.269 0.036 0.032 0.595 Note: P-values are given in parentheses and *, **, *** denote 10%, 5% and 1% significance levels. TO TV ILLIQ TPI R_IMP R_REL S_REL 65O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 With respect to the influence of the control variables on the (il)liquidity of individual stocks we find robust results that are generally in line with economic intuition (see Tables 6 and 7). A detailed description, however, would go beyond the scope of this paper. The panel regression includes several well-known determinants of stock liquidity as control variables. Nevertheless, there is still the possibility that other common factors of liquidity exist for which the control variables do not account for. Since unobserved common drivers could induce cross-sectional dependence we ran the panel regression including the laggedmarket liquidity (i.e. the equally weighted cross-sectional mean of the respective (il)liquidity measure). This is to some extent in analogy to Chordia et al. (2000) and controls for commonality induced by other factors thanmonetary policy. All results (not reported in the paper) are robust with regard to the sign of the coefficients and their statistical significance. (a) Euronext Paris Base money growtht−1 0.017*** 0.005*** −0.001*** −54.564*** −0.000*** −0.000*** −0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) (0.000) Interaction termt−1 0.002** −0.000 0.000*** −0.416 0.000*** −0.000** 0.000*** (0.025) (0.555) (0.000) (0.835) (0.000) (0.049) (0.000) EONIAt−1 −0.238*** −0.076*** 0.021*** 1180.609*** 0.002*** 0.001*** 0.002*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Interaction termt−1 0.036*** 0.009*** −0.002*** −43.644*** −0.000*** 0.000 −0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.215) (0.000) (b) Milan stock exchange Base money growtht−1 0.034*** 0.005*** 0.000 0.102 −0.000*** −0.000*** 0.000** (0.000) (0.000) (0.826) (0.989) (0.000) (0.000) (0.028) Interaction termt−1 0.005* 0.000 −0.000 −5.538** 0.000** −0.000 −0.000*** (0.098) (0.627) (0.508) (0.039) (0.019) (0.319) (0.000) EONIAt−1 −0.479*** −0.078*** 0.004*** 486.079*** 0.000*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Interaction termt−1 −0.024 0.002 −0.000 −15.717 −0.000** −0.000 0.000*** (0.228) (0.416) (0.317) (0.353) (0.010) (0.594) (0.000) Note: P-values are given in parentheses and *, **, *** denote 10%, 5% and 1% significance levels. Table 8 Summarized panel estimates for the Euronext Paris and the Milan stock exchange. Dependent variable ((il)liquidity measure) Trading activity Price impact of order flow Transaction costs In order to check the robustness of the results, we also carried out the above-presented panel estimations for French and Italian stock markets. Overall, the Italian and French stock markets seem to be comparable to the Xetra trading system. The distributions of the cross-sectional (il)liquiditymeasures (not shown) have very similar properties and the bivariate correlations not only have equal signs, but are also similar in magnitude across markets. Table 8 summarizes the results of the panel estimations of the model outlined in (2) in order to examine the impact of the monetary policy on the (il)liquidity of French and Italian stocks. For reasons of brevity, we only report the coefficients and the respective p-values for corresponding monetary policy variable and the interaction term. Panel (a) of Table 8 presents the results for the French stockmarket, which suggest that the influence of bothmonetary policy variables on the (il)liquidity of individual stocks is significant and in linewith the hypotheses. To a great extent, the coefficients of other control variables (not reported) are qualitatively similar to those reported for the German stock sample. Panel (b) of Table 8 presents the estimation results for the Italian stockmarket. With respect to the relationship between basemoney growth and the (il)liquidity of stocks we find a significant positive influence on traded volume (TV) and turnover (TO), aswell as a significant negative influence on the Roll impactmeasure (R_IMP) and the relative Roll variable (R_REL). While these findings are in line with our hypotheses, the impact of base money growth on the illiquidity ratio (ILLIQ) and the turnover price impact (TPI) are not significant for the Italian stockmarket. The relative bid-ask spread (S_REL) evenhas an unexpected sign. In contrast, the relationship between the EONIA and (il)liquidity at theMilan stock exchange is highly significant in all tested specifications and all of them confirm our expectations. While the empirical pattern for these two markets is somehow different and less clear for the interaction term, the broad picture is basically confirmed, at least for the French market. Overall, we argue that our panel estimation results from the French and Italian stock markets confirm the hypothesis that monetary policy interventions of the ECB determine the liquidity of stocks traded at the Euronext Paris and Milan stock exchange. 5. Summary and conclusion This study sheds light on the role of monetary policy as a potential determinant of stock liquidity. We examine whether an expansionary (restrictive) monetary policy of the ECB increases (decreases) the liquidity of stocks both at the macro and micro level for the Xetra trading system, Euronext Paris and Milan stock exchange. In order to measure (il)liquidity we employ seven variables that ca the tw Ou hypoth analys signifi like m liquidi Ou 66 O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 homogeneity of the slope parameters in the panel models has to be investigated and tested in future work, as assuming fixed effects and including an interaction of monetary policy and market capitalization of stocks in our panel estimations might not be sufficient to account for all potential forms of cross-sectional heterogeneity. Furthermore, a content-related extension could, for instance, take the bond market into consideration. The impact of central bank policies on bond markets seems rather obvious due to open market operations and other monetary policy instruments. Moreover, as suggested by Keynesian arguments, the final effect of monetary policy on liquidity depends on the relative attractiveness of other asset markets (i.e. the bond market). A tightening (easing) of the monetary policy would for instance make bonds relatively more attractive compared to equities and part of the effect of monetary policy on stock market liquidity would be channeled through the bond market (flight-to-quality or flight-to-liquidity episodes). Noteworthy, existing literature supports this conclusion, as shown by Goyenko and Ukhov (2009). In this respect, it should be noted that this effect does not change the causation direction observed in our study (from monetary policy to stock market liquidity), but solely concerns the transmission mechanism of monetary policy shocks to the stock market (potentially also through the bond market). Moreover, studying cross-market information could also turn out to be an interesting research question. Specifically, information across countries could play a role in determining to what extent co-movements towards low-liquidity levels across countries (for example in periods of global crisis) would determine the conduct of common monetary policy in the euro area. Such an extension would add information about a potential reverse causality in those periods. Appendix A. Definition of (il)liquidity and monetary policy measures Turnover rate The turnover rate (TOiym) of stock i in monthm of year y is computed by dividing the monthly sum (over Diym days) of the daily number of shares traded (VOiymd) by the number of shares outstanding (NOSHiym), TOiym ¼ ∑Diymd¼1 VOiymd NOSHiym : ð3Þ Traded volume in euro The traded volume in euro (TViym) of stock i in month m of year y is proxied by taking the natural logarithm of the monthly sum (over Diym days) of the daily product of the number of shares traded (VOiymd) and the respective prices (Piymd), TViym ¼ ln XDiym d¼1 VOiymdPiymd � �0@ 1 A: ð4Þ Amihud's (2002) illiquidity ratio The Amihud's (2002) illiquidity ratio (ILLIQiyd) of security i on day d of year y is computed as the absolute daily return of security i (|Riyd|) divided by the respective traded volume in euro (TViyd), ILLIQiyd ¼ Riyd ��� ��� TViyd : ð5Þ ral banks should be considered as a determinant of individual stock liquidity, whichmay help to explain observed commonality in ty, as well as variations in liquidity at the aggregated-market level. r study leaves several doors open to further research. From a methodological point of view, the issue of cross-sectional firms seem to be more responsive to liquidity effects of monetary policy. This observation is in line with a scenario in which monetary policy affects funding costs of firms. Especially in the context of bank-dependent systems like those of the euro area,monetary policymay impact stock fundamentals through an increase in the cost of bank financing. Our results thus highlight the importance of considering severalmeasures capturing different aspects of the concept of liquidity andmonetary policy variables.Moreover,monetary interventions of cent eses, though those for a shock to the EONIA are in general not statistically significant. Secondly, we complement the macro is by means of panel estimations with stock-fixed effects and find that an expansionary (restrictive) monetary policy also cantly leads to an increase (decrease) in the liquidity of individual stocks. Interestingly, we find that individual stock characteristics arket capitalization, which have not been considered by prior research, indeed play a role in the relationship. In general, smaller liquidity we use VARmodels in order to take potential endogeneities into account. The Granger-causality tests favor the conclusion that the central bank policy causes stock market liquidity, while evidence for a reversed relationship is rather weak. These observations are consistent with the fact that the ECB clearly focuses on price stability, thereby being less activist with regard to other objectives. The estimated impulse response functions confirm that an expansionarymonetary policy entailsmore liquid stockmarkets.Most signs of the responses of the aggregated market (il)liquidity measures after a shock in the monetary policy variables are well in line with our pture the aspects trading activity, price impact and transaction costs. Themonetary policy of the ECB is approximated either by elve-month growth rate of the monetary base or by the EONIA. r findings can be summarized as follows. Firstly, to examine the relationship between monetary policy and aggregated market Turnov Th securi RXIMP ¼ : ð7Þ Ro indepe Relativ Th Aggreg 67O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 LIQym ¼ 1 Nym XNym i¼1 LIQiym: ð11Þ LIQiym ¼ 1 Diym XDiym d¼1 LIQiymd; ð10Þ For the estimations we compute monthly averages of the individual daily (il)liquidity measures for each stock i (LIQiym) as well as the equally weighted average of the (il)liquidity measures across all stocks (LIQym). The replacement characters LIQ in (10) and (11) are each of the above described (il)liquidity measures, Diym in (10) is the number of daily observations of stock i in month m of year y, and Nym is the number of observed stocks in month m of year y in (11) SXRELiyd ¼ iyd iydPAiydþPBiyd 2 � � : ð9Þ ation of the (il)liquidity measures for the estimations RXRELiyd ¼ ROLLiyd Piyd ; ð8Þ where ROLLiyd is the Roll (1984) estimate and Piyd the end-of-day price of stock i. Relative bid-ask-spread The relative bid-ask-spread (S_RELiyd) of stock i at the end of trading day d of year y is the difference between the quoted end-of-day ask (PAiyd) and bid prices (PBiyd), divided by the mid price of stock i, PA −PB e Roll (1984) estimate e relative Roll (1984) estimate (R_RELiyd) of stock i on day d of year y as the ratio 1 2 S (i.e. half the bid-ask-spread) and the probability of a buy and sell order equals 0.5 and is also iid. The observed price at time t depends on whether a buy or sell order occurs and thus equals Pt ¼ mt þ Qt12S, where Qt=1 (Qt=−1) if the asset is bought (sold). Computing the covariance of consecutive price changes yields Cov ΔPt ;ΔPt−1ð Þ ¼−14S2. Inverting this relation gives the following proxy for the spread S ¼ 2 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi−Cov ΔPt ;ΔPt−1ð Þ p (see, for example, Hasbrouck, 2007, or Harris, 2002, for further explanations of the Roll, 1984, measure). For its empirical estimation, we compute the serial covariance of daily price changes each month. Whenever this covariance is negative ROLLiyd ¼ 2 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi −Cov ΔPt ;ΔPt−1ð Þ p , ROLLiyd=0 otherwise. iyd TViyd ll (1984) assumes that the fundamental value of an asset at period t (mt) follows a randomwalk, with innovation ut that are ndent and identically distributed (iid) with zero mean and σ standard deviation. Each transaction causes transaction costs Roll impact The Roll impact (R_IMPiyd) is computed by dividing the Roll (1984) estimate (ROLLiyd) by the traded volume in euro (TViyd) of stock i on day d of year y, ROLLiyd er price impact e turnover price impact (TPIiyd) of security i on day d of year y is computed by dividing the absolute daily return (|Riyd|) of ty i by its turnover rate (TOiyd), TPIiyd ¼ Riyd ��� ��� TOiyd : ð6Þ Base money growth We define the rolling twelve-month growth rate of base money BMt as base money growtht ¼ BMt−BMt−12 BMt−12 ⋅100: ð12Þ References Acharya, V.V., Pedersen, L.H., 2005. Asset pricing with liquidity risk. J. Financ. Econ. 77 (2), 375–410. Akaike, H., 1973. Information theory and an extension of the maximum likelihood principle. Proc. of the 2nd Int. Symp. on Information Theory, pp. 267–281. Amihud, Y., 2002. Illiquidity and stock returns: cross-section and time-series effects. J. Financ. Mark. 5 (1), 31–56. Amihud, Y., Mendelson, H., 1986. Asset pricing and the bid-ask spread. J. Financ. Econ. 17 (2), 223–249. Amihud, Y., Mendelson, H., Pedersen, L.H., 2005. Liquidity and asset prices (foundations and trends in finance). Now Publishers Inc. Brennan, M.J., Chordia, T., Subrahmanyam, A., 1998. Alternative factor specifications, security characteristics, and the cross-section of expected stock returns. J. Financ. Econ. 49 (3), 345–373. Brunnermeier, M.K., Pedersen, L.H., 2009. Market liquidity and funding liquidity. Rev. Financ. Stud. 22 (6), 2201–2238. Chordia, T., Roll, R., Subrahmanyam, A., 2000. Commonality in liquidity. J. Financ. Econ. 56 (1), 3–28. Chordia, T., Roll, R., Subrahmanyam, A., 2001. Market liquidity and trading activity. J. Finance 56 (2), 501–530. Chordia, T., Sarkar, A., Subrahmanyam, A., 2005. An empirical analysis of stock and bond market liquidity. Rev. Financ. Stud. 18 (1), 85–129. Constantinides, G.M., 1986. Capital market equilibrium with transaction costs. J. Polit. Econ. 4, 842–862. 68 O. Fernández-Amador et al. / Journal of Empirical Finance 21 (2013) 54–68 Copeland, T.E., Galai, D., 1983. Information effects on the bid-ask spread. J. Finance 5, 1457–1469. Datar, V.T., Naik, Y.N., Radcliffe, R., 1998. Liquidity and stock returns: an alternative test. J. Financ. Mark. 1 (2), 203–219. Dickey, D.A., Fuller, W.A., 1979. Distribution of the estimators for autoregressive time series with a unit root. J. Am. Stat. Assoc. 74 (366), 427. Eisfeldt, A.L., 2004. Endogenous liquidity in asset markets. J. Finance 59 (1), 1–30. Florackis, C., Gregoriou, A., Kostakis, A., 2010. Trading frequency and asset pricing: evidence from a new price impact ratio. SSRN eLibrary. Fujimoto, A., 2003. Macroeconomic sources of systematic liquidity. Working Paper, Yale University. Garcia, G., 1989. The lender of last resort in the wake of the crash. Am. Econ. Rev. 2, 151–155. Goyenko, R.Y., Ukhov, A.D., 2009. Stock and bond market liquidity: a long-run empirical analysis. J. Financ. Quant. Anal. 44 (1), 189–212. Goyenko, R.Y., Holden, C.W., Trzcinka, C.A., 2009. Do liquidity measures measure liquidity? J. Financ. Econ. 92 (2), 153–181. Granger, C.W.J., 1969. Investigating causal relations by econometric models and cross-spectral methods. Econometrica 37 (3), 424–438. Hameed, A., Kang, W., Viswanathan, S., 2010. Stock market declines and liquidity. J. Finance 65 (1), 257–293. Harris, L.E., 2002. Trading and Exchanges: Market Microstructure for Practitioners. Oxford University Press, USA. Hasbrouck, J., 2007. Empirical Market Microstructure: The Institutions, Economics, and Econometrics of Securities Trading. Oxford University Press, USA. Hasbrouck, J., 2009. Trading costs and returns for U.S. equities: estimating effective costs from daily data. J. Finance 64 (3), 1445–1477. Hasbrouck, J., Seppi, D., 2001. Common factors in prices, order flows, and liquidity. J. Financ. Econ. 59, 383–411. Huberman, G., Halka, D., 2001. Systematic liquidity. J. Financ. Res. 2, 161–178. Kempf, A., Mayston, D., 2008. Liquidity commonality beyond best prices. J. Financ. Res. 31 (1), 25–40. Levin, A., Chien-Fu Lin, A., Chu, C.-S.J., 2002. Unit root tests in panel data: asymptotic and finite-sample properties. J. Econ. 108 (1), 1–24. Næs, R., Skjeltorp, J.A., Ødegaard, B.A., 2011. Stock market liquidity and the business cycle. J. Finance 66 (1), 139–176. O'Hara, M., 1998. Market Microstructure Theory. Wiley. Pastor, L., Stambaugh, R.F., 2003. Liquidity risk and expected stock returns. J. Polit. Econ. 111 (3), 642–685. Pesaran, M.H., 2007. A simple panel unit root test in the presence of cross-section dependence. J. Appl. Econ. 22 (2), 265–312. Roll, R., 1984. A simple implicit measure of the effective bid-ask spread in an efficient market. J. Finance 39 (4), 1127–1139. Sauer, S., Sturm, J.-E., 2007. Using Taylor rules to understand European central bank monetary policy. Ger. Econ. Rev. 8, 375–398. Schwarz, G., 1978. Estimating the dimension of a model. Ann. Stat. 6 (2), 461–464. Sims, C., 1980. Macroeconomics and reality. Econometrica 48 (1), 1–48. Soederberg, J., 2008. Do macroeconomic variables forecast changes in liquidity? An out-of-sample study on the order-driven stock markets in Scandinavia. CAFO Working Paper, Vaexjoe University. Taylor, J.B., 1993. Discretion versus policy rules in practice. Carnegie-Rochester Conference Series on Public Policy, 39(1), pp. 195–214. Does monetary policy determine stock market liquidity? New evidence from the euro zone 1. Introduction 2. Data and hypotheses 2.1. Data set 2.2. Dependent variables: (Il)liquidity measures 2.3. Explanatory variables 2.3.1. Central bank policy measures 2.3.2. Individual effects: Does the impact of monetary policy depend on the market capitalization? 2.3.3. Control variables 3. The macro level: Aggregated stock market liquidity 3.1. Empirical method 3.2. Empirical results 4. The micro level: Individual stock liquidity 4.1. Empirical method 4.2. Empirical results 5. Summary and conclusion Appendix A. Definition of (il)liquidity and monetary policy measures Turnover rate Traded volume in euro Amihud's (2002) illiquidity ratio Turnover price impact Roll impact Relative Roll (1984) estimate Relative bid-ask-spread Aggregation of the (il)liquidity measures for the estimations Base money growth References