Stock Market Liquidity and Macro-Liquidity Shocks: Evidence from the Financial Crisis

Transcription

1 Stock Market Liqidity and Macro-Liqidity Shocks: Evidence from the Financial Crisis Chris Florackis *, Alexandros Kontonikas and Alexandros Kostakis Abstract We develop an empirical framework that links micro-liqidity, macro-liqidity and stock prices. We provide evidence of a strong link between macro-liqidity shocks and the retrns of UK stock portfolios constrcted on the basis of micro-liqidity measres between Specifically, macro-liqidity shocks, which are extracted on the meeting days of the Bank of England Monetary Policy Committee relative to market expectations embedded in 3- month LIBOR ftres prices, are transmitted in a differential manner to the cross-section of liqidity-sorted portfolios, with liqid stocks playing the most active role. We also find that there is a significant increase in shares trading activity and a rather small increase in their trading cost on MPC meeting days. Finally, or reslts emphatically docment that dring the recent financial crisis the shocks-retrns relationship has reversed its sign. Interest rate cts dring the crisis were perceived by market participants as a signal of deteriorating economic prospects and reinforced flight to safety trading. Keywords: Liqidity Shocks; Monetary Policy; Market Micro-Strctre; Stock Retrns. JEL Classification: G12; E43; E44; E51; E52 * Department of Economics, Finance and Acconting, University of Liverpool Management School, Chatham Street, Liverpool, L69 7ZH, UK, tel.: +44 (0) , c.florackis@liv.ac.k. Corresponding athor, Acconting and Finance Sbject Area, Adam Smith Bsiness School, University of Glasgow, Glasgow, G12 8QQ, UK, tel: +44 (0) , alexandros.kontonikas@glasgow.ac.k. Acconting and Finance Division, Manchester Bsiness School, University of Manchester, Booth Street, Manchester M15 6PB, UK, tel.: +44 (0) , alexandros.kostakis@mbs.ac.k.

2 1. Introdction The recent global financial crisis has highlighted the importance of liqidity for the wellfnctioning of financial markets. It is now well nderstood that a decline or, worse, evaporation of liqidity may case large falls in asset prices that are not jstified by their fndamentals. It may also case the initialization of a downward spiral in asset prices, amplified by fire sales and deleveraging to meet margin calls and higher haircts (see Brnnermeier, 2009, and Gorton and Metrick, 2010). Sch feedback mechanisms can eventally pose a major threat to the stability of the financial system (Pedersen, 2009). Liqidity plays a crcial role both at the macro-level and at the micro-level. Macroliqidity mainly refers to the money spply provision by central banks and the availability of fnds for financial markets participants, sch as financial intermediaries. Central banks possess a set of monetary policy tools for managing macro-liqidity. The policy rate they determine is considered to be the benchmark for the term strctre of interest rates. This is particlarly tre for the short-end of the yield crve (Kttner, 2001). Moreover, the terms of liqidity provision to financial intermediaries affect to a great extent the broad money spply in the economy. Crcially for the focs of or stdy, the pivotal role of intermediaries in the modern financial system also implies that macro-liqidity shocks indced by changes in the monetary policy stance of central banks can be transmitted throgh the entire intermediation chain, eventally affecting investors in the marketplace. 1 Most obviosly, the interbank market is crcially affected by monetary policy decisions. These are reflected in LIBOR flctations that inflence the flow of fnds among major intermediaries and determine the vale of their proprietary portfolio of assets and agreements as well as the borrowing ability of dealers. As a reslt, these intermediaries may have to rebalance their own portfolios and modify their risk exposre and degree of leverage to meet 1 Adrian and Shin (2010a) provide a detailed description of the long intermediation chain characterizing a modern financial system and the transmission of liqidity shocks across its links. See also Garcia (1989) for an accont of the monetary policy tools for liqidity provision to financial intermediaries. 1

3 reglatory reqirements and remain solvent (see e.g. Adrian and Shin, 2010b). At the same time, intermediaries pass on to their instittional or individal clients these new terms of fnds exchange by modifying their lending standards as well as their margin reqirements or call rates. This, in trn, may case major shifts in the composition of these clients portfolios and the trading conditions for the corresponding financial assets. 2 In sm, a shift in the qantity of available fnds and the price of liqidity at the macro-level can be spread along the intermediation chain, reaching investors and traders by altering their fnding conditions and investment decisions. Micro-liqidity mainly refers to these trading conditions, namely the cost, speed, volme and price impact of transforming cash into financial assets and vice versa (Chordia et al., 2005). It is now well nderstood that micro-liqidity is an important sorce of market frictions and it can have first-order effects on asset prices (see Amihd and Mendelson, 1980, 1986a, 1986b). Most importantly, micro-liqidity can be regarded as a risk factor leading to sbstantial risk premia in the cross-section of stock retrns (Pastor and Stambagh, 2003; Acharya and Pedersen, 2005). Motivated by this evidence, we arge that macro-liqidity shocks, to the extent that they affect liqidity conditions in the stock market, may also have a differential impact on the cross-section of liqidity-sorted portfolios retrns. Stocks with different microstrctre characteristics, and hence different exposre to micro-liqidity risk, may be differently affected by a common macro-liqidity shock. A nmber of recent stdies for the US find that expansionary macro-liqidity shocks improve micro-liqidity conditions, especially dring periods of financial distress (see e.g. Chordia et al., 2005, Goyenko and Ukhov, 2009, Jensen and Moorman, 2010). 3 Nyborg and Ostberg (2010) analyse the process throgh which banks attempt to recover liqidity when 2 See Brnnermeier and Pedersen (2009) for a model of the interaction between the availability of fnds for traders and microstrctre liqidity. Fortne (2000) explains the mechanics of margin lending and demonstrates the close relationship between the broker call money rate and the Fed Fnds rate. 3 See also Fernandez-Amador et al. (2013) for evidence sggesting that expansionary monetary policy by the Eropean Central Bank increases stock market liqidity in three Erozone markets (France, Germany and Italy). 2

4 they face tighter fnding conditions. In line with liqidity pll-back trading, they find that increased tightening in the interbank market is associated with greater trading activity in highly liqid stocks relative to less liqid ones. 4 Or stdy is also related to the wellestablished strand of the literatre that examines the impact of monetary policy shocks on stock retrns (see e.g. Thorbecke, 1997, and Bernanke and Kttner, 2005). 5 The aim of this stdy is to examine empirically the relationship between macro- and micro-liqidity conditions in the UK and to accont for the potential impact of the recent global financial crisis. We follow Bernanke and Kttner (2005) and adopt an event stdy methodology along with interest ftres based measres of macro-liqidity shocks in order to assess the impact of macro-liqidity on the retrns of stock portfolios that have been formed on the basis of micro-liqidity measres. 6 UK macro-liqidity shocks are extracted on the meeting days of the Bank of England s (BoE) Monetary Policy Committee (MPC) relative to market expectations embedded in 3-month LIBOR ftres prices. By examining the cross-section of liqidity-sorted portfolios, rather than broad stock market or sectoral indices, or approach can shed more light on the link between macro- and micro-liqidity. Specifically, we tilise the niverse of stocks listed on the London Stock Exchange (LSE) dring the period and constrct portfolios by sorting them according to their Retrn-to-Volme price impact ratio. This is the most commonly sed liqidity measre, originally sggested by Amihd (2002). To accont for the cross-sectional 4 Nyborg and Ostberg s (2010) empirical approach is based pon identifying the highest and lowest fnding illiqidity days in every month of their sample (based pon the LIBOR-OIS spread and the TED spread) and examining the patterns in the corresponding trading activity in more liqid verss less liqid stocks. 5 These stdies se varios empirical approaches, ranging from vector atoregressive models to event stdies, and find that US stocks react positively to expansionary monetary policy shocks, with a stronger reaction identified for small and vale stocks (see e.g. Kontonikas and Kostakis, 2013, Maio, 2013). Wongswan (2009) and Hasman and Wongswan (2011) investigate the impact of US monetary policy on international stock markets. Previos stdies on the relationship between monetary policy and stock retrns in the UK inclde Bredin et al. (2007) and Gregorio et al. (2009). 6 Bernanke and Kttner (2005) se the methodology sggested by Kttner (2001) to extract a measre of interest rate shocks from ftres contracts written on the Fed fnds rate. This approach has become qite poplar in the literatre becase these ftres contracts natrally embed market participants interest rates expectations; ths, one-day changes in their prices cleanly isolate the nexpected rate changes. 3

5 size bias indced by this measre, we also tilize the recently introdced Retrn-to-Trnover Rate price impact ratio (Florackis et al., 2011), which is free of size bias. 7 Ths, we contribte to the existing literatre by: (i) estimating the response of UK liqidity-sorted stock portfolios retrns to macro-liqidity shocks extracted on the BoE MPC meeting days; (ii) acconting for the effect of the financial crisis; (iii) testing whether this retrn response differs across the portfolios containing the most and the least liqid stocks; (iv) examining whether sch a potentially different response can be attribted to changes in stocks trading activity or trading cost conditions indced by these macroliqidity shocks; and (v) tilising also a micro-liqidity measre that is immne to size bias. Previewing or empirical reslts, we highlight the following main findings. First, the relationship between macro-liqidity shocks and liqidity-sorted portfolio retrns is sbject to an important strctral break dring the recent financial crisis; failing to accont for this break, one wold erroneosly conclde that macro-liqidity shocks have no effect on retrns. Interest rate cts dring the crisis not only failed to boost stock prices at MPC meeting days, bt they actally led liqid stocks to lower prices becase these were perceived by stock market participants as bad news, signals by the BoE of a worsening economic otlook. Second, and related to the previos point, interest rate cts dring the crisis reinforced flight to safety trading away from declining stocks and towards government bonds. Third, or reslts indicate that macro-liqidity shocks are transmitted to the cross-section of liqiditysorted portfolios albeit in a differential manner between the most liqid and the most illiqid portfolios. Interestingly, the effect is mch more statistically and economically significant for the most liqid portfolios. Forth, there is a significant increase in shares trading activity and a rather small increase in their trading cost on MPC meeting days. The increase in trading 7 Despite the considerable attention that micro-liqidity has attracted in prior literatre, it remains an elsive concept (Amihd, 2002, Pastor and Stambagh, 2003). This featre has led to the emergence of a vast literatre proposing a series of measres captring the for dimensions of liqidity (trading cost, qantity, speed and price impact). 4

6 activity is more prononced for the most illiqid shares, while there is no significant crosssectional difference in the trading cost effect. Or stdy is strctred as follows. Section 2 describes the datasets and discsses varios methodological isses. Section 3 examines the impact of macro-liqidity shocks on liqiditysorted stock portfolios, while Section 4 assesses stock market liqidity conditions, on MPC meeting days. Section 5 contains a series of robstness tests. Section 6 concldes. 2. Data and Methodology In line with the methodology sggested by Kttner (2001), we se data from interest rate ftres to extract macro-liqidity shocks on BoE s MPC meeting days. 8 For the UK, however, there are no ftres market instrments that track the BoE s policy rate (the 2-week repo rate). The closest sbstitte is the short sterling ftres contract that settles on the 3- month British Bankers Association (BBA) London Interbank Offer Rate (LIBOR). 9 This is one of the instrments sed by BoE to gage market expectations regarding ftre interest rates and is widely sed to hedge against and speclate on interest rate movements (Brooke et al., 2000, Bredin et al., 2007, Joyce et al. 2008). Since this ftres contract is written on the LIBOR, we arge that the changes of their prices on MPC meetings days can be more broadly considered as macro-liqidity shocks initiated by the central bank actions (or inactions) rather than being narrowly defined as monetary policy shocks. This is especially tre becase the LIBOR is not necessarily eqal to the BoE s policy rate; their spread, eqivalent to the LIBOR-OIS spread in the US, is actally time-varying and conveys significant information for the interbank market conditions in periods of liqidity draghts (Gorton and Metrick, 2010, 2012). 8 The list of meetings and decisions is available at 9 The settlement price is 100 mins the BBA LIBOR that prevails at 11:00 on the last trading day (third Wednesday of the delivery month) ronded to three decimal places. Contracts are standardised and traded between members of the London International Financial Ftres and Options Exchange (LIFFE). 5

7 The nanticipated interest rate change (macro-liqidity shock), i d, is defined as the change in the implied 3-month LIBOR rate on the MPC meeting day, d, relative to the previos day, d-1, i.e.: i f f d m, d m, d 1 (1) where f md, is the implied interest rate, 100 mins the LIFFE ftres contract price, extracted by the corresponding contract with delivery month m nearest to the MPC meeting day d. 10 The sample period nder investigation is Jne 1999-December 2012, yielding a total of 164 MPC meetings, and the sorce of LIBOR ftres prices is Thomson DataStream. 11 Moreover, we define the anticipated change in interest rate, e i d, as the actal change in the 3-month rate mins the nanticipated change: i i i e d d d (2) Descriptive statistics for the nexpected and expected LIBOR changes on MPC meeting days are provided in Panel A of Table 1, along with the corresponding statistics for the 5-year and 10-year UK Government bond yield changes, which are also sorced from Thomson DataStream. The average nexpected interest rate change is close to zero (-0.001%), ranging from a minimm of -0.39% to a maximm of 0.23%. Figre 1 plots the actal along with the nexpected change in LIBOR on MPC meeting days. It indicates significant interest rate volatility dring the financial crisis, especially following the Lehman Brothers collapse in September No adjstment is necessary for the nmber of days remaining in the month as in the US stdies, becase nlike the ftres on the Fed fnds rate whose settlement is based on the average Fed fnds rate of the last month in the ftres life, in the UK the settlement of the 3-month LIBOR ftres is based on the corresponding LIBOR of the last trading day. 11 We start or event stdy analysis from Jne 1999 becase LIBOR ftres contracts did not settle on a monthly basis before that date; only contracts with qarterly delivery existed. The lack of correspondence in freqencies between the event (MPC monthly meetings) and the instrment s settlement may lead to biased estimates of the shock before Jne

8 [Table 1 abot here] [Figre 1 abot here] For the constrction of liqidity-sorted portfolios, we consider an initial sample that consists of all common stocks listed on the LSE for the period from May 1999 to December Or analysis covers both presently listed shares and shares that were de-listed at some point dring the sample period, and hence or dataset is free of any potential srvivorship bias. We minimize the impact of otliers by exclding firms with a market vale less than 5 million. Finally, following conventional practice in UK stock market stdies (see e.g. Fletcher and Kihanda, 2005), we exclde nit trsts, investment trsts and ADRs. We obtain data from Thomson DataStream and constrct, on a daily basis for each stock, a series of micro-liqidity measres, namely bid-ask spread, trnover rate, trading volme, Retrn-to- Volme and Retrn-to-Trnover Rate price impact ratios, which captre different dimensions of liqidity (i.e. trading cost, trading qantity, trading speed and price-impact). We define bid-ask spread as the difference between the ask price qoted (PA) and the bid price offered (PB) at the close of the market. Trnover Rate is the ratio of nmber of shares traded on a day to the nmber of otstanding shares. Trading Volme is measred as the total vale (in ponds) of all shares traded on a particlar day. Retrn-to-Volme (RtoV) represents the price impact ratio for each share and it is calclated as the average ratio of the absolte daily retrn to the corresponding pond trading volme, sing 60 trading days prior to the MPC meeting day d in month m. To eliminate the impact of very thinly traded stocks, we reqire a non-zero trading volme for at least 45 ot of the 60 trading days. Formally, this price impact ratio is given by: RtoV R D i 1 id i Di d 1 V (3) id 7

9 where R id and V id are, respectively, the retrn and monetary trading volme of stock i on day d and D i is the nmber of non-zero trading days for stock i. Finally, the Retrn-to-Trnover Rate (RtoTR) is an alternative price impact ratio recently proposed by Florackis et al. (2011) and it is calclated as the average ratio of the absolte daily retrn to the corresponding trnover rate, again sing 60 trading days prior to the MPC meeting. This price impact ratio is given by: RtoTR i R Di 1 id (4) Di d 1 TRid where TR id is the Trnover Rate of stock i at day d, while D i and R id are as previosly defined. The se of RtoTR is motivated by the potential cross-sectional size bias that Amihd s RtoV ratio encompasses. In particlar, since the trading volme that appears in the denominator of this ratio is highly correlated with stocks market vale, ranking stocks according to RtoV is almost identical to ranking them according to their capitalization (see Florackis et al., 2011, for a more detailed analysis). On the other hand, RtoTR is not expected to exhibit a size pattern, becase trnover rates are not strongly correlated with market vales. Or final dataset comprises of an average of 780 shares in each month. Using alternatively RtoV and RtoTR, we sort listed firms on the trading day prior to each MPC meeting day d in month m and we constrct qintile portfolios (P1 to P5). In this way, we tilize the latest available information regarding shares liqidity characteristics and at the same time we ensre that these were also available to investors in real time. Portfolio 1 (P1) contains the most liqid shares while Portfolio 5 (P5) contains the most illiqid shares. Since we are interested in the portfolios retrn response de to a macro-liqidity shock on the MPC meeting days, we calclate (daily) portfolio retrns on each MPC meeting day. Or 8

10 benchmark reslts se eqally-weighted portfolio retrns, bt for robstness we also calclate vale-weighted retrns. Panels B and C of Table 1 provide descriptive statistics for the daily retrns of portfolios sorted on the basis of RtoV and RtoTR, respectively. In or stdy we also examine the impact of macro-liqidity shocks on stock market liqidity conditions. To this end, we investigate whether trading activity, as measred by trading volme and trnover rate, and trading cost, as measred by bid-ask spread, for the shares in each of the previosly described liqidity-sorted portfolios are affected on the MPC meeting day. To isolate the effect of the macro-liqidity shock on the MPC meeting day, we follow Nyborg and Ostberg (2010) by normalizing each share s trading activity or cost measre extracted on that day with its corresponding average vale in the 5 prior trading days. For example, the normalized trading volme for share i on MPC meeting day d is given by the following expression: Normalized Volme = id, d 1 j d k Volme id, Volme / k i, j (5) where k=5 in or benchmark reslts. Eqipped with these normalized measres, we calclate the average portfolio normalized trading volme, trnover rate and bid-ask spread for each liqidity-sorted portfolio and each MPC meeting day in or sample. 3. Response of stock retrns to macro-liqidity shocks The starting point of or analysis is to examine the relationship between expected and nexpected interest rate changes and the retrns of liqidity-sorted portfolios on BoE s MPC meetings. The benchmark model employed by Bernanke and Kttner (2005) is given by: r i i (6) e e p, d d d d 9

11 where r pd, is the daily retrn of the RtoV- or RtoTR-sorted portfolio p on the MPC meeting day d. Using this benchmark specification, we find in nreported reslts that neither anticipated nor nanticipated interest rate changes have a significant impact on liqidity-sorted portfolio retrns on MPC meeting days. Moreover, the explanatory power of these regressions is almost negligible. These reslts for the fll sample period Jne December 2012 are in sharp contrast to the evidence docmented in prior stdies on the inverse retrns-shocks relationship (see e.g. Bernanke and Kttner, 2005 for the US and Bredin et al., 2007 for the UK market). An inspection of portfolio retrns dring the crisis period and the preliminary evidence provided by Gregorio et al. (2009) for the UK market motivates s to examine whether strctral instability lies behind this pzzling finding. In particlar, the common perception and finding of previos stdies that share prices rise de to a larger than expected decrease in interest rates was not confirmed dring the recent financial crisis period. As a characteristic example, while the nexpected interest rate decrease on the MPC meeting of 6 th November 2008 was an astonishing 0.4%, relative to market expectations implied by the 3-month LIBOR ftres price, the FTSE All Share index plmmeted by 5.53% on that day. 12 Similarly, the most liqid portfolio (P1) constrcted on the basis of RtoV (RtoTR) yielded a negative retrn of -4.90% (-3.97%) on the same day. Moreover, the nexpected interest rate decrease of 0.1% on the meeting of 8 th October 2008 was associated with a FTSE All Share drop of 4.98%, while the corresponding retrn for the most liqid portfolio on the basis of RtoV (RtoTR) was -3.97% (-3.14%). This pzzling phenomenon has attracted considerable attention in the financial press; a plasible interpretation is that srprising the market dring the crisis by redcing rates more than 12 It is worth noting that the magnitde of this nanticipated decrease is so big becase the market was actally expecting an increase in LIBOR on that meeting day. Moreover, the BoE ct its policy rate by a historical record of 1.5%. Hence, it is an even more intriging fact that the stock market collapsed in the face of the largest nexpected interest rate ct in record. 10

12 expected was perceived as a signal for an even bleaker economic otlook by the central bank. Frthermore, conventional monetary policy becomes ineffective close to the zero lower bond and the redction of interest rates to historically low levels may have been seen as a sign of the desperation of the central banks. 13. Hence, interest rate cts at the peak of the financial crisis instead of boosting stock retrns, as previosly thoght, actally signalled bad news to the investment commnity and led stock prices to lower levels. On the other hand, rising interest rates cold have been regarded as good news, indicating the end of the crisis period. 14 In order to formally test for strctral change in the relationship between liqidity-sorted portfolio retrns and macro-liqidity shocks dring the financial crisis, we introdce a crisis period dmmy variable that spans late 2007 to early Specifically, the start of the financial crisis is dated to Agst 2007 when major dobts abot global financial stability emerged and the first major central bank interventions in response to increasing interbank market pressres took place, shortly followed by the bank rn at Northern Rock in September The end of the most intense period of the crisis is dated to early March 2009 when the stock market reached its lowest level and sbseqently started to recover. The dating scheme is also consistent with previos analyses of the recent financial crisis (see e.g. Brnnermeier, 2009, and Kontonikas et al., 2013). The crisis dmmy variable is interacted with the explanatory variables of the benchmark specification in (6), leading to the following regression model: r (1 D ) i D i (1 D ) i D i (7) Crisis Crisis e Crisis e e Crisis e p, d 1 d d 2 d d 1 d d 2 d d d 13 Bttonwood also tilizes this line of argmentation in Another paradox of thrift, The Economist, 18 th September See, for example, Why rising rates is good news, Financial Times, 14 th December For robstness, in Section 5.1 we alternatively se a narrower definition of the financial crisis period. 11

13 where DCrisis stands for the crisis period dmmy variable that takes the vale 1 from Agst 2007 to March 2009 and 0 otherwise. For estimation, we first se ordinary least sqares where t-vales are calclated sing Newey-West (1987) standard errors. For robstness prposes, and in order to accont for otliers, we also follow Basistha and Krov (2008), Krov (2010) and Kontonikas et al. (2013), employing the MM weighted least sqares procedre introdced by Yohai (1987), which yields estimates that are robst to the presence of otliers. Table 2 (Table 3) reports the least sqares (robst) estimates for RtoV- and RtoTR-sorted eqally-weighted portfolio retrns in Panels A and B, respectively. We also report least sqare estimates from model (7) sing vale-weighted, instead of eqally-weighted, portfolio retrns (see Table 4). [Tables 2, 3 and 4 abot here] The reslts reported in these three Tables lead to the following set of conclsions. Starting with the reslts covering the period otside the financial crisis, we recover the inverse relationship between interest rate shocks and stock retrns. This inverse relationship is both economically and statistically significant, especially for the portfolios containing the most liqid stocks. To illstrate the economic significance of the relationship, the estimated 1 coefficients in Table 3 (robst estimates) indicate that, otside the crisis period, an nexpected interest rate decrease of 25 basis points wold be associated with a positive daily retrn of 1.66% (1.17%) for the most liqid qintile portfolio (P1) constrcted on the basis of RtoV (RtoTR) price impact ratio. On the other hand, the economic and statistical significance of this inverse relationship is mch lower for the least liqid portfolio (P5), especially when RtoV is sed as a sorting criterion. This finding highlights that at MPC meeting days macroliqidity shocks are transmitted in a differential manner to stocks with different microliqidity characteristics. 12

14 To frther examine the link between the performance of liqidity-sorted stock portfolios and macro-liqidity shocks, we formally test whether the most liqid - least liqid differential response is also statistically significant. This is tre when least sqares estimates are sed (see Table 2). Under robst estimates (Table 3), the differential is also statistically significant when RtoV is sed as a sorting criterion, bt not when RtoTR is employed instead. A potential interpretation for the differential response is that an expansive macro-liqidity shock led to improved stock market liqidity conditions, rendering the liqid stocks even closer sbstittes to other highly liqid instrments, and hence the corresponding liqidity premim reqired by investors to withhold them was redced, boosting their prices. The opposite process wold take place in a contractionary macro-liqidity shock. On the other hand, the micro-liqidity conditions of the most illiqid shares were largely naffected by either expansionary or contractionary shocks, rendering their prices nresponsive. We formally test this explanation in Section 4. An alternative explanation for this differential response is that investors revise the premia they reqire relative to other highly liqid asset classes (e.g. government bonds and commercial paper) only for the most liqid shares; this process does not involve the most illiqid shares becase their liqidity characteristics classify them as a separate asset class. The second conclsion we derive from the reslts presented in Tables 2 to 4 is that both expected and nexpected interest rate changes can help explain the daily retrns of the liqidity-sorted portfolios on BoE MPC meetings once we adjst for the crisis effect. Thogh daily retrns are qite noisy by natre, macro-liqidity shocks exhibit a very high explanatory power. The adjsted R 2 of the model can be even higher than 20% for the most liqid portfolios and for both liqidity proxies sed to constrct them when we se robst regressions (see Table 3). On the other hand, the explanatory power of the model for the portfolios containing the least liqid shares is rather low, showing again that these shares 13

15 retrns are rather nresponsive to the shocks. Overall, these findings highlight the fndamental importance of macro-liqidity, confirming that sch shocks are directly transmitted to share prices via the channels we described in the Introdction, albeit in a differential manner between liqid and illiqid shares. Finally, the economic and statistical significance of the expected interest rate change coefficients, which was also reported in the seminal stdy of Bernanke and Kttner (2005) for the US market, contradicts the conjectre that this information wold have been already incorporated into stock prices. In terms of the reslts dring the crisis period, we emphatically docment that the interest rate changes - retrns relationship reversed its sign dring that period. In particlarly, e 2 and 2 estimates indicate that it trned into a positive relationship that is statistically as well as economically highly significant, for both nexpected expected and expected rate changes, respectively. In other words, an nexpected (or expected) decrease in the 3-month LIBOR led to a negative portfolio retrn response dring the crisis, while it wold have yielded a positive retrn otside the crisis period. The magnitde of the retrn responses is also noteworthy; for the portfolios containing liqid shares, the positive response to the shock dring the crisis was at least twice greater than the negative response (in absolte vale) docmented exclding the crisis. On the other hand, the corresponding response for the portfolio P5 containing the least liqid shares was mch lower, bt still positive dring the crisis period. In fact, the differential response between the most and the least liqid portfolio (P1-P5) dring the crisis period was economically and statistically significant for both liqidity proxies, for both econometric methodologies and for both eqally- and valeweighted retrns, as shown in Tables 2 to 4. A possible explanation for the reversal of the shocks-retrns relationship dring the crisis period is that a decrease in interest rates on BoE MPC meeting days was signalling worsening prospects for the financial system and the macroeconomy; hence investors fled the stock 14

16 market liqidating their positions to hoard cash or cash-like instrments, redce their risk exposre and meet margin calls. The flight to safety or flight to liqidity mechanism, according to which investors rebalance their portfolios towards less risky and more liqid assets dring times of economic and financial distress, is well-stdied in the previos literatre (see e.g. Longstaff, 2004, Chordia et al., 2005, Goyenko and Ukhov, 2009). Indeed, following the collapse of Lehman Brothers in September 2008, risk aversion peaked and eqities experienced major losses while the price of safe haven assets increased. 16 To formally test whether interest rate shocks dring the financial crisis reinforced flight to safety trading, we regress the daily changes in the 5-year and 10-year UK government bond yield on MPC meetings (Δyield) on the expected and nexpected interest rate changes: yield (1 D ) i D i (1 D ) i D i (8) Crisis Crisis e Crisis e e Crisis e d 1 d d 2 d d 1 d d 2 d d d The corresponding regression reslts that are reported in Table 5 validate this argment. While the effect of nexpected rate changes on the 5-year UK government bond yield is small and statistically insignificant otside the crisis, there is a positive relationship that is highly economically and statistically significant dring the crisis. In other words, an nexpected interest ct dring the crisis period was accompanied by a fall in the 5-year UK government bond yield, cased by investors fleeing the stock market and investing in government bonds, which were considered a safe haven at the time. Similar is the evidence from the 10-year UK government bond yield. [Table 5 abot here] Finally, the significantly differential retrn response between the portfolios containing the most and the least liqid shares that we previosly docmented dring the crisis period is also consistent with flight to safety trading. The rebalancing of investors portfolios dring 16 For example, while the FTSE 100 stock index declined by 50% between Agst 2007 and March 2009, the 5-year UK government yield fell from 5.25% to 2.41% over the same period. 15

17 the crisis period was mainly accomplished by selling off their most liqid shareholdings, becase these were easier to liqidate relative to the least liqid shares (consistent with the conjectre of Brnnermeier, 2009, and the evidence provided by Anand et al., 2010) and at the same time they ceased to be regarded as close sbstittes to cash-like instrments sch as fixed income secrities. Therefore, the reqired premia for the most liqid shares were increased, and hence their prices were mch more heavily penalized relative to the least liqid shares that were already penalized with a high premim reqired by investors to hold them. 4. Effects on stock market liqidity The reslts in the previos section indicate that the macro-liqidity shocks on MPC meetings days are transmitted to the cross-section of liqidity-sorted portfolio retrns in a differential way. In this section we examine how these shocks affect the trading activity and trading costs of these portfolios, testing also whether there is a differential effect between liqid verss illiqid shares. 4.1 Trading activity In this section we examine the trading activity of RtoV- and RtoTR-sorted portfolios on MPC meeting days. We proxy trading activity sing shares trading volme and, alternatively, trnover rate. To isolate the effect of the macro-liqidity shock on the MPC meeting day only, we follow Nyborg and Ostberg (2010) by normalizing each share s trading volme and trnover rate extracted on that day with its corresponding average vale in the 5 prior trading days, as in eqation (5). This normalization adjsts for the different levels of trading activity that each share exhibits prior to the MPC meeting and may be irrelevant to the effect of the specific macro-liqidity shock. In this way, we can also compare the relative 16

18 effect on trading activity between liqid and illiqid shares, as classified on the basis of the two price impact ratios we tilize in this stdy. Vales of the normalized measre above (below) 1 indicate that the trading activity on the MPC meeting day was higher (lower) than the prior average trading activity. We have also examined alternative short windows for the normalization; the reslts, which are readily available pon reqest, are qalitatively similar to the ones presented here. Table 6 presents the average normalized trading volmes for RtoV-sorted (Panel A) and RtoTR-sorted (Panel B) portfolios, while Table 7 presents the corresponding average normalized trnover rates. In addition to reporting these averages for all MPC meetings in or sample period, we separately report them for meetings with a negative i 0, positive i 0 or no i 0 nexpected interest rate change occrred. In this way, we can also examine whether the direction of the interest rate shock on the MPC meeting day affects shares trading activity. Finally, we also report the corresponding averages exclding the meetings that took place dring the financial crisis period, i.e. from Agst 2007 to March 2009, to examine whether there was any particlar crisis effect. [Tables 6 and 7 abot here] The reslts reported in Tables 6 and 7 lead to the following conclsions. First, trading activity considerably increases on MPC meeting days for the entire cross-section of liqiditysorted portfolios. This is tre for both proxies of trading activity. In particlar, the reported vales show that, in almost every case we examined, there is an increase of at least 15% relative to the average trading activity observed in the 5 prior trading days. Unreported t-tests show that this increase is also statistically significant at the 1% level or lower. Second, the observed increase in the trading activity remains remarkably robst across the different categories of MPC meetings. It does not depend on the sign of the interest rate shock and it cannot be attribted to the crisis period. This finding highlights that the MPC meeting is an 17

19 important date for investors calendar and that the informational content of MPC decisions is very important, leading to a significant increase in trading activity on this day. Third, the normalized trading activity of the least liqid shares, as classified by their price impact ratios prior to the meeting, increases mch more relative to the trading activity of the most liqid shares. The differential trading activity normalized increase between the most and the least liqid portfolios (P1-P5) can be even higher than 20% and it is highly significant. This finding highlights that investors trade the least liqid shares on MPC meetings mch more actively relative to the prior days. Therefore, the finding of the previos section that the retrns response of the least liqid shares to macro-liqidity shocks is rather mndane cannot be attribted to thin trading of these stocks; qite the opposite is tre on MPC meetings. Ths, it appears that higher trading in least liqid shares on MPC meeting days does not exhibit an overall direction that is as consistent as in the case of most liqid shares, and therefore does not lead to strongly positive or negative retrns. Conclding, the differential response between the most and the least liqid shares retrns to the common macro-liqidity shocks that we docmented in Tables 2 to 4 is not de to a differential impact on their microliqidity conditions. 4.2 Trading cost In Table 8 we report the average normalized bid-ask spread for RtoV-sorted (Panel A) and RtoTR-sorted (Panel B) portfolios on MPC meeting days. The normalization is performed sing each share s average bid-ask spread in the 5 trading days prior to the meeting. As with trading activity, we report the average bid-ask spreads for all meetings taking place in or sample period, for different signs of the interest rate shocks as well as exclding the meetings that took place dring the crisis period. [Table 8 abot here] 18

20 The reslts reported in Table 8 lead to the following conclsions. First, there is a small increase in shares bid-ask spreads on MPC meetings relative to the prior trading days. This increase is in the order of 5% and it is observed across all liqidity-sorted portfolios. Second, this small increase in the normalized bid-ask spreads remains intact when we separately examine MPC meetings associated with a positive, negative or no nanticipated interest rate change. This is an interesting and conterintitive finding since, for example, trading costs are not redced even when expansive macro-liqidity shocks (i.e. nexpected interest rate cts) occr. Possibly, this reslt shows that bid-ask spreads widen relatively to the prior days becase of the arrival of new information related to MPC decisions and the corresponding price discovery process that accompanies the considerably increased trading activity we previosly docmented. Third, this relative increase in the bid-ask spreads is similar across the liqidity-sorted portfolios. The differential relative increase between the most liqid and the most illiqid portfolios (P1-P5) is neither economically nor statistically significantly different from zero. 5. Robstness checks 5.1 Alternative definition of the crisis period The introdction of the slope dmmy variable for the recent crisis period in model (6) has played a crcial role for or analysis. As a reslt, it is legitimate to ask how an alternative definition of the crisis period may affect the reported reslts. For robstness prposes, we se September 2008, when Lehman Brothers collapsed, as an alternative starting point of the crisis. Ths, the slope dmmy variable DCrisis now takes the vale of 1 dring the period September March 2009 and 0 otherwise. Using the reslting narrower crisis period definition that essentially captres the most intense phase of the recent financial crisis, we reestimate model (7). 19

21 The estimation reslts, shown on Table 9, are very similar to the ones obtained sing the benchmark definition of the crisis period. The inverse shocks-retrns relationship exclding the crisis is economically and statistically significant. Most importantly, we also confirm the reversal in the sign of this relationship dring the crisis period. The economic as well as the statistical significance of the positive response of retrns to interest rate shocks remains intact and robst to the narrower definition of the crisis period. Examining the retrn responses across liqidity-sorted portfolio, or findings are very robst to the alternative characterization of the crisis period. In particlar, the most liqid portfolios retrns react more negatively than the most illiqid portfolios retrns to interest rate shocks before and after the crisis period. Moreover, dring the crisis period, the most liqid portfolios retrns react far more positively to these shocks relative to the most illiqid portfolios retrns. [Table 9 abot here] 5.2 Additional control variables Or analysis has focsed on the response of liqidity-sorted daily portfolios retrns to macro-liqidity shocks. Despite the se of an event stdy methodology, argably other factors may be driving or reslts. To take into accont potentially omitted variables that may affect UK daily stock retrns, we estimate the following agmented regression model: r (1 D ) i D i (1 D ) i D i ' X Crisis Crisis e Crisis e e Crisis e p, d 1 d d 2 d d 1 d d 2 d d d d (9) where X d represents the vector of additional explanatory variables. Following Bredin et al. (2007), we consider as additional control variables the daily change in the log sterling pond/ US dollar exchange rate, the daily change in the log sterling pond/ Ero exchange rate as 20

22 well as the retrn on the US market as proxied by the daily change in the log S&P 500 index. 17 Table 10 contains the response coefficients estimated from model (9). Panel A presents the reslts for the RtoV-sorted portfolios retrns, while Panel B shows the corresponding reslts for the RtoTR-sorted portfolios retrns. We can confirm the robstness of or benchmark reslts, even in the presence of additional explanatory variables. There is a differential response to interest rate shocks between the most liqid and the most illiqid portfolios retrns. This differential response is particlarly significant, both statistically and economically, dring the crisis period. More specifically, most liqid portfolios retrns exhibited a highly positive reaction to macro-liqidity shocks dring the crisis, while most illiqid portfolios retrns remained largely naffected. [Table 10 abot here] 5.3 Alternative definition of macro-liqidity shocks This stdy has tilized macro-liqidity shocks defined relative to market expectations embedded in the traded ftres contract written on the 3-month LIBOR. A series of previos US-based stdies have tilized the changes in the LIBOR-OIS spread as a measre of interbank market fnding conditions, with increases in the spread indicating macro-liqidity deterioration. 18 Therefore, in this sbsection we seek to examine the response of the microliqidity sorted portfolios retrns to changes in this alternative proxy of macro-liqidity conditions. For the UK market, we define the eqivalent spread as the difference between the 3-month LIBOR (L) and the BoE Base rate (B). An increase in the LIBOR-BoE rate spread on an MPC meeting day implies an adverse macro-liqidity shock, in the sense that fnding conditions for financial intermediaries and market participants deteriorate, either throgh an 17 Given the time lag between the US and the UK market close, we follow common practice in the literatre and se the lagged S&P 500 daily retrn. 18 The LIBOR-OIS spread is widely accepted as a barometer of fears of bank insolvency in the words of Alan Greenspan (see Thornton, 2009, and Gorton and Metrick, 2012, for an analysis of its featres). 21

23 increase in the cost of fnds or throgh a redction in their spply. The model we estimate is given by: spread rp, d ( LIBOR BoE rate) d d (10) where ( LIBOR BoE rate) d stands for the change in the spread on meeting day d over the previos trading day d-1. An important featre of the LIBOR-BoE rate spread is that it becomes very active mainly since the onset of the financial crisis in late 2007 (see Figre 2). As a reslt, we do not need to introdce a crisis period slope dmmy variable in model (10). This effect is inherently taken into accont by the behavior of the spread. Frthermore, we do not decompose between anticipated and nanticipated components of this spread change (see also Nyborg and Ostberg, 2010). Table 11 presents the estimated response coefficients from model (10). Panel A (Panel B) contains the reslts for the RtoV (RtoTR)-sorted portfolios retrns. Overall, the reslts show again a significantly different response between the most liqid and the most illiqid portfolios retrns. In particlar, we find that the retrns of the most liqid portfolios exhibit a significantly negative response to innovations in the LIBOR-Base rate spread. Moreover, this variable possesses very strong explanatory power with respect to liqid portfolios retrns. Regarding the most illiqid portfolios retrns, they are significantly less affected by innovations in this spread. [Table 11 abot here] 6. Conclsions The global financial meltdown has broght macro- and micro-liqidity to the center stage of analysis for asset prices flctations and the stability of the financial system. 22

24 This stdy examines the transmission of shocks that affect the fnding liqidity conditions of market participants and financial intermediaries to stock market retrns. In particlar, we examine the potential link between macro-liqidity shocks and the retrns of portfolios sorted on the basis of the shares micro-liqidity. In the process of doing so, we also accont for the impact of the recent financial crisis. Shocks are extracted relative to market expectations embedded in LIBOR ftres prices on BoE MPC s meeting days. Frthermore, we evalate the nderlying micro-liqidity conditions dring those days. There are for important conclsions from or empirical analysis. First, the relationship between macro-liqidity shocks and liqidity-sorted portfolio retrns has changed significantly, by reversing its sign, dring the recent financial crisis. Specifically, as nominal interest rates moved towards the zero lower bond, expansionary rate srprises led to lower prices for liqid stocks since investors perceived them as bad news for ftre economic prospects. Second, seen in this light, interest rate cts dring the crisis reinforced a rebalancing of portfolios away from stocks and towards safe haven assets sch as government bonds. Third, the impact of macro-liqidity shocks is mch more statistically and economically significant for the portfolio retrns of most liqid stocks. Forth, on MPC meeting days stocks trading activity increases significantly, especially for the most illiqid shares. Trading cost also increases bt not strongly, and withot exhibiting significant crosssectional differences. These empirical findings have several important implications for policy makers and market participants. The conventional wisdom that redcing interest rates can boost stock prices in the short-rn does not necessarily hold dring a severe crisis period. To the contrary, nder sch conditions, interest rate cts may actally send the wrong signal to investors who may perceive them as a message that macroeconomic prospects are even worse than previosly thoght. This signalling effect may actally exacerbate the impact of adverse 23

25 market conditions, eventally leading investors to flee the stock market, initializing a downward price spiral and posing a serios threat to the stability of the financial system. Moreover, the transmission of these macro-liqidity shocks may affect differentially stocks with different micro-liqidity conditions. Therefore, dissecting the cross-section of stock retrns can prove qite sefl for portfolio selection and diversification de to their differential exposre to liqidity risk. This differential risk exposre may lead to differential premia and less correlated price movements. Conclding, as Blinder (2012) recently noted, the redction of policy rates to levels close to the zero lower bond and the implementation of Qantitative Easing by BoE calls for a detailed investigation of the impact of nconventional monetary policies on the retrns of liqidity-sorted portfolios; this topic is left for ftre research. 24