Liquidity Commonality in Foreign Exchange Markets During the Global Financial Crisis and the Sovereign Debt Crisis: Effects of Macroeconomic and Quantitative Easing Announcements

North American Journal of Economics and Finance 42 (2017) 172–192

Contents lists available at ScienceDirect

North American Journal of Economics and Finance j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / e c o fi n

Liquidity Commonality in Foreign Exchange Markets During the Global Financial Crisis and the Sovereign Debt Crisis: Effects of Macroeconomic and Quantitative Easing Announcements Ya-Ting Chang a,⇑, Yin-Feng Gau b, Chih-Chiang Hsu a a b

Department of Economics, National Central University, 300 Jhongda Rd., Jhongli, Taoyuan, Taiwan Department of Finance, National Central University, 300 Jhongda Rd., Jhongli, Taoyuan, Taiwan

a r t i c l e

i n f o

Article history: Received 25 December 2016 Received in revised form 8 May 2017 Accepted 26 June 2017

JEL classification: F31 G15 C22 Keywords: Liquidity commonality Macroeconomic announcements Quantitative easing policies Foreign exchange market

a b s t r a c t Noting the time-varying dynamics in liquidity, we use a generalized dynamic factor model (GDFM) to identify market-wide liquidity across foreign exchange (FX) markets. Liquidity commonality across currencies increases during the 2008–2009 global financial crisis and the 2009–2011 European sovereign debt crisis, which affirms the spiral effect between funding liquidity and FX market liquidity. The effect of funding constraint on liquidity in the FX market may be through carry trade activities that link the FX market and other classes of asset markets, as suggested by Melvin and Taylor (2009) and Banti (2016). The shift in liquidity commonality around the release of macroeconomic announcements also can be related to the spurs of unwinding carry trade positions in response to unexpected macro shock that affects interest rate differential. In contrast, quantitative easing (QE) policies in the United States, which inject high capital inflows into financial markets, are associated with decreased liquidity commonality, implying that QE implementation actually improves the funding liquidity and weakens the spiral effect, ultimately inducing weaker commonality in FX liquidity. Ó 2017 Elsevier Inc. All rights reserved.

1. Introduction Liquidity shocks related to the 2008–2009 subprime mortgage debt crisis and the 2009–2011 European sovereign debt crisis stressed global financial markets, highlighting the importance of liquidity risk for asset returns. Understanding commonality in the liquidity that exists in global markets in turn is important for risk management and portfolio selection. For example, the foreign exchange (FX) market is the world’s largest in terms of trading volume (Bank for International Settlements, 2013), but as Mancini, Ranaldo, and Wrampelmeyer (2013) and Karnaukh, Ranaldo, and Soderlind (2015) note, relatively minimal research investigates FX liquidity, compared with studies of liquidity in equity and bond markets. But because approximately 60% of FX trading volume consists of major currencies, the liquidity in FX markets differs both by currency and over time, at intraday and daily frequencies (Mancini et al., 2013). Therefore, we undertake a dedicated investigation of liquidity commonality in FX markets to provide insights into its dynamics during liquidity crises. Commonality in liquidity reflects co-movement of one asset’s liquidity with aggregate market-wide liquidity. Previous research offers profound evidence of liquidity commonality in the stock market (e.g., Brockman, Chung, & Pérignon, 2009; ⇑ Corresponding author. E-mail addresses: [email protected] (Y.-T. Chang), [email protected] (Y.-F. Gau), [email protected] (C.-C. Hsu). http://dx.doi.org/10.1016/j.najef.2017.06.004 1062-9408/Ó 2017 Elsevier Inc. All rights reserved.

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

173

Chordia, Roll, & Subrahmanyam, 2000; Hasbrouck & Seppi, 2001).1 Several studies also document liquidity commonality in bond and FX markets. Chordia et al. (2005) analyze liquidity co-movements between the stock and bond markets. Banti et al. (2012) provide evidence of a common component in liquidity across currencies, such that dealers’ responses to incoming orders of different currencies have a common component, seemingly due to their inventory position choices. Moreover, Mancini et al. (2013) find that FX liquidity is not isolated from exchange rates, such that the market liquidities of individual currencies move together and are positively (to varying extents) related to market-wide FX liquidity. Furthermore, in their analyses of liquidity commonality during liquidity crises, Kamara, Lou, and Sadka (2008), Karolyi, Lee, and Van Dijk (2012), and Rösch and Kaserer (2013) suggest that when liquidity suddenly dries up in the market, due to a financial crisis, it may lead to stronger commonality for liquidity in the stock market. With investigations of liquidity co-movement in FX markets during the 2008–2009 financial crisis, Banti et al. (2012), Mancini et al. (2013), and Karnaukh et al. (2015) show that commonality in FX liquidity is stronger in distressed markets.2 According to Hameed, Kang, and Viswanathan (2010), liquidity commonality in stocks exists because liquidity providers withdraw market liquidity after market declines, consistent with a theoretical model by Brunnermeier and Pedersen (2009) that links asset liquidity and traders’ funding liquidity. Funding liquidity is a major factor that triggers commonality in liquidity. Carry trades are popular currency trading strategies that involve borrowing in low interest rate currencies and investing in high interest rate currencies. The returns to carry trade will affect the liquidity in the FX markets abruptly through the unwinding positions of carry trade caused by a burst of big losses (Melvin & Taylor, 2009). Osler (2012) interprets the Brunnermeier and Pedersen (2009) model of funding constraints in the FX context through the channel of carry trade. As the market reverses, carry traders may take a ‘‘wait-and-see” approach to timing the market and traders will not simultaneously and immediately liquidate positions. In effect, both carry-trade returns and carry-trade fragility may be selffulfilling. Brunnermeier, Nagel, and Pedersen (2009) predict a positive relation between crash risk and the extent of carry trading. Once market crashes, carry-trade unwinding increases. Traders will be forced to unwind carry trade positions as they get near to their own funding constraints. Then we may predict that currency markets will be less liquid during times of carry-trade unwinds and that the illiquidity should be more pronounced for investment currencies involved in the carry trade (Mancini et al., 2013). As the markets are relatively more volatile and illiquid, the tendency of unwinding of carry trade position becomes higher. Brunnermeier et al. (2009) find that a rising VIX index, which indicates heightened market risk and/or risk aversion, is indeed associated with carry-trade unwinds. Karnaukh et al. (2015) also find that FX liquidity tends to decline with the volatility and illiquidity of global equity and bond markets. To explore how public news arrival affects FX market liquidity, we also study changes in liquidity commonality around macroeconomic announcements and the impacts of quantitative easing (QE) monetary policy announcements. Liquidity shifts following the release of macroeconomic announcements, because the information environment changes (Andersen & Bollerslev, 1998; Andersen, Bollerslev, Diebold, & Vega, 2003, 2007; Bauwens, Omrane, & Giot, 2005; Evans & Lyons, 2005; Evans & Lyons, 2008). Information asymmetry in FX markets may be associated traders’ interpretation ability and sophistication about publicly released macro news. (Evans, 2010) The variation in bid-ask spreads can be affected by information asymmetry about market structure and current market condition. The shift in liquidity commonality around the release of macroeconomic announcements can be related to the spurs of unwinding carry trade positions in response to an unexpected macroeconomic shock. When carry trade positions are adjusted in response to unexpected macro shock, it is possible that a negative news surprise causes the unwinding of carry trade positions during crisis periods, and the commonality in FX liquidity may increase. Moreover, a negative news surprise may result in an increase in volatility, leading to the wider bid-ask spread and lower liquidity (Stoll, 1978). As argued in Menkhoff, Sarno, Schmeling, and Schrimpf (2012) that carry trade returns related negatively to FX volatility, we may expect that when the FX market volatility increases at the arrival of negative news, the co-movements in FX liquidity may be stronger when the major unwinding of carry trade happen in times of crisis. (Mancini et al., 2013). By using Electronic Broking Services (EBS) intraday data, we investigate factors that drive dynamic FX liquidity commonality.3 The liquidity measures calculated from the EBS data exhibit significant autocorrelation. We also use a generalized dynamic factor model (GDFM) to extract commonality in FX liquidity and address the potential effect of autocorrelation. Traditional principal components analyses ignore such autocorrelation and thus may lead to biased measures of the common component in individual currency market liquidities. With this approach, we find that liquidity commonality significantly varies over time; we also find ample evidence of strong commonality in liquidities during periods of financial crisis. Consistent with

1 The issue of commonality also has been investigated in bond and FX markets. Chordia, Sarkar, and Subrahmanyam (2005) analyze liquidity co-movements between the stock and bond markets. Banti, Phylaktis, and Sarno (2012) provide evidence of a common component in liquidity across currencies, such that dealers’ responses to incoming orders of different currencies have a common component, seemingly due to their inventory position choices. 2 As noted in Galati, Heath, and Mcguire (2007) and Karnaukh et al. (2015), with little and no margin requirements on FX spot transactions, currency traders can take highly leveraged positions. However, this feature induces currency liquidity to deteriorate in crisis periods or in times of unexpected high volatility because of the unwinding of leveraged positions in FX and related markets. 3 The EBS was established by several market-making banks to counter the dominant role of Reuters; when EBS acquired Minex in December 1995, it gained significant market share in Asia. For a detailed description of the structure of FX markets and electronic trading platforms, see Rime (2003).

174

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Mancini et al. (2013) and Rösch and Kaserer (2013), we find that increased systematic liquidity risk triggers illiquidity spillovers across markets and that liquidity commonality can be a source of financial contagion. To determine whether liquidity commonality differs with positive versus negative macroeconomic announcements, we also examine reactions to the U.S. Federal Reserve’s quantitative easing (QE) announcements. Using dummy variables that indicate an announcement release, we distinguish how liquidity commonality changes following a news release. The price volatility invoked by news releases affects the extent of FX liquidity commonality. In most cases, liquidity commonality tends to exhibit a stronger response to positive announcements than to negative ones during the crisis period. We also find that QE policies ease funding constraints and decrease FX liquidity commonality. With these efforts, our study contributes to extant literature in three main ways. First, our use of the GDFM mitigates the bias associated with measuring the common liquidity factor for FX markets, in light of significant autocorrelation in the liquidity of individual markets. Forni, Hallin, Lippi, and Reichlin (2000) provide some theoretical background, and Hallin, Mathias, Pirotte, and Veredas (2011) scrutinize the GDFM’s empirical properties for equity markets. However, previous studies of common components in FX liquidity use principal components analysis, ignoring the characteristic of autocorrelation. Even without rolling-sample estimation, we find time-varying common liquidity with the GDFM. This finding confirms that common liquidity in FX markets significantly increases during periods of crisis. Second, we analyze the effects of macroeconomic announcements on liquidity commonality and FX market-wide liquidity. By considering the effects of announcements on liquidity commonality, we enhance understanding of the link between market liquidity and macroeconomic announcements. Our findings reinforce the results published by Mancini et al. (2013) and Karnaukh et al. (2015), revealing that increased uncertainty or price volatility due to news releases can lead liquidity commonality to vary around the announcement. Moreover, the effect of expansionary QE policy announcements indicates that recovery speeds in systemic liquidity versus the liquidity of an individual currency pair differ. Third, with our relatively longer sample period, we can investigate two recent and pertinent financial crises. By including both the subprime mortgage financial crisis and the Eurozone sovereign debt crisis, we can analyze whether the effects of news announcements on liquidity differ between crisis and non-crisis periods. In addition, we can identify factors that determine FX market-wide liquidity. The results show that VIX of Chicago Board Options Exchange and U.S. unemployment news surprises are significantly related to aggregate market-wide FX liquidity. Section 2 describes the EBS and macroeconomic announcements data. In Section 3 we outline our measure of liquidity and approach to constructing common or market-wide liquidity. Section 4 presents the empirical results; Section 5 concludes. 2. Data 2.1. EBS data Electronic brokers are first introduced in the inter-dealer FX market as early as in 1992. Increasing competition from inter-dealer electronic platforms brings the high efficiency of transaction processing and settlement systems as well as lower transaction costs. Such lower costs make more trading strategies profitable, attracting more speculative activities and stimulating the entry of new investors in global FX markets. In the FX interdealer market, most electronic spot interdealer trading occurs on two competing platforms: Reuters and EBS. The EBS had achieved a market share of more than 60% and is the leading global marketplace for spot interdealer FX trading. For two major currency pairs, EUR/USD and USD/JPY, spot trading is mainly represented by EBS data. Furthermore, all dealers on the EBS platform are prescreened for credit and bilateral credit lines, which are then continuously monitored by the system, so counterparty risk is virtually negligible (Mancini et al., 2013). We also note rapidly growing literature that uses EBS data to analyze FX markets (e.g., Ito & Hashimoto, 2006; Karnaukh et al., 2015; Mancini et al., 2013). On the other hand, electronic brokers opened up to hedge funds and other customers via prime brokerage arrangements after 2004. These platforms became active areas for proprietary trading firms that are specialized in high-frequency trading (Bank for International Settlements, 2013; Banti, 2016). Since 2004, financial institutions trading scale through dealer’s order flow has grown explosively. These institutions include institutional investors (such as insurance companies and pension funds), hedge funds and proprietary trading firms. The market share of FX trading by institutional investors is up from 11% in 2013 to 16% in 2016, whereas the corresponding share of FX trading by proprietary trading firms and hedge funds declined from 11% in 2013 to 8% in 2016 (Bank for International Settlements, 2016). Our exchange rate data come from the brokered segment of the interdealer FX market, provided by the EBS. We consider nine currency pairs: USD/GBP, USD/CHF, USD/AUD, USD/JPY, USD/CAD, USD/EUR, EUR/GBP, EUR/JPY, and EUR/CHF. The data cover the period from January 7, 2008 to December 31, 2013. The data set contains the quote price and deal price for each 0.1 s time-slice. The quote price is a snapshot of the ten best levels of the book at the end of a time-slice (if a price or volume changed during that time-slice). The deal price lists the highest buying deal price and the lowest selling deal price (with dealt volumes) during the time-slice. We can classify the currencies studied into two types: the funding currency (including USD, EUR, JPY, and CHF) and the investment currency (including AUD, CAD, NZD, and GBP), according to Banti (2016) and Galati et al. (2007).

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

175

On the EBS trading platform, foreign currencies trade continuously, 24 h a day; however, transaction volume is relatively smaller on weekends. We therefore exclude the periods from Friday 24:00 to Sunday 24:00 GMT. To avoid extreme highfrequency noise and no-activity periods in very small time windows, we focus on 1 min frequencies, such that at the end of each minute of our sample, we record the last transaction price in each exchange rate. Each trade contains the volume of buyer-initiated trades in each minute, the volume of seller-initiated trades in each minute, and the transaction price. The volume of buyer- initiated trades refers to the total volume transacted, such that a quote to buy euros for dollars or dollars for yen represents an initiated transaction. The volume of seller-initiated trades is defined similarly for offers to sell. On the EBS trading platform, foreign currencies trade continuously, 24 h a day; however, transaction volume is relatively smaller on weekends. We therefore exclude the periods from Friday 24:00 to Sunday 24:00 GMT. To avoid extreme highfrequency noise and no-activity periods in very small time windows, we focus on 1 min frequencies, such that at the end of each minute of our sample, we record the last transaction price in each exchange rate. Each trade contains the volume of buyer-initiated trades in each minute, the volume of seller-initiated trades in each minute, and the transaction price. The volume of buyer- initiated trades refers to the total volume transacted, such that a quote to buy euros for dollars or dollars for yen represents an initiated transaction. The volume of seller-initiated trades is defined similarly for offers to sell. 2.2. Macroeconomic announcements data Many studies demonstrate how scheduled macroeconomic news announcements affect FX market volatility (e.g., Andersen & Bollerslev, 1998; Bauwens et al., 2005; Evans & Lyons, 2008), with a general consensus that U.S. domestic news announcements increase FX volatility between the dollar and other currencies. Andersen et al. (2003, 2007), Ehrmann and Fratzscher (2005), and Evans and Lyons (2005, 2008) also verify that U.S. macroeconomic information can affect other foreign exchange rates if it contains information about the state of other economies—perhaps because U.S. economic performance indicates the well-being of the ‘‘global economy.” We collected three relevant U.S. macroeconomic news announcements (CPI, GDP, and unemployment rate), for which we report the release date, time-stamped to the minute in GMT; the announced series value; and the median market survey expectation in Table 1.4 Then, following Balduzzi, Elton, and Green (2001), we constructed the standardized news surprise as:

Sk;t ¼

jActualk;t  Expectedk;t j

rðActualk;t  Expectedk;t Þ

ð1Þ

where Actualk;t is the announced actual value of indicator k at time t, Expectedk;t is the median forecast value of indicator k from the survey conducted by Econoday, and rðActualk;t  Expectedk;t Þ denotes the standard error of the difference between the actual and expected value of indicator k. In turn, we identify news surprises as favorable or adverse according to whether they lead the U.S. dollar to appreciate if the actual announced value is greater than the expected value. For example, a favorable surprise might announce that U.S. GDP is higher than expected, that is, SGDP > 0, and this ‘‘good” news may lead to appreciation of the U.S. dollar. If the actual value is lower than the expected value, SGDP < 0, this ‘‘bad” news likely leads to the depreciation of the U.S. dollar. We apply similar assessments to the unemployment rate and CPI. Finally, we define a dummy variable that is equal to 1 for a day that features a QE policy announcement in Table 2. The policies announcements include ‘‘Large Scale Asset Purchases,” such as ‘‘Quantitative Easing,” ‘‘Maturity Extension Program,” and tapering activities, i.e. the communication of the intended degree of future policy accommodation. Such events have been documented in Fawley and Neely (2013) and Altavilla and Giannone (2016), who detail the timeline of economic events that have prompted responses by the Fed. Therefore, our U.S. announcement data include three macroeconomic surprise variables and one announcement dummy variable related to QE. 3. Liquidity commonality 3.1. Liquidity measures We measure liquidity in several ways, following Mancini et al. (2013). The first liquidity measure is the proportional quoted spread, calculated as:

LðbaÞ ¼ ðPA  PB Þ=PM

ð2Þ

where P A , P B , and PM indicate the ask, bid, and mid quotes, respectively. The mid quote is defined as PM ¼ ðPA þ PB Þ=2. A market is liquid if the proportional quoted spread is low. The second measure is effective cost or effective spread. Because some traders in the electronic market may post hidden limit orders that are not reflected in quoted spreads, trades are not always executed at the posted bid or ask quotes. Yet the

4 The median market survey expectation is the median value of a survey conducted by Econoday, collected weekly and processed on the Friday prior to the announcement week. It represents the market consensus value. Several studies, such as Scholtus, van Dijk, and Frijns (2014) and Opschoor, Taylor, Van der Wel, and van Dijk (2014), affirm that the Econoday expectations contain valuable information about the expected value.

176

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Table 1 U.S. Macroeconomics Announcements. Announcements

Number

Sources

Dates

Favorable news if

GDP CPI Unemployment rate

24 72 72

BEA BLS BLS

01/2008–12/2013 01/2008–12/2013 01/2008–12/2013

Actual value higher than expected value Actual value lower than expected value Actual value lower than expected value

This table lists the U.S. macroeconomic announcement category, number of observations, source, and dates. The announcements considered include GDP, CPI, and unemployment rate. The sources are Bureau of Labor Statistics (BLS) and Bureau of Economic Analysis (BEA).

Table 2 Quantitative easing announcements by the federal reserve. Episode

Announcement date

Policy event

Event

QE1 QE1 QE1 QE1 QE1 QE1 QE1 QE1 QE1 QE2 QE2 QE2 QE2 QE2 QE2 Maturity Maturity Maturity Maturity Maturity Maturity QE3 QE3 QE3 Tapering Tapering Tapering

25-Nov-08 1-Dec-08 16-Dec-08 28-Jan-09 18-Mar-09 12-Aug-09 23-Sep-09 4-Nov-09 10-Aug-10 27-Aug-10 21-Sep-10 15-Oct-10 21-Oct-10 3-Nov-10 22-Jun-11 9-Aug-11 21-Sep-11 2-Nov-11 13-Dec-11 25-Jan-12 20-Jun-12 22-Aug-12 13-Sep-12 12-Dec-12 22-May-13 18-Jun-13 18-Dec-13

LSAP announcement Chairman Speech FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting Chairman speech FOMC meeting Chairman speech FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting FOMC meeting Chairman Speech FOMC meeting FOMC meeting

QE 1 Starts

Extension Extension Extension Extension Extension Extension

QE 1 Ends QE 2 Starts

QE 2 Ends Maturity Extension Program Starts

QE 3 Starts QE 3 Ends

This table reports the timing of Federal Reserve Quantitative Easing announcements from Fawley and Neely (2013) and Altavilla and Giannone (2016). Column ‘‘Episode” describes the relevant Federal Reserve policy statements; ‘‘Policy Event” indicates the source of policy Statements; ‘‘Event” reflects the beginning and ending of each QE period.

effective cost can be used to compare transaction prices with the quotes that prevail at the time of execution. The effective cost is:

LðecÞ ¼




ðP  PM Þ=P M

for buyer  initiated trades

ðPM  PÞ=P M

for seller  initiated trades

ð3Þ

where P denotes the transaction price. The daily average proportional quote spread and effective cost are calculated for each FX rate. The next two liquidity proxies are price impact and return reversal. The measure of price impact is based on contemporaneous relationship between returns and order flow (Evans & Lyons, 2002). The return reversal is related to the temporary price change due to order flow, proposed by Pástor and Stambaugh (2003). The intuition behind the Pastor-Stambaugh measure is that lower liquidity is accompanied by a higher volume-related return reversal. Banti et al. (2012) use a regression of exchange rate returns on contemporaneous and one-period-lagged order flows to estimate price impact and return reversal in the FX context. We follow Mancini et al. (2013) by including K-period lagged order flows in the regression of intraday return rti as follows:

rti ;t ¼ ht þ ut ðv b;ti  v s;ti Þ þ

K X

ct;k ðv b;ti k  v s;ti k Þ þ eti

ð4Þ

k¼1

For exchange rate i, rti ;t denotes the intraday return between t i  1 and t i on day t; v b;ti and v s;ti are the volume of buyerinitiated trades and the volume of seller-initiated trades at time ti during day t, respectively. Following Mancini et al. (2013),

177

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

we use one-minute data of returns and order flow, with the lag length K = 5, to obtain daily estimates of price impact and PK k¼1 ct;k , which is expected to be negative. The price

price reversal. The overall return reversal is calculated by LðrrÞ ¼ ct ¼

impact, LðpiÞ ¼ ut , is expected to be positive due to net buying pressure. Table 3 reports correlations in effective costs across individual exchange rates in the full period (Jan 2008–Dec 2013) and two sub-periods of crisis: Aug 2008–Mar 2009 and Oct 2009–July 2011. We observe all correlations are positive, with values ranging from 0.5006 to 0.8346. Overall, the correlation across individual currency liquidity is stronger during the subprime mortgage financial crisis period; particularly, the correlations between EUR/JPY and USD/JPY liquidities and between USD/ CHF and EUR/JPY liquidities are higher than 0.9. Although the liquidity correlations across exchange rates are consistently higher in the period of Eurozone sovereign debt crisis, we observe that most liquidity correlations between EUR/JPY and the other currency are higher than 0.6 during Oct 2009–July 2011. Table 4 contains the correlations between the common or market-wide FX liquidity according to various measures: return reversal, price impact, bid–ask spread, effective cost, and price dispersion. It also includes the common factor of the four liquidity measures we consider. To compute the correlations, we used 72 nonoverlapping monthly observations. The data cover the period from January 7, 2008, to December 31, 2013. The results show that the common factor of each liquidity measure is contemporaneously correlated with each other and with their common factor, at values ranging from 0.537 to 0.936. Thus, we identify important commonalities across various measures of liquidity, consistent with Mancini et al.’s (2013) finding of strong comovements among market-wide liquidity measures. We report the results using the measure of effective cost as a representative case. The summary statistics for the liquidity measures (price impact, return reversal, bid–ask spread, effective cost) are in Table 5. In line with Evans and Lyons (2002) and Berger, Chaboud, Chernenko, Howorka, and Wright (2008), more liquid assets should exhibit a lower price impact. Table 5 shows that the average of the price impact coefficient is positive, ranging from 0.00005 to 0.00022. Furthermore, USD/EUR has the smallest price impact, whereas investment currencies have the greatest, especially USD/AUD and USD/CAD. The average return reversal, that is, the temporary price change accompanying lagged order flows, is negative and ranges from 0.00005 to 0.00025, showing a slight reduction in current one-minute returns in response to order flows in the preTable 3 Correlation of effective costs. USD/GBP

USD/CHF

EUR/CHF

EUR/GBP

EUR/JPY

USD/AUD

USD/CAD

USD/JPY

USD/EUR

1 0.8050 0.7263 0.7710 0.7019 0.7174 0.7305 0.8346

1 0.5951 0.6514 0.5006 0.5227 0.5470 0.6505

1 0.7511 0.6412 0.6812 0.6708 0.7352

1 0.7098 0.6489 0.7661 0.7496

1 0.7442 0.7096 0.7276

1 0.7601 0.7626

1 0.8193

1

1 0.8250 0.9355 0.7163 0.7262 0.8793 0.8886

1 0.8333 0.6181 0.6828 0.7875 0.8191

1 0.7824 0.7263 0.9312 0.9232

1 0.6220 0.7553 0.7157

1 0.7097 0.6750

1 0.8845

1

1 0.2926 0.4887 0.1826 0.1702 0.2166 0.4347

1 0.6751 0.5486 0.7171 0.4456 0.6682

1 0.6046 0.6417 0.6511 0.6920

1 0.5414 0.5089 0.5016

1 0.5829 0.5239

1 0.4028

1

Panel A. All sample USD/GBP USD/CHF EUR/CHF EUR/GBP EUR/JPY USD/AUD USD/CAD USD/JPY USD/EUR

1 0.7734 0.5708 0.8124 0.7719 0.7503 0.7753 0.7639 0.7977

Panel B. Subprime mortgage financial crisis period USD/GBP USD/CHF EUR/CHF EUR/GBP EUR/JPY USD/AUD USD/CAD USD/JPY USD/EUR

1 0.8299 0.8518 0.8598 0.8535 0.6650 0.6707 0.8229 0.8191

1 0.9218 0.8248 0.9027 0.7055 0.6470 0.8468 0.8909

Panel C. Eurozone sovereign debt crisis period USD/GBP USD/CHF EUR/CHF EUR/GBP EUR/JPY USD/AUD USD/CAD USD/JPY USD/EUR

1 0.5802 0.1304 0.7863 0.7139 0.5238 0.7657 0.5809 0.6263

1 0.5548 0.5687 0.7039 0.4452 0.4890 0.4386 0.6200

This table presents the time-series correlations of the across-currencies effective cost over the whole period and over two subperiods: whole sample (Jan 2008–Dec 2013), the subprime mortgage financial crisis (Aug 2008–Mar 2009) and European sovereign debt crisis (Oct 2009–July 2011). Effective cost denotes the daily average of intraday effective cost estimates. Foreign exchange illiquidity is reported for the British pound (GBP), Swiss France (CHF), euro (EUR), Australian (AUD), Canada (CAD), and Japanese yen (JPY).

178

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Table 4 Correlation of market-wide liquidity measures.

Return reversal Price impact Bid-ask spread Efficient cost Common factor with all four measures

Return reversal

Price impact

Bid-ask spread

Efficient cost

Common factor with all four measures

1 0.8439 0.6266 0.6443 0.6329

1 0.6199 0.6397 0.6293

1 0.9667 0.9899

1 0.9895

1

Individual liquidity is measured by return reversal, price impact, bid–ask spread, and effective cost, as well as a common factor that combines all four measures. This table reports correlations between market-wide liquidity (i.e., common factor across currencies for each liquidity measure). We use daily standardized measures of liquidity to extract the corresponding common factors or market-wide liquidity, according to a generalized dynamic factor method (GDFM). The sample period spans from January 2, 2008, to December 30, 2013.

Table 5 Descriptive statistics for daily illiquidity measures.

Price impact Mean Std.dev. Skewness Kurtosis

q1 Q(20)

USD/GBP

USD/CHF

EUR/CHF

EUR/GBP

EUR/JPY

USD/AUD

USD/CAD

USD/JPY

USD/EUR

0.00015 0.00017 3.0129 16.3846 0.1347 478.94***

0.00008 0.0010 2.9190 15.9160 0.1395 566.33***

0.00007 0.00009 4.2144 33.7664 0.2504 1526***

0.00013 0.00013 2.4556 11.6556 0.0775 118.47***

0.00011 0.00013 2.9544 16.4213 0.1280 596.54***

0.00022 0.00037 5.7703 50.6673 0.2333 1563***

0.00019 0.00023 3.3321 19.6029 0.1630 1107***

0.00005 0.00007 3.8008 24.0750 0.2066 1712.3***

0.00005 0.00005 2.6867 12.9243 0.2408 1759***

0.00016 0.00019 3.6263 23.4872 0.0843 214.97***

0.00009 0.00010 2.9731 16.4556 0.1077 367.22***

0.00008 0.00014 5.9387 52.9624 0.1710 672.67***

0.00014 0.00016 3.8011 27.2871 0.0077 29.12***

0.00011 0.00014 3.9940 30.9321 0.1094 407.05***

0.00025 0.00045 5.8441 49.9698 0.1208 436.94***

0.00020 0.00023 3.2922 19.4240 0.1781 779.88***

0.00005 0.0006 3.2823 18.2359 0.2627 1862.4***

0.00005 0.0005 3.0905 16.2677 0.1873 916.87***

0.00029 0.00019 3.5954 22.0047 0.5735 9220***

0.00031 0.00025 3.8821 24.7391 0.0548 2060***

0.00026 0.00020 2.9492 16.7307 0.2702 4045***

0.00030 0.00011 4.2106 36.0977 0.6248 8102***

0.00034 0.00031 4.2342 26.5987 0.0110 1681***

0.00041 0.00040 4.7053 33.6777 0.3091 3373***

0.00036 0.00020 2.3659 12.7713 0.8176 18892***

0.00019 0.00010 2.4834 10.8747 0.0953 3730***

0.00013 0.00008 2.4872 10.2813 0.0649 3193***

0.00011 0.00006 3.1713 17.4174 0.6794 1235***

0.00011 0.00008 4.2254 29.1040 0.0938 1707***

0.00009 0.00006 2.7500 13.9781 0.2800 3781***

0.00011 0.00004 2.8602 16.5537 0.6512 8826***

0.00014 0.00011 4.5505 31.3487 0.0709 1279***

0.00017 0.00016 6.2188 56.3751 0.2373 2016***

0.00014 0.00007 2.4351 17.1576 0.7883 16963***

0.00007 0.00003 2.1451 9.5751 0.2556 44551***

0.00005 0.00003 2.0734 8.5586 0.2164 3809***

Return reversal Mean Std.dev. Skewness Kurtosis

q1 Q(20) Bid-ask spread Mean Std.dev. Skewness Kurtosis

q1 Q(20) Effective cost Mean Std.dev. Skewness Kurtosis

q1 Q(20)

This table contains summary statistics for the mean, standard deviation, skewness, and kurtosis for various daily measures of the liquidity of each currency. Bid-ask spread is the average bid-ask spread computed from intraday data for each trading day (Eq. (2)). Effective cost is the average relative difference between the transaction price and the bid-ask quote prevailing at the time of trade (Eq. (3)). Price impact is the estimated coefficient of contemporaneous order flow ut , in a regression of one-minute returns on contemporaneous and lagged order flow. Return reversal denotes the sum of the coefficients of lagged order flow, as shown in Eq. (4). In addition, q1 is the sample autocorrelation at lag 1 for each currency; the Ljung-Box Q tests for autocorrelation in time series data; and Q ð20Þ is the critical value for rejection of the null hypothesis of randomness up to lag 20. The data cover the period from January 7, 2008, through December 31, 2013. ⁄⁄⁄, ⁄⁄, and ⁄ denote the significance at the 1%, 5%, and 10% levels, respectively.

vious five minutes. As argued in Mancini et al. (2013). This reduction is economically significant, given the fact that average five-minute returns are near to zero. We also observe that effective costs are less than half the quoted bid-ask spread, similar to Mancini et al. (2013). In most cases, the illiquidity measures for USD/CAD and USD/AUD reveal significantly greater variability than those of other currency pairs. In addition, USD/EUR and USD/JPY have a lower mean and standard deviation than other exchange rates, suggesting that they are the most liquid exchange rates, a claim that corresponds with the perceptions of market participants and its largest market share in terms of turnover (Bank for International Settlements, 2013). However, USD/GBP trading takes one of the largest market shares in the world, and it appears to be the least illiquid currency market, likely because USD/GBP is mostly traded on the Reuters rather than the EBS platform (Chaboud, Chernenko, & Wright, 2007). In the last two columns of Table 5, we report the autocorrelation and Ljung-Box Q-test for serial correlation. For most markets,

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

179

Fig. 1. Plots daily standardized effective cost. The effective cost is defined as (P  P M Þ=P M for buyer-initiated trades and (P M  PÞ=P M for seller-initiated trades, where P denotes the transaction price and P M is the mid-quote price. The sample spans from January 2, 2008 to December 30, 2013. The first shaded area in Fig. 1 is subprime mortgage financial crisis period (Aug 2008–Mar 2009) and the second shaded area is European sovereign debt crisis (Oct 2009– July 2011).

the illiquidity measures exhibit strong autocorrelation. Illiquidity has a persistent phenomenon, suggesting that an illiquid day is likely to be followed by another illiquid day. To understand the dynamics of the illiquidity measure, in Fig. 1, we plot the effective cost, as defined in Eq. (3), for nine currency pairs. As Fig. 1 shows, the liquidity of most currency pairs is quite stable during the first half of 2008. Afterward, the illiquidity measure sharply increased during crisis episodes, suggesting FX liquidity dropped suddenly following the collapse of Bear Stearns in 2008 and at the start of the Eurozone sovereign debt crisis in this decline reflected the turmoil and uncertainty in financial markets caused by the Lehman Brothers bankruptcy and rising sovereign debt. Fig. 1 also suggests large cross-sectional differences in FX rate illiquidities, as measured by effective costs. The decline in USD/AUD, USD/CAD, and USD/GBP liquidity following the subprime mortgage financial crisis and Eurozone sovereign debt crisis were speedy and more significant than those of any other currency pairs. We find that the level of liquidity varies significantly across FX rates and over time. In times of crisis, the salient reduction in currency liquidity are consistent with Melvin and Taylor (2009), Mancini et al. (2013), Karnaukh et al. (2015), and Banti (2016) that argue extremely high volatility and liquidity risk lead to reduced liquidity in the FX market, channeled by carry trade activity of currency investors. In particular, the liquidity of investment currencies is more sensitive to market stress than that of funding currencies. 3.2. Common liquidity across exchange rates To study co-movement in liquidity across currencies, we must identify the market-wide or global common liquidity in FX markets. The approach of principal components analysis is commonly used to extract the liquidity commonality.5 Instead, we 5 For example, Hasbrouck and Seppi (2001) use a principal components analysis to retrieve the common component in liquidities of 30 stocks in the Dow Jones index. A canonical correlation analysis indicates that the common factor in order flows correlates strongly with the common factor in returns. Korajczyk and Sadka (2008) instead use a latent factor model to measure common liquidity and find that shocks to assets’ liquidity have a common component across measures, which accounts for most explained variation in the individual liquidity.

180

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

apply a generalized dynamic factor model (GDFM) to assess market-wide liquidity (Forni et al., 2000), which offers a parsimonious, realistic representation of the data. As Hallin et al. (2011) explain, liquidity might be significantly autocorrelated. If we overlook the lead-lag relationship in the liquidity, we might obtain an inefficient estimator for common market-wide liquidity. Previous studies use the GDFM to construct macroeconomic indexes (Forni, Hallin, Lippi, & Reichlin, 2003; Forni et al., 2000), perform forecasting (Stock & Watson, 2002a 2002b; 6 Forni, Hallin, Lippi, & Reichlin, 2005), and analyze financial markets (Corielli & Marcellino, 2006; Hallin et al., 2011; Luciani & Veredas, 2015; Ludvigson & Ng, 2007, 2009). The GDFM has become a common tool, used in many research fields, for several reasons. First, it can handle the autocorrelation in timeseries data. Second, the information inherent to a large set of variables is broader and can improve estimation efficiency. Both Stock (1998) and Forni et al. (2000) confirm that a dynamic factor analysis can be performed on a large database without suffering the curse of dimensionality. Third, the GDFM allows for a moderate correlation between idiosyncratic components, whereas a classical factor model must presume that the variables are mutually orthogonal. To overweighting some measures that are more volatile than others, simply due to their measurement scale, we first standardized our liquidity measures. For each exchange rate, a given liquidity measure is deducted by its time-series mean and then divided by its standard deviation. Thus, our observation is an n  T panel of FX liquidity measures, which is a finite realization

0

1 L11 ; L12 ;    ; L1t B L21; L22 ;    ; L2t C B C B C .. .. C B .. @ . . . A Ln1 ; Ln2 ;    ; Lnt of a double stochastic process sequence, of the form L = {Lit |i 2 N; t 2 Z}, where Lit is the liquidity measure at day t for exchange rate i. In the dynamic factor analysis, Lit for all i and t can be decomposed into a common component X it and an idiosyncratic component Z it with q factors. Then Lit can be written as:

Lit ¼ X it þ Z it ¼

q X cik ðBÞukt þ Z it

ð5Þ

k¼1

where ut = {(u1t ; u2t ; . . . ; uqt )0 |t 2 z is a q  1 of dynamic factors with white noise; Z it and ujtk are mutually orthogonal at any lead and lag for all i, j 2 N; B stands for backshift operator; and the polynomials in the backshift operator filters ci1 ðBÞ + . . . + ciq ðBÞ are square-summable and one-sided filters, following the in principle of infinite order. For any n 2 N, Lit can be written with a vector notation:

Ln ¼ X n þ Z n ¼ CðBÞut þ Z n ; t 2 z

ð6Þ

0

such that Ln = {(L1t , L2t ; . . . Lnt ) |t 2 z} is the n-dimensional vector subprocesses of L, and C(B) = (C 1 ðBÞ; . . . ; C q ðBÞ) is an n  q square-summable polynomial in the backshift operator B and q < n. The decomposition of Ln produces a decomposition CLn ¼ CX n þ CZ n of the cross-covariance matrices CLn ¼ E½Ln;t L0n;tk  of the various Ln . Denoted by the covariance matrix of Ln P P as L;n ðhÞ, it can be written as the sum of the spectral density matrix of common component X;n ðhÞ and the spectral density P P P P matrix of idiosyncratic component Z;n ðhÞ, that is L;n ðhÞ ¼ X;n ðhÞ þ Z;n ðhÞ. The statistical treatment of Lit is as follows: (1) According to the Brillinger concept of dynamic principal components (Forni et al., 2000), we use it to estimate a consistent (both n and T tend to infinity) reconstruction of Y n decomposition. (2) An information criterion for determining the number q of common shocks in the general dynamic factor model (Hallin & Liška, 2007). We use this process to estimate the common factor for each measure of liquidity individually. Each market-wide liquidity measure is estimated across all exchange rates and uses the first factor as a proxy for market-wide liquidity, LiM;t , combining most information across all exchange rates. We thus obtain four market-wide liquidity indicators: RR the market-wide liquidity indicator of price impact LPI M;t , the market-wide liquidity indicator of return reversal LM;t , the EC market-wide liquidity indicator of bid–ask spread LBA M;t , and the market-wide liquidity indicator of effective cost LM;t . For

interpretability, we only include the liquidity indicator of effective cost LEC M;t as our market-wide liquidity measure. Finally, we assess common factors across all measures of liquidity and across all exchange rates, following Korajczyk and Sadka’s (2008) suggestion to extract the common liquidity and going one step further by combining information from varðbaÞ ðecÞ ðpiÞ ðrrÞ 0 ious liquidity measures. We stack all four liquidity measures, across all exchange rates, into ~ LM;t ¼ ½~L ; ~ L ;~ L ;~ L  . Theret

t

t

t

6 Stock and Watson (1998) initially developed the dynamic factor approach using principal components analysis for forecasting. The Stock-Watson factor model (SW) differs from the GDFM method in three respects. First, the common factor calculation method is different. The SW factor model uses a least squares regression, whereas the GDFM approach relies on a non-parametric regression, accounting for differences between dynamic factors and their lagged values and imposing rank reduction onto the spectral frequency matrix. Second, the weights differ when common factors are calculated. The SW factor model employs a standard principal components model to acquire common factors; the GDFM’s estimation is based on a method of generalized principal components. This weighting scheme is a more efficient estimation method. Third, the methods differ in the way they forecast the idiosyncratic component. The SW model uses lagged values in the forecast calculation, and the GDFM forecasts the idiosyncratic component on the basis of an assumption of orthogonality across the common and idiosyncratic components.

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

181

fore, we assume a latent factor model of liquidity, which is estimated using GDFM. We refer to common factors extracted across the liquidity measures as an ‘‘across-measure factor” or ‘‘global-wide liquidity index,” ~ LM;t . 4. Empirical results In Section 4.1, we investigate the pervasiveness of co-movements of the individual FX liquidity and global-wide FX liquidity, following Mancini et al.’s (2013) commonality methodology for FX markets. Section 4.2 examines commonality in FX liquidity around macroeconomic and QE announcements. Finally, we study the potential determinants of FX liquidity commonality in Section 4.3. 4.1. Liquidity commonality for FX markets during crisis period As documented in Melvin and Taylor (2009), the crisis in the FX market came relatively later than the crisis in equity and fixed income markets since August 2007. The tremendous unwinding of the carry trade caused huge losses among currency investors, and induced a dramatic reduction in FX market liquidity in times of crisis. (Melvin & Taylor, 2009). Additionally, Mancini et al. (2013) and Karnaukh et al. (2015) also find evidence of strong linkage of illiquidity across currency markets during the financial crisis period. Melvin and Taylor (2009) documented that exchange rates experienced unprecedented levels of volatility after the failure of Lehman Brothers. The sharp rise in FX illiquidity is accompanied with extremely high levels of volatility and inventory risk. As a result, the dealer ultimately attempted to offset the inventory risk by using a widened bid-ask spread dramatically (Banti, 2016; Melvin & Taylor, 2009). Furthermore, investment currencies are more easily exposed to volatility and liquidity risk than funding currencies (Banti, 2016; Karnaukh et al., 2015; Mancini et al., 2013; Melvin & Taylor, 2009; Osler, 2012). In particular, the major unwinding of carry trade not only leads to deteriorated liquidity but also increased risk for currencies involved in carry trade activity. We extend Mancini et al.’s (2013) analysis to include a financial crisis factor. For this study, we focus on subprime mortgage financial crisis and Eurozone Sovereign Debt Crisis, to scrutinize whether the financial turmoil of the financial crisis had any impact on commonality in liquidity. Accordingly, we start by determining the crisis length, based on official timeline provided by Federal Reserve Bank of St. Louis (2009) and Filardo et al. (2010).7 We add two dummy variables to capture its impact: Icrisis1;t corresponds to the subprime mortgage financial crisis, such that it equals 1 for t during September 2008–March 2009; and Icrisis2;t corresponds to the Eurozone sovereign debt crisis, so it equals 1 for t during October 2009–July 2011. For each currency j, we test for commonality in liquidity using a time-series regression:

LEC i;t ¼ a0;i þ

J X EC EC EC hj;i LEC i;tj þ b0;i LM;t þ a1;i I crisis1;t þ b1;i Icrisis1;t LM;t þ a2;i Icrisis2;t þ b2;i Icrisis2;t LM;t þ ei;t

ð7Þ

j¼1 EC 8 where LEC i;t is the daily liquidity measure of effective cost for exchange rate i on day t ; LM;t is the market-wide liquidity indicator of effective cost on day t. To avoid potentially upward-biased estimates for b0;i , b1;i , and b2i , we exclude exchange rate i in

9 the computation of LEC M;t . The terms of b0;i þ b1;i and b0;i þ b2;i measure the magnitude of liquidity commonality during the subprime mortgage and Eurozone sovereign debt crisis periods, respectively. The b0;i parameter indicates the effect of liquidity commonality in the non-crisis period, which we refer to as the liquidity beta. Table 6 presents strong evidence for the existence of liquidity commonality. We find co-movement of individual currency liquidity with the aggregate FX liquidity. Liquidity co-movement thus is a pervasive phenomenon across all currencies, in support of our prediction. As predicted by Mancini et al. (2013), we also confirm that liquidity commonality exists in FX markets. The coefficients of the crisis dummy variables also are significant and positive across almost all currency markets studied. We find that the level of liquidity varies significantly across FX rates and over time, such that the relation between individual FX liquidity and aggregate common FX liquidity grows much stronger in times of crisis, is consistent with findings in the equity market (Kamara et al., 2008; Karolyi et al., 2012; Rösch & Kaserer, 2013) and the FX market (Karnaukh et al., 2015; Mancini et al., 2013; Melvin & Taylor, 2009). Meanwhile, we observe that investment currencies, such as USD/GBP, USD/AUD, and EUR/GBP, have a higher level of liquidity commonality than those of funding currencies. As Mancini et al. (2013) and Banti (2016) show, investment currencies

7 According to the Federal Reserve Bank of St. Louis (2009) and Filardo et al. (2010), the subprime mortgage financial crisis consists of four phases: (1) August 1, 2007 to mid-September 2008, which is ‘‘initial financial turmoil”; (2) September 16–December 31, 2008, defined as ‘‘sharp financial market deterioration”; (3) January 1–March 31, 2009, or ‘‘macroeconomic deterioration”; and (4) after April 1, 2009, is described as ‘‘stabilization and tentative signs of recovery” (post-crisis period) including a financial market rally. 8 The optimal value of J is determined by the Akaike information criterion (AIC). EC EC 9 As noted in Chordia et al. (2000) and Mancini et al. (2013), if LEC i;t is included in the calculation of LM;t , Li;t will enter both sides of Eq. (7), potentially leading to upward-biased estimates for commonality coefficients. We thank the referee for pointing out this adjustment.

182

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Table 6 Commonality in FX liquidity during crisis and non-crisis periods.

a0,i b0,i

a1,i b1,i

a2,i b2,i H0: b1,i=b2,i Adj:R2

USD/GBP

USD/CHF

EUR/CHF

EUR/GBP

EUR/JPY

USD/AUD

USD/CAD

USD/JPY

USD/EUR

0.0877*** (0.0181) 0.0288*** (0.0066) 0.3640*** (0.0872) 0.0765** (0.0252) 0.0807*** (0.0239) 0.0377* (0.0172) 0.1797 0.778

0.0574*** (0.0163) 0.0410*** (0.0099) 0.1525** (0.0551) 0.0354 (0.0234) 0.1021** (0.0330) 0.0165 (0.0248) 0.5448 0.780

0.0320* (0.0130) 0.0256*** (0.0057) 0.0283 (0.0439) 0.0154 (0.0207) 0.1255*** (0.0341) 0.0513* (0.0233) 0.2282 0.855

0.0629** (0.0191) 0.0533*** (0.0088) 0.2854*** (0.0705) 0.0158 (0.0198) 0.1488*** (0.0416) 0.0945** (0.0309) 0.0238* 0.628

0.1062*** (0.0196) 0.0126 (0.0091) 0.3585*** (0.0762) 0.0533 (0.0292) 0.1750*** (0.0360) 0.0735** (0.0226) 0.5563 0.692

0.0687*** (0.0177) 0.0196** (0.0069) 0.2868*** (0.0603) 0.0522** (0.0177) 0.0380 (0.0197) 0.0158 (0.0138) 0.0811 0.781

0.0461*** (0.0136) 0.0368*** (0.0060) 0.1971** (0.0752) 0.0129 (0.0228) 0.0581* (0.0229) 0.0149 (0.0140) 0.936 0.791

0.0765*** (0.0173) 0.0167* (0.0072) 0.2481*** (0.0569) 0.0181 (0.0146) 0.1378*** (0.0314) 0.0373* (0.0174) 0.3507 0.799

0.0976*** (0.0199) 0.0309*** (0.0079) 0.3166*** (0.0660) 0.0264 (0.0175) 0.1951*** (0.0383) 0.0621** (0.0231) 0.1863 0.792

PJ EC h LEC þ b0;i LEC This table reports the estimation results of the following model: LEC i;t ¼ a0;i þ M;t þ a1;i Icrisis1;t þ b1;i Icrisis1;t LM;t þ a2;i Icrisis2;t þ j¼1 j;i i;tj EC b2;i Icrisis2;t LEC þ e ; where L is the daily liquidity measure by minus the effective cost for each FX rate on day t; I ¼ 1 if t is during the subprime i;t crisis1;t M;t i;t mortgage financial crisis period, and Icrisis2;t ¼ 1 if t is during the Eurozone sovereign debt crisis period, and 0 otherwise. Moreover, b0;i indicates individual liquidity sensitivity to market-wide FX liquidity; b1;i measures the change in liquidity commonality during the subprime mortgage financial crisis period, compared with the tranquil period; and b2;i refers to the change in the liquidity commonality during the Eurozone sovereign debt crisis period, compared with the tranquil period. The optimal length of lagged LEC i;tj included is determined*** by the Akaike information criterion (AIC) and varies for each currencypair. The estimated coefficients of lagged LEC i;tj are not reported in the table for simplicity. The heteroscedasticity- and autocorrelation-consistent standard errors (Newey & West, 1987) are reported in parentheses. The data cover the period from January 7, 2008, through December 31, 2013. ***, **, and * denote the significance at the 1%, 5%, and 10% levels, respectively.

suffer greater exposure to the global risk resulting from the unwinds of carry trade position, which may lead to a sharp decrease in FX liquidity. This result corresponds with the spirit of currency crash. Since more participants conduct positive-feedback trading strategies in the times of a market downturn, it further strengthens the risk of a crash (Osler, 2012). As the market declines, the fluctuations in asset value affect a trader’s portfolio value, which increases the probability of margin calls and reduces the supply of funds to traders, resulting in greater funding constraints (Banti, 2016; Melvin & Taylor, 2009). A trader might be forced to partially liquidate the portfolio, imposing additional price pressures on the asset. A self-enforcing liquidity spiral might emerge, with an adverse effect on liquidity in carrying traders’ currencies. Therefore, during the financial crisis periods, most of investment currencies are affected in common, leading to increases in liquidity commonality than those of funding currencies. These results support our hypothesis that liquidity commonality exists and is time-varying. Almost all currencies we study relate strongly to the global-wide common liquidity component, especially for investment currencies. Overall, the results corroborate the view that the relation between liquidity in the individual currency market and the common component of liquidity is stronger during crisis periods. 4.2. Liquidity commonality around macroeconomic announcements We now move on to how FX liquidity responds to news. According to an inventory control model (Amihud & Mendelson, 1980; Fleming & Remolona, 1999; Ho & Stoll, 1983; O’Hara & Oldfield, 1986), the wide bid–ask spread at an announcement reflects the dealer’s reluctance to make markets during times marked by high price volatility. As Karolyi et al. (2012) show, higher market volatility affects either the liquidity demand (i.e., panic selling, risk aversion) or its supply (capital constraint) across many assets, leading to diminished market-wide liquidity or a common component of liquidity. The resulting decrease in an individual market’s liquidity causes further asset price pressures, creating an illiquidity spiral that strengthens the co-movement of liquidity (Brunnermeier & Pedersen, 2009). With this analysis, we aim to test whether a news announcement that invokes greater market uncertainly or volatility reinforces commonality in liquidity and whether market liquidity reacts asymmetrically to positive and negative news surprises, namely, with a more pronounced reaction to negative than to positive news. Moreover, we acknowledge that U.S. macroeconomic information might move exchange rates if it contains information about the state of other economies or the global economy (e.g., Andersen & Bollerslev, 1998; Bauwens et al., 2005; Evans & Lyons, 2008).10 Therefore, small changes in underlying U.S. news announcements might cause sharp price volatility during

10 Furthermore, examining how macroeconomic news surprises affect the two components of volatility (i.e., direct and indirect, where indirect volatility is induced by volatility spillovers), Omrane and Hafner (2015) find that European news surprises trigger significant boosts of the British pound and Japanese yen and also contribute to the volatility of the euro. Similar results emanate from British and Japanese news. However, U.S. macroeconomic announcements have the largest impact, in that both scheduled and unscheduled announcements have significant effects on the volatility of EUR/USD, GBP/USD, and JPY/USD exchange rates.

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

183

crises, leading to greater commonality liquidity. With this analysis, we aim to test whether US news surprises that foster greater risk reversal and unwind carry trade activity, reinforcing commonality in liquidity during the subprime mortgage crisis and Eurozone sovereign debt crisis. Therefore, we extend Eq. (7) by including macroeconomic announcement surprises. The revised regression is as follows:

LEC i;t ¼ a0;i þ

J 3 3 3 3 X X X X X EC hj;i LEC /0;j;i Sþj;t þ d0;j;i Sþj;t þ c0;j;i Sþj;t LEC k0;j;i Sj;t LEC i;tj þ b0;i LM;t þ M;t þ M;t j¼1

þ Icrisis1;t a1;i þ

j¼1

b1;i LEC M;t

þ

3 X j¼1

þ Icrisis2;t a2i þ b2;i LEC M;t

/1;j;i Sþj;t

j¼1

þ

3 X j¼1

d1;j;i Sj;t

j¼1

þ

3 X

þ EC 1;j;i Sj;t LM;t

c

j¼1

j¼1 3 X þ k1;j;i Sj;t LEC M;t j¼1

3 3 3 3 X X X X þ /2;j;i Sþj;t þ d2;j;i Sj;t þ c2;j;i Sþj;t LEC k2;j;i Sj;t LEC M;t þ M;t j¼1

j¼1

j¼1

!

! þ ei;t

ð8Þ

j¼1

where Icrisis1;t ¼ 1 if t is during the subprime mortgage financial crisis period, and Icrisis2;t ¼ 1 if t is during the Eurozone sovereign debt crisis period, and 0 otherwise; b0;i indicates individual liquidity sensitivity to systematic liquidity during days  þ without an announcement. Furthermore, Sþ j;t and Sj;t are positive and negative news surprises, respectively; Sj;t ¼ Sj;t if the surprise denotes good news for indicator j, and 0 otherwise; S ¼ S if the surprise denotes bad news for indicator j, and j;t j;t 0 otherwise. Let j = 1, 2, and 3 denote CPI, GDP, and unemployment rate announcements, respectively. The b0;j parameter represents the extent of commonality in liquidity during the non-crisis period, while b0;j þ b1;j and b0;j þ b2;j measure the extent of liquidity commonality during the subprime mortgage and Eurozone sovereign debt crisis periods in times of no announcements, respectively. At the release of positive news, /0;j;i captures the news effect during the tranquil period, while /1;j;i and /2;j;i measure the change in news effects during the subprime mortgage and Eurozone sovereign debt crisis periods. At the release of negative news, d0;j;i captures the news effect during the tranquil period, whereas d1;j;i and d2;j;i measure the change in news effects during the subprime mortgage and Eurozone sovereign debt crisis periods, respectively.  EC EC The interaction terms Sþ j;t LM;t and Sj;t LM;t allow us to examine whether the liquidity commonality is affected by positive and negative news surprises, respectively. c0;j;i , c0;j;i þ c1;j;i and c0;j;i þ c2;j;i guage how positive surprises affect the liquidity commonality during the tranquil period, the subprime mortgage crisis period, and Eurozone sovereign debt crisis period, respectively. Alternatively, k0;j;i k0;j;i þ k1;j;i and k0;j;i þ k2;j;i guage how negative surprises affect the liquidity commonality during the tranquil period, the subprime mortgage crisis period, and Eurozone sovereign debt crisis period, respectively. Tables 7 reports the results for the change in liquidity commonality in response to U.S. positive and negative announcement surprises. First of all, we observe that a negative U.S. announcement has no significant influence on individual liquidity during the tranquil period. Moreover, during times of crisis, both positive and negative announcements are significantly related to individual currency liquidity, although in different directions. Overall, negative news is negatively associated with individual FX liquidity. As for the liquidity commonality, we find that, during the subprime mortgage financial crisis period, liquidity commonality increases in response to U.S. positive news only for the CPI and unemployment, whereas it declines for GDP announcement. However, the good news about the CPI announcement has an opposite effect on liquidity commonality during the two periods of crisis. Overall, we find greater liquidity commonality during the subprime mortgage financial crisis period, compared to the Eurozone sovereign debt crisis period. As Brockman et al. (2009) show, U.S. macroeconomic announcements have significantly positive impacts on the liquidity commonality across global stock markets. A negative U.S. announcement generally may increase the market stress and may invoke greater volatility in the market. An increase in volatility may lead to a wider bid-ask spread and lower liquidity (Stoll, 1978). However, this effect cannot fully explain the potentially negative impact of bad news on FX liquidity commonality during periods of crisis. Menkhoff et al. (2012) relate carry trade returns to FX volatility and show that carry trade performs poor in times of high volatility. Therefore, when the market becomes more volatile at the arrival of negative macro news, the ill performance of carry trade activity may induce some traders to unwind their carry trade position, thus the co-movement in FX liquidity becomes stronger through the channel of funding liquidity during the crisis periods (Mancini et al., 2013). Overall then, these results show that FX liquidity commonality responds asymmetrically to investment currencies and safe-haven currencies. US news surprises significantly affect the FX liquidity co-movements, measured by the efficient spread, consistent with Brockman et al. (2009). Our findings also suggest that FX dealers trade more actively and faster in response to investment currencies than to safe-haven currencies; investment currencies exert a greater impact on liquidity commonality during the crisis period.11

11 We report results based on three major announcements (Brockman et al., 2009), but we actually considered announcements about other macroeconomic indicators too, including nonfarm production, consumer confidence index (CCI), durable orders, housing starts, industrial production, jobless claims, producer price index (PPI), personal spending, retail sales, and trade balance. Only the three focal announcements exert significant influences on liquidity commonality, so we limit our analysis to these announcements.

184

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Table 7 Impacts of U.S. Positive and negative macroeconomic announcements on liquidity commonality.

a0;i b0;i /0;1;i /0;2;i /0;3;i d0;1;i d0;2;i d0;3;i

c0;1;i c0;2;i c0;3;i k0;1;i k0;2;i k0;3;i

a1;i b1;i /1;1;i /1;2;i /1;3;i d1;1;i d1;2;i d1;3;i

c1;1;i c1;2;i c1;3;i k1;1;i k1;2;i k1;3;i

a2;i b2;i /2;1;i /2;2;i /2;3;i

USD/GBP

USD/CHF

EUR/CHF

EUR/GBP

EUR/JPY

USD/AUD

USD/CAD

USD/JPY

USD/EUR

0.1043*** (0.0208) 0.0355*** (0.0076) 0.0205 (0.0537) 0.0617 (0.0331) 0.0083 (0.0259) 0.0381 (0.0730) 0.0661 (0.0463) 0.0556 (0.0802) 0.0002

0.0664** (0.0205) 0.0538*** (0.0137) 0.0007 (0.0454) 0.0286 (0.0810) 0.0159 (0.0620) 0.1329 (0.0921) 0.0064 (0.0789) 0.0354 (0.0767) 0.0561***

0.0372* (0.0150) 0.0336*** (0.0067) 0.0213 (0.0700) 0.0206 (0.0298) 0.0389 (0.0756) 0.1161 (0.0878) 0.1312 (0.0941) 0.0973 (0.1010) 0.0016

0.0762*** (0.0222) 0.0555*** (0.0101) 0.0343 (0.0970) 0.3061 (0.1976) 0.1899* (0.0891) 0.1059 (0.0801) 0.0672 (0.1247) 0.0014 (0.2624) 0.0343

0.1028*** (0.0220) 0.0192 (0.0112) 0.0883 (0.0639) 0.0781 (0.0413) 0.0325 (0.0918) 0.1112 (0.1414) 0.0507 (0.1262) 0.1770 (0.1182) 0.1312***

0.0830*** (0.0217) 0.0230** (0.0087) 0.0118 (0.0399) 0.1292 (0.1001) 0.0473 (0.0403) 0.0064 (0.0495) 0.0684 (0.1159) 0.0244 (0.0802) 0.0403*

0.0638** (0.0202) 0.0431*** (0.0081) 0.0731 (0.0638) 0.1030 (0.0862) 0.1157 (0.0735) 0.0294 (0.0926) 0.0757 (0.1008) 0.1982 (0.1433) 0.0449

0.0766*** (0.0193) 0.0175* (0.0083) 0.0221 (0.0556) 0.0266 (0.0468) 0.0316 (0.0709) 0.0510 (0.0904) 0.0869 (0.1027) 0.3111*** (0.0669) 0.1085***

0.1161*** (0.0209) 0.0332*** (0.0089) 0.0597 (0.0396) 0.0090 (0.0520) 0.0492 (0.0396) 0.0726 (0.0503) 0.1095 (0.0827) 0.1903 (0.1246) 0.0985***

(0.0157) 0.0043 (0.0262) 0.0111

(0.0140) 0.0518 (0.0658) 0.0373

(0.0325) 0.0049 (0.0204) 0.0231

(0.0266) 0.1799 (0.1249) 0.0098

(0.0383) 0.0401 (0.0333) 0.0552

(0.0162) 0.0984 (0.0838) 0.0137

(0.0242) 0.0241 (0.0440) 0.0128

(0.0222) 0.0545 (0.0317) 0.0078

(0.0145) 0.0497 (0.0498) 0.0055

(0.0155) 0.0003 (0.0206) 0.0127 (0.0257) 0.0194 (0.0167) 0.4291*** (0.0996) 0.0651* (0.0255) 0.0073 (0.1658) 0.1138 (0.1738) – – 0.3920 (0.2514) 0.0241 (0.0901) 0.2109 (0.4585) 0.0443 (0.0266) 0.1811

(0.0280) 0.0341 (0.0454) 0.0313 (0.0388) 0.0504 (0.0195) 0.1839*** (0.0649) 0.0227 (0.0244) 0.0101 (0.0756) 0.1178 (0.1716) – – 0.0133 (0.1152) 0.0332 (0.0843) 0.0243 (0.2087) 0.0311 (0.0278) 0.1850

(0.0353) 0.0340 (0.0243) 0.0834 (0.0579) 0.0117 (0.0175) 0.3242*** (0.0494) 0.0118 (0.0225) 0.0033 (0.1474) 0.0196 (0.0916) – – 0.1439 (0.1230) 0.0354 (0.0964) 0.3402* (0.1350) 0.0060 (0.0349) 0.1499**

(0.0517) 0.0170 (0.0240) 0.0232 (0.0634) 0.0315 (0.0382) 0.3242*** (0.0802) 0.0101 (0.0198) 0.3461 (0.2059) 0.3853 (0.3109) – – 0.2618* (0.1095) 0.0183 (0.1313) 0.7948 (0.4381) 0.0688* (0.0334) 0.1741

(0.0376) 0.0740 (0.0439) 0.0548 (0.0777) 0.0073 (0.0165) 0.3352*** (0.0808) 0.0611 (0.0323) 0.4206 (0.3912) 0.3852 (0.3801) – – 0.1057 (0.1506) 0.1549 (0.2564) 0.2851 (0.1810) 0.1783*** (0.0526) 0.4893

(0.0119) 0.0331* (0.0158) 0.0960 (0.0767) 0.0094 (0.0122) 0.2804*** (0.0302) 0.0804* (0.0320) 0.6684 (0.3810) 0.4201*** (0.1256) – – 0.1462* (0.0731) 1.0690*** (0.2266) 0.4119 (0.2583) 0.0792 (0.0443) 0.0830

(0.0330) 0.0330 (0.0271) 0.0523 (0.0606) 0.0401 (0.0231) 0.2587*** (0.0950) 0.0148 (0.0244) 0.3513 (0.2109) 0.5366*** (0.1413) – – 0.4085 (0.3707) 0.1197 (0.1292) 0.8358 (0.5023) 0.0280 (0.0299) 0.3718***

(0.0339) 0.0589** (0.0214) 0.0284 (0.0464) 0.0356** (0.0133) 0.2180*** (0.0621) 0.0321* (0.0163) 0.4282** (0.1495) 0.3865 (0.3082) – – 0.0274 (0.1010) 0.0398 (0.3070) 0.5354** (0.2024) 0.1920*** (0.0261) 0.2289

(0.0254) 0.0565** (0.0195) 0.0604 (0.0551) 0.0295 (0.0231) 0.3719*** (0.0687) 0.0263 (0.0183) 0.1284 (0.1904) 0.4041 (0.4114) – – 0.0877 (0.0835) 0.0862 (0.0919) 0.2262 (0.1801) 0.1115*** (0.0224) 0.6761*

(0.1303) 0.7449 (0.0486) 0.5436*** (0.0811) 0.1334** (0.0460) 0.1650 (0.1463) 0.08211*** (0.0273) 0.0328 (0.0191) 0.0860 (0.0923) 0.0561 (0.0663) 0.0059 (0.0585)

(0.1132) 0.2665*** (0.0630) 0.4966*** (0.0766) 0.0712 (0.0415) 0.0312 (0.0698) 0.1167*** (0.0398) 0.0141 (0.0304) 0.0405 (0.1267) 0.0137 (0.1100) 0.1026 (0.1247)

(0.0491) 0.0997* (0.0415) 0.2145*** (0.0373) 0.1040 (0.0584) 0.0806* (0.0348) 0.1171*** (0.0395) 0.0363 (0.0271) 0.0581 (0.1044) 0.0729 (0.1199) 0.0495 (0.0894)

(0.1733) 0.0132 (0.0527) 0.4111*** (0.0352) 0.0609 (0.0643) 0.2124* (0.0982) 0.1346*** (0.0498) 0.0920* (0.0364) 0.4887** (0.1724) 0.2625 (0.2169) 0.0982 (0.1117)

(0.2769) 0.2623*** (0.0639) 0.0668 (0.0567) 0.2009 (0.1888) 0.0628 (0.0554) 0.1623*** (0.0413) 0.0616* (0.0250) 0.2145 (0.1353) 0.2482 (0.3251) 0.0588 (0.1010)

(0.1165) 0.2920*** (0.0422) 0.0209 (0.0474) 0.6212*** (0.1778) 0.0336 (0.0866) 0.0302 (0.0295) 0.0023 (0.0195) 0.2665** (0.0866) 0.0016 (0.1220) 0.0603 (0.0675)

(0.0746) 0.8394*** (0.1993) 0.2251 (0.1364) 0.2510*** (0.0734) 0.1317 (0.1152) 0.0536* (0.0281) 0.0005 (0.0151) 0.2417 (0.1489) 0.1716 (0.1494) 0.0379 (0.0828)

(0.1344) 0.4387*** (0.0449) 0.0665 (0.0298) 0.2134 (0.1776) 0.0648* (0.0306) 0.1379*** (0.0364) 0.0358 (0.0204) 0.0438 (0.0946) 0.1743 (0.1494) 0.1417 (0.1187)

(0.2903) 0.1811*** (0.0455) 0.0552 (0.0900) 0.1273* (0.0629) 0.0871 (0.0439) 0.1994*** (0.0433) 0.0516 (0.0266) 0.0927 (0.1119) 0.5044 (0.3176) 0.1392 (0.0851)

185

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192 Table 7 (continued)

d2;1;i d2;2;i d2;3;i

c2;1;i c2;2;i c2;3;i k2;1;i k2;2;i k2;3;i Adj. R2

USD/GBP

USD/CHF

EUR/CHF

EUR/GBP

EUR/JPY

USD/AUD

USD/CAD

USD/JPY

USD/EUR

0.0859 (0.1395) 0.0284 (0.0616) 0.2710 (0.1833) 0.1000*

0.2177 (0.1202) 0.1611 (0.0966) 0.0755 (0.1787) 0.0848

0.1409 (0.2334) 0.0335 (0.0998) 0.3868* (0.1929) 0.0030

0.0907 (0.1700) 0.1246 (0.1621) 0.7003 (0.4515) 0.2476*

0.1905 (0.3038) 0.0338 (0.1817) 0.2962 (0.3439) 0.2354***

0.0032 (0.2089) 0.1695 (0.1340) 0.1797 (0.1394) 0.1191***

0.0246 (0.1251) 0.2387 (0.1086) 0.0749 (0.2139) 0.0393

0.1752 (0.1851) 0.3525 (0.1220) 0.3878* (0.1952) 0.0015

0.0472 (0.1561) 0.0126 (0.2228) 0.1473 (0.3190) 0.0035

(0.0398) 0.1441 (0.0875) 0.1247*

(0.0508) 0.1648 (0.1150) 0.3198

(0.0529) 0.0884 (0.1179) 0.0192

(0.1039) 0.3087 (0.1752) 0.0398

(0.0703) 0.0824 (0.3387) 0.0936

(0.0319) 0.0621 (0.1048) 0.0834

(0.0528) 0.2206* (0.1092) 0.0727

(0.0432) 0.0381 (0.1367) 0.1752

(0.0769) 0.0210 (0.2752) 0.1735*

(0.0504) 0.0589 (0.1375) 0.0489 (0.0318) 0.0176 (0.1373)

(0.1657) 0.1384 (0.1005) 0.0385 (0.0488) 0.2775 (0.1512)

(0.0594) 0.1344 (0.1356) 0.0025 (0.0606) 0.1252 (0.1098)

(0.0719) 0.5010*** (0.1440) 0.1004 (0.0784) 0.5835 (0.4119)

(0.0580) 0.4518 (0.3440) 0.0490 (0.0966) 0.0137 (0.1540)

(0.0451) 0.1205 (0.2026) 0.1253 (0.0799) 0.0245 (0.0618)

(0.0435) 0.0080 (0.0879) 0.1420* (0.0648) 0.1015 (0.0616)

(0.1059) 0.4343* (0.1764) 0.0831 (0.0622) 0.0385 (0.1023)

(0.0799) 0.0345 (0.1070) 0.0831 (0.0984) 0.2019 (0.2140)

0.769

0.744

0.845

0.628

0.696

0.727

0.731

PJ

P3

0.800 P3

0.782 P3

þ  þ EC This table reports the estimation results of the following model: LEC h LEC þ b0;i LEC i;t ¼ a0;i þ M;t þ j¼1 /0;j;i Sj;t þ j¼1 d0;j;i Sj;t þ j¼1 c0;j;i Sj;t LM;t þ j¼1 j;i i;tj P3 P3 P3 P3 P3 P3 P3  EC þ  þ EC  EC þ  EC EC k S L þ I ð a þ b L þ / S þ d S þ c S L þ k S L Þ þ I ð a þ b L þ / S þ d crisis1;t 1;i crisis2;t 2;i 1;i M;t 2;i M;t j¼1 0;j;i j;t M;t j¼1 1;j;i j;t j¼1 1;j;i j;t j¼1 1;j;i j;t M;t j¼1 1;j;i j;t M;t j¼1 2;j;i j;t j¼1 2;j;i Sj;t þ P3 P3 þ EC  EC EC c S L þ k S L Þ þ e , where L is the daily liquidity measure by minus the effective cost for each FX rate on day t; I ¼ 1 if t is during 2;j;i i;t crisis1;t j;t M;t i;t j¼1 2;j;i j;t M;t j¼1

the subprime mortgage financial crisis period; Icrisis2;t ¼ 1 if t is during the Eurozone sovereign debt crisis period, and 0 otherwise; and b0;i indicates  individual liquidity sensitivity to systematic liquidity during days without an announcement. Sþ j;t and Sj;t are positive and negative news surprises,  respectively; Sþ j;t ¼ Sj;t if the surprise denotes good news for indicator j, and 0 otherwise; Sj;t ¼ Sj;t if the surprise denotes bad news for indicator j, and 0

otherwise. Let j = 1, 2, and 3 denote CPI, GDP, and unemployment rate announcements, respectively. The symbol of ‘‘–” indicates that there is no observation value during this period. The optimal length of lagged LEC i;tj included is determined by the Akaike information criterion (AIC) and varies for each currencypair. The estimated coefficients of lagged LEC i;tj are not reported in the table for simplicity. The heteroscedasticity- and autocorrelation-consistent standard errors (Newey & West, 1987) are reported in parentheses. The data cover the period from January 7, 2008, through December 31, 2013. ***, **, and * denote the significance at the 1%, 5%, and 10% levels, respectively.

4.3. Liquidity commonality around quantitative easing announcements Noting the strong variation in liquidity commonality during crisis periods, we explore the impact of quantitative easing (QE) announcements on liquidity commonality over time. According to Brunnermeier and Pedersen (2009) identify the connection between market liquidity and funding constraint: when funding liquidity is tight, traders become risk averse and shift their portfolios away from high risk assets toward insurance assets, thus leading to diminished market liquidity and higher volatility. In certain conditions, lower expected liquidity increases the risk of financing a trade across assets, thus increasing co-movements in liquidity among markets. Melvin and Taylor (2009) document the timeline of 2007–2008 crisis in FX markets, and address the crucial link between the FX market liquidity and stock market liquidity through the carry trade activity. Banti (2016) studies the illiquidity channel linking stock and FX markets during the 2000–2011 dot-come crisis, 2007–2009 financial crisis, and 2010–2014 European sovereign debt crisis, by focusing on illiquidity spirals. Banti (2016) show that currencies’ role in the carry trade affects the illiquidity linkages between stock and FX markets, as suggested by the finding of greater illiquidity linkages between stocks and investment currencies during the 2007–2009 financial crisis. Similarly, Brunnermeier et al. (2009) find that investment currencies are sensitive to crash risk, that is, the risk of a sudden unwinding of carry trade positions by funding-constrained traders. During the financial crisis, the Fed adopted QE policies to inject high capital inflows into the U.S. economy and improve the liquidity in asset markets. Because of the links between funding and market liquidity, we seek to understand whether QE policies can enhance FX liquidity and affect the extent of liquidity commonality in FX markets. Therefore, we follow Mancini et al. (2013) and estimate liquidity betas by adding dummy variables that indicate an announcement of a certain monetary policy. Thus we can explore the impact of the unconventional monetary policy on liquidity commonality. We rewrite the regression model as follows:

LEC i;t ¼ a0;i þ

J X EC EC EC hj;i LEC i;tj þ b0;i LM;t þ IQE;t  Icrisis1;t ða1;i þ b1;i LM;t Þ þ I QE;t  I crisis2;t ða2;i þ b2;i LM;t Þ þ ei;t

ð9Þ

j¼1

where IQE;t ¼ 1 if the QE policy is announced on day t, and 0 otherwise; b0 measures the extent of liquidity commonality during the normal period without a QE policy announcement; and Icrisis1;t ¼ 1 and Icrisis2;t ¼ 1 refer to days during the subprime

186

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Table 8 Effect of quantitative easing policy announcements on liquidity commonality during crisis and non-crisis periods.

a0,i b0,i

a1,i b1,i

a2,i b2,i H0: b1,i = b2,i Adj:R2

USD/GBP

USD/CHF

EUR/CHF

EUR/GBP

EUR/JPY

USD/AUD

USD/CAD

USD/JPY

USD/EUR

0.0036 (0.0112) 0.0619*** (0.0102) 0.3952 (0.8064) 0.1116 (0.0988) 0.0447 (0.0462) 0.0494 (0.0838) 0.6314 0.765

0.0006 (0.0106) 0.0567*** (0.0106) 0.3446* (0.1720) 0.0608 (0.0423) 0.2379 (0.1343) 0.0356 (0.0866) 0.3132 0.777

0.0022 (0.0088) 0.0349*** (0.0085) 0.0251 (0.4601) 0.0224 (0.0539) 0.0860 (0.0696) 0.0193 (0.0675) 0.9711 0.852

0.0042 (0.0137) 0.0741*** (0.0095) 0.1717 (0.6847) 0.0224 (0.0753) 0.0364 (0.0540) 0.2976* (0.1290) 0.0434* 0.620

0.0031 (0.0127) 0.0444*** (0.0121) 0.5588*** (0.1086) 0.0516** (0.0168) 0.1034 (0.1395) 0.0275 (0.1217) 0.5142 0.679

0.0048 (0.0098) 0.0384*** (0.0079) 0.6389*** (0.1207) 0.0032 (0.0144) 0.1254*** (0.0368) 0.2444*** (0.0532) 0.0000*** 0.772

0.0055 (0.0099) 0.0417*** (0.0084) 0.2760 (0.2386) 0.1568*** (0.0285) 0.1503 (0.1045) 0.1425 (0.1154) 0.0115* 0.792

0.0017 (0.0101) 0.0327*** (0.0068) 0.5155 (0.4191) 0.0246 (0.0474) 0.0277 (0.1254) 0.4115 (0.3162) 0.2263 0.796

0.0006 (0.0105) 0.0467*** (0.0087) 0.2127 (0.1423) 0.0350 (0.0232) 0.0145 (0.0876) 0.0386 (0.0524) 0.9482 0.785

PJ EC h LEC þ b0;i LEC This table reports the estimation results of the following model: LEC i;t ¼ a0;i þ M;t þ IQE;t  Icrisis1;t ða1;i þ b1;i LM;t Þ þ IQE;t  j¼1 j;i i;tj EC Icrisis2;t ða2;i þ b3;i LEC Þ þ e , where L is the daily liquidity measure by minus the effective cost for currency-pair i on day t; I ¼ 1 if a quantitative easing QE;t i;t M;t i;t policy announcement is announced during day t, and 0 otherwise; Icrisis1;t ¼ 1 if t is during the subprime mortgage financial crisis period, and Icrisis2;t ¼ 1 if t is during the Eurozone sovereign debt crisis period, and 0 otherwise; b0;j indicates individual liquidity sensitivity to FX systematic liquidity during days without a quantitative easing policy announcement; and b0;i þ b1;i and b0;i þ b2;i indicate the change in the liquidity commonality related to a quantitative easing policy announcement during the subprime mortgage financial crisis period and the Eurozone sovereign debt crisis period, respectively. Each currency owns different optimal lag-order variables. The optimal length of lagged LEC i;tj included is determined by the Akaike information criterion (AIC) and varies for each currency-pair. The estimated coefficients of lagged LEC i;tj are not reported in the table for simplicity. The heteroscedasticity- and autocorrelation-consistent (Newey & West, 1987) standard errors are reported in parentheses. The data cover the period from January 7, 2008, through December 31, 2013. ***, **, and * denote the significance at the 1%, 5%, and 10% levels, respectively.

mortgage or Eurozone sovereign debt crisis period, respectively, and 0 otherwise. The estimated coefficients of b0;i þ b1;i and b0;i þ b2;i represent the impact of the QE policy announcement on liquidity commonality during these two crisis periods, respectively. Table 8 contains the results regarding the responses of liquidity commonality to QE announcements. For most markets, the coefficient for the dummy variable of the QE announcement is negative, with the exceptions of USD/CAD, EUR/JPY, and USD/CHF. That is, a liquidity injection from a central bank can alleviate liquidity strains in other investment currencies and moderate both the sudden appreciation of funding currencies and the depreciation of investment currencies. According to Mancini et al. (2013), during the unwinding of carry trades, market-wide FX liquidity decreases enhance the selling pressure on investment currencies, which exacerbates currency crashes. Therefore, the QE announcement improves investors’ funding restrictions and stimulates more trading activity. The unconventional monetary policy also had international spillover effects that led to depreciation of the dollar (Fawley and Neely, 2013). This depreciation, resulting from the Fed’s QE policy, might lead risk-averse traders to shift their portfolio allocations away from dollar-denominated assets toward more foreign currency–denominated assets. Investors would demand a higher risk premium to compensate for their losses from holding USD, which appears more likely to depreciate rather than to appreciate sharply; that is, demand for USD decreases, while demand for non-dollar currencies increases. As a result, when the announcement of expansionary monetary policy causes an increase in market liquidity, liquidity commonality diminishes. Consistent with Banti (2016), due to a huge amount of ‘‘official” liquidity created by the policy responses to the subprime mortgage financial crisis and the Eurozone sovereign debt crisis, the QE policy temporarily relaxes leveraged traders’ funding constraints. Our evidence suggests that official liquidity may relax the pressures on asset value, reduce losses on carry trades, and improve currency liquidity, particularly for investment currencies. Therefore, investment currencies have a significantly lower commonality in liquidity than funding currencies. The significant and negative effect of QE announcements on FX liquidity commonality suggests that the funding liquidity channel has an important effect on market liquidity. 4.4. Determinants of FX global-wide systemic liquidity After detecting this stronger liquidity commonality during recent financial crisis periods, we investigate potential determinants that might affect the systemic liquidity, that is, ~ LM;t . We begin by estimating the time-series regression model of systematic liquidity as follows:

~LM;t ¼ a0 þ

4 1 X 3 X X other /j ~LM;tj þ bother f t1 þ cþ0;j;k Sþj;tk j¼1

þ

1 X 3 X k¼1 j¼1

k¼1 j¼1   0;j;k Sj;tk

c

þ et

ð10Þ

Table 9 Determinants of Systematic Liquidity in FX Markets.

Model (2) Model (3) Model (4)

Model (4)

Model (2) Model (3) Model (4)

ðþÞ

ðþÞ

ðþÞ

ðÞ

ðÞ

ðÞ

S1;t

S2;t

S3;t

S1;t

S2;t

S3;t

S1;t1

S1;t2

S1;t3

0.4203*** (0.1090) 0.4160*** (0.1086) 0.4090*** (0.1085) 0.7732*** (0.0940)

0.0061* (0.0027) 0.0061* (0.0027) 0.0060* (0.0027) 0.0066** (0.0024)

0.0575* (0.0263) 0.0582* (0.0264) 0.0566* (0.0262) 0.0474 (0.0246)

0.1050 (0.3805) 0.1559 (0.3935) 0.0733 (0.3834)

0.1438 (0.2497) 0.1064 (0.2471) 0.0939 (0.2437)

0.3437 (0.6438) 0.3380 (0.6281) 0.3540 (0.5683)

0.2708 (0.4057) 0.3079 (0.4039) 0.2274 (0.3236)

0.4543* (0.2528) 0.5655* (0.2494) 0.2192 (0.3233)

0.5281* (0.2845) 0.5234 (0.2850) 0.6384* (0.2793)

0.5559** (0.1989) 0.8114* (0.3570)

0.2941 (0.3317) 0.1151 (0.3119)

0.0689 (0.7498) 0.0457 (0.6878)

ðþÞ

ðþÞ

S1;t1

S2;t1

S3;t1

1.6867* (0.8335) 1.3405 (0.7910)

0.1071 (0.3790) 0.0273 (0.3994)

0.6805* (0.2787) 0.5141* (0.2534) AR(2)

AR(1) Model (1)

ðÞ

VIX t1

ðþÞ

Model (1) Model (2) Model (3)

ðÞ

TEDt1

0.4927*** (0.0687) 0.4924*** (0.0690) 0.4936*** (0.0694) 0.4452*** (0.0618)

0.0741 (0.0508) 0.0741 (0.0511) 0.0749 (0.0504) 0.0539 (0.0450)

ðÞ

ðÞ

ðÞ

ðþÞ

ðþÞ

ðþÞ

S1;tþ1

S2;tþ1

S3;tþ1

S1;tþ1

S2;tþ1

S3;tþ1

0.0874 (0.5832)

0.1299 (0.5596)

1.7748 (0.9921) AR(4)

0.3727 (0.6306)

0.0558 (0.2642)

0.3052 (0.2074)

AR(3) 0.0399 (0.0478) 0.0413 (0.0482) 0.0419 (0.0481) 0.0260 (0.0467) P1 P3

0.1125** (0.0372) 0.1120** (0.0373) 0.1128** (0.0369) 0.0844* (0.0353)

a

Adj:R2

6.6251*** (1.2983) 6.6172*** (1.2961) 6.5032*** (1.3002) 10.5981*** (1.2432)

0.862 0.861 0.862 0.875

P P ðÞ ðþÞ other This table reports the estimation results of the following model: ~LM;t ¼ a0 þ b0 D~ LM;t1 þ hother f t1 þ k¼1 j¼1 c0;j Sj;tk þ 1k¼1 3j¼1 /0;j Sj;tk þ et , where ~ LM;t is the market-wide FX liquidity index, and the AR(4) 0 other structure is determined according to the Akaike information criterion (AIC); and f t1 is the set of lagged control variables, including global FX volatility (VXY), Chicago Board Options Exchange Volatility Index þ þ (VIX), and TED spread (TED). Sj;tk and Sj;tk are positive and negative news surprises, respectively, where j = 1, 2, and 3 denotes CPI, GDP, and unemployment rate announcements, respectively. The index t  k (for k ¼ 1; 0; 1) indicates the overall observed window for news effect: a pre-announcement period (k ¼ 1), the news announcement (k ¼ 0), and a post-announcement period (k ¼ 1). The heteroscedasticity- and autocorrelation-consistent (Newey & West, 1987) standard errors are reported in parentheses. The data cover the period from January 7, 2008, through December 31, 2013. ***, **, and * denote the significance at the 1%, 5%, and 10% levels, respectively.

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Model (1)

ðÞ

VXY t1

187

188

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

where ~LM;t is the market-wide FX liquidity index based on minus the effective cost, and the AR(4) structure is determined other

according to the Akaike information criterion (AIC); f t1 is the set of lagged control variables, including global FX volatility þ (VXY),12 the Chicago Board Options Exchange Volatility Index (VIX), and TED spread (TED). Again, Sþ j;tk and Sj;tk are positive and negative news surprises, respectively, where j = 1, 2, and 3 denotes CPI, GDP, and unemployment rate announcements, respectively. The index t  k (for k ¼ 1; 0; 1) indicates the overall observed window for news effect: a pre-announcement period (k ¼ 1), the news announcement (k ¼ 0), and a post-announcement period (k ¼ 1). Table 9 displays the ordinary least square estimation results with robust standard errors (Newey & West, 1987). In Model (1), we regress systemic illiquidity on lagged VIX, lagged VXY, lagged TED, and news surprises variables. The result of significantly negative impact of VIX on the aggregate market-wide liquidity corroborate Mancini et al. (2013) that predict liquidity is negatively related to investors’ fear through carry trade activity that links stock and FX market illiquidity. An increase in VIX on day t  1 would yield an increase in FX systemic illiquidity on day t. We also find a significantly negative relationship between market-wide liquidity and TED. Since TED spread is a proxy for the level of credit risk and funding liquidity in the interbank market (Brunnermeier et al., 2009), this result implies a spillover effect between the funding liquidity and the market-wide FX liquidity. When traders face a tight funding situation, it might propagate to other assets, leading systemic FX liquidity to dry up. Consistent with Kyle and Xiong (2001), we find that if financial intermediaries providing liquidity in two markets endure trading losses in one market, they may reduce liquidity provision in both markets (Table 10). Furthermore, VXY is a proxy for perceived FX market volatility. Its estimated coefficient suggests a negatively significant association with the market-wide FX liquidity. This result is in line with insights by Stoll (1978) and Fleming and Remolona (1999), supporting that an increase in volatility leads to a wider bid–ask spread and lower liquidity. Dealers usually widen the spread or withdraw their quotes in response to the inventory risks that result from a sharp shift in price volatility. Moreover, in Models (2)–(4) of Table 9, we include the absolute value of positive and negative news surprises, to study potentially asymmetric reactions of systemic liquidity to news shocks. Here, FX systemic liquidity is more sensitive to economic fundamentals when positive news shocks occur, a finding that contests Riordan, Storkenmaier, Wagener, and Sarah Zhang’s (2013) results, in which liquidity increases with news that is associated with positive or neutral sentiment, but news with negative sentiment is associated with decreased liquidity. In Model (2), we observe that two of the three U.S. macroeconomic announcements (GDP and unemployment rate) have negative, significant impacts on global systemic liquidity. The lagged news surprise effect in Model (3) indicates that the coefficient of lagged unemployment rate has a positive, significant impact on market-wide systemic liquidity. In Model (4), we further observe that if the transaction occurs on the day before or of U.S. news announcements, market participants are reluctant to trade, seemingly due to the high price volatility. The result may be a decrease in systemic liquidity, particularly for negative news surprises. However, on the day after U.S. news announcements, the lower market uncertainty is accompanied by lower volatility, likely driving market participants to trade and indirectly raise market liquidity. Thus, the change in systematic liquidity may be caused by U.S news surprises. As Fleming and Remolona (1999) show, dealers react to news releases by offering a new quote or withdrawing quotes in response to inventory risks or sharp price changes during the news announcement period. As a result, spread increases evidently are driven by inventory risk. Finally, we consider whether this result might vary during financial crisis periods, such that we incorporate the crisis dummy variable into the regression model:

~LM;t ¼ a0 þ

4 X

1 X 3 X

/j ~LM;tj þ hother f t1 þ 0

cþ0;j;k Sþj;tk

other

j¼1 1 X 3 X

c

  0;j;k Sj;tk

k¼1 j¼1

þ Icrisis1;t a1 þ

other hother f t1 1

k¼1 j¼1

þ

1 X 3 X

þ þ 1;j;k Sj;tk

c

k¼1 j¼1 other

f t1 þ þ Icrisis2;t a2 þ hother 2

!   1;j;k Sj;tk

c

k¼1 j¼1

1 X 3 X

1 X 3 X

k¼1 j¼1

k¼1 j¼1

cþ2;j;k Sþj;tk þ

þ

1 X 3 X

!

c2;j;k Sj;tk þ et

ð11Þ

where Icrisis1;t ¼ 1 if day t is in the subprime mortgage financial crisis period, and 0 otherwise; and Icrisis2;t ¼ 1 if day t is during the period of Eurozone sovereign debt crisis, and 0 otherwise. Compared with non-crisis periods, the sensitivity of the control variables during a financial crisis differs; they have no significant influence on FX systematic liquidity, except for news surprises and VXY. The coefficients of the lagged VIX and TED are not significantly different from 0 in times of crisis. This finding is inconsistent with theory that suggests liquidity spiral effects are stronger during crisis periods (Brunnermeier et al., 2009). Still, news surprises can affect systemic liquidity in FX markets during periods of financial crisis; GDP announcements in particular have significant effects on the market-wide FX liquidity. The impact of positive news surprises on the market-wide

12 The VXY denotes the JP Morgan Implied Volatility Index for a basket of G7 currencies. It launched in December 2006, by JP Morgan, with intraday updates reported on Bloomberg. Daily data from Datastream are available from April 12, 2007. The VXY index follows aggregate volatility in currencies using a turnoverweighted index of G7. The weightings of different currencies in the indices are based on option turnover taken from the BIS’s Triennial Central Bank Survey of FX and derivatives markets.

189

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192 Table 10 Determinants of systematic liquidity in FX markets during crisis periods.

Model (1) Model (2) Model (3) Model (4)

Model (1) Model (2) Model (3) Model (4)

Icrisis1;t

Icrisis2;t

VXY t1

TEDt1

VIY t1

Ic1;t VXY t1

Ic2;t VXY t1

Ic1;t TEDt1

Ic2;t TEDt1

8.3327** (2.5513) 8.4372** (2.5765) 8.4743** (2.6531) 8.1797** (2.5008)

6.7773*** (1.7714) 6.8302*** (1.7828) 6.7944*** (1.7918) 6.1817*** (1.6753)

0.2835* (0.1243) 0.2800* (0.1237) 0.2657* (0.1236) 0.3192** (0.1142)

0.0024 (0.0025) 0.0022 (0.0025) 0.0025 (0.0024) 0.0017 (0.0023)

0.0453 (0.0314) 0.0445 (0.0315) 0.0468 (0.0314) 0.0327 (0.0273)

0.6437* (0.2678) 0.6450* (0.2716) 0.6358* (0.2774) 0.5814* (0.2532)

0.6202** (0.1912) 0.6299** (0.1921) 0.6413*** (0.1905) 0.5516** (0.1787)

0.0141 (0.0072) 0.0145* (0.0073) 0.0146* (0.0073) 0.0164* (0.0073)

0.0171 (0.0135) 0.0130 (0.0135) 0.0142 (0.0135) 0.0181 (0.0131)

Ic1;t VIX t1

Ic2;t VIX t1

S1;t

Icrisis1;t S1;t

Icrisis2;t S1;t

S2;t

Icrisis1t S2;t

Icrisis2;t S2;t

S3;t

0.0185 (0.0776) 0.0172 (0.0784) 0.0168 (0.0801) 0.0091 (0.0776)

0.0365 (0.0674) 0.0257 (0.0674) 0.0221 (0.0681) 0.0409 (0.0650)

0.1773 (0.2653) 0.2029 (0.2729) 0.2132 (0.2409)

0.6219 (1.9852) 1.5009 (2.7155) 1.4358 (2.7435)

2.2054 (1.9207) 2.1115 (1.9323) 2.2012 (1.9722)

0.0863 (0.3480) 0.1809 (0.3516) 0.2228 (0.3535)

0.1822 (0.4520) 0.1040 (0.4617) 0.0607 (0.4785)

0.9405 (0.5132) 1.0653* (0.5132) 1.0181* (0.5063)

0.0341 (0.2660) 0.0451 (0.2623) 0.0992 (0.3478)

ðÞ

Model (1) Model (2) Model (3) Model (4)

Model (3) Model (4)

Model (4)

Model (4)

(1) (2) (3) (4)

ðþÞ

ðÞ

ðþÞ

S2;t

Icrisis1;t S2;t

Icrisis2;t S2;t

S3;t

0.0211 (1.4847) 0.1312 (1.4321) 0.8803 (1.2428)

1.1902 (1.3598) 1.3023 (1.3053) 1.4505 (1.3178)

0.1466 (0.6021) 0.0914 (0.5935) 0.0845 (0.6003)

0.1456 (0.8423) 0.0841 (0.8589) 0.0967 (0.8869)

1.2946 (0.8934) 1.1819 (0.9039) 1.2612 (0.9233)

0.5169* (0.2490) 0.6751* (0.2623) 0.6431* (0.2561)

0.7063 (0.8575) 0.8589 (0.8583) 0.8295 (0.8495)

0.1933 (0.5017) 0.0221 (0.5103) 0.0612 (0.5006)

0.4390 (0.3975) 0.4479 (0.4034) 0.3840 (0.4021)

ðþÞ

ðÞ

ðÞ

ðÞ

ðÞ

ðÞ

ðÞ

ðÞ

Icrisis1;t S3;t

Icrisis2;t S3;t

S1;t1

Icrisis1;t S1;t1

Icrisis2;t S1;t1

S2;t1

Icrisis1;t S2;t1

Icrisis2;t S2;t1

S3;t1

11.4433*** (1.7920) 11.4018*** (1.8577) 11.4831*** (1.8599)

0.3677 (0.4205) 0.3385 (0.4241) 0.4317 (0.4239)

0.5255*** (0.1514) 0.5566** (0.1801)

0. (0.7436) 2.8149 (1.5424)

2.1328* (0.8805) 2.2124* (0.8718)

0.3085 (0.4233) 0.2574 (0.4596)

0.4475 (0.8315) 0.4685 (0.8336)

0.8068 (0.5403) 0.6908 (0.5600)

0.1976 (0.4669) 0.1756 (0.4730)

ðÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

Icrisis1;t S3;t1

Icrisis2;t S3;t1

S1;t1

Icrisis1;t S1;t1

Icrisis2;t S1;t1

S2;t1

Icrisis1;t S2;t1

Icrisis2;t S2;t1

S3;t1

1.3524 (1.8615) 1.3385 (1.8888)

0.8458 (0.5643) 0.7884 (0.5621)

2.3123 (1.6952) 2.3567 (1.6968)

1.4741 (1.8708) 1.8906 (1.7495)

2.2001 (1.8775) 2.1338 (1.8756)

0.1195 (0.1857) 0.1193 (0.1867)

0.2921 (1.5237) 0.1046 (1.2522)

2.0651** (0.6375) 1.9996** (0.6418)

0.2056 (0.3089) 0.2480 (0.2639)

ðþÞ

ðÞ

ðÞ

ðÞ

ðÞ

ðÞ

ðÞ

ðÞ

Icrisis1;t S3;t1

Icrisis2;t S3;t1

S1;tþ1

Icrisis1;t S1;tþ1

Icrisis2;t S1;tþ1

S2;tþ1

Icrisis1;t S2;tþ1

Icrisis2;t S2;tþ1

S3;tþ1

18.8667*** (1.4616) 19.1638*** (1.5074)

0.6601 (0.5204) 0.5944 (0.4892)

0.9443 (0.7499)

6.3728*** (1.4076)

1.9124 (1.1686)

0.3622 (0.2994)

0.6489 (0.4687)

2.2496 (1.6800)

0.3069 (0.3098)

ðÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

ðþÞ

Icrisis1;t S3;tþ1

Icrisis2;t S3;tþ1

S1;tþ1

Icrisis1;t S1;tþ1

Icrisis2;t S1;tþ1

S2;tþ1

Icrisis1;t S2;tþ1

Icrisis2;t S2;tþ1

S3;tþ1

7.0466 (2.8509)

1.4931 (1.2946)

0.3856 (0.3780)

1.9505 (1.1695)

0.1990 (0.7502)

0.0872 (0.2674)

0.6300 (0.6395)

1.0151* (0.4771)

0.5366* (0.2148)

AR(1)

AR(2)

AR(3)

AR(4)

a

Adj:R2

0.4476*** (0.0656) 0.4489*** (0.0671)

0.0534 (0.0474) 0.0486 (0.0486)

0.0283 (0.0466) 0.0404 (0.0471)

0.1016** (0.0362) 0.0974** (0.0364)

5.3261*** (1.3028) 5.2741*** (1.3090)

0.869

ðþÞ

Model (2)

ðþÞ

ðÞ

Icrisis2;t S1;t

Icrisis1;t S3;tþ1 Model (1)

ðþÞ

ðÞ

Icrisis1;t S1;t

ðÞ

Model Model Model Model

ðþÞ

ðÞ

S1;t

ðþÞ

Model (1) Model (2) Model (3)

ðþÞ

ðÞ

Icrisis2;t S3;t

ðÞ

Model (1) Model (2) Model (3)

ðþÞ

ðÞ

Icrisis1;t S3;t

ðþÞ

Model (1) Model (2)

ðÞ

ðÞ

ðþÞ

Icrisis2;t S3;tþ1

0.869 (continued on next page)

190

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Table 10 (continued) Icrisis1;t

Icrisis2;t

VXY t1

TEDt1

VIY t1

Ic1;t VXY t1

Ic2;t VXY t1

Ic1;t TEDt1

0.0524 (0.0481) 0.0403 (0.0486)

0.0341 (0.0475) 0.0285 (0.0471)

0.1139*** (0.0336) 0.1271*** (0.0335)

5.1735*** (1.3153) 5.3909*** (1.3214)

0.870

0.4228 (0.3303)

0.4455*** (0.0676) 0.4552*** (0.0683)

Model (3) Model (4)

1.7387 (2.0059)

Ic2;t TEDt1

0.874

P P P P P other  This table reports the estimation results of the following model: ~ LM;t ¼ a0 þ 4j¼1 /j ~ LM;tj þ hother f t1 þ 1k¼1 3j¼1 cþ Sþ þ 1k¼1 3j¼1 c 0 0;j;k Sj;tk þ 0;j;k j;tk P1 P1 P1 P1 P3 þ þ P3   P3 þ þ P3   other other other other Icrisis1;t ða1 þ h1 f t1 þ k¼1 j¼1 c1;j;k Sj;tk þ k¼1 j¼1 c1;j;k Sj;tk Þ þ Icrisis2;t ða2 þ h2 f t1 þ k¼1 j¼1 c2;j;k Sj;tk þ k¼1 j¼1 c2;j;k Sj;tk Þ þ et , where other ~ LM;t is the market-wide FX liquidity index, and the AR(4) structure is determined according to the Akaike information criterion (AIC); f t1 is the set of

lagged control variables, including global FX volatility (VXY), the Chicago Board Options Exchange Volatility Index (VIX), and TED spread (TED). Again, Sþ j;tk

and Sþ j;tk are positive and negative news surprises, respectively, where j = 1, 2, and 3 denotes CPI, GDP, and unemployment rate announcements, respectively. The index t  k (for k ¼ 1; 0; 1) indicates the overall observed window for news effect: a pre-announcement period (k ¼ 1), the news announcement (k ¼ 0), and a post-announcement period (k ¼ 1). Icrisis1;t ¼ 1 if day t occurs during the subprime mortgage financial crisis period, and 0 otherwise; and Icrisis2;t ¼ 1 if day t is during the period of Eurozone sovereign debt crisis, and 0 otherwise. The heteroscedasticity- and autocorrelationconsistent (Newey & West, 1987) standard errors are reported in parentheses. The data cover the period from January 7, 2008, through December 31, 2013. ***, **, and * denote the significance at the 1%, 5%, and 10% levels, respectively.

FX liquidity also is more influential than that of negative news surprises. For positive news surprises, on the day after or of the news announcement, lower market uncertainty increases market transactions and then indirectly improves the marketwide FX liquidity. However, we find no conclusive results around negative news announcements. Before the announcements, higher uncertainty prevails in the market, Hutchison and Sushko (2013) find that the downside risk for carry trade activity is higher. To relieve the risk of carry trade, currency investors may consider to unwind their positions, resulting in a lower liquidity in the FX market. On the other hand, the higher level of information asymmetry before the announcement, as addressed in Riordan et al. (2013), would also explain the lower liquidity before the announcement. Moreover, the market-wide FX liquidity during periods of financial crisis becomes even lower before the announcement of macroeconomic news. 5. Conclusions Using high-frequency EBS data over six years and nine major currency pairs, from January 5, 2008, through December 31, 2013, we study FX liquidity in depth. Extending Mancini et al.’s (2013) analysis, we also consider financial crises factor, in the form of the subprime mortgage financial crisis and Eurozone sovereign debt crisis. This study thus advances understanding of the dynamics of FX systemic liquidity in times of crisis. We also apply a generalized factor model, the GDFM, to extract systematic liquidity among FX markets. By using this method, we resolve issues caused by the salient autocorrelation in liquidity. Our four main findings offer new insights. First, we provide evidence of stronger liquidity commonality in times of crisis, indicating stronger comovement between liquidity in individual currency pairs and the aggregate market-wide liquidity among many currency pairs during times of crisis. Moreover, most of investment currencies have a stronger commonality with the market-wide liquidity than do safe-haven currencies during crisis periods. Consistent with Mancini et al. (2013), our results support that investment currencies suffer a greater exposure to the global risk resulting from the unwinding of carry trade position and a dramatic reduction in liquidity, and are linked tighter with the reduced market-wide currency than the other currencies. Second, we show that increased uncertainty or price volatility invoked by news releases may induce liquidity commonality to vary around the announcement. Separating news surprises into positive and negative news, we can better understand variation in FX liquidity commonality around the announcement. Positive news has a stronger impact on the FX liquidity commonality than negative news during crisis periods. Third, the announcement of a QE policy affects FX liquidity commonality. This policy may improve investors’ funding restrictions and stimulate trading activities, so liquidity commonality decreases. This empirical evidence corroborates the influence of funding constraints on liquidity, as addressed by Brunnermeier and Pedersen (2009) and Banti (2016). Although the QE policy might improve overall systemic liquidity and lower the adverse effect of a liquidity spiral on individual liquidity, the weaker comovement between systemic liquidity and individual liquidity in response to a QE announcement also suggests that the recovery speeds for systemic liquidity and the liquidity of an individual currency pair are different. Fourth, we examine factors that affect systemic liquidity over time. Our results show that systemic liquidity significantly depends on either supply-side forces, related to funding liquidity (i.e., TED spread), or increased stock market volatility (i.e., VIX), or increased FX volatility (i.e., VXY), related to currency investors’ risk of leveraged position, or dealers’ information asymmetry costs (i.e., news surprises). These results accordingly are in line with the notion that news releases have important functions in times of financial crisis.

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

191

Acknowledgements The authors gratefully acknowledge the editor and an anonymous referee for helpful comments and suggestions. Ya-Ting Chang gratefully acknowledges support from a PhD Dissertation Fellowship from the Ministry of Science and Technology. Yin-Feng Gau gratefully acknowledges a research grant from the Ministry of Science and Technology (104-2410-H-008014-MY3). References Altavilla, C., & Giannone, D. (2016). The effectiveness of non-standard monetary policy measures: Evidence from survey data. Journal of Applied Econometrics (forthcoming). Amihud, Y., & Mendelson, H. (1980). Dealership market: Market-making with inventory. Journal of Financial Economics, 8(1), 31–53. Andersen, T. G., & Bollerslev, T. (1998). Deutsche mark-dollar volatility: Intraday activity patterns, macroeconomic announcements, and longer run dependencies. Journal of Finance, 53(1), 219–265. Andersen, T. G., Bollerslev, T., Diebold, F. X., & Vega, C. (2003). Micro effects of macro announcements: Real-time price discovery in foreign exchange. American Economic Review, 93(1), 38–62. Andersen, T. G., Bollerslev, T., Diebold, F. X., & Vega, C. (2007). Real-time price discovery in global stock, bond and foreign exchange markets. Journal of International Economics, 73(2), 251–277. Balduzzi, P., Elton, E. J., & Green, T. C. (2001). Economic news and bond prices: Evidence from the US Treasury market. Journal of Financial and Quantitative Analysis, 36(04), 523–543. Bank for International Settlements (2013). Foreign exchange turnover in April 2013: Preliminary global results. Monetary and Economic Department, September. Bank for International Settlements (2016). Foreign exchange turnover in April 2016. Monetary and Economic Department, September. Banti, C. (2016). Illiquidity in the stock and foreign exchange markets: An investigation of their cross-market dynamics. Journal of Financial Research, 39(4), 411–436. Banti, C., Phylaktis, K., & Sarno, L. (2012). Global liquidity risk in the foreign exchange market. Journal of International Money and Finance, 31(2), 267–291. Bauwens, L., Omrane, W. B., & Giot, P. (2005). News announcements, market activity and volatility in the euro/dollar foreign exchange market. Journal of International Money and Finance, 24(7), 1108–1125. Berger, D. W., Chaboud, A. P., Chernenko, S. V., Howorka, E., & Wright, J. H. (2008). Order flow and exchange rate dynamics in electronic brokerage system data. Journal of International Economics, 75(1), 93–109. Brockman, P., Chung, D. Y., & Pérignon, C. (2009). Commonality in liquidity: A global perspective. Journal of Financial and Quantitative Analysis, 44(04), 851–882. Brunnermeier, M. K., Nagel, S., & Pedersen, L. H. (2009). Carry trades and currency crashes. NBER Macroeconomics Annual, 23(1), 313–347. Brunnermeier, M. K., & Pedersen, L. H. (2009). Market liquidity and funding liquidity. Review of Financial Studies, 22(6), 2201–2238. Chaboud, A., Chernenko, S., Wright, J. H. (2007). Trading activity and exchange rates in high-frequency EBS data. FRB International Finance Discussion Paper No. 903. Chordia, T., Roll, R., & Subrahmanyam, A. (2000). Commonality in liquidity. Journal of Financial Economics, 56(1), 3–28. Chordia, T., Sarkar, A., & Subrahmanyam, A. (2005). An empirical analysis of stock and bond market liquidity. Review of Financial Studies, 18(1), 85–129. Corielli, F., & Marcellino, M. (2006). Factor based index tracking. Journal of Banking and Finance, 30(8), 2215–2233. Ehrmann, M., & Fratzscher, M. (2005). Exchange rates and fundamentals: New evidence from real-time data. Journal of International Money and Finance, 24 (2), 317–341. Evans, M. D., & Lyons, R. K. (2002). Time-varying liquidity in foreign exchange. Journal of Monetary Economics, 49(5), 1025–1051. Evans, M. D., & Lyons, R. K. (2005). Do currency markets absorb news quickly? Journal of International Money and Finance, 24(2), 197–217. Evans, M. D., & Lyons, R. K. (2008). How is macro news transmitted to exchange rates? Journal of Financial Economics, 88(1), 26–50. Evans, M. D. (2010). Order flows and the exchange rate disconnect puzzle. Journal of International Economics, 80(1), 58–71. Fawley, B. W., & Neely, C. J. (2013). Four stories of quantitative easing. Review-Federal Reserve Bank of St. Louis, 95(1), 51–88. Federal Reserve Bank of St. Louis (2009). The Financial Crisis: Full Timeline. https://www.stlouisfed.org/financial-crisis/full-timeline. Filardo, A., George, J., Loretan, M., Ma, G., Munro, A., Shim, I., & Zhu, H. (2010). The international financial crisis: Timeline, impact and policy responses in Asia and the Pacific. BIS Papers, 52: 21–82, Bank for International Settlements. Fleming, M. J., & Remolona, E. M. (1999). Price formation and liquidity in the US Treasury market: The response to public information. Journal of Finance, 54 (5), 1901–1915. Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2000). The generalized dynamic-factor model: Identification and estimation. Review of Economics and Statistics, 82(4), 540–554. Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2003). Do financial variables help forecasting inflation and real activity in the euro area? Journal of Monetary Economics, 50(6), 1243–1255. Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2005). The generalized dynamic factor model. Journal of the American Statistical Association, 100(471), 830–840. Galati, G., Heath, A., & Mcguire, P. (2007). Evidence of carry trade activity. BIS Quarterly Review, 3, 27–41. Hallin, M., & Liška, R. (2007). Determining the number of factors in the general dynamic factor model. Journal of the American Statistical Association, 102(478), 603–617. Hallin, M., Mathias, C., Pirotte, H., & Veredas, D. (2011). Market liquidity as dynamic factors. Journal of Econometrics, 163(1), 42–50. Hameed, A., Kang, W., & Viswanathan, S. (2010). Stock market declines and liquidity. Journal of Finance, 65(1), 257–293. Hasbrouck, J., & Seppi, D. J. (2001). Common factors in prices, order flows, and liquidity. Journal of Financial Economics, 59(3), 383–411. Ho, T. S., & Stoll, H. R. (1983). The dynamics of dealer markets under competition. Journal of Finance, 38(4), 1053–1074. Hutchison, M., & Sushko, V. (2013). Impact of macroeconomic surprises on carry trade activity. Journal of Banking and Finance, 37(4), 1133–1147. Ito, T., & Hashimoto, Y. (2006). Intraday seasonality in activities of the foreign exchange markets: Evidence from the electronic broking system. Journal of the Japanese and International Economies, 20(4), 637–664. Kamara, A., Lou, X., & Sadka, R. (2008). The divergence of liquidity commonality in the cross-section of stocks. Journal of Financial Economics, 89(3), 444–466. Karnaukh, N., Ranaldo, A., & Soderlind, P. (2015). Understanding FX liquidity. Review of Financial Studies, 28(11), 3073–3108. Karolyi, G. A., Lee, K.-H., & Van Dijk, M. A. (2012). Understanding commonality in liquidity around the world. Journal of Financial Economics, 105(1), 82–112. Korajczyk, R. A., & Sadka, R. (2008). Pricing the commonality across alternative measures of liquidity. Journal of Financial Economics, 87(1), 45–72. Kyle, A. S., & Xiong, W. (2001). Contagion as a wealth effect. Journal of Finance, 56(4), 1401–1440. Luciani, M., & Veredas, D. (2015). Estimating and forecasting large panels of volatilities with approximate dynamic factor models. Journal of Forecasting, 34 (3), 163–176. Ludvigson, S. C., & Ng, S. (2007). The empirical risk-return relation: A factor analysis approach. Journal of Financial Economics, 83(1), 171–222. Ludvigson, S. C., & Ng, S. (2009). Macro factors in bond risk premia. Review of Financial Studies, 22(12), 5027–5067. Mancini, L., Ranaldo, A., & Wrampelmeyer, J. (2013). Liquidity in the foreign exchange market: Measurement, commonality, and risk premiums. Journal of Finance, 68(5), 1805–1841.

192

Y.-T. Chang et al. / North American Journal of Economics and Finance 42 (2017) 172–192

Melvin, M., & Taylor, M. P. (2009). The crisis in the foreign exchange market. Journal of International Money and Finance, 28(8), 1317–1330. Menkhoff, L., Sarno, L., Schmeling, M., & Schrimpf, A. (2012). Carry trades and global foreign exchange volatility. Journal of Finance, 67(2), 681–718. Newey, W. K., & West, K. D. (1987). Hypothesis testing with efficient method of moments estimation. International Economic Review, 28(3), 777–787. O’Hara, M., & Oldfield, G. S. (1986). The microeconomics of market making. Journal of Financial and Quantitative analysis, 21(4), 361–376. Omrane, Walid Ben, & Hafner, Christian (2015). Macroeconomic news surprises and volatility spillover in foreign exchange markets. Empirical Economics, 48 (2), 577–607. Opschoor, A., Taylor, N., Van der Wel, M., & van Dijk, D. (2014). Order flow and volatility: An empirical investigation. Journal of Empirical Finance, 28, 185–201. Osler, C. L. (2012). Market microstructure and the profitability of currency trading. Annual Review of Financial Economics, 4, 469–495. Pástor, L., & Stambaugh, R. F. (2003). Liquidity risk and expected stock returns. Journal of Political Economy, 111(3), 642–685. Rime, D. (2003). New electronic trading systems in the foreign exchange market. In The New Economy Handbook (pp. 469–504). USA: Elsevier Science. Riordan, R., Storkenmaier, A., Wagener, M., & Sarah Zhang, S. (2013). Public information arrival: Price discovery and liquidity in electronic limit order markets. Journal of Banking and Finance, 37(4), 1148–1159. Rösch, C. G., & Kaserer, C. (2013). Market liquidity in the financial crisis: The role of liquidity commonality and flight to quality. Journal of Banking and Finance, 37(7), 2284–2302. Scholtus, M., van Dijk, D., & Frijns, B. (2014). Speed, algorithmic trading, and market quality around macroeconomic news announcements. Journal of Banking and Finance, 38(1), 89–105. Stock, J. H., Watson, M. W. (1998). Diffusion Indexes. NBER Working Papers No. 6702. Stock, J. H., & Watson, M. W. (2002a). Macroeconomic forecasting using diffusion indexes. Journal of Business and Economic Statistics, 20(2), 147–162. Stock, J. H., & Watson, M. W. (2002b). Forecasting using principal components from a large number of predictors. Journal of the American Statistical Association, 97(460), 1167–1179. Stoll, H. R. (1978). The supply of dealer services in securities markets. Journal of Finance, 33(4), 1133–1151.