Price-setting in the foreign exchange swap market: Evidence from order flow

Syrstad, Olav; Viswanath-Natraj, Ganesh Working Paper Price-setting in the foreign exchange swap market: Evidence from order flow Working Paper, No. 16/2020 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Syrstad, Olav; Viswanath-Natraj, Ganesh (2020) : Price-setting in the foreign exchange swap market: Evidence from order flow, Working Paper, No. 16/2020, ISBN 978-82-8379-170-9, Norges Bank, Oslo, https://hdl.handle.net/11250/2690236 This Version is available at: https://hdl.handle.net/10419/246118 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/deed.no Price-setting in the foreign exchange swap market: Evidence from order flow NORGES BANK RESEARCH 16 | 2020 OLAV SYRSTAD AND GANESH VISWANATH-NATRAJ WORKING PAPER NORGES BANK WORKING PAPER XX | 2014 RAPPORTNAVN 2 Working papers fra Norges Bank, fra 1992/1 til 2009/2 kan bestilles over e-post: [email protected] Fra 1999 og senere er publikasjonene tilgjengelige på www.norges-bank.no Working papers inneholder forskningsarbeider og utredninger som vanligvis ikke har fått sin endelige form. Hensikten er blant annet at forfatteren kan motta kommentarer fra kolleger og andre interesserte. Synspunkter og konklusjoner i arbeidene står for forfatternes regning. Working papers from Norges Bank, from 1992/1 to 2009/2 can be ordered by e-mail: [email protected] Working papers from 1999 onwards are available on www.norges-bank.no Norges Bank’s working papers present research projects and reports (not usually in their final form) and are intended inter alia to enable the author to benefit from the comments of colleagues and other interested parties. Views and conclusions expressed in working papers are the responsibility of the authors alone. ISSN 1502-8190 (online) ISBN 978-82-8379-170-9 (online) Price-setting in the Foreign Exchange Swap Market: Evidence from Order Flow Olav Syrstad and Ganesh Viswanath-Natraj∗ Abstract This paper investigates price discovery in foreign exchange (FX) swaps. Using data on inter-dealer transactions, we find that a 1 standard deviation increase in order flow (i.e. net pressure to obtain USD through FX swaps) increases the cost of dollar funding by up to 4 basis points after the 2008 crisis. This is explained by increased dispersion in dollar funding costs and quarter-end periods. We find central bank swap lines reduced the order flow to obtain USD through FX swaps, subsequently affecting the forward rate. In contrast, during quarter-ends and monetary announcements we observe high frequency adjustment of the forward rate. Keywords: interest rate parity, exchange rates, currency swaps, order flow, dollar funding JEL Classifications: E43, F31, G15 ∗Norges Bank (Ola[email protected]) and Warwick Business School (ganesh.viswanathnatra[email protected]) respectively. This working paper should not be reported as representing the views of Norges Bank . The views expressed are those of the author and do not necessarily reflect those of Norges Bank. We would like to thank Saleem Bahaj, Barry Eichengreen, Gerardo Ferrera, Yuriy Gorodnichenko, Pierre-Olivier Gourinchas, Ingomar Krohn, Richard Lyons, Dagfinn Rime, Andrew Rose, Andreas Schrimpf, Kjetil Stiansen, Ingebjørg Hansen Sævareid, Saskia Ter-Ellen and seminar participants at the Australasian Banking and Finance Conference, the Bank of England, the Bank of International Settlements, BI Business School Oslo, Norges Bank and UC Berkeley. 1 Introduction Foreign Exchange (FX) swaps allow market participants to hedge exchange rate risk arising from currency mismatch between assets and liabilities. After growing steadily over the past decade, FX swaps are now the most traded foreign exchange instrument worldwide, with a daily turnover of approximately $3.2 Trillion USD (2019 BIS triennial survey). In theory, the pricing of FX swaps is pinned down by Covered Interest Parity (CIP) - a renowned noarbitrage relationship in international finance. CIP states that the rate of return on equivalent domestic and foreign assets should equalize after covering exchange rate fluctuations in the FX swap market. However, the FX swap market has been subject to considerable scrutiny since the global financial crisis, as the pricing no longer obeys the iron law of CIP. Since 2008, CIP deviations have been large and persistent, and have implied a systematic premium to swap EUR, CHF and JPY into USD via FX swaps (Figure 1). In this paper we focus on the mechanisms that govern price-setting in the FX swap market. While much of the recent literature focuses on why CIP deviations exist, and range from explanations that center on limits to the supply of dollars in the FX swap market due to bank regulations (Du et al.,2018;Cenedese et al.,2019;Bräuning and Puria,2017) and funding constraints (Rime et al.,2017;Liao,2020), as well as factors that lead to an excess demand for dollars in the FX swap market (Borio et al.,2016;Sushko et al.,2017), less is understood about the role of price-setting in the FX swap market. This paper aims to fill this gap. To this end, we examine order flow - the net of buyer and seller initiated transactions - as a fundamental signal used by dealers to update the forward rate of the FX swap contract. In particular, we investigate the price impact of order flow before and after the financial crisis and how dealers respond to different types of information. When a no-arbitrage relationship like CIP holds tightly, the role of order flow is confined to correcting short-lived periods of mispricing. In the pre-crisis period, money markets were characterized by low heterogeneity in funding costs where Libor (London interbank offered rate) acted as an accurate representation of banks’ marginal cost across currencies. This made price-setting in the FX swap market a straightforward process, where the dealer takes interest rates in respective currencies as given. The dealer then sets the forward rate according to CIP so that the returns are equalized after covering the exchange rate risk. Hence, we hypothesize that the price impact of order flow is small prior to 2008. In contrast, the post-crisis period is characterized by a large dispersion in funding costs, differences in funding availability across currency areas and tighter constraints on banks’ balance sheets. This makes it difficult for dealers to determine the equilibrium price. We hypothesize that dealers use order flow in the post-crisis period as a signal to set the forward rate. By exploiting time variation in the dispersion of funding costs and balance sheet constraints we examine how these factors can 1 account for an increase in the price impact of order flow. Our order flow measure is based on transaction level data from the Thomson Reuters D20002 platform. This platform registers inter-dealer transactions in the FX swap market where each trade is signed as either a buyer or a seller initiated transaction.1The daily net of buyer and seller initiated transactions constitutes our order flow measure. In our sign convention, we interpret a positive order flow as net buying pressure to obtain USD through FX swaps. Due to superior market depth, we use 1-week maturity as our preferred tenor and base our empirical analysis on the 1-week FX swap order flow and 1-week deviation from CIP.2 We start by developing a microstructural model of the FX swap market. The model has three key agents; customers, arbitrageurs and dealers. Customers are managing the currency exposure on their balance sheets, for example by swapping foreign currency into USD. Arbitrageurs provide funds through the FX swap market when arbitrage opportunities appear. Dealers act as intermediaries and match the flows of customers and arbitrageurs and typically try to keep their positions flat to avoid financing inventories (Lyons,1995;Bjønnes and Rime,2005). Dealer aversion to inventory accumulation yields a price-setting condition in which the forward rate is set to correct order imbalances. The model’s primary contribution is to map a linear relationship between order flow and the price-setting of FX swaps. We can use this framework to study how the price impact of order flow is governed by shocks to arbitrageurs. For example, the model predicts that a tightening of funding and balance sheet constraints leads to an inelastic supply of arbitrage capital, with dealers adjusting the price substantially to avoid order flow and balance inventories. Guided by our model, we first estimate the price impact of order flow, and find it has increased substantially after the global financial crisis in 2008. Since the crisis, a positive 1 standard deviation shock to order flow, i.e. demand to borrow USD through the FX swap market, causes a widening of CIP deviations by up to 4 basis points. That is, when the demand for USD in the FX swap market increases, the cost of obtaining dollars through FX swaps increases as well. We then test why the price impact of order flow is a post-crisis phenomenon by examining potential differences in the price impact of order flow in periods when the heterogeneity in US funding costs is large and when balance sheet expansion is particularly costly for banks.3Our estimates reveal that up to three quarters of the increased price impact 1Note that the signing comes directly from the data provider and hence is slightly different from the measure used in Rime et al. (2017) which apply the Ready-Lee algorithm. The main difference is that our measure sign all trades accurately, while the Ready-Lee algorithm may not be able to sign trades conducted at the mid price or in rare occasions even misclassify trades. 2Note that our aim is not to precisely measure CIP arbitrage opportunities, but rather price-setting in the FX swap market. We are therefore using 1-week Libor rates as the benchmark rate in our CIP calculation. Importantly, by examining CIP deviations instead of the forward rate directly, we control for movements in the forward rate that relates to changes in the interest rate differential. 3We define periods of funding heterogeneity by the daily cross sectional dispersion in 3-month US Libor panel 2 can be attributed to periods when funding heterogeneity in USD is high and when the FX swap contract crosses regulatory reporting dates at quarter-ends. Higher funding heterogeneity reduces the number of potential arbitrageurs as an increasing share of market participants face funding costs exceeding the threshold necessary to reap arbitrage profit. Similarly, regulatory reporting at quarter-ends give arbitrageurs incentives to reduce their provision of arbitrage capital. Consequently, a larger price adjustment is necessary for dealers to balance inventories. In addition to the time varying price impact of order flow, we also test whether news is impounded in the price contemporaneously, or through trading (see Evans and Lyons (2005) for analysis on the FX spot market). Our model predicts that private information is revealed through order flow meaning that prices adjust as a result of trading activity. For example, suppose in response to a shock to its access to dollar funding, a Euro area bank now obtains dollars via the FX swap market. If the bank’s information is private and not known to the dealers before the order appears, these excess demands translate to order flow in the interdealer market, which can then be used by dealers to update the forward rate. Alternatively, we hypothesize public information is impounded in the price contemporaneously. For example, consider a scheduled monetary announcement of a central bank, where the outcomes of the meeting are conveyed to all market participants simultaneously. If the announcement implies a change in the interest rate differential between two currencies, the dealer can reset the forward rate to match the change in the interest differential. In this setting, the monetary news is impounded in the price, suggestive of efficient price-setting in the FX swap market. We test whether the public or private information view is relevant in price-setting by studying three different types of events. First, we examine the effect of Federal Reserve Swap lines during the period 2007-2010. The swap lines allowed foreign central banks to provide US dollar funding directly to their own eligible counterparties. By doing so, a larger set of counterparties were able to access USD directly from the central bank rather than via the FX swap market.4Although it is publicly announced when these auctions take place, the dealers do not have detailed information on whether individual counterparties would draw on the swap line.5 Therefore, we expect swap lines will reduce the demand for USD through FX swaps and lower the order flow into USD. Second, we look at dates when the FX swap contract crosses quarterends. A large number of banks report quarter-end balance sheet snapshots to regulators. This quotes. 4Alternatively, the swap line also relaxes arbitrageur balance sheet constraints and increases arbitrageurs’ ability to supply dollars in the FX swap market. The effects on both customers and dealers will have an equivalent effect of reducing the relative demand for dollar funding in the FX swap market. 5We stress that the private information is not the announcement of the swap line itself, which is known to dealers, but the details of counterparties that use the swap line. For example, only a subset of banks that draw on the swap line may have previously been relying on dollar funding via FX swaps. Similarly, banks may now start using these dollar funds to supply dollars in the FX swap market. Both of these outcomes are unanticipated by dealers until they are revealed as positive order flow. 3 implies incentives to reduce the size of the balance sheet leading to a more inelastic supply of arbitrage capital and significant price effects over reporting dates, as documented in Du et al. (2018). Given quarter-ends are public information and known to dealers in advance, we hypothesize contemporaneous price adjustment. Third, we identify monetary policy surprises to test whether the adjustment in the forward rate following monetary policy announcements happens through order flow. Following our example of a monetary announcement constituting public information, our theory points towards monetary news being impounded in the forward rate contemporaneously. Turning to the empirical evidence, we find evidence that the swap lines reduced the order flow into USD which in turn affected the forward rate, supporting the private information hypothesis. In response to quarter-ends, we utilize high frequency data of forward rates to show a large contemporaneous price adjustment exactly at the hour the FX swap contract crosses quarter-ends, with the full price adjustment priced within two hours. Finally, in line with the hypothesis of public information we find no effect on order flow of monetary policy announcements. These results highlight that dealers efficiently adjust the price according to publicly available information. Roadmap. The paper is outlined as follows. Section 2provides an overview of related literature. In section 3, we outline definitions of covered interest rate parity, FX swaps and order flow and describe the data. In section 4, we develop a model of the microstructure of the FX swap market and derive a price-setting rule that relates the forward rate of the swap to order flow observed in the inter-dealer market. In section 5, we first provide baseline estimates of the price impact of order flow. In section 6, we empirically test the microstructure hypotheses of how prices are determined in response to public and private sources of information, using the response of the FX swap market in response to quarter-end bank regulations and central bank swap lines. In section 7we conclude. 2 Related literature The literature on post 2008 CIP violations naturally centre on theories of what are the supply and demand fundamentals in the FX swap market that explain persistent violation of deviations. Theories on limits to the supply of dollars in the FX swap market include rising balance sheet costs and regulatory requirements (Du et al.,2018;Liao,2020;Bräuning and Puria,2017), the role of the dollar in constraining leverage (Avdjiev et al.,2016), and rising bid/ask spreads due to limited dealer capacity (Pinnington and Shamloo,2016), costs to leverage such as shareholder risk (Andersen et al.,2019) and rising counterparty or liquidity risk (Baba and Packer,2009;Mancini Griffoli and Ranaldo,2009). Other factors affecting agents demands for dollars in the FX swap market include declines in bank quality, declines in 4 short-term funding, unconventional monetary policies, and central bank swap lines (Sushko et al.,2017;Bahaj et al.,2018;Ivashina et al.,2015;Iida et al.,2016). This paper contributes to understanding CIP violations by understanding how constraints on the supply of dollars in the FX swap market can lead to price discovery through order flow. This is a critical component of the FX swap market microstructure and we show empirically that dealers use order flow as a fundamental signal to update the forward rate of the FX swap. The seminal work on market microstructure in FX has typically examined the price impact of order flow on spot foreign exchange markets (Evans and Lyons,2002,2005,2006;Berger et al.,2008;Rime et al.,2010;Ranaldo and Somogyi,2019). Microstructure models in Evans and Lyons (2002) have typically used simultaneous trade models in which dealers set prices, and use inter-dealer order flow following a trading round as information to reset prices. In developing our model framework of the FX swap market, we share many of the elements in trading, however we note two clear differences in FX swaps. The first is that customers in the FX swap market are trading for hedging purposes. In contrast, investors in the FX spot market are composed of informed and uninformed traders, with informed traders having an information advantage in the price of the spot exchange rate, which is treated as a speculative asset. Second, we add arbitrageurs to the framework as they attempt to make systematic profits from the mispricing of the forward rate. Using our framework, we derive a price-setting relation in which price adjustment of the FX swap, which we denote by the change in the CIP deviation, is linearly related to order flow. Finally, we relate to a recent interest in understanding the microstructure and impact of order flow in the FX swap market. Krohn and Sushko (2017) examine how the market structure of the FX swap market has led to a reduction in market liquidity and rising bid/ask spreads during quarter-end periods. Cenedese et al. (2019) and Rime et al. (2017) find evidence that order flow has price impact in the post-crisis period. We extend their work in several ways. First, our measure of order flow is based on each trade being marked as buyer or seller initiated within the data from the trading platform, not the Ready-Lee algorithm. This eliminates potential errors and enable us to sign each trade. Second, through a model framework, we derive the price impact of order flow on the FX swap market through an inter-dealer market that sets the forward rate to minimize inventory accumulation. Third, in contrast to the existing literature we provide an explanation on how price impact of order flow has changed after the financial crisis. Our model framework enables us to link two factors, increased dispersion in dollar funding costs, and the tightening of leverage constraints, that explain up to three quarters of increased price impact we observe empirically in the post-crisis period. Fourth, we find that the source of information matters: in response to public announcements, dealers set the forward rate contemporaneously. In contrast, order flow plays a significant role in price-setting of the forward rate in response to private information, and this is substantiated through the allotment 5 x∗ $,j,t =         ∆t−c$,j,t−φ0 jx$,j,t Wt ρθ2σ2, c$,j,t < c∗ $ 0, c$,j,t ≥c∗ $ (6) To summarize, arbitrageur supply of dollars is positively related to the forward premium (and CIP deviation ∆), and negatively related to dollar funding spreads and leverage constraints. Increased heterogeneity in dealer funding spreads, for example, will cause more arbitrageurs to have too high funding spreads c$,j,t ≥c∗ $, and to not supply dollars in the FX swap market. We will discuss the implications of funding spread heterogeneity for price-setting in the inter-dealer market in a following section. Customers Customers, typically banks, use the FX swap market to hedge their currency balance sheet mismatch. We capture customer demands by the following stylized function, where banks are in a continuum [0,1] indexed by bank quality θband the CIP deviation ∆. Importantly, xD $,t is a measure of the net demand for USD at the spot leg of the FX swap. xD $,t =Z1 0 f(θb,∆)db (7) The first determinant of net demand for USD in the FX swap market is θb, which measures counterparty quality. All else equal, counterparties with higher quality are more likely to obtain dollars directly via commercial paper markets or bank deposits. Therefore, demands for dollar funding via FX swaps is inversely related to counterparty quality. The increase in counterparty risk is a key determinant of the increased demand for dollar constrained banks in the FX swap market in 2008 (Baba and Packer,2009). The second determinant of net demand is the CIP deviation ∆. All else equal, a higher CIP deviation implies an increase in the net cost of swapping euros, swiss francs and yen into dollars. Evidence in Eguren Martin et al. (2018) suggests that in response to shocks to the CIP deviation, banks’ net demand for dollars in the FX swap market declines. Inter-Dealer Market We have defined customers and arbitrageurs. Each of these agents are price-takers, and go to a market-maker to find a counter-party to take the other side of the trade. The market-maker is the dealer in our model. The dealer’s objective is to match flows of swapping domestic currency into dollars with opposing flows. This is consistent with theories of market microstructure 12 where dealers are sufficiently risk averse to holding inventory (see Lyons (1995) and Bjønnes and Rime (2005) for empirical evidence on dealers minimizing inventory). Moreover, all dealers in FX swaps are associated with a bank. Therefore, the dealer faces the same funding costs and constraints as the mother bank that indeed can be part of the group of arbitrageurs. We denote the net dollar demands by customers to dealer jby xD $,j. Denote the net supply of dollars by arbitrageurs to dealer jby x∗ $,t,j. Unmatched flows in dollars are submitted to the inter-dealer market. We illustrate the unmatched flows of a dealer in Figure 7. The dealer submits the excess demand for dollar funding to the inter-dealer market, and this is observed as OFt,j in the Figure. Aggregating across all dealers, we obtain an expression for inter-dealer order flow OFt, in equation 8. Inter-dealer order flow is equal to the net buying pressure of swapping EUR, CHF or JPY (domestic currency) into USD. Net customer demands for USD at the spot leg of the FX swap is equal to xD $,t. Net supply of dollars by Nsymmetric arbitrageurs in the FX swap market is given by PN i=1 x∗ $,t, where we denote Nas the number of arbitrageurs that have sufficiently low funding spreads in order to supply dollars in the FX swap market. OFt=xD $,t − N X i=1 x∗ $,t (8) To illustrate the timing of customer-dealer trades and price-setting, Figure 8depicts a two period model, in which customers and dealers trade at the beginning of each period. Immediately after each period of trading, the inter-dealer market observes order flow. Dealers then set the forward rate of the FX swap, and hence the CIP deviation ∆, to set expected order imbalances to zero for the next period of trading. Definition [Price setting]: The inter-dealer market sets a forward rate to set inter-dealer order flow to be zero, based on an information set that includes information on current and past prices, and customer and arbitrageur fundamentals. E t[OFt(∆t)|It]=0 (9) The price-setting condition is implicitly assuming an inter-dealer market that sets a common price for all dealers. This is a reasonable assumption, as if dealers set different prices, this would not be a sustainable equilibrium as other dealers will only execute swap trades with the dealer that sets the most favorable rate.13 Combining equations 8and 9, we can rewrite the order flow in period tas the unanticipated components of customer demand and dealer supply of dollars 13Moreover, inter-dealer trades are secured by daily margining practically eliminating potential differences in counterparty risk across dealers. 13 in the FX swap market. OFt=xD $,t − E hxD $,t|Iti− N X j=1 x∗ j,t − E hx∗ j,t|Iti (10) In the model, order flow responds to changes to demand fundamentals that are not forecast by dealers. This provides a simple decomposition of order imbalances into unexpected idiosyncratic shocks to customers and dealers, shown in equation 11. The first term reflects unanticipated shocks to customer type and funding spreads. For example, the inter-dealer market may not directly observe customer types, such as credit ratings and their ability to borrow dollars in alternative markets. The second term reflects unanticipated changes in funding spreads. The third term reflects rises in the cost of leverage. OFt=Z1 0 f(θb, .)− E [f(θb, .)|It]db | {z } customer type + 1 ρθ2σ2 N X j=1      c$,j,t − E [c$,j,t|It] | {z } funding spreads +φ0 j,t x W− E φ0 j,t x W|It | {z } leverage constraints      (11) Finally, we can solve for the equilibrium CIP deviation ∆, can be derived from setting expected order flow to zero, in equation 12. Intuitively, an increase in customer demand, an increase in dollar funding spreads, or a tightening of leverage constraints on arbitrageurs, leads to a widening of the CIP deviation. ∆t= E [c$,j,t|It] + E φ0 j,t x W|It+ρθ2σ2 NZ1 0 E [f(θb, .)|It](12) We can use the framework to study the price impact of order flow, as well as the propagation of shocks to demand and supply on price-setting in the FX swap market. Proposition 1: price impact of order flow A positive shock to order flow in period timplies a widening of CIP deviations, with the price sensitivity β=ρθ2σ2 N. ∆t−∆t−1=βOFt(13) The price impact of order flow is seen in equation 13 is governed by β, which is related positively to variance of the exchange rate, counterparty risk, and negatively related to the 14 number of arbitrageurs N.14 We can further decompose price impact of order flow into unanticipated shocks to customer demand and arbitrageur supply in equation 14. An unanticipated change in customer demands for dollars in the FX swap market, due to a change in quality, or an unanticipated change in arbitrageur funding spreads and leverage constraints, has price impact through order flow. ∆t−∆t−1=βZ1 0 f(θb, .)− E [f(θb, .)|It]db | {z } customer type + 1 N N X j=1      c$,j,t − E [c$,j,t|It] | {z } funding spreads +φ0 j,t x W− E φ0 j,t x W|It | {z } leverage constraints      (14) In an efficient market with no limits to arbitrage, N→ ∞ and there is a zero price impact of order flow. In this case, there is an elastic supply of arbitrage capital to take the other side of customer trades, and dealers are able to minimise inventory and match flows. This leads us to proposition 2, which states that the price impact of order flow is more sensitive in periods of increased dispersion in funding spreads of arbitrageurs. Proposition 2: (i) Heterogeneity in funding spreads and (ii) the tightening of leverage constraints, increases the price impact of order flow Assume the distribution of arbitrageur funding costs is given by c$,j,t ∼N( ¯c$, σ2 c). An increased dispersion in funding costs (σc↑) and an increase in the marginal cost of leverage (φ0 j(x W)↑) leads to a decline in the number of arbitrageurs supplying dollars in the FX swap market (N↓) and an increased price impact of order flow, β↑). Proof: The number of arbitrageurs supplying dollars in the FX swap market is proportional to the probability that the investor will conduct arbitrage, which is when the funding cost c$,j,t is less than the threshold value c∗ $. Utilizing the fact that the threshold value c∗ $= ∆ −φ0 j(x W), we can write the probability of conducting arbitrage in equation 15. Prob[c$,j,t < c∗ $]=Φ ∆−φ0 j(x W)−¯c$ σc!(15) 14This contrasts to the βin microstructure models of the spot FX market, which typically measures the relative share of informed traders (Evans and Lyons,2002). We differentiate our price impact equation in that FX swaps do not feature uninformed traders, and rely on customers that use FX swaps for largely hedging purposes. 15 The expected price impact of order flow is then given by E [β] = ρθ2σ2 N×Φ∆−φ0 j(x W)−¯c$ σc. An increase in σcand an increase in the marginal cost of leverage φ0 j(x W)lowers the probability of conducting arbitrage, and increases the expected price impact of order flow. Heterogeneity of funding spreads is consistent with the empirical findings of Rime et al. (2017). When heterogeneity in funding costs increases, the remaining arbitrageurs that still face arbitrage opportunities have to supply more USD. 15 Limits to arbitrage capital are particularly pronounced during quarter-end regulations, and there is micro level evidence suggesting dealers that are more leveraged are more sensitive to order imbalances and demand a higher forward premium on the contract (Du et al.,2018;Cenedese et al.,2019). Finally, we can use the framework to decompose shocks to private and public information, that gives rise to our two views of price-setting in the FX swap market in proposition 3. Proposition 3: Public versus Private information view Consider a shock to customer quality, arbitrageur funding spreads or leverage constraints. Denote these shocks t= [Q,t, C,t, L,t]. Define the dealer information set It, and private information I− tis the complementary set. The price-setting equation can then be expressed in equation 16, where OFt=Q,t +C,t +L,t|I− t ∆t−∆t−1=βOFt+β E [Q,t +C,t +L,t|It](16) Proof: For illustration, let us partition shocks to customer quality into public and private information shocks. Q,t =Q,t|I− t+Q,t|It. Order flow is then defined as shocks to customer quality unanticipated by dealers: OF =Q,t −Q,t|It. Using the price-setting equation, and re-arranging terms, gives us equation 16. ∆t−∆t−1=βQ,t =β(Q,t|I− t+Q,t|It) =β(Q,t −Q,t|It) + β(Q,t|It) =β OFt |{z} private +β(Q,t|It) | {z } public Proposition 3 states that the source of information matters for price-setting. Public infor15Arbitrageurs face higher funding costs when they need to raise large amounts of debt to fund the arbitrage trade. For instance Money Market Funds that lend dollars are limited by regulation not to invest more than 5 per cent of their assets in a single issuer. 16 mation shocks are impounded in the price contemporaneously. In contrast, order flow reflects private information. We provide examples of public and private information shocks in the context of the model. Public information shocks: Examples include monetary announcements and quarter-end reporting requirements. A testable implication of our framework is that we expect to observe contemporaneous adjustment of the forward rate in response to public announcements. Private information shocks: An example of a shock to bank quality that we test empirically is the introduction of central bank swap lines. Central bank swap lines by the Federal Reserve provide incremental dollar liquidity to sufficiently dollar constrained banks. As banks of low quality are more likely to use central bank swap lines as a way to meet dollar funding, we can interpret this as reducing customer demand for dollars via FX swaps. Crucially, if the swap line auctions to dollar constrained banks are private information, this results in a decline in order flow, causing a decline in the forward premium of the swap trade. To conclude, the model has provided a framework to show how unanticipated shocks to customer demand, funding spreads and leverage constraints can translate to an increase in inter-dealer order flow. This causes dealers to reset the forward premium of the FX swap to offset order flow, resulting in a widening of the CIP deviation. This is consistent with microstructure theories on inventory control; dealers are sufficiently averse to holding inventory and update the forward rate as a response to avoid inventory accumulation. We test three predictions in our empirical evidence. First, based on proposition 1, we measure the price impact of order imbalances. We then test proposition 2; which states that heterogeneous dollar funding costs and leverage constraints during quarter-ends lead to increased price impact of order flow. In proposition 3, we test the public and private information views with an analysis of central bank swap lines, quarter-ends and monetary announcements. 5 Price Impact of Order Flow Baseline specification In this section, we examine the price impact of order flow. In proposition 1 of the model, we concluded that an increase in order flow in the inter-dealer market is consistent with excess demands for swapping other currencies into USD, i.e. borrowing USD through FX swaps. As dealers are averse to holding inventory, the inter-dealer market resets the forward rate to offset order flow. This leads to an increase in the forward premium and a widening of CIP deviations. Our baseline specification for testing the price impact of order flow is outlined in equation 17. ∆CIPt=α+β1OFt+βjXj,t +t(17) 17 The outcome variable is the daily change in 1-week CIP deviations, where negative values indicate that it is more costly to obtain USD through FX swaps relative to the direct borrowing rate in USD. Our variable of interest, β1, measures the price impact of order flow (OF). X is a vector of control variables including the change in the U.S. Libor-OIS spreads for 1-week and 3-month maturities, the VIX index, and the USD Trade weighted exchange rate. We run the specification for all currency pairs as a panel and for the EUR/USD, CHF/USD and JPY/USD pairs separately, and divide our sample into two periods, a pre 2008 period (January 2005 to December 2007), and a post 2008 period (January 2008 to September 2017). By dividing our sample into before and after 2008 we are able to uncover potential changes in the price impact of order flow after the global financial crisis. Our justification is that during the pre 2008 period, CIP held tightly, indicating an elastic supply of arbitrage capital. Order flow is not expected to have any significant price impact, as there are relatively short-lived periods of mispricing in the FX swap market (Akram et al.,2008). However, in the post-crisis period, arbitrage capital becomes scarce. We hypothesize that dealers adjust prices more aggressively during this period to attract the necessary arbitrage capital and balance dealer inventories. Our results are presented in Table 3. In columns (1) through to (4), we test for price impact in the pre-crisis period, and in columns (5) through to (8), we test for price impact in the post-crisis period. We find that order flow has significant price impact in the post 2008 period for all 3 pairs, with a one standard deviation change in order flow widening CIP violations by up to 4 basis points based on the panel specification in column (5). In contrast, there is no significance in the pre 2008 period. Dynamic effects In addition to the contemporaneous price impact of order flow, we test for dynamic effects using a structural vector autoregression (VAR) framework. Following the work of Hasbrouck (1991) and Ranaldo and Somogyi (2019), we estimate the following bivariate VAR, illustrated in equations 18 and 19. In equation 18, a contemporaneous shock to daily order flow is impounded in the price the same day, which is consistent with the price-setting equation derived in our model framework. Conversely, we only allow for shocks to prices to affect order flow with a lag. The identification assumption is consistent with causality running from order flow to price-setting of the FX swap. CIPt=α1+ L X k=1 γ1,kCIPt−k+ L X k=0 β1,kOFt−k+1,t (18) OFt=α2+ L X k=1 γ2,kCIPt−k+ L X k=1 β2,kOFt−k+2,t (19) 18 Based on our specification with 7 lags, we test the effects of a 1 standard deviation shock to order flow on the CIP deviations in Figure 9. On the left panel, we test for effects during the pre 2008 period, and observe no systematic effect of order flow on the CIP deviation for all pairs of the EUR/USD, CHF/USD and JPY/USD. In the post 2008 period, we find the CIP deviation widens by approximately 4 basis points contemporaneously, with the price impact decaying to zero approximately 3-5 days following the shock. This response is intuitive; order flow has maximum impact contemporaneously, and decays over time as dealers update their information sets. We now turn to two factors that can restrict arbitrage capital, an increase in funding cost heterogeneity and regulatory reporting during quarter-ends, that account for the increase in price impact observed in the post 2008 period. Dispersion in funding costs and quarter-ends Proposition 2 of the model in section 4predicts that the price impact of order flow increases when heterogeneity in U.S. funding costs is large and when banks’ balance sheet constraints are more binding. When funding cost dispersion increases in USD less market participants are in the position to conduct the arbitrage trade. The remaining arbitrageurs with favourable funding costs may face limits to the scalability of the arbitrage trade. Important reporting dates, such as quarter-ends, represent an opportunity to test how the price impact of order flow responds to tighter regulatory constraints. An increasing marginal cost of leverage suggests the supply of arbitrage capital becomes less elastic and the price impact of order flow increases. To jointly test these hypotheses we run the following regression specification in equation 20. The variables FundingHet and Qend represent dummy variables for funding cost heterogeneity and quarter-ends, respectively. We are interested in the interaction between order flow and these variables to examine if the price impact of order flow changes when dispersion in funding costs increases and the FX swap contract crosses quarter-ends. The dummy FundingHet captures days with high cross sectional dispersion among U.S. Libor panel banks in their individual submissions and when the 1-week FX swap contract matures after quarter-ends.16 The dummy Qend captures an increase in balance sheet constraints as regulatory authorities in most jurisdictions rely on quarter-end snapshot of banks’ balance sheets. In addition to the control variables mentioned in the baseline specification, we also include the two dummies FundingHet and Qend. ∆CIPt=α+β1OFt+β2OFt×FundingHett+β3OFt×Qendt+βjXj,t +t(20) 16The dispersion dummy takes value 1 when the cross sectional dispersion (difference between the maximum and the minimum submitted quote) is within the highest quartile of the distribution on the respective day and zero otherwise 19 Table 4presents the results. Columns (1) to (4) measure the price impact of order flow during the pre-crisis period, and columns (5) to (8) measure price impact during the post period. Consistent with our theory, we find that the price impact of order flow is significantly higher during periods of high dispersion in Libor quotes in the post crisis period. The results suggest that larger heterogeneity in funding spreads implies that the forward rate has to adjust more aggressively to attract the necessary arbitrage capital to balance the market. Moreover, there is a substantial increase in price impact during quarter-end periods. This is also consistent with our theory of more inelastic supply of arbitrage capital when balance sheet constraints are more binding. Based on the panel regression estimates in column (5), days with high funding cost heterogeneity and quarter-end periods account for approximately three quarters of the increase in price impact after 2008.17 Direction of order flow We now test for asymmetric price impact of order flow. We expect that the price impact of order flow when it is positive, i.e. there is net pressure for swapping domestic currency into USD in the inter-dealer market. This is because high funding cost heterogeneity in USD leads to a shortage of arbitrage capital in USD. However, for negative shocks to order flow, the availability of arbitrage capital in other currencies is what matters for price impact. We present our results in Table 5. Columns (1) through to (4) depict the results from regressing positive and negative order flow on changes in CIP deviations and these two variables interacted with a dummy that takes the value of 1 after 2008, and zero otherwise. As expected, neither positive nor negative order flow have any price impact prior to 2008. After 2008, both negative and positive order flow is highly significant across all currencies. This result serves as an indication that insufficient arbitrage capital in USD as an important constraint after 2008. In addition, we run a similar regression as specified in equation 20 on the panel of currencies, but now with the order flow split between positive and negative order flow. Columns (5) and (6) depict the pre and post 2008 results respectively. In the post 2008 sample, shown in column (6), the price impact of positive order flow is significantly larger than for negative order flow, and during periods when funding heterogeneity is high. This is consistent with our hypothesis that high funding cost heterogeneity in USD leads to a shortage of arbitrage capital in USD, making dealers more sensitive to positive order flow (net demand for swapping domestic currencies into USD) in the FX swap market. During quarter-ends, we find both positive and negative order flow have large price impact. Dealers are aggressively adjusting the price to balance order flow independent of the direction. They are using the order flow as a signal to update the forward 17To arrive at this calculation, we note that the aggregate price impact of a one standard deviation order flow shock is approximately 4 basis points. After controlling for funding heterogeneity and quarter-ends, the coefficient β1≈1, implying that three quarters of the price impact is explained by these two factors. 20 rate of the swap in periods when balance sheets are particularly constrained. Bid/ask spreads and price volatility To further substantiate the results on the price impact of order flow we examine bid/ask spreads and price volatility, calculated from high frequency quotes in the FX swap market. Bid/ask spreads are a proxy for market liquidity, and constructed as the daily intraday average of 1-week bid and ask for each currency pair. High bid/ask spreads can either indicate inventory risk for the dealer, or asymmetric information in the FX swap market. We calculate the daily standard deviation of price changes from high frequency intraday FX swap quotes as a proxy for price volatility. We hypothesize that bid-ask spreads and intraday volatility increase during periods of funding heterogeneity and when the FX swap contract trades over quarter-ends. Tables 6and 7 present results from regressing the bid/ask spread and price volatility on the quarter-end and funding heterogeneity variables used in specification 20. Columns (1) through to (3) test for effects in the pre-crisis period. Columns (4) to (9) test for effects in the post-crisis period, with additional dummies capturing the post 2015 period. First, the constant in the regression indicates that the bid/ask spreads are lowest for EUR/USD and highest for CHF/USD. This is in line with the interpretation that the EUR/USD is the most liquid currency pair while the CHF/USD is the least liquid. Similarly, price volatility increases during periods of large funding cost heterogeneity and during quarter-end periods. This is in line with order flow having stronger price impact and wider bid/ask spreads during these periods. Most of the increase during quarter-ends have taken place after banks started to report leverage ratio to the public in 2015. In addition to lower market liquidity over quarter-ends, this may be related to leverage ratio increasing the costs of inventory for dealers. Wider bid/ask spreads act as a compensation for higher costs of being a dealer, and is consistent with empirical evidence in Krohn and Sushko (2017) which find bid/ask spreads rise during quarter-end periods.18 6 Public vs Private Information Shocks In this section we empirically test the microstructural hypotheses of public and private information, outlined in proposition 3 of the model in section 4. We examine how price-setting in the FX swap market is determined in response to three different types of announcements, central bank swap lines, quarter-end reporting requirements and monetary policy surprises. 18They make an additional point that market structure matters for dealer pricing. In particular, the role of smaller dealers providing arbitrage capital during quarter-ends leads to an increase in the observed bid/ask spreads 21 References Akram, Q Farooq, Dagfinn Rime, and Lucio Sarno, “Arbitrage in the foreign exchange market: Turning on the microscope,” Journal of International Economics, 2008, 76 (2), 237–253. Andersen, Leif, Darrell Duffie, and Yang Song, “Funding value adjustments,” The Journal of Finance, 2019, 74 (1), 145–192. Avdjiev, Stefan, Wenxin Du, Catherine Koch, and Hyun Song Shin, “The dollar, bank leverage and the deviation from covered interest parity,” 2016. Baba, Naohiko and Frank Packer, “Interpreting deviations from covered interest parity during the financial market turmoil of 2007–08,” Journal of Banking & Finance, 2009, 33 (11), 1953–1962. Bahaj, Saleem, Ricardo Reis et al., “Central Bank Swap Lines,” Bank of England Working Paper 2018. Berger, David W, Alain P Chaboud, Sergey V Chernenko, Edward Howorka, and Jonathan H Wright, “Order flow and exchange rate dynamics in electronic brokerage system data,” Journal of international Economics, 2008, 75 (1), 93–109. Bjønnes, Geir Høidal and Dagfinn Rime, “Dealer behavior and trading systems in foreign exchange markets,” Journal of Financial Economics, 2005, 75 (3), 571–605. Borio, Claudio EV, Robert N McCauley, Patrick McGuire, and Vladyslav Sushko, “Covered interest parity lost: understanding the cross-currency basis,” 2016. Bräuning, Falk and Kovid Puria, “Uncovering covered interest parity: the role of bank regulation and monetary policy,” 2017. Cenedese, Gino, Pasquale Della Corte, and Tianyu Wang, “Currency mispricing and dealer balance sheets,” Available at SSRN 3327088, 2019. Du, Wenxin, Alexander Tepper, and Adrien Verdelhan, “Deviations from covered interest rate parity,” The Journal of Finance, 2018, 73 (3), 915–957. Evans, Martin DD and Richard K Lyons, “Order flow and exchange rate dynamics,” Journal of political economy, 2002, 110 (1), 170–180. and , “Do currency markets absorb news quickly?,” Journal of International Money and Finance, 2005, 24 (2), 197–217. 28 and , “Understanding order flow,” International Journal of Finance & Economics, 2006, 11 (1), 3–23. Griffoli, T Mancini and A Ranaldo, “Deviations from Covered Interest Parity during the crisis: A story of funding liquidity constraint,” Technical Report, Working Paper, Swiss National Bank 2009. Hasbrouck, Joel, “Measuring the information content of stock trades,” The Journal of Finance, 1991, 46 (1), 179–207. Iida, Tomoyuki, Takeshi Kimura, Nao Sudo et al., “Regulatory reforms and the dollar funding of global banks: Evidence from the impact of monetary policy divergence,” Technical Report, Bank of Japan 2016. Ivashina, Victoria, David S Scharfstein, and Jeremy C Stein, “Dollar Funding and the Lending Behavior of Global Banks,” The Quarterly Journal of Economics, 2015, 130 (3), 1241–1281. Kozhan, Roman and Mark Salmon, “The information content of a limit order book: The case of an FX market,” Journal of Financial Markets, 2012, 15 (1), 1–28. Krohn, Ingomar and Vladyslav Sushko, “FX spot and swap market liquidity spillovers,” WBS Finance Group Research Paper, 2017, (243). Lee, Charles MC and Mark J Ready, “Inferring trade direction from intraday data,” The Journal of Finance, 1991, 46 (2), 733–746. Liao, Gordon Y, “Credit migration and covered interest rate parity,” Journal of Financial Economics, 2020. Lyons, Richard K, “Tests of microstructural hypotheses in the foreign exchange market,” Journal of Financial Economics, 1995, 39 (2-3), 321–351. Martin, Fernando Eguren, Matias Ossandon Busch, and Dennis Reinhardt, “Global banks and synthetic funding: the benefits of foreign relatives,” 2018. Pinnington, James and Maral Shamloo, “Limits to arbitrage and deviations from covered interest rate parity,” Technical Report, Bank of Canada staff discussion paper 2016. Ranaldo, Angelo and Fabricius Somogyi, “Asymmetric Information Risk in FX Markets,” 2019. 29 Rime, Dagfinn, Andreas Schrimpf, and Olav Syrstad, “Segmented money markets and covered interest parity arbitrage,” 2017. , Lucio Sarno, and Elvira Sojli, “Exchange rate forecasting, order flow and macroeconomic information,” Journal of International Economics, 2010, 80 (1), 72–88. Sushko, Vladyslav, Claudio EV Borio, Robert N McCauley, and Patrick McGuire, “The failure of covered interest parity: FX hedging demand and costly balance sheets,” 2017. 30 Figures Figure 1: 1 Week CIP Deviations for EUR/USD, JPY/USD and CHF/USD pairs 2006 2008 2010 2012 2014 2016 2018 700 600 500 400 300 200 100 0 CIP Deviation (Basis Points) EUR CHF JPY Note: This figure plots the 1 Week CIP deviation measured in basis points, obtained from Thomson Reuters Tick History. This provides a measure of CIP deviations based on a LIBOR benchmark rate. Negative deviations indicate a dollar borrowing premium for the EUR/USD, CHF/USD and JPY/USD pairs. Sample period is 01/2005-09/2017. 31 Figure 2: Foreign exchange swap Customer Dealer Customer Dealer X Euros SX USD X Euros FX USD Spot Leg Forward Leg Note: FX swap is typically for maturities at less than 3m. At the spot leg, domestic currency and dollars are swapped at the prevailing spot rate. At maturity, the principals are then re-exchanged at the forward rate. 32 Figure 3: Daily Order Flow measureEUR/USD, CHF/USD and JPY/USD 2006 2008 2010 2012 2014 2016 2018 Date 20 10 0 10 20 30 Net Buyer Initiated Transactions OF: EUR 2006 2008 2010 2012 2014 2016 2018 Date 10 5 0 5 10 Net Buyer Initiated Transactions OF: CHF 2006 2008 2010 2012 2014 2016 2018 Date 10 5 0 5 10 Net Buyer Initiated Transactions OF: JPY Note: Daily count order flow for EUR/USD, JPY/USD and CHF/USD pairs using the TR D2000-2, for FX swap maturities at 1 week. Order flow is given as the net of buyer initiated transactions, where buyer initiated transactions are signed +1 and seller initiated transactions are signed −1.OF count t=Pk=t k=t0 1 [Tk=B]− 1 [Tk= S]Sample period is 01/2005-09/2017. 33 Figure 4: Range of Libor Fixing quotes 2006 2008 2010 2012 2014 2016 2018 Date 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 % (per annum) Note: The figure depicts the daily difference in percentage points between the highest and lowest submission among the contributing banks in USD Libor with 3-month maturity. The data are obtained from Bloomberg and Intercontinental Exchange (ICE). Sample period is 01/2005-09/2017. 34 Figure 5: Bid/ask spreads for EUR, CHF and JPY 1 Week FX swaps 2006 2008 2010 2012 2014 2016 2018 0 50 100 150 200 Basis Points EUR CHF JPY Note: The figure shows the daily average between the bid and the ask quotation based on hourly data from Thomson Reuters tick history. The bid/ask spread is expressed in basis points. Sample period is 01/200509/2017. 35 Figure 6: Intra-day forward rate volatility for EUR, CHF and JPY 1 Week FX swaps 2006 2008 2010 2012 2014 2016 2018 0 25 50 75 100 125 150 175 Basis Points EUR CHF JPY Note: The figure shows the daily standard deviation in swap points based on hourly data from Thomson Reuters tick history. The standard deviation is expressed in basis points. Sample period is 01/2005-09/2017. 36 Figure 7: Schematic of the interactions between customers, dealers and the inter-dealer market InterDealer Customer Dealer 1 Arbitrageur Dealer 2 Arbitrageur Customer XD $,1 X∗ 1 OF1 OF2 XD $,2 XD $,2 Note: This schematic illustrates the structure of the dealer-customer and inter-dealer market. Each customer has a net demand for dollar funding in the FX swap market, which we denote xD $. The excess demands for dollar funding that cannot be met by the dealer’s supply of dollars, is in turn submitted to the inter-dealer market. Aggregating net orders for swapping domestic currency into dollars gives rise to inter-dealer order flow OF which is observed as a public signal by the inter-dealer market for setting the forward rate. 37 Table 2: Summary Statistics count Order Flow. Pre 2008 Post 2008 mean sd min max mean sd min max EUR/USD 0.00 3.91 -16 29 -.09 3.25 -24 18 CHF/USD 0.08 1.59 -8 8 0.11 1.01 -10 7 JPY/USD 0.06 1.63 -7 8 0.00 1.41 -9 8 Note: This table records summary statistics of order flow based on trades in 1 week FX swaps using interdealer trades in Thomson Reuters D2000-2 Platform. Order flow is constructed as the net of buyer initiated transactions, where a transaction is signed +1 if it is swapping euros, swiss francs and yen into dollars at the spot leg of the FX swap contract. The sample period is from 01/2005-09/2017. 44 Table 3: Price impact of order flow before and after GFC (1) (2) (3) (4) (5) (6) (7) (8) Pre 2008 Post 2008 ∆CIP panel ∆CIP eur ∆CIP chf ∆CIP jpy ∆CIP panel ∆CIP eur ∆CIP chf ∆CIP jpy OF -0.16 -0.42** -0.41 -0.01 -3.64*** -3.74*** -2.98*** -4.42*** (0.33) (0.18) (0.41) (0.96) (0.55) (0.89) (0.85) (1.05) Constant -0.01 -0.10 0.05 0.05 -0.02 -0.22 0.24 -0.13 (0.35) (0.24) (0.50) (0.93) (0.27) (0.45) (0.48) (0.47) Observations 2,084 713 712 659 6,703 2,238 2,234 2,231 R-squared 0.00 0.01 0.00 0.01 0.08 0.11 0.04 0.09 Controls Yes Yes Yes Yes Yes Yes Yes Yes Note: This table regresses order flow for 1 week EUR/USD, CHF/USD and JPY/USD FX swaps on daily changes in 1 week CIP deviations based on Libor rates. Standardized order flow OF is measuring the net buyer transactions of swapping euros, swiss francs and yen into dollars, and is sourced from TR D2000-2 inter-dealer trades for 1 Week FX swaps. The 1 Week CIP deviation is calculated using TR tick history quotes on 1 week spot and forward rates with close at 5 pm London time. Controls include the changes in USD Libor-OIS spreads for 1 week and 3 month maturities, the VIX index, and the USD Trade weighted exchange rate. The full sample from Jan 1, 2005 to Sep 1, 2017 is split into pre and post 2008. Data is daily. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 45 Table 4: Price impact of order flow; funding constraints and quarter-ends (1) (2) (3) (4) (5) (6) (7) (8) Pre 2008 Post 2008 ∆CIPpanel ∆CIP eur ∆CIP chf ∆CIP jpy ∆CIP panel ∆CIP eur ∆CIPchf ∆CIP jpy OF -0.46 -0.47** -0.50 -0.57 -1.04** -1.20** -2.02* 0.06 (0.31) (0.20) (0.45) (0.94) (0.48) (0.47) (1.03) (0.60) OF ×FundingHet 6.00 0.91 0.32 37.08 -4.14*** -4.84** -0.34 -6.36*** (6.46) (4.30) (2.30) (26.70) (1.18) (2.17) (1.57) (1.68) Qend ×OF 1.54 -0.13 0.29 3.16 -9.27*** -2.83 -9.09* -15.97*** (1.40) (0.57) (0.82) (3.88) (3.26) (2.95) (5.32) (6.06) Constant 0.13 0.03 -0.04 0.48 0.39* 0.24 0.92** 0.19 (0.36) (0.25) (0.52) (0.96) (0.21) (0.28) (0.41) (0.37) Observations 2,084 713 712 659 6,703 2,238 2,234 2,231 R-squared 0.01 0.01 0.01 0.04 0.11 0.13 0.07 0.16 Controls Yes Yes Yes Yes Yes Yes Yes Yes Note: This table regresses order flow for 1 week EUR/USD, CHF/USD and JPY/USD FX swaps on daily changes in 1 week CIP deviations based on Libor rates. Standardized order flow OF is measuring the net buyer transactions of swapping euros, swiss francs and yen into dollars, and is sourced from TR D2000-2 inter-dealer trades for 1 Week FX swaps. The 1 Week CIP deviation is calculated using TR tick history quotes on 1 week spot and forward rates with close at 5 pm London time. FundingHet is a dummy variable that takes the value 1 when the daily dispersion in individual panel bank’s 3-month Libor quotes is among the 25 per cent largest values, and zero otherwise. Qend is a dummy variable taking the value 1 when the 1 week contract is settled prior to quarter-end and matures after quarter-end. Controls include the changes in USD Libor-OIS spreads for 1 week and 3 month maturities, the VIX index, the USD Trade weighted exchange rate. Additionally, the following variables are included in the regression specification but not shown in the Table; Qend and FundingHet. Data is daily. The sample runs from Jan 1, 2005 to Sep 1, 2017. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 46 Table 5: Price impact of order flow; direction of flow (1) (2) (3) (4) (5) (6) ∆CIP panel ∆CIP eur ∆CIP chf ∆CIP jpy ∆CIP panel ∆CIP panel OF × 1 [OF > 0] -0.12 -1.22* -0.28 0.97 -0.88 -1.42* (0.67) (0.67) (0.35) (2.13) (0.60) (0.72) OF × 1 [OF < 0] 0.38 0.02 -0.25 1.26 0.01 -0.86 (0.47) (0.51) (0.95) (0.97) (0.47) (0.86) OF × 1 [OF > 0] ×post2008 -4.39*** -5.07** -2.58* -6.04* (1.23) (2.31) (1.38) (2.77) OF × 1 [OF < 0] ×post2008 -3.30*** -1.32 -2.99** -5.32 (1.01) (1.03) (1.51) (2.58) OF × 1 [OF > 0] ×FundingHet 11.25* -5.31** (6.23) (2.20) OF × 1 [OF < 0] ×FundingHet 5.89 -2.83* (7.46) (1.65) OF × 1 [OF > 0] ×Qend 3.43 -6.98* (2.42) (4.24) OF × 1 [OF < 0] ×Qend -0.96 -12.33* (1.28) (7.48) Constant 0.10 0.36 -0.02 -0.02 0.41 0.50 (0.51) (0.50) (0.52) (1.35) (0.52) (0.34) Observations 8,787 2,951 2,946 2,890 2,084 6,703 R-squared 0.05 0.10 0.03 0.04 0.01 0.11 Controls Yes Yes Yes Yes Yes Yes Post2008 No Yes Note: This table regresses order flow for 1 week EUR/USD, CHF/USD and JPY/USD FX swaps on daily changes in 1 week CIP deviations based on Libor rates. Standardized order flow OF is measuring the net buyer transactions of swapping euros, swiss francs and yen into dollars, and is sourced from TR D2000-2 inter-dealer trades for 1 Week FX swaps. The 1 Week CIP deviation is calculated using TR tick history quotes on 1 week spot and forward rates with close at 5 pm London time. 1 [OF > 0] takes the order flow value if the order flow is positive, zero otherwise. Positive order flow implies a pressure to obtain USD spot and sell USD forward (i.e. borrow USD). 1 [OF < 0] takes the order flow value if the order flow is negative, zero otherwise. FundingHet is a dummy variable that takes the value 1 when the daily dispersion in individual panel bank’s 3-month Libor quotes is among the 25 per cent largest values, and zero otherwise. Qend is a dummy variable taking the value 1 when the 1 week contract is settled prior to quarter-end and matures after quarter-end. Post 2008 is a dummy that takes the value 1 after Jan 1 2008, and zero otherwise. The Table only shows the relevant coefficients. Controls include the changes in USD Libor-OIS spreads for 1 week and 3 month maturities, the VIX index, the USD Trade weighted exchange rate. Additionally, the dummies for FundingHet, Qend and Post2008 are included, but not shown. Data is daily. The sample runs from Jan 1, 2005 to Sep 1, 2017. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 47 Table 6: Bid/ask spreads (1) (2) (3) (4) (5) (6) (7) (8) (9) Pre 2008 Post 2008 EUR CHF JPY EUR CHF JPY EUR CHF JPY FundingHet 0.02 14.05*** 2.21** 10.95*** 5.07*** 13.39*** 17.40*** 10.13*** 18.67*** (0.19) (2.08) (1.09) (0.95) (0.89) (1.13) (1.59) (1.32) (1.95) Qend 0.01 -0.41 0.27 3.64*** 3.59** 4.81** 2.69 0.61 3.25 (0.05) (0.53) (0.34) (1.33) (1.70) (1.91) (1.63) (1.48) (2.44) post2015 -2.16*** 10.74*** 2.32*** (0.28) (1.40) (0.40) post2015 ×Qend 2.57 10.68** 5.36* (2.36) (4.97) (2.98) post2015 ×FundingHet -13.53*** -17.89*** -13.50*** (1.63) (2.01) (2.01) Constant 2.51*** 9.56*** 4.52*** 5.61*** 16.54*** 6.11*** 6.06*** 14.53*** 5.71*** (0.01) (0.15) (0.07) (0.18) (0.42) (0.19) (0.22) (0.39) (0.23) Observations 756 745 691 2,434 2,437 2,438 2,434 2,437 2,438 R-squared 0.00 0.12 0.02 0.10 0.02 0.10 0.17 0.07 0.13 Note: This table regresses bid/ask spreads for 1 week FX swap quotes based on high frequency data from Thomson Reuters Tick History database on dummies for Funding Heterogeneity (FundingHet) and dates when the 1 week contract crosses quarter-ends (Qends) for three currency pairs (EUR/USD, CHF/USD and JPY/USD). Column (1) to (3) depict the results from a sample that runs from January 1,2005 to December 31, 2007, while column (4) to (6) are based on a sample period that runs from January 1, 2008 to September 1, 2017. In column (7) to (9) the interaction terms between a dummy that takes the value 1 from January 1, 2005 and onwards (zero otherwise) and funding heterogeneity and quarter-ends are added. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 48 Table 7: Price volatility (1) (2) (3) (4) (5) (6) (7) (8) (9) Pre 2008 Post 2008 EUR CHF JPY EUR CHF JPY EUR CHF JPY FundingHet 0.04*** 0.03*** 0.07** 0.04*** 0.02*** 0.03*** 0.06*** 0.04*** 0.03*** (0.01) (0.01) (0.03) (0.01) (0.01) (0.00) (0.01) (0.01) (0.00) Qend 0.00 0.01 0.01* 0.04*** 0.03*** 0.05*** 0.02** 0.01 0.02** (0.00) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) post2015 -0.01*** 0.02*** 0.01*** (0.00) (0.01) (0.00) post2015 ×Qend 0.08*** 0.07*** 0.14*** (0.03) (0.03) (0.03) post2015 ×FundingHet -0.04*** -0.07*** -0.02*** (0.01) (0.01) (0.01) Constant 0.01*** 0.02*** 0.02*** 0.02*** 0.05*** 0.02*** 0.02*** 0.05*** 0.02*** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Observations 756 745 691 2,434 2,437 2,396 2,434 2,437 2,396 R-squared 0.05 0.03 0.05 0.05 0.02 0.08 0.08 0.05 0.15 Note: This table regresses daily price volatility for 1 week FX swap quotes based on high frequency data from Thomson Reuters Tick History database on dummies for Funding Heterogeneity (FundingHet) and dates when the 1 week contract crosses quarter-ends (Qends) for three currency pairs (EUR/USD, CHF/USD and JPY/USD). Column (1) to (3) depict the results from a sample that runs from January 1,2005 to December 31, 2007, while column (4) to (6) are based on a sample period that runs from January 1, 2008 to September 1, 2017. In column (7) to (9) the interaction terms between a dummy that takes the value 1 from January 1, 2005 and onwards (zero otherwise) and funding heterogeneity and quarter-ends are added. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 49 Table 8: Effect on order flow; Swap Lines (1) (2) (3) (4) OFpanel OFeur OFchf OFjpy SwapLines -0.21** -0.28* -0.04 -0.50* (0.11) (0.15) (0.19) (0.26) Constant 0.03 -0.14* 0.18*** 0.07 (0.04) (0.08) (0.06) (0.07) Observations 1,444 480 482 482 R-squared 0.02 0.06 0.01 0.04 Controls Yes Yes Yes Yes Note: This table illustrates the impact of quarter-end on for 1 week order flow. Standardized order flow OF is measuring the net buyer transactions of swapping euros, swiss francs and yen into dollars, and is sourced from TR D2000-2 inter-dealer trades for 1 Week FX swaps. SwapLine is a dummy variable that takes the value 1 on days when there was initial take up in any of the swap lines between the Fed and foreign central banks, zero otherwise. Controls include the changes in USD Libor-OIS spreads for 1 week and 3 month maturities, the VIX index, the USD Trade weighted exchange rate. Data is daily. The sample runs from Jan 1, 2008 to Dec 31, 2009. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 50 Table 9: Price adjustment quarter-end for EUR/USD, CHF/USD and JPY/USD EUR/USD CHF/USD JPY/USD Hour 2005-2007 2008-2013 2013-2015 2015-2017 2005-2007 2008-2013 2013-2015 2015-2017 2005-2007 2008-2013 2013-2015 2015-2017 -5 0 -0.5 0 0.3 -0.1 2.1 -0.2 -1.8 -0.2 0.1 -0.3 -0.3 -4 0.3 1 0 0.2 -0.2 1.8 -0.3 -0.1 0.2 1 -2.1 -0.3 -3 -0.1 -0.8 0.2 0.2 -0.1 -1 0.1 -3.3 -0.8 -0.9 -0.3 -0.5 -2 0.1 -0.2 2.1 1.7 0.2 -4.7 1.3 2.4 -0.8 2.2 -0.4 15.6 -1 0 0.9 0 9 1.3 2.7 -1.4 1.6 0.7 0.1 0.4 2.7 0 0.8 0.8 -1.6 22.7 0.2 2.3 -2.8 22.3 14.3 5.1 16.3 55.2 1 0.5 0 -0.4 6 -3.2 -3.2 2 16.2 -4.3 4.5 10.8 36 2 0.9 2.2 -0.3 0.8 0 1.1 -1.1 5.2 -1.2 4 -2.8 32.9 3 0.2 2.9 -0.2 0 -0.1 3.7 0.2 12.2 1.1 1.3 0.4 0 4 -0.1 0 -0.4 -0.7 0.4 -2 0.5 -2.2 -0.3 1 1.1 0.9 5 0.2 -0.4 0 0.1 -0.1 1.3 -0.7 -6.5 -0.2 0.2 0.7 10.3 Note: This table illustrates the hourly change in the FX swapped (synthetic) USD rate calculated from Libor from 5 hours before to 5 hours after the 1 week FX swap contract matures after quarter-end. 0 denotes the hour when the contract first matures after quarter-end. The numbers are in basis points and represent the average of all quarter-ends within the sample period. 51 Table 10: Effect on order flow; Quarter-end (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Pre 2008 Post 2008 OFpanel OFeur OFchf OFjpy OFpanel OFeur OFchf OFjpy OFpanel OFpanel Qend 0.14 -0.03 0.61*** -0.17 0.13*** 0.15* 0.16** 0.08 0.22*** 0.14** (0.10) (0.14) (0.18) (0.19) (0.05) (0.09) (0.07) (0.08) (0.08) (0.06) post2013 0.03 (0.03) post2013 ×Qend -0.29*** (0.10) post2015 -0.01 (0.02) post2015 ×Qend -0.02 (0.09) Constant 0.05** -0.04 0.11*** 0.07* 0.03** -0.04* 0.11*** 0.02 0.02 0.03** (0.02) (0.04) (0.04) (0.04) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) Observations 2,095 724 712 659 6,723 2,240 2,241 2,242 6,723 6,723 R-squared 0.01 0.01 0.03 0.04 0.01 0.01 0.00 0.01 0.01 0.01 Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Note: This table illustrates the impact of quarter-end on 1 week order flow. Standardized order flow OF is measuring the net buyer transactions of swapping euros, swiss francs and yen into dollars, and is sourced from TR D2000-2 inter-dealer trades for 1 Week FX swaps. Qend is a dummy variable taking the value 1 when the 1 week contract is settled prior to quarter-end and matures after quarter-end, zero otherwise. Post2013 and post2015 are dummy variables that are 1 from Jan 2013 and Jan 2015, respectively, zero otherwise. Controls include the changes in USD Libor-OIS spreads for 1 week and 3 month maturities, the VIX index, the USD Trade weighted exchange rate. Data is daily. The sample runs from Jan 1, 2008 to Sep 1, 2017. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 52 Table 11: Effect on order flow; Monetary Policy surprises (1) (2) (3) (4) OFpanel OFeur OFchf OFjpy ∆ois -1.86 -2.28 -1.18 13.86 (1.59) (3.84) (1.85) (10.07) Constant -0.05 0.02 -0.03 -0.09 (0.05) (0.07) (0.11) (0.09) Observations 363 136 87 122 R-squared 0.01 0.04 0.02 0.09 Controls Yes Yes Yes Yes Post2008 Yes Yes Yes Yes Note: This table illustrates the impact of monetary policy surpirses on 1 week order flow. Standardized order flow OF is measuring the net buyer transactions of swapping euros, swiss francs and yen into dollars, and is sourced from TR D2000-2 inter-dealer trades for 1 Week FX swaps. ∆OIS is the 30 min change in the 1-month OIS rate (Overnight Index Swaps - a proxy for the risk free rate) in the respective currency around the central bank policy announcement. Monetary announcements in EUR, CHF, JPY and USD are considered. In the case of US announcements the sign of the change in the OIS is switched so that a positive change in the OIS always proxy an increase in the interest rate differential towards the US (i.e foreign currency rate minus the US rate). Controls include the changes in USD Libor-OIS spreads for 1 week and 3 month maturities, the VIX index, the USD Trade weighted exchange rate. Data is daily. The sample runs from Jan 1, 2008 to Sep 1, 2017. *** denotes significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level. 53