Divergent monetary policies and international dollar credit: Evidence from bank-level data

He, Dong; Wong, Eric; Ho, Kelvin; Tsang, Andrew Working Paper Divergent monetary policies and international dollar credit: Evidence from bank-level data ADBI Working Paper, No. 741 Provided in Cooperation with: Asian Development Bank Institute (ADBI), Tokyo Suggested Citation: He, Dong; Wong, Eric; Ho, Kelvin; Tsang, Andrew (2017) : Divergent monetary policies and international dollar credit: Evidence from bank-level data, ADBI Working Paper, No. 741, Asian Development Bank Institute (ADBI), Tokyo This Version is available at: https://hdl.handle.net/10419/163240 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/3.0/igo/ ADBI Working Paper Series DIVERGENT MONETARY POLICIES AND INTERNATIONAL DOLLAR CREDIT: EVIDENCE FROM BANK-LEVEL DATA Dong He, Eric Wong, Kelvin Ho, and Andrew Tsang No. 741 May 2017 Asian Development Bank Institute The Working Paper series is a continuation of the formerly named Discussion Paper series; the numbering of the papers continued without interruption or change. ADBI’s working papers reflect initial ideas on a topic and are posted online for discussion. ADBI encourages readers to post their comments on the main page for each working paper (given in the citation below). Some working papers may develop into other forms of publication. ADB recognizes “Hong Kong” as Hong Kong, China. Suggested citation: He, D., E. Wong, K. Ho, and A. Tsang. 2017. Divergent Monetary Policies and International Dollar Credit: Evidence from Bank-Level Data. ADBI Working Paper 741. Tokyo: Asian Development Bank Institute. Available: https://www.adb.org/publications/divergent-monetarypolicies-and-international-dollar-credit Please contact the authors for information about this paper. Email: [email protected]; [email protected]; [email protected]; [email protected] The views expressed in this paper are those of the authors, and do not necessarily reflect those of the International Monetary Fund, the Hong Kong Monetary Authority, the Hong Kong Institute for Monetary Research, its Council of Advisers, or the Board of Directors. Dong He is deputy director of the Monetary and Capital Markets Department at the International Monetary Fund. Eric Wong is senior manager and Kelvin Ho is manager at the Research Department of the Hong Kong Monetary Authority. Andrew Tsang is manager at the Hong Kong Institute for Monetary Research. The views expressed in this paper are the views of the author and do not necessarily reflect the views or policies of ADBI, ADB, its Board of Directors, or the governments they represent. ADBI does not guarantee the accuracy of the data included in this paper and accepts no responsibility for any consequences of their use. Terminology used may not necessarily be consistent with ADB official terms. Working papers are subject to formal revision and correction before they are finalized and considered published. Asian Development Bank Institute Kasumigaseki Building, 8th Floor 3-2-5 Kasumigaseki, Chiyoda-ku Tokyo 100-6008, Japan Tel: +81-3-3593-5500 Fax: +81-3-3593-5571 URL: www.adbi.org E-mail: [email protected] © 2017 Asian Development Bank Institute ADBI Working Paper 741 He, Wong, Ho, and Tsang Abstract This paper uses a comprehensive and detailed bank-level data set to study how the divergence of central bank balance sheet policy in the US vis-à-vis the euro area and Japan affects the supply of international US dollar loans by global banks. Our empirical findings support the view that the contractionary effect of US monetary normalization on global dollar liquidity would be offset by an expansionary effect from a continued supply of US dollar loans by euro area and Japanese banks. The net effect, however, is crucially dependent on the stability of global foreign exchange markets and investor perceptions of the default risks of global banks. The analysis shows that US monetary policy shocks are one of the most important explanatory variables for the deviations from the covered interest rate parity (CIP) in the major foreign exchange (FX) markets. We also demonstrate a tail risk scenario of the contraction of the supply of international US dollar loans if and when the US monetary normalization coincides with a dislocation of the FX swap market and a rise of bank default risks. Our results are robust to alternative model specifications and different data sets. JEL Classification: E44, E52, E58, E63 ADBI Working Paper 741 He, Wong, Ho, and Tsang Contents 1. INTRODUCTION ......................................................................................................... 1 2. THEORETICAL DISCUSSIONS.................................................................................. 3 3. EMPIRICAL MODELS AND DATA .............................................................................. 4 3.1 The Baseline Model ......................................................................................... 4 3.2 Extended Models ............................................................................................. 5 3.3 The HKMA Data Set ........................................................................................ 7 4. EMPIRICAL FINDINGS ............................................................................................... 8 4.1 The Net Impact of Divergence of BSPs on the Supply of International Dollar Loans ......................................................................... 10 5. ROBUSTNESS ANALYSIS ....................................................................................... 16 5.1 Robustness Analysis using the BIS Data Set ................................................ 16 5.2 Other Robustness Checks ............................................................................ 18 6. CONCLUSION .......................................................................................................... 19 REFERENCES ..................................................................................................................... 20 APPENDICES 1 Details for the Theoretical Discussions ......................................................... 23 2 Description of Variables ................................................................................ 25 3 Economic Significance of the Impact of Balance Sheet Factors on ∆ L* ........ 26 4 Detailed Description of the Monte Carlo Simulations .................................... 27 ADBI Working Paper 741 He, Wong, Ho, and Tsang 1. INTRODUCTION The US dollar is the premier currency for international trade and investment. According to statistics from the Bank for International Settlements (BIS), around half of international claims by banks were US dollar denominated at the end of 2015 (Figure 1). The supply of international dollar credit1 is largely influenced by the behavior of non-US international banks, particularly those headquartered in Europe and Japan (McCauley, McGuire and Sushko 2015; Ivashina, Scharfstein and Stein 2015), as they provide the lion’s share of international dollar credit. Figure 1: US Dollar International Claims by Nationality of Banks Notes: 1. The claims are vis-à-vis all sectors and include interoffice bank claims. 2. US dollar international claims include US dollar cross-border claims and local credit extended in US dollars in countries other than the US. 3. European banks include those headquartered in Belgium, France, Germany, Italy, the Netherlands, Spain, Sweden, Switzerland, and the UK. Source: BIS locational banking statistics (by nationality). The strong presence of European and Japanese banks in the international dollar loan market raises interesting questions about the role of their respective home central banks relative to that of the US Federal Reserve (Fed) in influencing global dollar liquidity. For example, how does a divergence of central bank balance sheet policies (BSPs) in the US vis-à-vis the euro area and Japan affect the supply of international dollar credit? Policymakers are particularly concerned about the potential disruption of global liquidity arising from the Fed’s monetary normalization. Indeed, as evidenced during the 2007–08 Global Financial Crisis (GFC), a global shortage of US dollar liquidity contributed to a significant tightening of global financial conditions, hampering economic activities not only for advanced economies but also for emerging market economies where the dollar is used extensively to finance domestic economic activities. More recently, the expected tapering of the Fed’s large-scale asset purchase 1 Throughout this paper, “international dollar credit” refers to US dollar-denominated credit by banks to nonbanks outside the US. 1 ADBI Working Paper 741 He, Wong, Ho, and Tsang program in 2013 also induced instability at the global level (Aizenman, Mahir and Hutchison 2014; Eichengreen and Gupta 2014; Mishra et al. 2014). However, there is a counterargument that the Fed’s monetary normalization will not necessarily lead to a significant contraction in the supply of international dollar loans if the Bank of Japan (BOJ) and the European Central Bank (ECB) continue to expand their balance sheets through asset purchase programs. At the heart of this argument is that with the ample home-currency liquidity provided by their respective central banks, Japanese and euro area banks can fund their international dollar loans continuously through foreign exchange (FX) or cross-currency swaps. In principle, this can narrow or close the US dollar liquidity gap arising from the Fed’s monetary normalization. This paper attempts to answer this important policy question by empirically investigating the net impact of the divergence of central bank BSPs on the supply of international dollar loans through the bank lending channel. Drawing on the theoretical work by Ivashina, Scharfstein and Stein (2015), we specify our empirical models to study how a global bank’s supply of international dollar loans would be affected by central bank BSPs in the US and its home country, the functioning of the FX swap market, the bank’s default risk, and balance sheet characteristics. Since our primary objective is to understand how these factors affect the supply of international dollar loans, we apply the fixed-effects approach advocated by Khwaja and Mian (2008)2 to disentangle the demand-side effect. We carry out the empirical study using a unique confidential panel data set from the Hong Kong Monetary Authority (HKMA) of US dollar loans of foreign banks in Hong Kong, China. We conduct various robustness checks, including re-estimating the empirical models using an alternative confidential data set from the BIS. Our empirical findings suggest that from a global bank’s perspective, the expansion of central bank balance sheets in the US and in the home country would produce expansionary effects on its supply of international dollar loans. This finding is consistent with the evidence of monetary policy spillovers on cross-border bank capital flows through a risk-taking channel provided by Bruno and Shin (2015). The functioning of the FX swap market and bank default risk are also found to be significant determinants of the supply of international dollar loans, which is in line with the findings of Baba and Packer (2009) and McGuire and von Peter (2009) that the impairment of the FX swap market and heightened default risk of global banks contributed to a prolonged global US dollar shortage during the GFC. Finally, we find that global banks’ risk-taking attitude, credit risk exposure, and the business model of their overseas branches matter for how central banks’ BSPs are transmitted internationally. This finding echoes the conclusion of Brunnermeier et al. (2012) that the financial and organizational structure of global banks plays a vital role in transmitting imbalances of cross-border funding flows. On the net impact of divergent central bank BSPs on the supply of international dollar loans, we find that the expansionary effect of continued asset purchases by the ECB and BOJ would offset the contractionary effect of US monetary normalization. The net effect, however, is crucially dependent on whether normalization of monetary policy in the US coincides with risk aversion by global investors and serious dislocation of the foreign exchange markets. Specifically, our tail risk analysis shows that the supply of 2 This approach identifies the supply effect using a special data set that contains loan data on multiplebank firms. By using firm-specific fixed effects to control for the change in loans of a firm from the preand post-event periods of liquidity shocks, any differences in loans provided to the same firm by different banks are attributable to the supply effect. See recent studies by Cetorelli and Goldberg (2011) and Aiyar et al. (2014). 2 ADBI Working Paper 741 He, Wong, Ho, and Tsang international US dollar credit could decline sharply if FX swap market dislocation leads to a sharp spike in the deviation from CIP conditions and the perceived banks’ default risks increase significantly. Indeed, estimates for the sample period show that US monetary policy shocks were one of the most important explanatory variables for the deviations from the CIP conditions in the markets that trade yen and euro against the US dollar. This paper contributes to the literature on the international transmission of financial shocks through the bank lending channel. Early studies include Peek and Rosengren (2000), who examine the effect of the bursting of the asset bubble in Japan in the early 1990s on the loan supply of Japanese banks in the US commercial real estate market. Chava and Purnanandam (2011) and Schnabl (2012) examine the effect of the 1998 Russian crisis on the supply of bank loans in the US and Peru, respectively. More recent studies focus on the transmission of funding stress during the GFC through the balance sheets of global banks (Cornett et al. 2011; Cetorelli and Goldberg 2011, 2012a, and 2012b; Buch and Goldberg 2015; Ivashina, Scharfstein and Stein 2015). A few recent studies examine how unconventional monetary policies (UMPs) are transmitted through the bank lending channel. However, they mainly focus on the impact on the domestic economy (Bowman et al. 2011; Joyce and Spaltro 2014). Cross-border transmission of UMPs through the banking channel remains an underexplored research topic (McCauley, McGuire and Sushko 2015).3 The rest of the paper is organized as follows. We first specify one channel through which central bank BSPs affect the supply of international dollar loans of global banks in Section 2 to support our regression specifications. Section 3 specifies the empirical models and describes the HKMA data set. Section 4 presents the empirical findings, while Section 5 conducts robustness checks. Section 6 concludes. 2. THEORETICAL DISCUSSIONS This section discusses the transmission of central bank BSPs to global banks’ supply of international dollar loans. Starting with a similar theoretical framework by Ivashina, Scharfstein and Stein (2015), we take the position of a global bank that provides home-country currency loans (L) in the local market and US dollar loans (L*) in the international market, with decreasing marginal returns for both L and L*. The bank has an initial amount of costless home-currency funding denoted by D and dollar funding denoted by D*. The bank can raise additional home-currency and dollar funding in the respective markets by any amount denoted by F and F*, respectively, but incurring increasing marginal costs. The bank cannot take any FX risk. So, for any level of L* exceeding D*, the bank needs to acquire dollar funding in the US (i.e. F*) or convert its home-currency funding into US dollars in the FX swap market by paying a swap cost (w). We denote the amount of swap funding by S. Following Ivashina, Scharfstein and Stein (2015), it is assumed that the bank has a default probability p and that it cannot pay off all its debt if it defaults. We further assume that only home-currency funding is insured. As a result, fund providers in the US will demand a risk premium equivalent to p to compensate the bank’s default risk. 3 See also He and McCauley (2013). There is another stand of literature focusing on the impact of UMPs on financial markets. D’Amico and King (2013) study the stock and flow effects of the Fed’s 2009 asset purchase program on the yield curve. Chen et al. (2012 and 2016) find that expansionary central bank balance sheet policies affect a broad range of asset prices in emerging markets. Fratzscher, Duca and Straub (2013) find that the Fed’s UMP has a significant spillover effect on financial markets in EMEs through a portfolio balancing channel. Neely (2015) and Bauer and Neely (2014) find sizable effects of the Fed’s UMP on sovereign yields in advanced economies. 3 ADBI Working Paper 741 He, Wong, Ho, and Tsang Based on the above setting, we can identify one channel through which central bank BSPs affect global banks’ supply of international dollar loans. Using the BOJ’s quantitative and qualitative program as an example to illustrate, when the BOJ purchases Japanese government bonds from a firm that has a bank account in a Japanese bank, the proceeds of the purchase will be reflected initially in the Japanese bank’s liability side as “current deposits,” while its asset side also expands by the same amount in “reserves at the central bank.” From the vantage point of the Japanese bank, the BOJ’s bond purchase could be taken as an exogenous positive shock on D. On the funding side, the bank will react by selecting less expensive home-currency funding by substituting some costly homecurrency funding F with D, leading to a lower marginal cost of F. On the asset side, the lower marginal cost of F induces the bank to increase its home-currency loans (L) until the marginal return of L equates to the marginal cost of F (which is lower now). Since F can alternatively finance dollar loans through the FX swap market, the lower marginal cost of F also implies that the bank increases its L* in equilibrium. Finally, the lower marginal cost of F leads the bank to substitute part of F* with S to finance L*. By the same logic, an expansion of the Fed’s balance sheet (i.e. a positive shock on D*) can be shown to produce an expansionary effect on both L and L*. It can also be shown that a higher w and p, which implies a higher US dollar funding cost, would reduce the supply of L* (see Appendix 1 for details). The above example shows that a global bank transmits central bank BSPs internationally driven by its profit-maximization decisions on cross-border loan allocations, which is consistent with the evidence of monetary policy spillovers on cross-border bank capital flows through a risk-taking channel demonstrated by Bruno and Shin (2015). More broadly, the mechanism described above is consistent with the consensus view that monetary policy affects the supply of credit by financial intermediaries through both the funding channel and the risk-taking channel (IMF 2015, 2016). It is true that excess reserves held with the central bank are an asset item on the balance sheet of the banking system, and do not affect the liability structure of the banks directly. However, the change in excess reserves as a result of asset purchases by the central bank would lead to changes in both the term and risk premiums, thereby affecting the broader funding conditions and the risk-bearing capacity of financial intermediaries. Therefore, a monetary policy shock as measured by an exogenous change in the size of the central bank balance sheet can be interpreted as an exogenous shock to the bank’s funding cost. 3. EMPIRICAL MODELS AND DATA 3.1 The Baseline Model We follow the discussion in the previous section to specify a baseline regression model by equation (1). The model will be estimated using the confidential panel data set from the HKMA, which records quarterly flows of US dollar-denominated loans of foreign bank branches in Hong Kong, China vis-à-vis more than 70 destination countries. ∆ L*ijt = β 1 ∆ HCBjt + β 2 ∆ FEDt *USFj + β 3 ∆ CDSjt + β 4 ∆ CIPjt-1 + β 5 ∆ GDPjt + µ it + ε ijt (1) 4 ADBI Working Paper 741 He, Wong, Ho, and Tsang between the Fed and the BOJ is likely to increase the cost of swapping yen for US dollars. Ignoring this potential impact on the swap cost would overestimate the supply of international dollar loans. We employ vector autoregressive (VAR) analysis to estimate first-order VAR models for two country pairs (i.e. US-Japan and US-euro area) to aid the identification of exogenous shocks of central bank BSPs in the US, the euro area, and Japan, and the empirical responses of ∆ CDS and ∆ CIP to the central bank BSP shocks. Each VAR model includes ten variables. Taking the US-Japan pair as an example, each country contains two macroeconomic variables (real industrial outputs and the consumer price index) and the central bank balance sheet. Their first differences of log seasonally adjusted time series are used in the estimations (denoted respectively by ∆𝑦𝑡 𝑈𝑆, ∆𝜋𝑡 𝑈𝑆, and ∆𝐹𝐸𝐷𝑡 for the US; and ∆𝑦𝑡 𝐽𝑃, ∆𝜋𝑡 𝐽𝑃, and ∆𝐻𝐶𝐵𝑡 𝐽𝑃 for Japan). We follow Gambacorta, Hofmann and Peersman (2014) to include the change in the VIX index (∆𝑉𝐼𝑋𝑡) in the model, arguing that central bank BSPs may react to financial market risks. The remaining three variables are financial market variables, namely the change in the cost of swapping yen for US dollars (∆𝐶𝐼𝑃 𝑡 𝐽𝑃), the change in the average CDS spread for major Japanese banks (∆𝐶𝐷𝑆𝑡 𝐽𝑃), and the change in the spot exchange rate of yen per US dollar (∆𝐸𝑋𝑅𝑡 𝐽𝑃). These variables are separated into US and Japanese blocks with the following ordering: {∆𝑦𝑡 𝑈𝑆 , ∆𝜋𝑡 𝑈𝑆 , ∆𝑉𝐼𝑋𝑡, ∆𝐹𝐸𝐷𝑡} and {∆𝑦𝑡 𝐽𝑃 , ∆𝜋𝑡 𝐽𝑃 , ∆𝐶𝐼𝑃 𝑡 𝐽𝑃, ∆𝐶𝐷𝑆𝑡 𝐽𝑃, ∆𝐸𝑋𝑅𝑡 𝐽𝑃, ∆𝐻𝐶𝐵𝑡 𝐽𝑃 }. The ordering of variables is largely consistent with the literature on monetary policy shock identification.11 To reduce the number of parameters needed to estimate, we follow Cushman and Zha (1997) to impose block exogeneity restrictions in the estimation such that shocks in the US block are assumed to have effects on variables in the Japanese block but any shock from Japan has no effects on the US. Apart from this statistical consideration, imposing the exogeneity restrictions can allow us to obtain an identical identification of US monetary policy shocks in the US-Japan and US-euro area models so that our estimates are self-consistent. We also conduct a robustness check by relaxing the exogeneity restrictions (see the next section). The model is estimated using the seemingly unrelated regression method with monthly data from August 2007 to December 2015. Once the model is estimated, the recursive identification scheme (i.e. the Cholesky decomposition of the variance-covariance matrix of the reduced-form disturbances) is adopted to identify the central bank BSP shocks with the mentioned ordering of the variables. The US-euro area model is also specified and estimated in a similar fashion. Figure 3 presents the estimated central bank BSP shocks for the Fed, the BOJ, and the ECB (in Panels A to C, respectively). The sizes of the BSP shocks measured by the one-year standard deviation are estimated to be 14.4%, 6.3%, and 8.5% for the Fed, the BOJ, and the ECB, respectively. These compare with 26.1%, 18.3%, and 13.3% for their respective average annual growth rate of balance sheets since the implementation of their BSPs after the GFC. This indicates a significant difference between the changes in central banks’ balance sheets observed and the exogenous central bank BSP shocks. 11 See Gambacorta, Hofmann and Peersman (2014) and Chen et al. (2016). 11 ADBI Working Paper 741 He, Wong, Ho, and Tsang Figure 3: Changes on Central Bank Balance Sheets and Identified BSP Shocks Sources: Board of Governors of the Federal Reserve System, Bank of Japan, the European Central Bank, IMF International Financial Statistics, and authors’ estimates. 12 ADBI Working Paper 741 He, Wong, Ho, and Tsang Our VAR estimations also show that financial market variables are responsive to central bank BSPs. In particular, Table 3 summarizes the variance decompositions for ∆𝐶𝐼𝑃𝑡 for the US-Japan and US-euro area models. The US central bank BSP shock is found to be the most important factor in explaining the forecast error variance for ∆𝐶𝐼𝑃𝑡, followed by its own shocks. Home-country central bank BSP shocks, however, are found to give little explanatory power. Both models show a similar picture. Table 3: Variance Decomposition Analysis for ∆CIP Decomposition of total variance of the forecast error for ∆CIP for the US-Japan VAR model Period ∆𝑽𝑰𝑿𝒕 ∆𝑭𝑬𝑫𝒕 ∆𝑪𝑰𝑷𝒕 𝑱𝑷 ∆𝑯𝑪𝑩𝒕 𝑱𝑷 Others 1 8.4 28.2 53.0 0.0 10.4 2 12.0 35.7 38.9 0.0 13.3 3 11.2 38.5 35.2 0.9 14.2 4 11.1 38.0 34.8 0.9 15.3 5 11.0 38.3 34.4 1.0 15.4 6 11.0 38.3 34.3 1.0 15.4 7 11.0 38.3 34.3 1.0 15.4 8 11.0 38.3 34.3 1.0 15.4 9 11.0 38.3 34.3 1.0 15.4 10 11.0 38.3 34.3 1.0 15.4 11 11.0 38.3 34.3 1.0 15.4 12 11.0 38.3 34.3 1.0 15.4 Note: Figures represent the percentage share of the total variance of the forecast error for ∆CIP attributable to the variance of each structural shock. Decomposition of total variance of the forecast error for ∆CIP for the US-euro area VAR model Period ∆𝑽𝑰𝑿𝒕 ∆𝑭𝑬𝑫𝒕 ∆𝑪𝑰𝑷𝒕 𝑬𝑼 ∆𝑯𝑪𝑩𝒕 𝑬𝑼 Others 1 24.0 40.0 29.6 0.0 6.4 2 19.3 32.9 26.6 3.7 17.5 3 16.5 37.4 22.5 3.4 20.3 4 15.6 37.1 21.5 3.5 22.3 5 15.4 36.6 21.2 3.8 23.0 6 15.3 36.5 21.2 3.8 23.2 7 15.3 36.5 21.1 3.9 23.2 8 15.3 36.5 21.1 3.9 23.3 9 15.3 36.4 21.1 3.9 23.3 10 15.3 36.4 21.1 3.9 23.3 11 15.3 36.4 21.1 3.9 23.3 12 15.3 36.4 21.1 3.9 23.3 Note: Figures represent the percentage share of the total variance of the forecast error for ∆CIP attributable to the variance of each structural shock. 13 ADBI Working Paper 741 He, Wong, Ho, and Tsang The finding that a significant part of the forecast error variance of ∆𝐶𝐼𝑃𝑡 can be explained by its own shocks is consistent with the finding by Sushko et al. (2016) that there remains a significant part of the CIP deviation that cannot be explained by factors identified in the literature (e.g. crisis and banks' default risks). In order to estimate the net impact of divergence of central bank BSPs on the supply of international dollar loans of the euro area and Japanese banks, we conduct Monte Carlo (MC) simulations based on the estimated VAR models and the identified central bank BSP shocks. Appendix 4 details the procedure of the MC simulations. For the case of Japanese banks, we impose divergent central bank BSP shocks in the VAR model by considering simultaneously negative shocks to the Fed’s balance sheet and positive shocks to the BOJ’s balance sheet. We then simulate the distributions for the four determinants of the supply of international dollar loans (i.e. ∆ HCBjt, ∆ FEDt, ∆ CIPjt-1, and ∆ CDSjt). Table 4 presents their distributional statistics for the cumulative 12-month changes from December 2015 based on 10,000 simulation trials. Although we impose negative shocks on ∆ FEDt, the simulation results show that the Fed’s balance sheet may expand or contract, depending on the simulated movements of other factors. By contrast, the balance sheet of the BOJ would be more likely to increase with an average growth rate of 17.7%. For the cost of swapping yen for US dollars ( ∆ CIPjt-1), the simulated distribution shows that it is more likely to increase than decrease, which is consistent with research findings that the current divergent monetary policy environment could push up the dollar funding cost in cross-currency funding markets (Iida, Kimura and Sudo 2016; Sushko et al. 2016). Finally, the direction of change in the default risk of Japanese banks is somewhat uncertain based on the simulated distribution for ∆ CDSjt. Table 4: Distributions of Simulated Changes for Key Variables that Affect the Supply of International Dollar Loans of Japanese Bank Branches under the Scenario of Divergence of Central Bank BSP Shocks Distributional Statistics ∆FED t (%) ∆HCB t (%) ∆CDS jt (bps) ∆CIP jt-1 (bps) 90th percentile 31.7 28.3 72.2 76.3 75th percentile (upper quartile) 19.8 23.2 43.6 55.7 Median 6.5 17.7 12.2 32.0 25th percentile (lower quartile) –6.3 12.3 –18.2 8.6 10th percentile –17.7 7.0 –46.5 –12.7 Mean 6.6 17.7 12.7 32.1 S.d. 19.3 8.3 46.2 34.4 We further decompose the contribution of these factors to the supply of ∆ L*ijt of Japanese banks by using the estimation results for Model 4 in Table 2. The decomposition results are presented in the upper panel of Table 5. We first focus on the lower and upper quartile estimates in order to reveal the expected range of contributions to the growth of international dollar loans of different factors. For the contribution of ∆ FEDt, the lower and upper quartile estimates are found to be –1.3 and 4 percentage points respectively, suggesting that ∆ FED may increase or reduce the supply of ∆ L* of Japanese banks. The possible contractionary effect of the Fed’s BSP on the supply of ∆ L*, however, would be offset by the expansionary effect of the BOJ’s BSP, as the contribution of ∆ HCB to the supply of ∆ L* is found to be positive for 14 ADBI Working Paper 741 He, Wong, Ho, and Tsang both the lower and upper quartile estimates (i.e. 3.9 and 7.4 percentage points, respectively). It is worth pointing out that the net impact would be largely dependent on financial market responses, particularly in the FX swap market. Specifically, the swap cost could contribute to a significant decline in the supply of dollar loans of Japanese banks, as both lower and upper quartile estimates are negative (i.e. –0.7 and –4.3 percentage points). Indeed, the last column of the upper panel of Table 5, which shows the estimates for the combined contributions of the four factors to changes in international dollar loans, 12 confirms this conjecture, as the estimated distribution shows that the supply of ∆ L* of Japanese banks could increase or decrease. Table 5: Distributions of Estimated Contribution of Factors to the Growth of International Dollar Loans of Japanese Bank Branches and Tail Risk Estimates under the Scenario of Divergence of Central Bank BSP Shocks Contribution to the Growth of Japanese Banks’ Dollar Loans ∆FEDt*USFj (%) ∆HCBjt (%) ∆CDSjt (%) ∆CIPjt-1 (%) ∆Loanijt (%) Upper panel 90th percentile 6.3 9.1 4.7 1.0 12.2 75th percentile (upper quartile) 4.0 7.4 1.8 –0.7 8.0 Median 1.3 5.7 –1.2 –2.5 3.3 25th percentile (lower quartile) –1.3 3.9 –4.4 –4.3 –1.6 10th percentile –3.6 2.3 –7.3 –5.9 –5.9 Mean 1.3 5.7 –1.3 –2.5 3.2 S.d. 3.9 2.6 4.7 2.7 7.1 Lower panel Tail risk estimate –3.3 3.5 –4.6 –4.2 –8.6 We also analyze the tail risk by estimating how these factors might contribute to an extreme decline in the supply of ∆ L* of Japanese bank branches in Hong Kong, China. We measure the tail risk by an expected shortfall estimate defined as the average estimated credit growth in the worst 10% of the 10,000 simulation trials. Among the worst 10% of trials, we compute the average changes of the factors and their contributions to the supply of ∆ L* of Japanese banks. We present the expected shortfall estimate in the lower panel of Table 5, which shows that the supply of dollar loans of Japanese banks could fall by 8.6%. Although the contractionary effect of the Fed’s BSP would be offset by the expansionary effect of the BOJ’s BSP (i.e. –3.3% vs 3.5%), the rising default risks for Japanese banks and the swap cost would lead to a significant decline in the supply of international dollar loans of Japanese bank branches in Hong Kong, China. We repeat the same analysis for euro area bank branches and report the results in Tables 6 and 7. The results are qualitatively similar to those for Japanese bank branches reported in Tables 4 and 5. 12 For the distributional statistics for the combined contribution estimates, we first drive a combined contribution estimate for each simulation trial by adding up the contributions of the four factors to dollar loan growth in that trial. We then use the 10,000 combined contribution estimates to obtain the distributional statistics. 15 ADBI Working Paper 741 He, Wong, Ho, and Tsang Table 6: Distributions of Simulated Changes for Key Variables that Affect the Supply of International Dollar Loans of Euro Area Bank Branches under the Scenario of Divergence of Central Bank BSP Shocks Distributional Statistics ∆FED t (%) ∆HCB t (%) ∆CDS jt (bps) ∆CIP jt-1 (bps) 90th percentile 31.7 29.0 61.9 64.0 75th percentile (upper quartile) 19.8 22.6 32.2 38.4 Median 6.5 16.0 –1.6 9.0 25th percentile (lower quartile) –6.3 9.5 –35.5 –19.1 10th percentile –17.7 3.5 –66.1 –44.9 Mean 6.6 16.1 –1.7 9.5 S.d. 19.3 9.9 50.1 42.4 Table 7: Distributions of Estimated Contribution of Factors to the Growth of International Dollar Loans of Euro Area Bank Branches and Tail Risk Estimates under the Scenario of Divergence of Central Bank BSP Shocks Factor Contribution ∆FED t *USF j (%) ∆HCB jt (%) ∆CDS jt (%) ∆CIP jt-1 (%) ∆Loan ijt (%) Upper panel 90th percentile 4.8 9.3 6.7 3.5 14.3 75th percentile (upper quartile) 3.0 7.2 3.6 1.5 10.2 Median 1.0 5.1 0.2 –0.7 5.5 25th percentile (lower quartile) –1.0 3.0 –3.2 –3.0 0.9 10th percentile –2.7 1.1 –6.3 –5.0 –3.1 Mean 1.0 5.1 0.2 –0.7 5.6 S.d. 2.9 3.2 5.1 3.3 6.8 Lower panel Tail risk estimate –0.3 5.6 –4.4 –5.1 –4.2 Taken together, the empirical findings in this section point to the same conclusion: The contractionary effect of US monetary normalization on global liquidity would be offset by an effect of central bank balance sheet expansion in Japan and the euro area. The net effect, however, is crucially dependent on whether the US monetary normalization coincides with risk aversion by global investors and leads to serious financial market dislocations. Specifically, our tail risk analysis shows that there remains a small risk that the supply of international US dollar credit will decline sharply if dislocations in FX swap markets occur and banks’ default risks increase as the US normalizes its monetary policy. 5. ROBUSTNESS ANALYSIS 5.1 Robustness Analysis using the BIS Data Set Our first robustness test re-estimates the baseline model using a confidential data set from the BIS. Based on the estimation results for the BIS data set, we then obtain the tail risk estimates for Japanese and euro area banks to assess the extent to which the main conclusion drawn in the final paragraph of the previous section is sensitive to an alternative data set. 16 ADBI Working Paper 741 He, Wong, Ho, and Tsang The BIS data set is constructed from the locational banking statistics by nationality. The BIS recently refined the data collection exercise and as a result, since June 2012, a breakdown of the statistics by 12 core global bank nationalities has been available for the BIS quarterly data on dollar-denominated external claims vis-à-vis 76 counterparty countries. The breakdown by reporting bank nationality makes it possible to identify the effect of central bank BSP in the home country on the supply of cross-border dollar credit by global banks. One advantage of the BIS data set is that it covers a major part of the aggregate position of reporting banks for the BIS location statistics, which is by far the most comprehensive data set available for analyzing international dollar loans. However, there are some caveats for the analysis using the BIS data set. First, the sample period of the BIS data set is short (i.e. our sample period covers seven time points only from June 2012 to March 2014), although there are a sufficiently large number of observations (more than 4,000). Second, we cannot analyze precisely international flows of dollar-denominated cross-border loans as we did in the previous section, as the exact variable is not available from the BIS data set. Only dollar-denominated cross-border claims that contain much broader assets than loans are available from the BIS data set. Finally, since the BIS data set is only available at the aggregate level by nationality of banks, we cannot consider bank-specific balance sheets as determinants of the transmission of central bank BSPs. Therefore we can only estimate the baseline model specified in equation (1) for the BIS data set. These caveats together suggest that the estimation results using the BIS data set could be significantly different from those using the HKMA data set. Table 8 presents the estimation results using the BIS data set, which are broadly in line with the discussion in Section 2, as the estimated coefficients are statistically significant and with the expected signs. This suggests that the baseline model specification has adequate explanatory power on the aggregate flow of international dollar loans. Table 8: Estimation Result for the BIS Data Set Variable ∆HCBjt 0.48*** (3.21) ∆FEDt*USFj 4.10*** (2.67) ∆CDSjt –9.12* (–1.86) ∆CIPjt-1 –23.97** (–2.11) ∆GDPjt –1.00* (–1.80) Country-time fixed effects for destination country i Yes R-squared 0.13 RMSE 0.58 No. of observations 9,161 Notes: 1. j = home country j, i = destination country i. 2. Figures in parentheses are t-statistics. 3. Standard errors are clustered by home country and destination country. 4. ***, **, and * respectively indicate significance at the 1%, 5%, and 10% level. 17 ADBI Working Paper 741 He, Wong, Ho, and Tsang Based on the results presented in Table 8, we obtain the tail risk estimates of credit growth for euro area and Japanese banks with the same procedure described in the previous section. The estimation results are presented in Table 9. The results are found to be qualitatively similar to those reported in the previous section, and the conclusion that there remains a tail risk that financial market responses to the divergence of central bank BSPs could lead to a sharp decline in the supply of international dollar loans remains. However, quantitatively, the FX swap cost is found to become an even more important contributor to the tail risk for the estimations using the BIS data set than for those using the HKMA data set. Table 9: Tail Risk Estimates based on Estimation Results from the BIS Data Set Tail Risk Estimates ∆FED t *USF j (%) ∆HCB jt (%) ∆CDS jt (%) ∆CIP jt-1 (%) ∆Loan ijt (%) Japanese banks –15.2 4.3 –0.6 –13.5 –25.1 Euro area banks –2.5 8.1 –4.0 –15.2 –13.5 5.2 Other Robustness Checks Apart from the above robustness analysis using the BIS data set, we also conduct the following analysis using the HKMA data set. First, we add shadow policy rates to the VAR models, arguing that shadow policy rates may contain different information from central banks’ balance sheets in respect of the unconventional monetary policy stance. We obtain our shadow policy rate estimates from Lombardi and Zhu (2014). More specifically, we employ updated Lombardi-Zhu shadow rate estimates for the US and preliminary estimates for the euro area and Japan from the authors to conduct the analysis. Panel A of Table 10 presents the tail risk estimates, which are found to be similar to those presented in Tables 5 and 7. Second, we relax the block exogeneity restrictions in the VAR models using the same recursive scheme with the same ordering of variables to identify monetary policy shocks. The tail risk estimates (see Panel B of Table 10) turn out to be qualitatively similar to those presented in the previous section. Table 10: Robustness Tests: Tail Risk Estimates for Alternative Model Specifications Tail Risk Estimates ∆FED t *USF j (%) ∆HCB jt (%) ∆CDS jt (%) ∆CIP jt-1 (%) ∆Loan ijt (%) Panel A (inclusion of shadow policy rates in the VAR models) Japanese banks –2.9 4.6 –4.2 –4.2 –6.7 Euro area banks –1.8 4.3 –9.1 –4.6 –11.1 Panel B (relaxing block exogeneity restrictions in the VAR models) Japanese banks –0.4 6.4 –4.8 –5.3 –4.2 Euro area banks –2.8 4.8 –5.5 –5.1 –8.6 18 ADBI Working Paper 741 He, Wong, Ho, and Tsang 6. CONCLUSION Our findings show that although continued asset purchases by the ECB and BOJ are expected to cushion the negative impact of US monetary policy normalization on the supply of international dollar credit, the Fed’s monetary policy shocks are found to be the principal factor driving the tail risks. In particular, we show a tail risk scenario in which the US monetary normalization could widen the CIP in the major FX swap markets (i.e. the yen and euro against the US dollar), leading to a sharp decline in the supply of international dollar liquidity. Similarly to findings by Stefan et al. (2016),13 our empirical findings point to an unmatched role of the US dollar in driving global financial stability through its impacts on cross-currency funding markets and thus international bank lending. 13 Stefan et al. (2016) documented that a stronger dollar goes hand in hand with larger deviations from CIP and lower growth of cross-border dollar-denominated lending. The paper attributes this triangular relationship to the role of the dollar as a proxy for the shadow price of bank leverage. 19 ADBI Working Paper 741 He, Wong, Ho, and Tsang REFERENCES Aiyar, S., C. W. Calomiris, J. Hooley, Y. Korniyenko and T. Wieladek (2014), “The International Transmission of Bank Capital Requirements: Evidence from the UK?” Journal of Financial Economics, 113(3): 368–82. Aizenman, J., B. Mahir and M. Hutchison (2014), “The Transmission of Federal Reserve Tapering News to Emerging Financial Markets,” NBER Working Paper No.19980. Baba, N. and F. Packer (2009), “From Turmoil to Crisis: Dislocations in the FX Swap Market Before and After the Failure of Lehman Brothers”, Journal of International Money and Finance, 28(8): 1350–74. Bauer, M. and C. Neely (2014), “International Channels of the Fed’s Unconventional Monetary Policy,” Journal of International Money and Finance, 44: 24–46. Bowman, D., F. Cai, S. Davies and S. Kamin (2011), “Quantitative Easing and Bank Lending: Evidence from Japan,” Board of Governors of the Federal Reserve System, International Finance Discussion Papers, No. 1018. Brunnermeier, M., J. Gregorio, B. Eichengreen, M. El-Erian, A. Fraga, T. Ito, P. Lane, J. Pisani-Ferry, E. Prasad, R. Rajan, M. Ramos, H. Rey, D. Rodrik, K. Rogoff, H. Shin, A. Velasco, B. Mauro, and Y. Yu (2012), “Banks and Cross-Border Capital Flows: Policy Challenges and Regulatory Responses,” Washington DC: Brookings Institutions. Bruno, V. and H. S. Shin (2015), “Capital Flows and the Risk-taking Channel of Monetary Policy,” Journal of Monetary Economics, 71: 119–132. Buch, C. M. and L. S. Goldberg (2015), “International Banking and Liquidity Risk Transmission: Lessons from across Countries,” IMF Economic Review, 63: 377–410. Cetorelli, N. and L. S. Goldberg (2011), “Global Banks and International Shock Transmission: Evidence from the Crisis,” IMF Economic Review, 59: 41–76. Cetorelli, N. and L. S. Goldberg (2012a), “Follow the Money: Quantifying Domestic Effects of Foreign Bank Shocks in the Great Recession,” American Economic Review, 102(3): 213–8. Cetorelli, N. and L. S. Goldberg (2012b), “Liquidity Management of US Global Banks: Internal Capital Markets in the Great Recession,” Journal of International Economics, 88: 299–311. Chava, S. and A. Purnanandam (2011), “The Effect of Banking Crisis on Bankdependent Borrowers,” Journal Financial Economics, 99: 116–135. Chen, Q., A. Filardo, D. He and F. Zhu (2012), “International Spillovers of Central Bank Balance Sheet Policies,” BIS Paper No. 66p. Chen, Q., A. Filardo, D. He and F. Zhu (2016), “Financial Crisis, Unconventional Monetary Policy and International Spillovers,” Journal of International Money and Finance, 67: 62–81. Cornett, M. M., J. J. McNutt, P. E. Strahan and H. Tehranian (2011), “Liquidity Risk Management and Credit Supply in the Financial Crisis,” Journal of Financial Economics, 101: 297–312. 20 ADBI Working Paper 741 He, Wong, Ho, and Tsang APPENDIX 4: DETAILED DESCRIPTION OF THE MONTE CARLO SIMULATIONS The procedure is illustrated using the case of Japanese banks. We first impose divergent central bank BSP shocks in the VAR model by considering simultaneously negative shocks to the Fed’s balance sheet and positive shocks to the BOJ’s balance sheet. Specifically, for the ∆𝐹𝐸𝐷𝑡 and ∆𝐻𝐶𝐵𝑡 𝐽𝑃 equations, their innovation terms in each of the consecutive 12 months starting from January 2016 are assumed to be –1.2% and 0.53%, respectively. The central bank BSP shocks are set thus as the sum of innovation terms for the 12 months is consistent with the one-year standard deviation of the BSP shocks as identified in Figure 3. Innovation terms for other variables in the VAR model are obtained by the MC simulation method. Hence, in each simulation trial a 12-month simulated path for each of the ten variables of the VAR model can be obtained. We focus on the simulated paths for ∆𝐹𝐸𝐷𝑡,∆𝐻𝐶𝐵𝑡 𝐽𝑃,∆𝐶𝐼𝑃 𝑡 𝐽𝑃,∆𝐶𝐷𝑆𝑡 𝐽𝑃, and ∆𝐸𝑋𝑅𝑡 𝐽𝑃, as these paths, together with the estimation results in Table 2, allow us to decompose the contribution of different factors to the supply of international dollar loans by Japanese banks under the divergence of BSP shocks of the Fed and BOJ. Based on the 10,000 simulation trials, we can compute the distributions for the four key variables in equation (1) that affect the supply of international dollar loans (i.e. ∆ HCBjt, ∆ FEDt, ∆ CIPjt-1, and ∆ CDSjt) for Japanese banks. We can further decompose the contribution of these factors to the supply of ∆ L*ijt of Japanese banks by using the estimation results for Model 4 in Table 2. The contribution of a factor is obtained by multiplying the estimates in Table 4 by the corresponding estimated coefficients from Model 4 in Table 2 except for that of ∆ FEDt.14 Since the specification for Model 4 assumes that bank-specific balance sheet characteristics affect the transmission of central bank BSPs, the decomposition analysis needs assumptions on the value for the balance sheet characteristics for Japanese banks. We hence assume a hypothetical Japanese bank whose balance sheet characteristics are the average values of Japanese banks in the estimation sample for the period Q12014 to Q42014. 14 The contribution of ∆ FEDt is derived by multiplying the estimated coefficient on ∆ FEDt *USFj and the value of USFj. 27