The Exchange Rate Disconnect and the Bank Lending Channel: Evidence from Switzerland

Transcription

1 The Exchange Rate Disconnect and the Bank Lending Channel: Evidence from Switzerland Isha Agarwal June 4, 2018 Abstract: Using the January 2015 episode of the Swiss franc appreciation as an exogenous exchange rate shock, this paper investigates the role of bank-lending channel in explaining the exchange rate disconnect puzzle. I construct a novel dataset on foreign currency exposure of Swiss banks by reading their annual reports and find that banks with a net liability exposure in foreign currency experience a higher loan growth in the post-shock period relative to banks with a net foreign currency asset exposure. This credit supply shock gets transmitted to the real economy by affecting investment of bank-dependent firms. I find that exporters investment goes down after the currency appreciation shock but the decline is offset by an increase in investment of firms in cantons which had a net foreign currency liability exposure before the shock. This finding suggests that the bank lending channel of exchange rates offsets the negative effect of currency appreciation on exporting firms investment, and can explain why currency appreciations are not always contractionary. To investigate the importance of the bank-lending channel more generally, I collect historical data on foreign currency exposure of banking sectors in advanced and emerging markets, and find that the negative response of GDP growth to nominal exchange rate appreciation is offset by the positive effect from the bank lending channel. However, this channel is relatively more important for emerging markets as compared to advanced economies. JEL Codes: G01; G21; J23; J63. Keywords: Bank lending; Exchange Rates; Open Economy; Currency Mismatch Cornell University. ia233@cornell.edu.

2 1 Introduction Competitive currency devaluation can be an important policy tool for central bankers to stimulate economic activity by affecting relative price of exports. However, empirical evidence seems to suggest weak co-movement between nominal exchange rates and macroeconomic fundamentals, also known as the exchange rate disconnect puzzle (Obstfeld and Rogoff, 2000). While there are many studies which explore the role of weak pass-through of currency depreciation to export prices to explain this disconnect (Devereux and Engel, 2003; Gopinath et al., 2010; Casas et al., 2016), little attention has been paid to the role of bank-lending channel of exchange rates, which could potentially play an important role in the transmission process. In this paper, I use the Swiss currency appreciation episode of January 2015 to understand the role of bank-lending channel in explaining the disconnect. Banks differ in their exposure to currency fluctuations as they have varied positions in terms of foreign-currency denominated assets and liabilities. For instance, a bank which borrows heavily in foreign currency, and is not entirely hedged against currency movements (either naturally or through currency derivatives), is likely to face a negative shock to its net-worth if domestic currency depreciates, and may contract lending. 1 This in turn, can have a negative impact on the investment activity of the firms that are major clients of such banks. Hence, the increased competitiveness of exporters due to currency depreciation can be offset by tightening of financial constraints for non-financial firms, and can potentially explain the weak response of output to currency devaluations (figure 1). Similarly, the response of macroeconomic variables to episodes of currency appreciation can be mitigated if some banks gain because of domestic currency appreciation. Hence, the bank-lending channel can act as a financial mitigator" in the transmission of international shocks to the real economy. The goal of this paper is to investigate whether there is a bank-lending channel of exchange rates and, if so, whether it can explain the weak response of macroeconomic variables to changes in exchange rates. It is difficult to estimate the causal impact of exchange rates on real economy using macroe- 1 Even if a financial intermediary actively hedges its portfolio, it may be unable to fully insulate its balance sheet from sudden movements in exchange rates. The following statement from the annual report of UBS reflects the inability of banks to fully protect their balance sheets from unfavorable movements in market prices: As seen during the financial crisis of , we are not always able to prevent serious losses arising from extreme or sudden market events that are not anticipated by our risk measures and systems. Value-at-risk, a statistical measure for market risk, is derived from historical market data, and thus by definition could not have anticipated the losses suffered in the stressed conditions of the financial crisis." 2

3 conomic data. Changes in exchange rates are often cyclically motivated. This leads to a problem of reverse causality and can produce biased estimates. Using lagged values of exchange rates may partially address the issue but can not completely eliminate it as the policy changes may be driven by anticipation of future economic activity. Another issue with using macroeconomic data is the simultaneity bias. A positive productivity shock may simultaneously lead to an appreciation in real exchange rate and an increase in current and future output. Hence, a time series analysis using macroeconomic data may lead us to falsely conclude that real exchange rate appreciations lead to higher economic growth. To circumvent these endogeneity concerns, this paper exploits a natural experiment to study the causal impact of changes in exchange rates on real economy. On January 15, 2015, the Swiss National Bank (SNB) surprised the market by abandoning the peg between the Swiss franc and Euro that had been in place since September The announcement led to approximately 20 percent appreciation in the value of the Swiss franc, which is considered huge in currency markets with daily average fluctuations of not greater than 1 percent. 2 As the announcement was completely unanticipated by the market and not driven by current economic situation in Switzerland, this episode allows me to address the standard endogeneity concerns associated with identifying the bank-lending channel of monetary policy shocks. 3 Moreover, the value of the Swiss franc remained elevated for approximately a year after the announcement. The persistence of this shock can potentially have economically significant effects on the banking sector in Switzerland, which has substantial foreign currency exposure. Besides, the Swiss economy has high reliance on exports, with a share of exports to GDP greater than 50 percent. This implies that the appreciation shock should have a large negative impact on economic activity if the trade channel is dominant. Hence, this episode provides a useful laboratory to test the importance of both the trade channel as well as the bank-lending channel. The empirical strategy relies on differential exposure of banks to foreign currencies in the preshock period. Since data on foreign currency exposure is not publicly available at the bank level, I construct a novel dataset on foreign currency denominated assets and liabilities of banks by reading their annual reports. This is the first paper to document bank-level measures of foreign 2 The average quarterly appreciation in nominal effective exchange rate for advanced economies in the sample was 0.14 percent and the 95th percentile was 3.8 percent for the time period 1994 to The following statement from the annual report of Bank Linth shows the SNB s decision was not expected by the market: Like all market participants, we were surprised by the decision of the Swiss National Bank on 15 January, 2015." I will argue more about the exogeneity of the event in section

4 currency mismatch. I collect data for the largest 100 banks in Switzerland, which account for approximately 90 percent of the banking system. The exposure measure is constructed for each bank by computing the difference between its foreign currency denominated liabilities and assets in December For further evidence, I also construct a narrative measure of exposure for each bank by reading their annual reports. For instance, the following statement from the annual report of Union Bank of Switzerland (UBS) is considered to reflect a net asset foreign currency exposure: The portion of our operating income denominated in non-swiss Franc currencies is greater than the portion of operating expenses denominated in non-swiss Franc currencies...therefore, the appreciation of the Swiss Franc against other currencies generally has an adverse effect on our earnings and equity.." Baseline results, from a diff-in-diff analysis, suggest that banks which had a net foreign currency asset exposure (foreign currency assets greater than foreign currency liabilities) experienced a fall in their profits and non-interest income, driven by a decline in their trading income. As a result, these banks contracted lending more as compared to banks which had a net liability exposure to foreign currency before the shock. This result suggests that there is a financial (banklending) channel of exchange rates, which can potentially mitigate or exacerbate the traditional trade channel of exchange rates. To explore the real effects of the currency appreciation event, I investigate the response of investment of non-financial firms after the appreciation shock. I find that exporters reduced their investment more than non-exporters after the shock. This is indicative of the trade channel. A stronger currency leads to weaker demand for goods produced by exporters, and this leads to reduced investment by exporting firms. To test for the bank-lending channel, I construct a measure of exposure to foreign currency of each canton in Switzerland as the average exposure of banks in that canton. Bank-firm matching is based on geographical proximity between banks and non-financial firms. In particular, I assume that firms in a given canton are more likely to have banking relationships with intermediaries in that canton. Geographical proximity to a financial intermediary has been used in the literature as an approximate measure of banking relationships. Hence, if we assume that non-financial firms in a given canton have banking relationships with intermediaries in that canton, we should expect investment of firms in cantons with positive average foreign currency exposure to increase 4

5 after the appreciation shock as banks increase their credit supply in that canton. Results from a triple-difference analysis suggest that bank dependent firms in cantons with a net foreign currency liability exposure see an increase in their investment after the shock as compared to bank dependent firms in cantons with a net foreign currency asset exposure. This suggests that the trade channel had a negative impact on investment of non-financial firms, but the negative effect was mitigated by the positive effect from the bank-lending channel as banks which were positively affected by the shock were able to expand lending. These results are robust to a battery of standard checks in the difference-in-difference literature. In particular, I test whether the results become stronger in a smaller window around the event and dissipate as the event window becomes longer. I also compute different measures of exposure and check if my results are dependent on the definition of the exposure variable. Results also satisfy the parallel trends assumption in the absence of the shock, the outcome variables for banks with positive and negative foreign currency exposure should not deviate from their pre-shock trend. While this natural experiment in Switzerland suggests the presence of bank-lending channel that offsets the trade channel of currency appreciations, it does not lend support to the importance of this channel in a more general setting. For external validation, I collect data on foreign currency exposure of banking sectors in a sample of 44 emerging and advanced economies. Using quarterly data on macroeconomic aggregates and nominal effective exchange rates for the period , I find that on an average, the elasticity of GDP growth to nominal currency appreciation is negative, but its magnitude is dampened by the presence of net foreign currency liability exposure on banks balance sheets. Further, the bank-lending channel is more potent for emerging markets as compared to advanced economies. This could perhaps be explained by the fact that financial markets in advanced economies are more liquid and deep, hence, the banking sector can actively hedge its foreign currency exposure by buying foreign exchange swaps or forward contracts. Banks in emerging markets may not be able to insulate their balance sheets from foreign currency movements as effectively as banks in advanced economies. These findings suggest that currency devaluation may not stimulate growth if the banking sector has large foreign currency exposure that is not perfectly hedged and has implications for monetary policy setting in small open economies. Rest of the paper is organized as follows. Section 2 discusses the related literature and main contribution of the paper. Section 3 provides background on the Swiss currency appreciation 5

6 episode and the Swiss banking sector. Section 4 presents data and summary statistics. Section 5 discusses the empirical strategy and results and section 6 concludes. 2 Related Literature This paper lies at the intersection of banking and open economy macroeconomics literature. It speaks to the literature documenting the lack of co-movement between nominal exchange rates and macroeconomic fundamentals in the data. For instance, Obstfeld and Rogoff (2000) refer to the weak correlation between exchange rates and macroeconomic aggregates as the exchange rate disconnect puzzle. 4 One of the ways in which the exchange rate disconnect puzzle manifests itself is the Meese-Rogoff Puzzle (Meese and Rogoff, 1983), which shows that a random walk model performs as well as standard open economy structural models in out-of-sample forecasting of exchange rates, and even contemporaneously there is little correlation between macroeconomic aggregates and nominal exchange rates. Another example of the exchange rate disconnect is illustrated by Baxter and Stockman (1989). They find that when countries transition from fixed to floating exchange rate regimes, the real and nominal exchange rates experience increased volatility, however, macroeconomic variables, such as trade balance, industrial production, and consumption, experienced little variability, suggesting a disconnect. Subsequent studies have tried to explain the exchange-rate disconnect puzzle by incorporating various features, such as local currency pricing, pricing-to-market, and imported intermediate inputs, in the standard open economy models. For instance, Devereux and Engel (2003) show that if export prices are set in importer s currency, and are sticky, changes in exchange rates will not have any impact on relative prices of consumer goods, leading to incomplete pass-through and, hence, a weakening of the expenditure switching effect of home currency depreciation. This channel is called the local currency pricing channel in the literature. 5 Exchange rate pass-through to relative prices also depends on whether firms price to the market, i.e., firms charge different prices for the same product in different markets. For instance, Fitzgerald and Haller (2013) use data on plant-level producer prices and find that firms price to the market. Finally, the presence of imported inputs (in global value chains) can also explain the low co-movement between 4 More broadly, this puzzle is related to the question on whether exchange rates should primarily be seen as asset prices or should they have any impact on the relative prices in the goods market. 5 Subsequent papers in this area, such as (Gopinath et al., 2010), (Casas et al., 2016), show that most of the traded good are invoiced in few dominant currencies, including the dollar. Hence, the exchange rate pass-through to relative prices depends on the price stickiness in dollar. 6

7 exchange rates and real variables. For instance, Amiti et al. (2014) show that large exporters are simultaneously large importers and hence the gain from currency depreciation via cheaper exports is offset by an increase in production cost due to higher cost of imported inputs. Rodnyansky (2018) shows that the muted response of the real economy to the large depreciation induced by the Abenomics QE episode in Japan can be explained by the presence of imported intermediate inputs. While the literature has primarily focused on the exchange rate pass-through to relative prices of exported goods (or the trade channel) to understand the exchange rate disconnect puzzle, few studies have explored the bank-lending channel of exchange rates, which is the main contribution of this paper. Bruno and Shin (2014) show that global banks lever up in response to local currency appreciation. Their model focuses on the link between exchange rates and financial stability and abstracts from the real side of the economy. Also, the focus is on cross-border transmission of shocks through global banks, rather than on studying the effects on domestic banks. 6 Kearns et al. (2016) investigate, using aggregate data in a cross-country setting, whether the financial channel of exchange rates offsets the trade channel. While their results do not provide direct causal evidence on the bank-lending channel, they provide suggestive evidence of the existence of a financial channel. 7 Recent literature has also discussed the importance of foreign currency exposure on banks balance sheets as a driver of international capital flows. For instance, Krogstrup and Tille (2017) show that the preexisting currency exposure of financial intermediaries determines the sign and magnitude of the response of their foreign currency funding to global risk factors. Similarly, Bräuning and Ivashina (2017) show that in response to domestic monetary easing, global banks increase their foreign reserves in currency areas with highest interest rates and reduce lending in these areas. These papers do not study how exchange rate shocks affect domestic lending and the subsequent real effects from change in supply of bank credit. This paper is also complementary to the literature on currency devaluations and exposure to foreign currency debt of non-financial firms, i.e., the firm balance sheet channel. There is mixed empirical evidence on the effect of foreign currency debt on investment in the aftermath of currency depreciation. For instance, Aguiar (2005) shows that after the Mexican peso crisis of 6 (Hofmann et al., 2016) is another recent paper which explores the risk taking channel of currency appreciations. 7 They define the financial channel of exchange rates as the share of foreign currency debt of the banking sector, and do not consider the asset side of the banking sector. 7

8 1994, exporters have higher sales and profits relative to non-exporters but their relatively high exposure to foreign currency debt has a negative impact on investment. On the contrary, Bleakley and Cowan (2008) fail to find any significant difference between the investment of firms which are highly exposed to foreign currency debt and those which are not. 8 More recently, Bruno et al. (2018) show that the working capital channel of a stronger dollar may offset the positive effect of domestic currency depreciation. These papers do not explore the effect of currency devaluation on bank lending in the presence of foreign currency denominated assets and liabilities in the banking system. 9 Finally, this paper contributes to the growing empirical literature on the bank-lending channel of monetary policy shocks, starting from the influential work by Kashyap and Stein (2000). Recent literature on banking and monetary policy has focused on cross-border transmission of monetary shocks, including papers by Cetorelli and Goldberg (2012), Schnabl (2012), Ongena et al. (2015). By focusing on the role of foreign currency mismatch on the balance sheet of banks, this paper studies a different channel of transmission of exchange rate shocks, mostly ignored by the literature on monetary policy and bank lending. 3 Background 3.1 The Currency Appreciation Episode - January 2015 Switzerland is a small open economy with high reliance on exports the share of exports in GDP has been greater than 50 percent since Europe is the most important destination of Swiss exports. In 2016, exports to Europe accounted for 54 percent of the total exports of Switzerland. Hence, the Swiss Euro exchange rate is an important policy variable for the Swiss National Bank (SNB) as changes in this exchange rate can potentially have a big impact on the real economy by affecting exports. In the early 2000s, the exchange rate between the Swiss Franc and the euro fluctuated between CHF 1.45 and 1.7 to the Euro. However, since 2008, as the global economic turmoil gathered pace, demand for the Swiss franc as a safe haven currency increased and by 8 Other papers related to foreign currency exposure and currency devaluation include (Desai et al., 2007), (Aghion et al., 2000), (Céspedes et al., 2002) 9 It is important to note that the presence of foreign currency denominated debt on the balance sheets of non-financial firms as compared to financial firms/banks will have different implications for the effect of currency devaluations on the real economy. If the banking sector is exposed to foreign currency debt, domestic currency devaluation will affect the real economy only in the presence of financial market frictions. Insofar as smaller firms have a higher reliance on bank debt than larger firms, they are likely to be affected relatively more if currency depreciation tightens financial constraints. 8

9 August 2011, the Swiss franc had appreciated by around 30 percent as compared to its value in To prevent further appreciation of the Swiss franc, the SNB introduced a floor of CHF 1.20 per Euro in September According to a press statement released in September 2011, the SNB announced that it was willing to purchase foreign exchange in unlimited quantities to defend the floor of CHF 1.20 to the Euro. On January 15, 2015, the SNB surprised the markets by discontinuing the minimum exchange rate of CHF 1.20 per euro. The announcement was even more surprising especially after the Vice Chairman of the SNB said in an interview on January 12, 2015 that the cap on the swiss-euro exchange rate must remain a cornerstone of our monetary policy". The announcement on January 15 led to approximately 20 percent increase in the value of the Swiss franc immediately after the announcement (figure 2). This was a huge shock by historical standards and led to an immediate collapse in the broad stock market index. Even after one year from the announcement, the Swiss Franc maintained its high value at 1.09 CHF per Euro. There is some narrative evidence to support that the decision of the SNB to discontinue the minimum exchange rate was completely unanticipated. For instance, the following statement from the annual report of Bank Linth shows that the decision of the SNB took the market by surprise: Like all market participants, we were surprised by the decision of the Swiss National Bank on 15 January, 2015 ". Moreover, the one-month forward exchange rate between the Swiss franc and the Euro did not change in December 2014, which should have been the case if the announcement was anticipated (figure 3). What makes this episode particularly useful to study the bank-lending channel of exchange rates is that the decision to discontinue the minimum exchange rate was not driven by the economic scenario in Switzerland. The SNB believed that the Swiss franc had weakened recently and hence the peg was no longer justified. According to a press release on January 15, 2015: The minimum exchange rate was introduced during a period of exceptional overvaluation of the Swiss franc and an extremely high level of uncertainty on the financial markets... While the Swiss franc is still high, the overvaluation has decreased as a whole since the introduction of the minimum exchange rate. The economy was able to take advantage of this phase to adjust to the new situation. The euro has depreciated substantially against the US dollar and this, in turn, has caused the Swiss franc to weaken against the US dollar. In these circumstances, the SNB has concluded that enforcing and maintaining the minimum exchange rate for the Swiss franc against 9

10 the euro is no longer justified." Hence, this shock can be considered largely exogenous and is pertinent to study the causal effect of home currency appreciation on the banking sector. 3.2 The Swiss Banking Sector The banking sector in Switzerland is divided into six broad categories of banks big banks, foreign banks, cantonal banks, regional and savings banks, Raiffeisen banks, and stock-exchange banks. UBS and Credit Suisse are the two banks in the big banks category, with a market share of 48 percent in Switzerland. 11 Cantonal banks are government-owned banks, one in each canton of Cantonal banks had a share of 17 percent in Foreign banks in Switzerland exist mostly as legal entities (controlled by their parent bank), with a share of 10 percent. Branches of foreign banks, which are not separate legal entities, had a market share of 2 percent in Most of the foreign banks in Switzerland are European banks. Regional and savings banks are small banks and focus on retail lending. They are geographically concentrated, with limited foreign exposure. The regional and savings banks had an asset share of 3.5 percent in Raiffeisen banks are structured as co-operatives, and have a regional presence, with an asset share of 6 percent. Stock exchange banks are primarily involved in providing asset management services to domestic and foreign clients, and have an asset share of 7 percent. Other banks have an asset share of 6 percent in The banking sector in Switzerland has a large exposure to international currencies. As of December 2014 (one month before the central bank announced its decision to abandon Swiss Euro peg), foreign currency assets accounted for 47 percent of the total assets of the banking system while foreign currency liabilities accounted for 48 percent of the total liabilities. Overall, the foreign currency exposure on the liability side exceeded the asset side by one percentage point. This is not trivial given that the banking sector in Switzerland was approximately 5 times the GDP in A breakdown by bank groups reveals that there is heterogeneity in the net foreign currency exposure, defined as the difference between the liability and asset share of foreign currency denominated liabilities and assets respectively. Big banks (UBS and Credit Suisse) had a net negative foreign currency exposure as of December 2014 their asset exposure exceeded their 10 The share of assets in total assets of the banking system in 2014 is calculated using the list of reporting banks and their balance sheet size, downloaded from the SNB s website. 11 There are 26 cantons in Switzerland and 24 cantonal banks. The cantonal bank in the canton of Appenzel Ausserrhoden was sold to UBS, and the one in the canton of Solothurn was privatized in

11 liability exposure by 4 percentage points. 12 While cantonal banks, regional banks, and Raiffeisen banks had a net positive foreign exposure their liability exposure exceeded their asset exposure. 4 Data and Summary Statistics A principal contribution of this paper is to construct a new dataset on foreign currency exposure (including off-balance sheet exposures) of banks and to link major bank- and firm-level datasets to observe the real effects of currency appreciation via changes in credit supply. For the bank-level analysis, I use the balance sheet, income statement, and cash flow data for banks in Switzerland from the Bankscope Database, accessed through the Wharton Research Data Services. This database is compiled by Bureau van Dijik (BvD) and provides information on bank balance sheet variables for banks in many countries around the world. It sources microdata on banks in different countries from their financial statements and presents the final data in a consistent and harmonized format across countries. Banks in this database account for approximately 90 percent of the total banking sector assets in each country (especially for advanced economies). 13 This database, however, does not have information on foreign currency exposure of banks. Hence, I hand-collect data on foreign currency denominated assets and liabilities from the annual reports of the biggest 100 banks, which account for approximately 80 percent of the total size of the banking system in Switzerland in For the bank-level analysis, I use annual data on loan growth, profits, net income, net interest income, non-interest income, bank size, liquidity, capital base, and loans to deposits for the period from Bankscope. 15 Bankscope does not have a good coverage of lending broken down by domicile, and by types (mortgage vs industrial). Hence, I hand-collect data on mortgage loans, inter-bank loans, customer loans, and deposits broken down by domicile, for 2013, 2014 and I also hand-collect data on trading income. Sample selection is based on availability of data on foreign currency exposure. All variables are winsorized at 5 and 95 percent to remove outliers. Panel A of table 1 reports summary statistics for the main variables used in the bank-level analysis. Data on accounting variables and operating performance of firms in Switzerland is from the 12 The gross asset exposure of the big banks to foreign currencies is huge it was 71 percent in The high gross exposure makes these banks particularly vulnerable to foreign exchange shocks. 13 BvD no longer publishes the Bankscope Database. It has replaced Bankscope database with ORBIS Bankfocus database, which is a close but not perfect substitute for Bankscope database. For my analysis, I have combined Bankscope database with Orbis Bankfocus database. 14 According to the Swiss National Bank s website, there are 268 reporting banks in Switzerland in 2014, including branches of foreign banks and excluding private bankers who do not actively seek deposits from the public. 15 See Data Appendix for definitions of variables used in the analysis. 11

12 Worldscope database. This database provides information on financial and income statements of publicly listed firms in Switzerland. I use the segments files of Worldscope database for information on foreign sales of Swiss firms. Data on bank-firm linkages comes from the Dealscan database. It contains deal/loan level information on financial transactions between firms and banks. I supplement this dataset with hand-collected data on bank-firm linkages from the financial statements of non-financial firms. 16 For the firm-level analysis, I use annual data on assets, market capitalization, pre-tax income, leverage, investment, and sales for the period The sample contains only non-financial firms firms with SIC codes in the range are droped. Following Almeida and Campello (2007), I exclude firms for which annual asset growth is higher than 100 percent. This practice is standard in the empirical corporate finance literature. Annual asset growth rates higher than 100 percent can be indicative of mergers or reorganizations. Hence, this rule ensures such firms are excluded from the sample. I also drop firms with negative values for sales. All variables are winsorized at 5 and 95 percent to exclude outliers. I characterize a firm as exporter if its ratio of foreign sales to total sales is in the top 25th percentile in To compute a measure of bank dependence of non-financial firms, I merge Capital IQ data with Worlsdsope data using ticker symbol as the identifier. Capital IQ database has detailed information on capital structure of publicly listed firms. For each firm, I compute the average ratio of bank debt to total capital between 2011 and A firm is classified as bank dependent if its ratio of bank debt to total capital is in the top quartile of the bank debt to capital distribution. Panel B of Table 1 presents summary statistics for key firm-level variables for the entire sample. Daily data on bilateral exchange rate between the Swiss franc and the euro is from the Global Financial Database (GFD). Data on Nominal Effective Exchange Rate and Real Effective Exchange Rate Index are from the Bank for International Settlements (BIS). Aggregate data on foreign currency assets and liabilities of banking sector in different countries are from the International Financial Statistics database (IFS) of the IMF and the Locational Banking Statistics database of the BIS. Quarterly data on real GDP for advanced and emerging markets is from the IFS database. 17 Data on central bank policy rates is from the Global Financial Database. 16 This is still work in progress. 17 See table 13 for definitions of policy rates for each country in the sample. 12

13 5 Empirical Analysis This section discusses the identification strategy and presents results. 5.1 Identification Strategy It is difficult to estimate the causal impact of exchange rates on real economy using macroeconomic data. Changes in exchange rates are often cyclically motivated. This leads to a problem of reverse causality and can produce biased estimates. Using lagged values of exchange rates may partially address the issue but can not completely eliminate it as the policy changes may be driven by anticipation of future economic activity. Another issue with using macroeconomic data is the simultaneity bias. A positive productivity shock may simultaneously lead to an appreciation in real exchange rate and an increase in current and future output. Hence, a time series analysis using macroeconomic data may lead us to falsely conclude that real exchange rate appreciations lead to higher economic growth. To circumvent this endogeneity issue, I use micro-data on bank balance sheets and estimate the causal impact of currency appreciation using a difference-in-difference estimation strategy. In particular, I exploit the differential exposure of bank balance sheets to foreign currency denominated assets and liabilities. A bank which has a higher share of foreign currency assets (in total assets) relative to foreign currency liabilities should be negatively affected by the appreciation of the Swiss franc as the value of its assets goes down and the value of it liabilities goes up. Similarly, a bank which borrows heavily in foreign currency but does not have a proportional foreign currency exposure on the asset side of the balance sheet should be positively affected by home currency appreciation since the appreciation reduces the debt burden of the bank. The identification strategy is two pronged. First, it relies on an exogenous change in exchange rate to study the impact of the shock. This exogeneity allows me to get around the reverse causality bias associated with using macroeconomic data. Second, it exploits cross-sectional variation in currency exposure of banks. This ensures that other macroeconomic events do not confound the causal mechanism. 5.2 Lending Outcomes The identification strategy, as discussed in section 5.1, hinges on differential exposure of banks to foreign currency assets and liabilities before the shock. 13

14 I construct the foreign currency exposure for each bank in the sample as the difference between foreign currency denominated liabilities and assets (as a share of total liabilities and assets respectively) in December 2014: 18 [ liabilitiesforiegn exposure liabilities tot assets ] foreign assets tot 2014 This measure will be negative if the share of foreign currency assets exceeds the share of foreign currency liabilities. I hypothesize that banks with negative values for this exposure measure should be negatively affected by the currency shock of 2015 and can reduce lending. To test this formally, I estimate the following model: LOANS b,t = α + βp ost t + γt reated b + δ(t reated b P ost t ) + θbank b,t 1 + ɛ b,t (1) where, LOANS bt is the annual growth rate of gross loans of bank b in time t, P ost t is a dummy variable that takes a value 1 for years 2015 and 2016, T reated b is an indicator variable that takes a value 0 if the foreign currency exposure of the bank is negative (i.e., if foreign currency asset share exceeds foreign currency liability share) and 1 otherwise, T reated b P ost t is an interaction term between the post-shock period (i.e., for years 2015 and 2016) and the treatment variable. This variable estimates the differential impact of the currency shock on the treated group relative to the control group and is our main variable of interest. BANK b,t 1 is a vector of bank-level control variables including lagged value of total assets (size), ratio of equity to assets, bank liquidity, and loans to deposits ratio. Since the treatment variable varies at the bank level, standard errors are clustered at the bank level to allow for intra-bank auto-correlation of the residuals within banks. This equation is estimated for the time period 2011 to The main variable of interest is δ as it reflects the difference between loan growth of treated and control group in the post-shock period. Table 2 reports the results for equation 1. From column (1), we see that the average loan growth was approximately 3 percentage points lower in the post-shock period as compared to the pre-shock period. Column (2) adds the main variable of interest the interaction between post and treated. We see that the coefficient on the interaction term is positive and significant, which implies that banks which had a net short position in foreign currency (more foreign liabilities as compared to foreign assets) had a significantly higher growth 18 This measure is constructed using information on foreign liabilities and assets based on domicile. For banks which report assets and liabilities broken down by currency and positions in the derivatives market, I use net positions in foreign currencies after incorporating off-balance sheet exposure arising from positions in forex derivative markets. 14

15 of lending as compared to the banks which had a net long position in foreign currency. Also, note that the coefficient on treated is not significant. This implies that on an average, the loan growth of banks with different pre-shock exposure is not statistically different. In column (3), bank-level variables which can affect loan growth are added. In terms of bank-level controls, banks size and liquidity are important determinants of loan growth. We find that the coefficient on the interaction term remains positive and statistically significant. Finally, to account for the fact that these differences could be driven by different business models of banks and other unobserved differences in bank characteristics across banks with positive and negative exposures, I control for bank fixed effects in column (4). The coefficient on the interaction term becomes slightly smaller but remains positive and statistically significant. Results are also economically significant. Banks which had a net short exposure in foreign currency experience approximately 8 percentage point higher loan growth in the post-shock period as compared to banks with a net long currency exposure. In column (5), I use the continuous measure of exposure as defined above (instead of using a dummy variable). The coefficient on the interaction term remains positive and statistically significant. One standard deviation increase in the measure of exposure leads to approximately 3.7 percentage point increase in loan growth of banks in the post-shock period. This is economically significant as the average loan growth in the sample period is close to 7 percent. A possible concern with the identification strategy could be that there are pre-existing trends that are driving the difference in lending growth of positively versus negatively exposed banks after the shock. To address this concern, I check for parallel trends. The parallel trends assumption requires that in the absence of treatment, the difference between treated and control group should be constant over time. I test this assumption formally using the following specification: LOANS bt = α b + t β tt t + t γ t(t reated b T t ) + θbank b,t 1 + ɛ b,t (2) t {2012, 2013, 2014, 2015, 2016} where, α b captures bank fixed effects, T t is a dummy variable for each time period (year) in the sample (2011 is the omitted year), T reated is a dummy variable as described above, (T reated b T t ) is an interaction term between time dummies and the dummy variable for exposure to foreign currency, BANK b,t 1 is a vector of lagged bank-level controls. Standard errors are clustered at the bank-level to allow for auto-correlation of residuals across time. The main coefficients of interest are γ t, which show the difference in loan growth of treated 15

16 and control groups in each period in the sample. Figure 4 shows the 95 percent confidence interval plots for the estimated coefficients γ t from equation 2. It is evident from the figure that there are no statistically significant differences in loan growth of banks with positive and negative foreign currency exposure in the years prior to the currency appreciation episode in 2015, which is consistent with the parallel trends assumption for the difference-in-difference methodology. In 2015, however, the banks with a positive foreign currency exposure (foreign currency liability share higher than foreign currency asset share) experience a higher loan growth as compared to banks with a negative foreign currency exposure and the difference is statistically significant Types of Lending It may be important to understand how the currency shock affected lending in different markets. For instance, adjusting inter-bank lending vs industrial lending can have different macroeconomic implications. While it is not possible to distinguish between consumer loans and industrial loans, banks do report their loans broken down as inter-bank loans, mortgage loans, and costumer loans. Customer loans can be short-term consumer loans as well as industrial loans. Mortgage lending has the biggest share in bank lending, with an average share of roughly 72 percent. Customer lending and inter-bank lending have average shares of 22 percent and 18 percent in total lending. Bankscope does not have good coverage of different types of lending, hence I hand-collect data on inter-bank, customer, and mortgage loans for 2014 and To explore the effect on different types of lending, I estimate the following equation: Y b = α + βexposure + θbank b + ɛ b (3) where, the dependent variable can be loan growth of mortgage loans, inter-bank loans, and customer loans of bank b in BANK b is a vector of lagged bank-level controls as discussed in the baseline specification. This equation is estimated for a cross-section of banks in Results are reported in table 3. Columns (1), (2), and (3) show results for mortgage lending, inter-bank lending and customer lending, respectively. We see that exposure is significant and positive for mortgage loans and customer loans. In terms of economic magnitude, a one standard deviation increase in exposure variable leads to 5.6 percentage point increase in growth rate of mortgage loans and 3.3 percentage point increase in the growth rate of customer loans. The coefficient on exposure is insignificant for inter-bank loans. 16

17 Another interesting dimension is to examine the effect on secured vs unsecured loans. When banks suffer losses, do they respond by changing the risk profile of their portfolio? To answer this question, I hand-collect data on loans secured by collateral and those without any collateral and investigate how the loan growth of secured vs unsecured loans reacts to the currency shock. Results are reported in table 3. The dependent variable is annual loan growth of unsecured loans in column (4) and of secured loans in column (5). We see that the foreign currency exposure does not matter for unsecured loans but it is positive and significant for secured loans. One way to interpret this result is that banks which have net long exposure in foreign currency reduce their collateralized lending when they are hit by the currency shock while they keep their unsecured lending unchanged. This implies that the share of unsecured loans increases in the lending portfolio of banks with a net long foreign currency exposure and has implications for risk shifting as an unintended consequence of currency appreciation. 5.3 Channels In this section, I investigate what are the channels through which the Swiss franc appreciation episode affects bank lending. Financial intermediaries try to hedge foreign exchange risk on their balance sheets by buying foreign exchange forward contracts, swaps or other forex derivatives. However, a financial intermediary that actively hedges its portfolio may be unable to fully insulate its balance sheet from sudden movements in exchange rates as their Value-at-Risk models typically calculate expected losses based on historical events. The following statement from the annual report of UBS reflects the inability of banks to fully protect their balance sheets from unfavorable movements in market prices: As seen during the financial crisis of , we are not always able to prevent serious losses arising from extreme or sudden market events that are not anticipated by our risk measures and systems. Value-at-risk, a statistical measure for market risk, is derived from historical market data, and thus by definition could not have anticipated the losses suffered in the stressed conditions of the financial crisis. Given that the January 2015 appreciation shock was largely unprecedented, it is likely that it affected banks profits. Banks report losses arising from derivatives and foreign currency translation as net trading income. If banks were not perfectly hedged against the appreciation shock, we should expect the trading income of banks with a net liability foreign currency exposure to go up relative to banks with a net asset foreign currency exposure. To test this formally, I investigate the response of profits and net income of banks around 17

18 the shock. More importantly, I want to explore the response of the non-interest component of net income, which includes trading income. To do this, I estimate equation 3, with P rof it, N etincome, and N on interest Income and Interest Income and T radingincome as dependent variables. 19 Results are reported in table 4. In column (1), the dependent variable is one-year change in profits, in column (2), the dependent variable is one-year difference in net income, in column (3), the dependent variable is one-year difference in non-interest income, in column (4), the dependent variable is one-year difference in net interest income and in column (5), the dependent variable is one-year difference in trading income. All outcome variables are scaled by bank assets in All columns include bank-level controls. From table 4, we find that banks with more foreign currency liabilities see a higher increase in profits and net income in 2015 (column 1 and 2). It is useful to examine whether this increase in net income is driven by an increase in interest income or non-interest income. In contrast to the conventional monetary policy transmission channel that works by affecting net interest income of banks, we see that the bank lending channel of currency appreciation works by affecting noninterest income. From column (3) and (4), we see that the exposure variable is insignificant for interest income but is positive and statistically significant for non interest income. Further, trading income is positively affected for banks with a net foreign currency liability exposure (column 5). Since trading income is part of non-interest income, it explains why the coefficient on exposure is statistically significant and positive for non-interest income. 5.4 Robustness and Additional Evidence In this section, I conduct a battery of robustness checks to address a number of concerns with the identification strategy and also supplement the main analysis with a narrative measure of exposure Narrative Evidence To provide additional support to the importance of foreign currency mismatch as a mechanism for transmission of currency shocks, I conduct an in-depth study of the annual reports of Swiss banks to gather narrative evidence on their foreign currency exposure and on the likely impact of 19 I am using a cross-section analysis for this section because I use hand-collected data on trading income (for which I collect data for 2014 and 2015). 20 Note that I am not using growth rates since profits can be negative, hence using growth rates may lead to misinterpretation of results. Since I scale absolute differences in all variables by lagged bank assets, I do not include bank size as a control variable in this regression. 18

19 the exchange rate shock on their balance sheet. I create a qualitative measure of exposure which takes a value 0 if the bank was negatively affected by the currency shock and mentions so in its annual report. For instance, the following statement from the annual report of Banque Privee BCP (Suisse) reflects a negative impact of the shock:... the decision of the Swiss National Bank to discontinue the minimum exchange rate of CHF 1.20 per euro impacted negatively the results of the bank, given that costs are expressed in CHF while the asset and revenue base are split between EUR and USD." The qualitative measure takes a value 0 for this bank. While Aargauische Cantonalbank has the following statement on the impact of the currency shock in its 2015 annual report: A massive increase is reflected in the result from trading activities. The previous year s figure of CHF 32.2 million was increased by 52.4% to CHF 49.1 million. The additional income was accrued primarily in foreign exchange and foreign currency trading due to the decisions of the SNB in January 2015." The qualitative measure of exposure takes a value 1 for this bank. Using this measure, I investigate the response of bank lending to the change in exchange rate for a cross-section of banks by estimating the following model: Y b = α + βqualitativemeasure + θbank b + ɛ b (4) where, Y b could be loan growth, change in profit or change in income of bank b between 2014 and 2015, QualitativeM easure is an indicator variable which takes a value 0 if the bank mentioned in its 2015 annual report that it was negatively affected by the shock and 1 otherwise, and BANK b is a vector of bank-level control variables including total assets (size), capital, liquidity, and loans to deposits in The main variable of interest is β, which provides an estimate of the differential impact of the currency shock on positively affected vs negatively affected banks. Equation 4 is estimated for a cross section of 53 banks in Table 5 reports the estimates from equation 4. Column (1) shows the estimate of β without including bank-level controls. We see that banks which reported as being positively affected by the shock experienced a 4.4 percentage point higher growth rate of loans as compared to banks which reported as being negatively affected by the shock. After controlling for bank-level variables like 21 I studied more than 100 annual reports to create this measure. Not all banks attribute changes in their balance sheets in 2015 to the currency shock. Sample selection (53 banks) for this analysis is based entirely on the existence of statements in banks annual reports on the effect of the shock. 19