«Floating with a Load of FX Debt? by Tatsiana Kliatskova and Uffe Mikkelsen IMF Working Papers describe research in progress by the author(s) and are ...»
External vs. domestic FX borrowing We separate our analysis into domestically and externally funded FX exposures. Central banks are likely to be more concerned about FX exposures in the non-financial private sector if these are financed from lending by the domestic banking system. The reason is that real sector negative consequences of depreciation may spill over to the financial sector. The real economic effects work through lower consumption and investment demand as firms and households experience a deterioration of their balance sheets and would subsequently start deleveraging (by cutting back their consumption and investment). The financial instability effect will be more pronounced if these FX exposures are financed by the domestic banking system as depreciation will then affect banks’ balance sheets negatively through nonperforming FX loans. Our analysis confirms this as shown in the Appendix. When we include only domestically financed FX debt, the coefficient on the interaction term increases in absolute values for the policy rate and FX intervention regressions whereas for the external debt, the results become insignificant.
14 Appreciation vs. depreciation of exchange rates Next, we differentiate between exchange rate depreciation and appreciation. Central banks may be more sensitive to depreciation of exchange rates as sharp currency depreciations threaten financial stability when the non-financial private sector has FX liabilities. On the other hand, the central bank may also be induced to counteract appreciations as real appreciations negatively affect export performance. To account for potential asymmetries, we multiply the interaction term of FX debt and exchange rate changes by a dummy variable that takes the value 1 if the exchange rate change is negative (appreciation). As shown in the Appendix, both FX interventions and policy rate changes are mostly driven by reaction to depreciation of exchange rates. On average, if a country has 10 percent FX debt to GDP it reacts to 10 percent depreciation by decreasing its NFA by 0.4 percent of GDP and increasing its policy rates by 0.15 percentage points (Figure 9). The reaction to appreciation is much smaller and not statistically significant.
Robustness The definition of FX interventions We check if our results are robust to a number of alternative specifications of FX interventions. Apart from using changes in net foreign assets (NFA) relative to GDP, we use changes in NFA relative to M2 and change in reserves minus gold relative to GDP as dependent variables in equation 1. Using these alternative specifications does not change the results qualitatively (see Appendix).
Our proxy for FX interventions – the change in NFA of the central bank – could change for reasons other than FX interventions. Importantly, we implicitly assume that the reserves are denominated in USD, while in reality it is a mix of currencies. To correct for valuation effects, we use the Currency Composition of Official Foreign Exchange Reserves (COFER) database for Emerging Markets and calculate the changes in NFA adjusted for valuation effects. Additionally, we correct for movements in the net position of derivatives as central banks may intervene in the FX currency market by engaging in forwards or futures 15 operations which do not show in the NFA of the central banks. None of these adjustments change the overall conclusion.10 The choice of exchange rate measure For our main results, we use the nominal bilateral exchange rate with the USD. While we believe this is likely to be the variable of concern for most of the countries in our sample, we explore different specifications for robustness check. We use real bilateral exchange rate with the USD, nominal effective exchange rate, and mixed series where we choose bilateral euro exchange rate for European countries and USD exchange rate for the remaining countries (see Appendix). Again, our results are robust to these alternative specifications.
Instruments Finally, while in our baseline result we use VIX change as the only instrument, we also estimate regressions using the EMBI change and a combination of both as additional instruments (see Appendix). The differences in the results from using these alternative specifications are also small and the main conclusions hold.
System of equations Apart from estimating equations separately we evaluate them as a system of equations assuming correlation of the disturbances for the first case and interdependence of FX interventions and monetary policy rates for the second case. The results do not differ from what we get by estimation of separate equations; that is, countries with high FX debt react more strongly to exchange rate movements using both FX interventions and policy rates.
Additionally, countries that use policy rates to stem exchange rate depreciation intervene less in the FX market. At the same time, the effect of FX interventions on policy rates is not statistically significant (see Appendix).
Extensions Next, we present a number of extensions to the results in section IV. We include in the analysis banks’ (on-balance sheet) net open FX positions as well as the government’s external borrowing; we analyze whether reactions to exchange rate changes depend on FX debt and exchange rates in a non-linear fashion and whether low reserves changes the reaction pattern; and, finally, we discuss and make a (rough) attempt to control for potential natural hedges of the corporate sector.
Including the FX debt of banks and government Throughout the paper we assume that banks’ FX exposures do not affect central bank policies. To account for banks’ FX exposures we include the net open position of the banking system in addition to the FX debt of non-financial corporates and households. An important 10 The data on currency composition of reserves is available only on an aggregated level for emerging markets and advanced economies. By using the average composition for EMs, we thus assume that the currency composition is the same across the countries in our sample. It is highly likely that there are large variations across countries (e.g. countries in Europe may have a larger share of euros than Asian or Latin American countries) and the valuation adjustment may add more noise than information. For our baseline regressions we therefore use the unadjusted NFA series.
16 caveat here is that we are only able to take into account the on-balance sheet net open FX positions of the banks and thus ignore off-balance sheet hedges. These can be large as it is the case, for example, for Turkey at the end of the sample period.11 With this caveat in mind, the results remain robust when we include the banking system exposures. Adding government external debt to the non-financial private sector FX debt also does not alter the conclusion that higher FX debt is associated with a stronger policy reaction to exchange rate changes (see Appendix). However, when we run regressions separately for government debt and bank net open position as the only FX exposures, the reaction of both FX reserves and policy rates to exchange rate changes interacted with FX debt becomes statistically insignificant. This suggests that the FX debt of non-financial corporates and households is indeed more important than that of the government and banks in affecting the policy reaction to exchange rate changes.
Non-linearities We analyze two types of non-linearities. The first is a non-linearity with respect to FX debt;
i.e., is it the case that the reaction of FX interventions and policy rates to exchange rate changes not only increases with the level of FX debt but becomes much stronger as FX debt rises? The second is a non-linearity with respect to the exchange rate changes; i.e., is policy reaction more sensitive to large changes in the exchange rates? We do not find support for any of these non-linearities as shown in the Appendix.
What happens when reserves are low?
When FX reserves are low, the central bank’s ability to use them to stem exchange rate pressure is limited and the interest rate becomes the predominant tool. The main regression for FX interventions includes the level of reserves coverage of M2 and imports as control variables. In addition, we define a dummy that takes the value 1 if reserves are low and interact it with our variable of interest. Reserves are defined as being low if they are less than 5 percent of GDP (7 percent of the total observations). We find that the reaction to exchange rate changes using reserves declines when reserves are low (the effect is negligible for higher reserve threshold values). The interest rate reaction to exchange rate volatility increases (as expected) when reserves are limited but the effect is not statistically significant (see Appendix).
Endogeneity of FX debt One potential bias in the estimates may arise from endogeneity in the level of FX debt of the non-financial private sector. As mentioned earlier, the choice of whether to borrow in FX may depend on expectations of the future policy reactions to exchange rate movements.
However, the empirical evidence for this is unclear. For example, Berkmen and Cavallo (2009) find that floating exchange rate regimes, by themselves, do not promote dedollarization. It suggests that the differences in FX indebtedness across countries and time 11 Turkey shows as the country with the largest banking sector on-balance sheet net open FX position in Q1, 2015 (about 8 percent of GDP). However, due to off-balance sheet hedges (for which data exists for Turkey) of roughly the same amount, the overall FX exposures of the banking system in Turkey are almost negligible as of Q1, 2015.
17 are unlikely to be mainly the result of differences in policy reactions to exchange rate changes.
Changes in FX debt are likely to be less affected by expectations of central bank policies far into the future. To limit potential endogeneity of FX debt we therefore run the regressions using a much longer lag (3 years) for FX debt. Our results are robust to this specification (see Appendix). However, we acknowledge that if high FX exposures 3 years ago are driven by expectations (which can be self-fulfilling) that today’s policy will react strongly to exchange rate movements, the use of longer lags will not guarantee that the results are unbiased. This would require measurement of the exogenous component of the level of FX exposure in the non-financial private sector, which is not a part of this paper.
One more potential source of endogeneity is institutional aspects of monetary and exchange rate policy that may affect both monetary policy design and choice of the currency of debt denomination. In particular, three countries in our sample agreed on Flexible Credit Line Agreement (FCL) with the IMF in 2009 (Mexico, Poland, and Colombia). This augments the access to official liquidity and thus supplements the potential FX reserves available to dampen exchange rate volatility. To account for this we introduce a dummy variable for FCL agreement and its interaction with exchange rate changes multiplied by FX debt. Both coefficients are statistically insignificant for the NFA regression but are significant (with the opposite sign) for the policy rate regression. While the overall results remain unchanged we find that after the FCL agreement the three countries use policy rates less to manage exchange rates.
Accounting for natural hedging Non-financial corporates and households may be hedged against exchange rate movements if they hold FX assets abroad or domestically, use financial hedging instruments, or they can be naturally hedged via FX income from exports. Testing the first two types of hedging requires micro level data on FX assets and financial hedges of individual households and firms in order to match such hedges with individual FX debt. Such data is generally not available.
Similarly, for natural hedging, the relevant measure is a micro level data matching of FX liabilities and FX revenues, which is also not available. As a rough proxy for FX revenues at the aggregate level we use exports and interact it with our variable of interest (exchange rate changes interacted with FX debt). A significant coefficient of the opposite sign as compared to an interaction term of FX debt and exchange rates changes would signal that in countries with high exports the concern about exchange rates could be smaller as firms are naturally hedged. The coefficients for an interaction term of exports and exchange rate changes are statistically insignificant in both equations (not reported). It most likely confirms that micro level data is necessary to account for the actual FX exposures of the non-financial private sector as the non-financial companies holding FX debt need not be the same as those that have export revenues.
Countries with floating exchange rate regimes are often reluctant to allow their currencies to float freely. In this paper, we show that balance sheet currency mismatches are important for producing fear of floating. We find that policymakers react more to exchange rate 18 movements – depreciations in particular – when FX debt in the non-financial private sector is high by using FX interventions and monetary policy rates. For FX interventions, we find that for every additional 10 percent of GDP FX debt in the non-financial private sector, the reaction to 10 percent depreciation increases by 0.2 percent of GDP. For monetary policy rates, we claim that 10 percent additional FX debt to GDP increases the monetary policy reaction to 10 percent depreciation by 0.08 percentage points in the next month and by about
0.2 percentage points cumulative over the following three months. Moreover, the funding source of these FX exposures matters. Non-financial private sector FX debt financed from the domestic banking sector seems to be more important than FX debt obtained directly from abroad.