«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 ...»
We use the change in the net foreign asset (NFA) position of the central bank in percent of GDP as proxy for FX interventions. As a robustness check we clean this measure for valuation effects. In addition, we use other proxies such as the change in official reserves minus gold in percent of GDP and the change in NFA relative to M2. For the policy rate we use the interbank rates, where possible, and short term government bond yields in the remaining cases.3 The average FX exposure of the countries in our sample increased during the global financial crisis to almost 25 percent of GDP on average from less than 20 percent before the crisis (Figure 1). Since then it has remained constant at 20-25 percent of GDP. The domestic part of FX debt increased over the observed period. In 2003, FX exposures from the domestic banking system only accounted for about ¼ of total FX exposures; since 2009 domestic and external FX debt has been roughly of the same size. Countries have not managed to bring down the overall FX exposures of the non-financial private sector and in many countries the exposures are large enough that exchange rate volatility (depreciations in particular) can have significant implications for corporate and household balance sheets. As shown in Figure 2, the size of FX exposures varies widely across countries from less than 5 percent of GDP to over 50 percent of GDP. In addition, the composition varies across countries.
3 We also use the actual official monetary policy rates of the countries as dependent variables. However, for countries where the monetary policy framework is not based on only one policy rate to affect market rates (e.g.
Turkey) but instead on several policy rates, we use market rates as a better indication of the monetary stance.
7 of the countries with large FX exposures, most of it is financed from domestic sources.
Finally, the data shows that external FX debt shows less cross-country variation than domestic FX debt.
In countries with large FX exposures there is a stronger correlation between FX interventions and exchange rate changes. This is shown in Figure 3, where exchange rate changes are plotted against FX interventions (defined as the change in central bank net foreign assets as a proxy) using monthly data from 2002-15 for 15 countries. Splitting the sample into high FX exposure observations (more than 20 percent of GDP total FX exposure for a given country at a specific time)4, FX interventions are generally larger when FX exposures in the nonfinancial private sector are large. Moreover, interventions are more negatively correlated when FX exposure is large. This correlation is driven by selling FX in the market when the currency depreciates rather than by buying FX during appreciation episodes.
Figure 3. Exchange Rate Changes and FX Interventions
For policy rates, Figure 4 shows that in countries with low FX exposures policy rate changes are smaller than in countries with high FX exposures and the largest exchange rate depreciations occur in highly-FX-indebted countries with the correlation being more positive in these countries.
the fraction of months where changes in exchange rates, reserves, and interest rates exceed a certain threshold.5 Countries with higher FX exposure show higher volatility in reserves and policy rates. However, exchange rate changes are not lower in these countries. This could indicate that countries with high FX exposures face larger exchange rate pressure. So despite their attempts to limit exchange rate volatility they experience as large exchange rate changes as countries that intervene less. When calculating two different intervention indexes, the results are confirmed. Countries with higher FX exposures show a higher degree of exchange rate management even though their official exchange rate regime is floating.
Approach We assume that emerging market central banks use policy rates and foreign exchange reserves as their two instruments for managing exchange rates.6 We suppose that these instruments work independently of each other.7 Therefore, we estimate two separate equations with FX interventions and policy rates as dependent variables and analyze whether
policy reactions to exchange rate movements depend on the level of FX debt in the nonfinancial private sector. We estimate the following equation for FX interventions (FXI):
where the first term on the right hand side is a country specific fixed effect, the second is the percent change of country ’s exchange rate, the third is the exchange rate change interacted with the FX debt to GDP (FXL). Control variables include trade openness, the current account balance, the change in the money stock (M2) to GDP, reserves relative to imports and M2, and the change and the level of FXL. The interaction term allows the coefficient on the exchange rate to vary with the level of FX debt and the expected negative sign of would indicate that countries with high FX debt react more strongly to exchange rates using FX interventions. The sign on the exchange rate ( ) is ambiguous as this can be interpreted as the reaction to exchange rate movements of a country with zero FX debt.
The second equation is an extended Taylor rule equation following Mohanty and Klau (2004), which is extended to include the interaction between exchange rate changes and nonfinancial private sector FX debt:
%∆ %∆ 2,,,,,,,,, %∆,,,,, 1, …,15; 10,2002, …, 3,2015 where, is the nominal policy rate of country at time. As in a standard Taylor rule, the central bank is expected to react to inflation,, and the output gap,, with the lagged policy rate included as explanatory variable to allow for persistence in adjusting policy rates. We assume that countries have different policy rules with regard to inflation, output gap, and lagged policy rates by allowing these coefficients to be country-specific.8 We include lags of
exchange rate changes and the interaction term to take into account that central banks’ policy rate reaction to exchange rate changes may happen with a lag. The parameter of interest is.
As before, if it is significant – and now with an expected positive sign – it indicates that higher FX debt leads to a stronger reaction to exchange rates by increasing (lowering) policy rates when the exchange rate depreciates (appreciates).
Instrumental variables to account for endogeneity Both equations suffer from an endogeneity problem as FX interventions and changes in the policy rate affect exchange rates. If countries intervene to stabilize exchange rates, exchange rate changes will be smaller and changes in interventions and policy rates will be larger.
Thus, the coefficient for the exchange rate changes interacted with FX debt will be biased towards a larger reaction to exchange rate changes. However, while quantitatively, estimates will be biased, the significance of the interaction term is informative as there should be no reason why the coefficient for exchange rate changes should vary with the level of FX debt due to endogeneity in a simple regression since the bias relates to the exchange rate changes and not the FX debt.
To address the issue of endogeneity we use instruments for the exchange rate changes. A good instrument is one which is correlated with the exchange rate changes but not with FX interventions and policy rates. We use the change in the EMBI spread and the VIX separately and in combination.9 The baseline results are reported using VIX as the only instrument for exchange rate changes.
VIX is associated with capital flows (Rey, 2015) and EMBI with the ability of debt repayment and both of them are used as proxies for risk aversion and uncertainty. Changes in VIX and EMBI show high degree of co-movement with changes in bilateral exchange rates.
Correlation of the exchange rate changes with the change in VIX and the change in the EMBI spread is 0.35 and 0.48 respectively; that is, the rise in VIX and EMBI tends to exert depreciation pressure on the bilateral exchange rates. As shown in Figure 5, both indexes capture exchange rate changes well especially during periods of financial stress (highlighted in red).
At the same time, changes in VIX and EMBI spread are considered to be exogenous events which are not directly affected by the FX interventions and monetary policy decisions of the individual countries in our sample. We also assume that FX interventions and monetary policy rates are affected by VIX and EMBI spread only through changes in exchange rates.
varies over time is challenging as 1) measuring the natural real interest rate over time for the countries is highly uncertain and 2) not all countries in the sample are inflation targeters.
9 VIX is the Chicago Board Options Exchange Market Volatility index. It is a measure of the implied volatility of S&P 500 index options. EMBI is the J.P. Morgan Emerging Markets Bond index that measures the total return performance of international government issued by emerging market countries.
11 Other estimation approaches Apart from estimating separate equations we estimate the equations as a system of seemingly unrelated regressions (SUR). This approach offers some efficiency gains as it incorporates potential correlation in error terms between the equations. Even though separate equations are still unbiased SUR provide more efficient estimates in a case of correlation of the disturbances. The system of equations is estimated by GLS using the variance-covariance matrix of the disturbances as a weighing matrix.
Further, we estimate a simultaneous equation model assuming that FX interventions and monetary policy rates are interdependent instruments in exchange rate management. We assume that FX interventions influence policy rates, and vice versa and estimate the model using GMM where policy rates and FX interventions are included as explanatory variables for the first and the second equation respectively. Excluded exogenous variables from the first equation are used as instruments for FX interventions in the second equation. The same approach is employed for the second equation. Finally, we apply the cross-equation correlations of the disturbances adjusted for heteroskedasticity as a weighing matrix for estimation of the system of equations (Greene, 2012).
Main results Table 1 presents the results of the regression with FX interventions as a dependent variable (equation 1) for a simple fixed effect regression and an IV estimation with fixed effects. The coefficient for the percent change in the nominal exchange rate is -0.02, i.e. for a country with no FX debt, a 10 percent nominal depreciation is associated with a 0.2 percent of GDP reduction of central bank net foreign assets. The coefficient is significant at 10 percent level 12 when controlling for the endogeneity. The coefficient of the interaction term is highly significant and with the right sign. The coefficient of -0.16 (for the IV estimation) implies that for 10 percent depreciation FX interventions increase by 0.16 percent of GDP for every additional 10 percent of GDP FX debt. Figure 7 shows the importance of FX debt in determining the FX intervention reaction to depreciation. Using the latest available level of FX debt of the non-financial private sector, the chart illustrates the level of FX intervention as a response to 10 percent depreciation.
For the monetary policy rate, higher FX debt is associated with a stronger reaction to exchange rate changes (see table 2). Again, the reaction to the exchange rate change interacted with FX debt is highly significant. The results indicate that in the absence of FX exposures, the policy rate reaction to exchange rate movements is limited. This is in line with a traditional Taylor rule approach where monetary policy reacts to inflation and the output gap. When exchange rate changes are interacted with FX debt, the reaction of policy rates to exchange rate changes (with a one month lag) is significant. The coefficient of 0.08 implies that a country with 10 percent of GDP FX debt will react to 10 percent depreciation by increasing its policy rate by 0.08 percentage point in the following month. The reason why the coefficient is significant with a one month lag – instead of contemporaneously – is likely that usually policy rate decisions are decided in planned policy consultation meetings, which occur with lower frequency than decisions to do FX interventions. Moreover, since policy rate inertia is high for all countries in the sample (the coefficient for the lagged policy rate is 0.80-0.99 for all countries but Indonesia) a longer lasting depreciation will lead to further increases in policy rates.
In Figure 8 we show the total increase in the policy rate for the countries in our sample over the first month and quarter (cumulative) following 10 percent deprecation based on their level of FX debt as of Q1, 2015. The effect is the largest for countries with high levels of FX debt in their non-financial private sector.
Note: Estimates are based on most recent data for FX debt (Q1, 2015 for most countries).