Journal Issue

2. Does the Exchange Rate Cause or Absorb Shocks in the EAC Countries?

Paulo Drummond, Ari Aisen, Emre Alper, Ejona Fuli, and Sébastien Walker
Published Date:
July 2015
  • ShareShare
Show Summary Details

When joining a currency union, countries give up control of their national currencies and can no longer count on the nominal exchange rate to stabilize their economies in the face of country-specific shocks. Nonetheless, joining a currency union may entail benefits if the exchange rate is itself a significant source of shocks. Whether the nominal exchange rate has a greater tendency to absorb or cause shocks is therefore a central consideration in weighing up the costs and benefits of a currency union.

Results from the previous section suggest that the EAC economies have been subject to asymmetric shocks. Hence, the exchange rate could potentially play a role as an important shock absorber to mitigate the macroeconomic impact of external shocks. Following Artis and Ehrman (2006), we assess whether exchange rates in the EAC have tended to absorb shocks or contributed to fluctuations in output using structural VARs.1 This methodology is applicable for cases in which the exchange rate regime is not necessarily a pure float. In what follows we review the extent of exchange-rate management since 1990 and before proceeding with our structural VAR analysis.

Our empirical results indicate that EAC exchange rates2 mostly absorb real asymmetric shocks evident in the previous section with the exception of Burundi. This highlights the need for additional tools to stabilize the EAC economies once country-specific (nominal) exchange rates are no longer available as shock absorbers. We also report that while exchange-rate shocks do not seem to affect output, they are a source of disturbances to inflation. This suggests that EAC countries should press ahead on their journey toward modernizing their monetary policy frameworks, to help anchor inflation expectations in the interim before the currency union is established.

The Extent of Exchange-Rate Management in EAC Countries

Previous empirical works on the exchange-rate regime assessment in the EAC agree that there have been periods of exchange-rate intervention to various degrees in all five EAC countries in recent decades.3 Indicators of how the de facto exchange-rate regimes have evolved are notably provided in the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions and in Ilzetzki, Reinhart, and Rogoff (2010) (IRR, Appendix I).

Figure 5 shows exchange-rate regime classifications and actual nominal and real exchange rate movements for the five EAC countries.4 For both classifications, a higher code indicates a more flexible exchange rate. Based on the two classifications, all EAC currencies have been subject to significant interventions until very recently. The clear trend has been for all five currencies to depreciate against the U.S. dollar in nominal terms since 1990 while in real terms, the EAC currencies have been more stable against the U.S. dollar5.

Figure 5.Exchange-Rate Classification in EAC Countries

Source: IMF staff calculations.

Even limited intervention in the form of “volatility management” is problematic for investigating whether the exchange rate has tended to cause or absorb shocks if one does not take explicit account of such interventions. Therefore, there is an overwhelming case to be made for using an approach that does take explicit account of exchange-rate interventions.


Our methodology closely follows that of Artis and Ehrmann (2006), which itself draws upon the contributions of Blanchard and Quah (1989) and Smets (1997). The methodology assumes capital mobility and functioning financial markets and provides a useful framework to study macroeconomic effects of shocks under various exchange-rate regimes.

The vector of endogenous variables xt is given by

in which Δyt is the change in log output; rt* is the nominal interest rate of the foreign country; rt is the domestic nominal interest rate; Δpt is the change in log prices; and Δet is the change in the log of the nominal exchange rate, defined in terms of foreign currency units per domestic currency, so that an increase in et is a domestic appreciation. Output is measured by GDP; the nominal interest rate is measured by the three-month Treasury bill rate6 for all countries except the United States, for which we use the federal funds rate; and prices are measured by the consumer price index (CPI). We use a quarterly data set with seasonally adjusted variables, so the variables in first differences of logs correspond to quarter-over-quarter growth rates. In line with Artis and Ehrmann, we include a linear trend and oil price shocks in the model (the latter are assumed exogenous for the countries under consideration, so this does not require any further identifying restrictions).

The structural VAR model is given by

with εtN(0, Σε), and in which A0 is a 5 × 5 matrix of coefficients and A(L) is the matrix lag polynomial. The vector of structural shocks is

in which ɛts is a supply shock, ɛtd is a demand shock, ɛtm* is a foreign monetary policy shock, ɛtm is a domestic monetary policy shock, and ɛte is an exchange-rate shock. The supply and demand shocks are referred to as “real” shocks and the remainder as “nominal” shocks. The reduced form of the structural model is

The residuals from the reduced form can be used to retrieve the structural shocks in the structural model once a sufficient number of identifying restrictions is imposed. A first series of restrictions are imposed by assuming that the five structural shocks are uncorrelated and have unit variance, so that Σε = I. Following Blanchard and Quah (1989), the supply shock is then identified as the only one that can have a permanent effect on output. The demand shock and the supply shock are assumed to be the only shocks that affect output contemporaneously, identifying the demand shock. The nominal shocks may thus affect output temporarily and with a lag. The foreign monetary policy shock is identified by assuming that the foreign interest rate does not react to domestic monetary policy shocks or to the exchange rate between the currencies of the foreign and the home country. This means, for example, that monetary policymakers at the Federal Reserve do not take into account Ugandan interest rates or the exchange rate between the U.S. dollar and the Ugandan shilling. It is sufficient to impose this restriction only contemporaneously for the model to be just-identified, once the remaining restrictions have been imposed.

A final restriction is imposed following Smets (1997) to distinguish between the domestic monetary policy and the exchange rate shock. To do so, we estimate the weight ω which central banks place on the exchange rate when setting monetary policy (Appendix II). Once all other shocks are identified, and the exchange rate and interest rate (reduced form) residuals have been “purged” of the effect of these shocks, the remaining unexplained components of the exchange rate and interest rate are determined entirely by the domestic monetary policy shock and the exchange rate shock.7 Thus,


in which utr and ute are, respectively, the unexplained parts of the domestic interest rate and the exchange rate once the effects of the supply, demand, and foreign monetary policy shocks have been removed from the exchange-rate and interest-rate residuals. Solving for the structural domestic monetary policy shock gives

and normalizing the sum of the weights on utr and ute to unity leads to


The extreme case of ω = α2 = 0 corresponds to pure interest rate targeting (so that no weight is placed on exchange-rate shocks in the central bank’s short-run reaction function), while the other extreme of ω = 1 and β2 = 0 corresponds to pure exchange-rate targeting (where there is a one-to-one correspondence between domestic monetary policy shocks and exchange-rate innovations, as exchange market disturbances are prevented from affecting the exchange rate). In between 0 and 1, there exists a range of intermediate cases.

Following Smets (1997), ω is estimated by rearranging the equation for ɛtm to give the non-linear regression model:

in which the (observable) residual utr is regressed on the (observable) residual ute, and 1/(1ω)ɛtm is an error term. Since ɛtm is a component of ute, the explanatory variable and the error term are correlated in this regression model. We therefore use Hansen’s (1982) generalized method of moments (GMM) estimator, with instruments chosen following the same logic as Smets (1997). When estimating the SVAR in U.S. dollars, we use shocks to the Canadian three-month Treasury bill rate and shocks to the Canadian dollar-U.S. dollar exchange rate as instruments.8 These shocks are obtained by regressing the variables on their own lags and lags of the variables in the VAR9, and on the contemporaneous estimated supply and demand shocks.

Empirical Results

Impulse responses and variance decompositions

The theoretical framework underpinning Artis and Ehrmann’s analysis is a Mundell-Fleming-Dornbusch model as in Obstfeld (1985). In such a model, under capital mobility and functioning financial markets, the predicted responses of the endogenous variables to the various shocks are as follows.

  • A positive supply shock should be followed by higher output, lower domestic interest rates, lower prices, and a real exchange rate depreciation (Clarida and Galí, 1994). However, the effect on the nominal exchange rate is ambiguous (Borghijs and Kuijs, 2004).10 When the foreign economy is subject to a symmetric supply shock, the foreign interest rates should also fall in response to the domestic supply shock. (Note that this does not in any way imply that domestic shocks cause foreign interest rate responses.)

  • A positive demand shock should be followed by higher output, higher domestic interest rates, higher prices, and a real exchange rate appreciation (Clarida and Galí, 1994). The effect on the nominal exchange rate should also be ambiguous, given that a nominal depreciation accompanied by a large enough increase in relative domestic prices would still yield a real appreciation. In the case where the foreign economy is subject to a symmetric demand shock, the foreign interest rates should also rise in response to the domestic demand shock. (The same caveat as in the previous bullet point applies.)

  • A positive foreign monetary policy shock (contraction) should lead to a rise in the foreign interest rate. The response of the other variables depends on the response of the domestic monetary authorities (Artis and Ehrmann, 2000).

  • A positive domestic monetary policy shock (contraction) should be followed by lower output, lower prices, and a nominal appreciation (Clarida and Galí, 1994).

  • An exchange-rate shock that leads to an appreciation should be followed by lower output, lower domestic interest rates, and lower prices.

The impulse response functions for our SVARs are plotted in Appendix III, and the corresponding forecast error variance decompositions are shown in Appendix IV. Almost all of the responses that are statistically significant (according to the bootstrapped 90 percent confidence bands around the impulse response functions) are as predicted above.

Supply and demand shocks

Symmetric supply or demand shocks require symmetric responses of foreign and domestic monetary authorities. Hence, if foreign and domestic interest rates react in a similar way to the domestic supply or demand shocks, these real shocks should be symmetric between the two countries; otherwise, they are asymmetric. The impulse responses for the EAC countries compared with the United States (Figure 6) show asymmetric responses of foreign and domestic interest rates to supply and demand shocks (if the responses are significant at all). 11

Figure 6.Response of the Exchange Rate to Supply and Demand Shocks in EAC

Source: IMF staff calculations.

Where real shocks are asymmetric, there is a potential role for the exchange rate to act as an absorber of such shocks. The nominal exchange rate does not react to either supply or demand shocks in the case of Burundi. For Kenya, the impulse response functions show a clear appreciation following a positive supply shock and a clear nominal depreciation following a positive demand shock, and the variance decompositions (Figure 7) show that supply and demand shocks together account for about 80 percent of the forecast error variance of the exchange rate at all horizons up to five years ahead. The same is true of Rwanda, albeit with a lower initial share of the forecast error variance for supply and demand shocks in the first year, and a higher share thereafter. For Tanzania and Uganda, the exchange rates seem to exhibit a similar tendency, although the responses are of borderline statistical significance; nonetheless, supply and demand shocks account for a material share of the exchange rate forecast error variance at longer time horizons.12

Figure 7.Variance Decomposition of Output, Price, and the Exchange Rate

Source: IMF staff calculations.

Overall, these results suggest that the exchange rate does act as an absorber of real shocks in EAC countries, with the exception of Burundi.

Exchange rate shocks

The exchange rate does not react significantly to its own shocks, displaying low persistence, in the case of Kenya or Rwanda (Figure 7). Burundi, Tanzania, and Uganda’s exchange rates all seem vulnerable to their own shocks, albeit to varying degrees. Whichever country is considered, the exchange-rate shock consistently has little-or-no effect on output; a possible exception appears to be Tanzania’s exchange rate, but the exchange rate shock accounts for at most only 12.25 percent of the forecast error variance of output. Additionally, the exchange-rate shock does generally have a significant impact on the price level, again to varying degrees. On the whole, this finding suggests that the exchange rate can be a source of shocks to inflation in EAC countries, but not to output (Figure 8).

Figure 8.Response of Output and Price to Exchange-Rate Shocks in the EAC

Source: IMF staff calculations.

This section investigates whether bilateral exchange rates have tended to cause or absorb shocks in the EAC countries. Our results suggest that real (supply and demand) shocks are asymmetric in all cases. Moreover, the exchange rate appears to absorb these real asymmetric shocks in all cases save that of Burundi. These results highlight the need for additional tools to stabilize the EAC economies once country-specific (nominal) exchange rates are no longer available as shock absorbers. Specifically, they highlight the need to progress on further integration especially in implementing labor mobility in the region, and institutional changes to promote risk sharing among members. Our results also indicate that, while exchange rate shocks do not seem materially to affect output, they are a source of disturbances to inflation, suggesting that EAC countries should press ahead on their journey toward inflation targeting.

We also present results treating Kenya as the “anchor” country within the EAC, and evaluating the role of the exchange rate between the Kenyan shilling and the currencies of the other four countries (Appendix 2).

We use bilateral nominal exchange rates in the EAC with respect to the U.S. dollar. The EAC currencies are the Burundian franc, Kenyan shilling, Rwandan franc, Tanzanian shilling, and Ugandan shilling.

For instance, see Slavov (2013).

The complete list of classification codes is given in Appendix I.

The currencies of the other four countries have mostly been depreciating against the Kenyan shilling over the period considered, but these exchange rates have generally been more stable than those between the EAC currencies and the U.S. dollar. In real terms the four EAC currencies have been more volatile against the Kenyan shilling.

Specifically, 90- or 91-day Treasury bill rate, depending on the country.

To purge the reduced form residuals (RFRs) we regressed the RFRs on the shocks that we wish to “purge out,” and keep the residuals from this equation for the subsequent analysis. In other words, we take the RFRs and regress them on the demand shock, the supply shock, and the foreign interest-rate shock. We then use the residuals from this last equation and use in the final regression to estimate ω.

When estimating for the four EAC countries to Kenya, we use shocks to the U.S. federal funds rate and shocks to the U.S. dollar–Kenyan shilling exchange rate as instruments.

Discussion of the number of lags is in Appendix II.

Note that in the in the Balassa-Samuelson framework, a supply shock leads to a permanent appreciation of the real exchange rate. For example, see Gauthier and Tessier (2002).

This result also holds when Kenya is taken as the foreign country, for the remaining EAC countries with the exception of Tanzania. This, in itself, warns of possible risks to Burundi, Rwanda, and Uganda once they join a currency union with Kenya and Tanzania (and vice versa).

In the estimations involving the four EAC countries’ bilateral exchange rates with the Kenyan shilling, neither Burundi’s nor Tanzania’s exchange rate reacts to real shocks reflecting that shocks to Kenya and Tanzania appear broadly symmetric. Rwanda and Uganda’s exchange rates exhibit a significant reaction to these shocks for at least part of the periods shown on the impulse response functions, and supply and demand shocks account for a material share of the exchange rate forecast error variance at most time horizons.

    Other Resources Citing This Publication