Chapter 4. External Assessments for Special Cases
- Anna Ter-Martirosyan, Sally Chen, Lawrence Dwight, Mwanza Nkusu, Mehdi Raissi, and Ashleigh Watson
- Published Date:
- January 2014
Using the Toolkit to Deal with Special Cases
Staff is developing an extension to the toolkit that takes into account differences across country groups with concentrated sources of external income.14 To illustrate the issues and the potential applications of the toolkit, we have conducted an exercise that begins with the CGER regression specifications and systematically examines possible differences in slope or intercept values and finally selects one modified regression specification. Results from this exercise (which are further discussed in Annex 4) suggest the value of modeling—in some manner—that allows for structural differences, particularly in special case countries. The exercise suggests significant differences in a number of slope coefficients (though none in intercepts):15
Nonrenewable commodity exporters: The slope coefficients on the fiscal balance (in the MB regression) and on the terms of trade and government consumption (in the ERER regression) are higher for nonrenewable commodity exporters (see Tables 1 and 2 in Appendix 4). This is not surprising because the fiscal balance in commodity-exporting countries is dominated by swings in oil revenues and hence strongly correlated with the CA, which is also dominated by such swings. Furthermore, given that their exports are less diversified, commodity-exporting countries are exposed to large terms of trade fluctuations that can induce large business cycle fluctuations, passing them through to domestic prices and the RER. Finally, fiscal transfers from oil revenues to the domestic economy in oil-exporting countries tend to raise domestic prices and appreciate the RER. These results are consistent with the findings of Bems and de Carvalho Filho (2009) for oil exporters.
Aid-dependent economies: The slope coefficients on the oil trade balance and aid inflows in the MB regression are higher for aid-dependent economies. This likely reflects their high vulnerability to terms of trade fluctuations and high sensitivity to aid inflows, given the limited scope for intertemporal consumption smoothing due to low savings and credit constraints (see Table 1 in Appendix 4). The sign of the coefficient on aid inflows generally depends on whether inflows are official grants included in the CA or loans that are excluded.16
Financial centers: In the MB regression, the slope coefficient on relative income is higher for financial centers, possibly a result of high earnings.
Remittance-dependent economies: Evidence indicates that the slope coefficient on relative productivity in the ERER regression is higher for remittance-dependent economies. Remittances may support a higher relative price of nontraded goods than relative productivity alone would imply, raising domestic prices and appreciating the RER (see Appendix 4).
This exercise illustrates that allowing for differences across country groups can have implications for exchange rate assessments. Table 4 shows that when slope coefficients are not allowed to differ across country groups, the regression residuals may be larger, suggesting that MB- and ERER-based estimates may overstate the exchange rate gap.17 For example, the average estimated exchange rate gaps for nonrenewable commodity exporters, under both the MB and ERER approaches, are about 6 percentage points further from zero in the absence of slope heterogeneity. For other special cases, the impact of allowing slope heterogeneity is smaller on average. As such, while this exercise is only illustrative and does not summarize all possible methods or outcomes, it does suggest that some form of customization may be appropriate, particularly if linked to economic theory and done within a multilaterally consistent framework that covers a large set of countries. Of course, other approaches to allowing for country heterogeneity can also be considered, including approaches that seek to directly measure countries’ structural differences.18
|MB (REER)||ERER (REER)|
|Exporters of nonrenewable commodities||5.8||5.6|
|Exporters of financial services||1.7||2.6|
|Expoters of tourism services||1.1||2.5|
|Recipients of aid||2.5||1.9|
|Recipients of remittances||0.5||0.7|
|Average (Customized groups)||0.9||2.7|
|Average (All other countries)||2.3||0.1|
However, there is a trade-off between allowing for country differences when modeling exchange rates, and with sample selection. Table 3 suggests that restricting the sample to a limited number of structurally similar countries may understate the magnitude of misalignment. By contrast, in Table 4 we see that not allowing coefficients to vary across structurally different economies may increase somewhat the residuals and thus overstate the magnitude of misalignment. The question of which issue is more meaningful comes down to understanding the nature of the effect evidently common to the country subsample. If the origin of that common effect is an undesirable distortion, then the focus on a subsample clearly weakens the analysis by understating gaps. If instead restricting the sample serves as a simple but effective way of controlling for omitted common structural factors appropriately influencing the CA and exchange rate, it may serve to enhance the assessment. Ideally, one would prefer to work with a wide country sample and to measure and model the omitted structural characteristics directly, if feasible, or otherwise to employ a dummy variable for the country group in question. In any event, given these considerations, adjustments to regression models must be accompanied by plausible hypothesis and grounded in economic theory.