Exchange rates are notoriously difficult to predict. Various theoretical and empirical papers have been written on the topic, most notably by Meese and Rogoff (1983a, 1983b). In this regard, several institutions – in particular the IMF – have invested significant resources in developing state-of-the-art models to estimate equilibrium exchange rates, assess their current values for overvaluation or undervaluation, and predict their future paths (e.g. Lee et al. 2008, Phillips et al. 2013).

In a recent paper, I evaluate the predictive performance of one particular vintage of IMF exchange rate models for subsequent exchange rate movements over a variety of horizons (Yeşin 2016). The research question is motivated by the fact that the IMF’s exchange rate assessments during 2006–2011 resembled predictions for future exchange rate movements, because the calculated equilibrium exchange rate was the medium-term outlook for the exchange rate. In my paper, I empirically analyse whether the IMF assessments could correctly predict subsequent exchange rate movements in terms of direction as well as magnitude.

The empirical question that I address is important for various reasons. First, exchange rate assessments were – and still are – an important input for the IMF’s regular surveillance activities and policy advice to its member countries. They tend to constitute the crux of the annual Article IV consultations. Second, two of the IMF models link exchange rates to developments in the current account balance and the net foreign asset position. In the context of the global imbalances debate, these models are pivotal for external adjustment discussion. Third, the sample period includes the ex ante unanticipated Global Crisis that led to significant exchange rate movements. And, last but not least, the IMF assessments were strictly confidential and not shared publicly during 2006–2011. Therefore, they could not have influenced markets.

## IMF exchange rate models

During 2006–2011, the IMF’s Consultative Group on Exchange Rate issues (CGER) employed three state-of-the-art models to assess real trade-weighted exchange rates for 27 advanced and emerging market economies on a semi-annual basis. These models were the macroeconomic balance (MB) model, which estimates an equilibrium current account based on the absorption approach; the equilibrium real exchange rate (ERER) model, which relies on a reduced form equation of the real effective exchange rate; and the external sustainability (ES) model, which estimates an equilibrium current account stabilising the net international investment position.

Each model yielded an independent misalignment value for the real effective exchange rate in the medium term for each currency – that is, the percentage overvaluation or undervaluation. Furthermore, these misalignment values were globally consistent.

The IMF’s final assessment of the currency was generally defined as a simple average of the three misalignment values. In my analysis, these variables are called *MB misalignment, ERER misalignment, ES misalignment* and *IMF misalignment*, respectively.

## Predictive power

To analyse the predictive power of these misalignment variables for future exchange rate movements, I conduct various empirical tests. Figure 1 summarises the directional accuracy and the linear fit in bivariate models in the two-year horizon when country-specific factors are not taken into account. Each panel shows a scatter plot of the two-year-ahead changes in the logarithm of the real effective exchange rate (REER) after an assessment together with a misalignment variable. The top left panel illustrates the *IMF misalignment.* If the* IMF misalignment* can predict the directional movements of future exchange rates, then observations should be only in the upper-left and lower-right quadrants. This area accounts for 61% of the observations. Thus, 61% of the time, the *IMF misalignment* diagnosis was accurate regarding the direction of the REER after two years. A more stringent criterion specifies that if the IMF is spot on with its assessment, and if the whole adjustment takes place in 2 years, the slope of the linear fit should equal -1. The simple linear fit without a constant and with robust errors clustered on the country level has a slope of -0.18, and is statistically significant at the 5% level. This means that, on average, only 18% of the* IMF misalignment* gap was closed after two years, when country-specific variables were not taken into account.

The other panels of Figure 1 show similar results. Among the different models, the* ERER misalignment* gap has the highest predictive power with 59% directional accuracy and a linear fit statistically significant at -0.16. Furthermore, there are many outlier observations in the upper-right quadrants for the *MB* and *ES* *misalignments*. In other words, several currencies were diagnosed as significantly overvalued by these two approaches and still appreciated in the next two years following the assessment. This is less of an issue for the *ERER misalignment*.

**Figure 1**. IMF assessment and subsequent REER changes

*Source*: IMF, BIS, and own calculations.*Note*: The figures show scatter plots of the different misalignment measures and the two-year-ahead changes in the natural logarithm of the REER following the assessments. A linear fit without a constant and with robust standard errors clustered at country level is also depicted in each figure.

Further analysis shows that the directional accuracy and the linear fit of the assessments increases with the length of the horizon for all misalignment variables. It is highest for the *IMF misalignment* variable for all horizons, followed by the* ERER misalignment. *Moreover, for a few currencies the models seem to have made persistent errors in predicting the direction of the future exchange rate movements also for longer horizons.

Next, in order to control for the cross-country differences in the sample, panel regressions with fixed effects are conducted. The analysis shows that all coefficients on the misalignment variables have negative signs and are statistically significant for all horizons. Controlling for fixed effects increases the predictive power of the assessments significantly. For example, 58% of the* IMF misalignment* gap is closed after two years when country-specific factors are taken into account. Furthermore, the *ERER misalignment* gap has the highest coefficient value in absolute value, together with the highest R-squares. According to the estimation, 59% of the misalignment gap diagnosed by the *ERER misalignment* is closed after two years if we control for country-specific factors. On the other hand, the predictive power of the *MB* and the *ES* *misalignments* is considerably weaker, with lower R-squares and lower coefficient values. This finding suggests that the predictive power of the *IMF misalignment *is due to the high predictive power of the *ERER misalignment.*

Using other empirical models in the paper, I also show that the assessments are better at predicting future exchange rate movements in advanced economies than in emerging market economies. Controlling for the exchange rate regime does not yield different results. Furthermore, the assessments have higher predictive performance in open economies than in closed economies. Last but not least, safe haven currencies close the misalignment gap predicted by the models faster than other currencies.

## Conclusion

These findings suggest that while movements of nominal exchange rates in the short term are notoriously difficult to predict, exchange rate models based on macroeconomic fundamentals can explain real effective exchange rate movements in the medium term surprisingly well.

## References

Lee, J, G M Milesi-Ferretti, J Ostry, A Prati, and L A Ricci (2008) “Exchange rate assessments: CGER methodologies”, IMF Occasional Paper, No 261.

Meese, R A and K Rogoff (1983a) “Empirical exchange rate models of the seventies: Do they fit out of sample?”, *Journal of International Economics,* 14: 3-24.

Meese, R A and K Rogoff (1983b) “The out-of sample failure of empirical exchange rates: Sampling error or misspecification?”, in J Frenkel (ed), *Exchange rates and international macroeconomics*, pp 67–105, Chicago: NBER and University of Chicago Press.

Phillips, S, L Catão, L Ricci, R Bems, M Das, J Di Giovanni, D F Unsal, M Castillo, J Lee, J Rodriguez and M Vargas (2013) “The external balance assessment (EBA) methodology”, IMF, Working Paper 13/272.

Yeşin, P (2016) “Exchange rate predictability and state-of-the-art models [4]”, Swiss National Bank, SNB Working Paper 16-02.