De Facto Exchange Rate Regime Classifications: An Evaluation

Abstract

There exist several statistically-based exchange rate regime classifications that disagree with one another to a disappointing degree. To what extent is this a matter of the quality of the design of these schemes, and to what extent does it reflect the need to supplement statistics with other information (as is done in the IMF’s de facto classification)? It is shown that statistical methods are good at the basics (distinguishing some type of peg from some type of float), but less helpful in other respects, such as determining whether a float is managed, particularly for countries that are not very remote from their main trading partners. Different measures of exchange rate volatility have been used but are not primarily responsible for differences between classifications. The theoretical underpinning of particular classification schemes needs to be more explicit.

Appendix – Detailed Comparison of Classification Schemes

In this Appendix we summarise and evaluate a number of classification schemes that are exclusively based on exchange rate data.

Ghosh et al. (2002) [hereafter GGW]. A set of annual scores (s) is constructed based on the mean (μ) and standard deviation (σ) of monthly rates of exchange rate depreciation against each of a number of reference currencies; the minimum of these values of s identifies the one to which the currency is potentially pegged, and all the above-minimum values are discarded. Specifically the statistic is:

$$ s=\sqrt{\mu^2+{\sigma}^2} $$

Countries’ regimes are recorded as pegged, intermediate or floating based on whether this statistic is low, intermediate or high. The distribution of regimes is recorded, and the thresholds for s in each year are chosen so as to yield exactly the same distribution as in the IMF de jure classification in that year.

Evaluation: a horizontal single-currency peg should yield a low value of s, as intended, but a realignment during the year would raise both μ ² and σ ², in which case a realignment year might be classified as intermediate or even floating. Basket pegs and crawling pegs will also tend to have higher values of s than single-currency pegs. The use of different thresholds in different years is unsatisfactory, as is the choice of the de jure classification as the basis for determining them, since the main point of a de facto classification is to establish if the de jure classification is correct.

Shambaugh (2004) [hereafter termed JS]. A reference currency is identified. If the maximum and minimum of the log of the exchange rate against the reference currency (the US dollar being the default) do not differ by more than 0.04 over the calendar year, that observation is a peg. Alternatively, if the 0.04 threshold is exceeded, the observation is still a peg if the log of the exchange rate is unchanged in eleven months out of twelve. Thus effectively the level of the exchange rate is allowed to vary by ±2%, or alternatively by a realignment of any size in one month and 0% in the remaining eleven months, for a peg to be coded.

Evaluation: the statistical criterion is the range of variation about the central rate, which has the merit that it is closely related to the width of the target zone. There is the same problem with basket pegs and crawling pegs as in the case of Ghosh et al. (2002). The switch to a zero range in the event of a realignment implies that many pegs may not be identified as such in realignment years. For these reasons the frequency of pegs is likely to be underestimated.

Reinhart and Rogoff (2004) [hereafter termed RR]. Movements of the log of the exchange rate against various reference currencies are analysed, and as in GGW the reference currency that yields the lowest volatility is used. Where available, the exchange rate in the parallel market rather than the official rate is used. If, over a five-year period from years T–4 to T, more than 80% of monthly changes in the log of the exchange rate against any of the reference currencies fall within the range ±0.01, the exchange rate regime in all of the years T–4 to T is classified as some form of peg. Alternatively, even if this criterion is not met, if the change in the exchange rate is zero for four months or more, it is classified as a peg for those months. If fewer than 80% of monthly changes fall within the range ±0.01, but more than 80% fall within the range ±0.02, the regime is classified as a band. If the exchange rate moves by more than 40% in a year, that observation is placed in a separate “freely falling” category (these observations are omitted from the comparison with other schemes). Thus the scheme focuses on the upper tail of the distribution of monthly exchange rate movements, and specifically the proportion that exceed either 1 or 2% in absolute value.

Evaluation: The statistic used represents the general idea that exchange rate volatility is lower under pegs and therefore that there are fewer large movements, and its relationship to the width of a target zone is unclear since it concerns the frequency distribution of exchange rate changes rather than a range for the level. Basket pegs may well not meet the criteria for a peg, but crawling pegs should do so if the crawl is slow enough.¹² Realignments, if not too frequent, should not cause any particular problem, since they represent just another tail observation. The proportion of pegs recorded will depend entirely on the threshold chosen for the statistic.

Bleaney and Tian (2017) [hereafter termed BT]. The scheme is based on the root mean square residual (RMSE) from a regression similar to that of Frankel and Wei (1995) for identifying basket pegs. For each calendar year, the change in the log of the official exchange rate against the Swiss franc (the chosen numéraire currency) is regressed on the change in the log of the US dollar and of the euro against the Swiss franc. Occasionally, other reference currencies are added.¹³ If the RMSE from this regression is less than 0.01, that country-year observation is coded a peg. If the RMSE is greater than 0.01, twelve new regressions are estimated each including a dummy variable for a particular month as a test for a realignment. If the F-statistic for the most significant of these dummy variables (April, say) is less than 30, the regime is coded a float. If the F-statistic for April is greater than 30, and the RMSE is less than 0.01, the observation is coded a peg with a realignment; otherwise it is a float.

Evaluation: the regression approach should cater for basket pegs (through the regression coefficients) or crawls (through the intercept), but errors may arise from the small number of degrees of freedom in each regression. The use of dummy variables solves the problem of realignments, provided that there is not more than one per year. The statistic used is different from that of Reinhart and Rogoff (2004), since it captures the average size of movements in the exchange rate rather than the proportion of large ones, but as in their case the statistic is not closely related to the width of the target zone.

Summary: the Shambaugh system comes closest to the notion of a peg as a narrow target zone for the exchange rate, but underestimates the proportion of pegs for various reasons. The other three systems use statistics which capture the general idea that pegs have lower volatility, but whose relationship to the width of the target zone is unclear.

Table 4 Algorithms for a binary classification using RANGE and RMSE

Table 5 Sample of countries (182)