Classifying de facto exchange rate regimes of financially open and closed economies: A statistical approach

ABSTRACT

This paper offers a new de facto exchange rate regime classification that draws on the strengths of three popular classifications. Its two hallmarks are the careful treatment of a nexus between an exchange rate regime and financial openness and the use of formal statistical tools (the trimmed k-means and k-nearest neighbour methods). It is demonstrated that our strategy minimises the impact of differences between market-determined and official exchange rates on the ‘fix’ and ‘float’ categories. Moreover, it is more suited to assess empirical relevance of the Mundellian trilemma and ‘irreconcilable duo’ hypotheses. Using comparative analysis we find that the degree of agreement between classifications is moderate: the null of no association is strongly rejected, but its strength ranges from low to moderate. Moreover, it is shown that our classification is the most strongly associated with each of the other classifications and as such can be considered (closest to) a centre of a space of alternative classifications. Finally, we demonstrate that unlike other classifications, ours lends more support to the Mundellian trilemma than to the ‘irreconcilable duo’ hypothesis. Overall, our classification cannot be considered a variant of any other de facto classification. It is a genuinely new classification.

Highlights

• We develop a statistically-based de facto exchange rate regime classification

• The trimmed k-means and k-nearest neighbour methods are used

• Fix and float categories are identified with the correction for financial openness

• The new classification is a centre of a space of alternative classifications

• Our classification supports the trilemma rather than irreconcilable duo hypothesis

KEYWORDS:

• exchange rate regime
• financial openness
• macroeconomic trilemma
• cluster analysis

JEL CLASSIFICATIONS:

• F33
• F31
• C38
• C82

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The IMF classification is updated in the Annual Report on Exchange Arrangements and Exchange Restrictions. The other two have recently been updated in Ilzetzki, Reinhart, and Rogoff (2019) and Levy Yeyati and Sturzenegger (2016).

2 For more on the evolution of categories used in the IMF’s classification see e.g. Klein and Shambaugh (2010, 31–36) or Habermeier et al. (2009).

3 Some other exchange rate regimes classifications include: a de facto classification (Bubula and Ötker- Robe 2002), a ‘consensus classification’ (Ghosh, Gulde, and Wolf 2003), a bivariate classification (Shambaugh 2004), a classification based on a ‘regression method’ (Bleaney and Tian 2017).

4 For a different approach based on statistical methodology see e.g. Bleaney and Tian (2017), Bleaney and Tian (2020). They, however, use simple regressions for the nominal exchange rates only.

5 See e.g. Eichengreen and Razo-Garcia (2013) or Klein and Shambaugh (2010).

6 For a recent discussion of this trilemma (and others) see e.g. Bordo and James (2017).

7 Reinhart and Rogoff (2004) admit that dual/parallel rates are usually accompanied by exchange controls.

8 See e.g. Edwards (2015) for evidence from Latin American countries with flexible exchange rates.

9 See also the updated classification by Ilzetzki, Reinhart, and Rogoff (2019) and more recent classification by Klein and Shambaugh (2010).

10 See also the updated classification in Levy Yeyati and Sturzenegger (2016).

11 See e.g. Stevenson (2002, 250) who considers systematic interventions in the foreign exchange market as a defining feature of the fixed exchange rate regime.

12 For an alternative exposition of this point see Frankel and Wei (2008). They developed a simple framework that illustrates the importance of variability of both the exchange rate and foreign exchange reserves.

13 Given that exchange rate fluctuations are not regular, systematic changes in foreign reserves could stabilise the exchange rate only by chance. Smooth changes in reserves could be related, for example, to funds received from international organisations.

14 To be precise they distinguished between two intermediate regimes: ‘crawling peg’ and ‘dirty float'. The former was the one with high reserves variability, high exchange rate variability, but low volatility of exchange rate changes, whereas the latter was characterized by high variability of all three variables.

15 We thank an anonymous referee for drawing our attention to this issue.

16 Inclusion of the euro area member states in the group of peggers is consistent with the lack of separate (national) legal tender. The IMF, however, classifies the European Monetary Union as ‘free floating’ on the basis of ‘the behaviour of the common currency’ (IMF 2007).

17 According to the IMF under the conventional peg arrangement ‘the exchange rate may fluctuate within narrow margins of less than ±1% around a central rate or the maximum and minimum value of the spot market exchange rate must remain within a narrow margin of 2% for at least six months’ (IMF 2019, 45).

18 Their process of anchor currency selection is different than ours. In principle, they rely on the monthly, one-year moving average of the absolute value of the change in the bilateral exchange rates relative to all candidate anchor currencies (Ilzetzki, Reinhart, and Rogoff 2019).

19 Due to data availability we were unable to estimate an anchor currency for 14 countries that were included in the set of countries used by Ilzetzki, Reinhart, and Rogoff (2019). These countries are very small, e.g. Andorra, Palau, Tuvalu, or/and the data are of poor quality (if available at all), e.g. Somalia, Turkmenistan, Zimbabwe.

20 The third, minor reason of differences was that Ilzetzki, Reinhart, and Rogoff (2019) used the domestic currency as an anchor for some countries (Australia, Japan, the United Kingdom, the United States). We cannot follow that approach since our objective is to identify the relevant exchange rate which by definition requires two currencies.

21 If just one correction is introduced the agreement is also higher: 88.0% if the first correction is made and 89.7% if the second correction is made.

22 We experimented with greater thresholds comparing the results with those with the threshold set at 0.4. Setting the threshold at greater levels resulted in little changes in our classification. The adjusted Rand index dropped when the threshold was set above 0.7, though it remained at 0.6 which was rather high level. In fact, such a decrease was a natural thing, because the number of country-years got smaller as the threshold was set at a higher level and at the same time the number of groups was kept fixed. We thank an anonymous referee for drawing our attention to this point.

23 We tried less restrictive ceilings, i.e. 0.1%, 0.25% and 0.5%, and found that they made it possible to reclassify additionally 38, 90, and 176 observations, respectively.

24 See articles 30 and 48 of the Statute of the European System of Central Banks and of the European Central Bank (Official Journal of the European Union, 2016/C 202/01).

25 The upper bound was motivated by the IMF’s definition of conventional peg.

26 We thank Eduardo Levy-Yeyati and Federico Sturzenegger for making their updated classification available to us.

27 The IMF’s classification has not been taken into account here due to the shorter period it covers.

28 There is not much difference in association if the lambda statistic is used.

29 One should add that all these regimes were uncovered in the second round of clustering. Interestingly, in all three countries the exchange rate regime in 2008 was a less flexible one, i.e. a fixed exchange rate in Estonia, a dirty float in Latvia and a crawling peg in Lithuania. In 2010 the exchange rate regimes found were the same as those in 2008 (with the exception for Latvia for which there was a missing value in 2010).

30 Data were obtained from Ethan Ilzetzki’s website (accessed on 18 July 2018).

31 Ilzetzki, Reinhart, and Rogoff (2019) observed, however, that that share increased to almost 20% in 2015 and 2016.

32 Alternatively, one can observe that countries with dual/multiple/parallel exchange rate were those with relatively closed capital accounts. See Figure A7 in the Appendix.

33 Data were obtained from Hiro Ito’s website (accessed on 1 May 2017).

34 Only country-year observations present in both classifications are included in Figure 2. Thus, the differences between distributions are not due to the different coverage of country-years. Distributions without the correction for the common coverage look very much the same and are available upon request.

35 Distributions of monetary independence index under floating rate irrespective of the degree of openness to capital flows are presented in Figure A8 in the Appendix. The results are similar to those in Figure 2(a).

36 Distributions of monetary independence index under fixed rate irrespective of the degree of openness to capital flows are presented in Figure A9 in the Appendix. The results are similar to those in Figure 2(b).

37 In order to check formally the similarity of distributions we have run the Wilcoxon and Mann-Whitney test of equality of medians. In all cases the null of equality was strongly rejected (at the 1% significance level). Detailed descriptive statistics and test results are depicted in Table A7 in the Appendix.

38 We thank an anonymous referee for pointing out this limitation.

Additional information

Funding

The authors gratefully acknowledge financial support from the National Science Centre Poland [grant number 2015/17/B/HS4/02681].