Inflation targeting and exchange rate volatility in emerging markets

Recent analysis by Ahmed et al. (2017), for example, finds evidence that differences in economic fundamentals played a role in explaining the heterogeneous EME financial market responses during the global financial crisis of 2008, and that the role of fundamentals appeared to progressively increase through the European crisis in 2011 and subsequently in the so-called taper-tantrum episode of 2013 triggered by the US monetary policy.

Modifying the model to nominal interest rates and foreign exchange reserves, Calvo and Reinhart (2002) have values ranging from zero (there is a peg or a very high degree of commitment to IT) to one when “seignorage” has a high weight in the policymaker’s objective function.

By the ex-ante uncovered interest parity (UIP) condition in international finance, higher-yielding currencies are expected to depreciate against the (foreign) lower paying counterpart. In practice, however, this condition is often violated. Flood and Rose (2002) investigate this for a period of turbulence in financial markets and conclude that UIP performs relatively better during the crisis of the 1990s than other times. Frankel and Poonawala (2010) confirm that the UIP condition is not observed (ex-post) in emerging markets, although the rejection is not as severe as in industrial countries.

Exchange rates may lead to inflationary effects, which forms an empirical matter of its own. Nogueira and León-Ledesma (2009), for example, use monthly data from 1995M1 to 2007M12 for Brazil and estimate inflation as a function of its past, real output growth, changes in nominal exchange rates, and changes in the price of foreign imports. They find that, after the policy change and adoption of IT in 1999M1 long-run exchange rate pass-through declined dramatically. Under the peg, exchange rate changes are more fully transmitted to prices. Also, under a credible IT nominal shocks are expected to have only limited effects on inflation.

In Aizenman et al. (2011), real GDP growth was relatively close across samples: 1.11 versus 1%. However, their figures were similar to ours on monetary policy: Inflation targeters show lower average inflation (5.40%) relative to the non-inflation targeters (9.60%) and the group of targeters present lower nominal interest rates (8.98%) than their counterparts (12.68%). Therefore, in addition to slower economic growth for ITers in our sample, the rate of change of exchange rates is much lower for ITers on average: 0.28 versus 0.73%.

For a general comparison of monetary policy rules in emerging markets, please refer to Mehrotra and Sánchez-Fung (2011). Canuto and Cavallari (2013) discuss whether central banks should incorporate indicators of financial stability into the reaction function. While one may associate asset prices with home and equity markets, private sector credit could indicate as well the extent of bank lending to the private sector. We might also consider alternative factors to the exchange rate such as private credit, money growth and other financial conditions.

In a previous version, we also reported “static SGMM”, which referred to as inflation and exchange rata variations as endogenous but without the lagged interest rate. We later considered output gap also as endogenous in the more general SGMM. In the current version, and to minimize confusion as implied by a comment from an anonymous referee, we include the lagged interest rate in all versions of SGMM. In both previous and current versions, all estimations of SGMM are performed under the xtabond2 command in STATA as in Roodman (2009b).

For dynamic panel data models, second-order serial correlation test is most important together with testing of instrument validity. The standard reference on specification tests for dynamic panel data model is Arellano and Bond (1991), who develop three such tests: a direct test on the second-order residual serial correlation coefficient [originally named m₂, commonly referred to as AB(2) more recently], a Sargan test of over-identifying restrictions and a Hausman specification test. Their conclusions were that “the robust m₂ statistics perform satisfactorily as do the two-step Sargan and difference—Sargan test, but the two-step Hausman test must be considered suspect in samples of this size.” Arellano and Bond (1991, p. 293). The influential article by Roodman (2009a) on instrument count for SGMM reports AB(2), together with Hansen and difference Sargan tests for the validity of instruments, but no AB(1) tests. Recent applied research with SGMM tends to report AB(2) together with Hansen’s J, but some also report AB(1) tests with mixed results. For example, for real exchange rates and productivity growth Caglayan and Demir (2014) usually report p values of 0 for AB(1) but those changes with alternative variables in the model with p values close to 0.10 or even 0.20 in some specifications. Gopalan and Rajan (2016) compare their findings on foreign aid for water supply and find no serial correlation of orders 1 and 2 for SGMM estimators but do find some serial correlation of order 1 for DGMM estimators. Rather than focusing on AB(1), the key consideration for the correct model specification is its stability, such as the lagged dependent variable coefficient being less than 1 and statistically significant, as well as satisfactory serial correlation tests of order 2 and some instrument validity tests as originally proposed by Arellano and Bond (1991) and revisited by Roodman (2009a).

The caveat for the subsample of ITers with only 9 countries in columns (4)–(6) of Table 4 is that some serial correlation shows up at around the 5% significance level.