RDP 2004-04: Inflation Convergence Across Countries 2. Inflation Convergence

Ball and Sheridan examine various aspects of the economic performance of OECD countries, including output growth, inflation variability and inflation persistence. They also investigate changes in several bivariate relationships. This paper, however, only focuses on Ball and Sheridan's inflation convergence results, which are reviewed in this section.

2.1 Ball and Sheridan

Ball and Sheridan (forthcoming) introduce their inflation convergence argument in a paper evaluating the comparative performance of countries who have adopted inflation-targeting regimes. They initially produce results suggesting benefits from inflation targeting, namely that the seven OECD countries that adopted inflation targets experienced larger falls in inflation than the thirteen OECD countries that did not adopt the regime.^[2] However, Ball and Sheridan argue that these results merely reflect the fact that inflation targeters had higher initial inflation, and ‘there is regression to the mean’. In other words, countries with high inflation will experience a larger degree of disinflation just by returning to some underlying cross-country mean rate of inflation. Ball and Sheridan use the following simple OLS regression to illustrate the dominance of inflation convergence (regression (2) in Ball and Sheridan):^[3]

where is the average inflation rate for country i after the adoption of inflation targeting, is the average inflation rate for country i before the adoption of inflation targeting, and ITDummy_i is a dummy variable equal to one for the inflation-targeting countries.^[4] The results produced by Ball and Sheridan are replicated in Table 1.^[5] These show that when only the ITDummy_i variable is considered, as in Regression 1, there is evidence to suggest that inflation-targeting countries had higher disinflation, with β₂ being negative and significant. Implicitly, this is the specification considered in the earlier empirical evidence on inflation targeting. However, drawing on the insight from Ball and Sheridan, if one instead uses the initial inflation rate in the regression, as in Regression 2, the coefficient on initial inflation, β₁ is negative and highly significant, and explanatory power of the regression is much higher than that of Regression 1. When both variables are included in Regression 3, the results suggest that initial inflation is more important in explaining the change in inflation over the 1990s, with β₁ significantly negative, while the β₂ coefficient is still negative, but no longer significant. These results suggest that higher inflation in the earlier period is associated with a larger fall in inflation between the two periods, and that for a given level of initial inflation, there is no difference in disinflation between inflation targeters and non-inflation targeters. Together with similar findings on inflation volatility, these results cause Ball and Sheridan to conclude that ‘inflation targeting has no beneficial effects’.

This inflation convergence phenomenon is illustrated in Figure 1, which plots average inflation in the post-inflation-targeting period less the average inflation in the pre-inflation-targeting period on the y-axis against average inflation in the pre-inflation-targeting period on the x-axis. The strong negative relationship is readily apparent.

The finding that the inflation-targeting dummy is significant by itself, but no longer significant once the initial inflation variable is included in Equation (1) is important. It suggests that the initial inflation variable and the inflation-targeting dummy are positively correlated, which is consistent with Ball and Sheridan's finding that the inflation-targeting countries had higher initial inflation. This correlation shows up in a significantly positive coefficient from regressing the inflation-targeting dummy on initial inflation (not shown).^[6] The positive correlation between the two variables is problematic, as it raises the possibility of endogeneity in the choice to adopt inflation targeting.

Indeed, in his comments on the Ball and Sheridan paper, Gertler (forthcoming) suggests that the inflation convergence argument is effectively an attempt to account for the potential endogeneity of inflation targeting. In particular, he argues that it is possible that the history of high inflation (as shown in Ball and Sheridan) induced certain countries to adopt inflation targeting. Thus there are two possible explanations for the favourable performance of inflation-targeting countries: either inflation targeting caused better performance, or the better performance merely reflects the fact that countries with a history of high inflation chose to adopt inflation targeting and that the inflation rates of these countries converged independently of the adoption of inflation targeting. Interpreted in this manner, it does seem that Ball and Sheridan have raised a valid criticism of the earlier empirical evidence on the impact of inflation targeting. However, Gertler goes on to argue that Ball and Sheridan's convergence framework is not ‘sharp enough’ to control for the potential endogeneity, disagreeing with their conclusion that inflation targeting is irrelevant.^[7] He points out that Ball and Sheridan's results are also consistent with an alternative interpretation, whereby inflation targeting indeed facilitated the disinflation.

2.2 An Investigation of the Reasons for Inflation Convergence

Ball and Sheridan highlight the inadequate treatment of potential endogeneity as a flaw in the existing literature on inflation targeting. However, the inflation convergence framework they use to control for ‘endogeneity’ would appear flawed in that ‘regression to the mean’ has no solid theoretical foundation. While there is a clear theoretical justification for the phenomenon of price level convergence based on the law of one price, it has been shown that price level convergence across countries can actually lead in the short term to some dispersion in inflation rates (see, for example, Rogers 2002).

In their paper, Ball and Sheridan do not discuss a theoretical model underlying their inflation convergence argument. Their explanation for why ‘regression to the mean’ occurs is captured by the following quote:

Poor performers in the pre-targeting period tend to improve more than good performers simply because initial performance depends partly on transitory factors.

This explanation is rather specific in that it assumes these ‘transitory factors’ as the cause for higher inflation in the preceding period, implying that there is little inertia in the inflation outcomes of countries.^[8] Moreover, the explanation generally ignores the role played by monetary policy. In particular, it is widely accepted that regardless of the framework, monetary policy is one of the most important determinants of inflation in the long run.^[9] It is highly likely that poor inflation performance is at least partially caused by poor policy. So while ‘transitory factors’ are relevant, arguably it is the response of monetary policy to these factors which is more likely to cause the inflation performance of countries to vary. If indeed poor policy leads to poor inflation performance, it would seem likely that an improvement in the way policy is conducted would lead to better inflation outcomes. In other words, policy could indeed be the reason for the observed mean-reversion result.

In the absence of a convincing theoretical case for inflation convergence, Ball and Sheridan's explanation appears to rely on convergence being a statistical or mechanical property of cross-country performance. If so, to attribute the observed fall in inflation in the inflation-targeting countries to the convergence phenomenon, one would ideally be able to show that the phenomenon is a stable property in cross-country inflation data over time. If it is not, ‘regression to the mean’ seems more like an ex-post observation about inflation in the 1990s, rather than an explanation for the better performance of inflation-targeting countries.

This paper evaluates two explanations for inflation convergence. The first is that inflation convergence is a mechanical property of cross-country inflation performance, as implied by Ball and Sheridan. The alternative explanation considered is that inflation convergence is brought about by monetary policy, more specifically, convergence in the objectives of policy.

Footnotes

The inflation targeters are Australia, Canada, Finland, New Zealand, Sweden, Spain and the United Kingdom. The other 13 OECD countries are Austria, Belgium, Denmark, France, Germany, Ireland, Italy, Japan, the Netherlands, Norway, Portugal, Switzerland and the United States. Ball and Sheridan wish to examine ‘major developed, moderate inflation economies’, and hence their sample excludes the emerging-market economies that joined the OECD after 1990 (Czech Republic, Hungary, Mexico, Poland, Slovak Republic and South Korea), countries that experienced annual inflation over 20 per cent since 1984 (Greece, Iceland and Turkey) and countries that lacked an independent currency before the introduction of the euro (Luxembourg). [2]

Ball and Sheridan's specification is rearranged so that the coefficient on the initial inflation variable is β₁. This is done to avoid any confusion in the case of results from Equation (2). Also, subscript i is added to the specification for clarity. [3]

Countries that did not adopt inflation targeting are assumed to start the post-inflation-targeting period at the average adoption date of inflation targets for the seven targeters. Ball and Sheridan calculate this is as the September quarter 1993. [4]

Table 1 presents the results of Equation (1) for three different specifications: the first with the dummy variable omitted, the second with the initial inflation variable omitted and the third being the full specification, Regressions 1, 2 and 3, respectively. The results of Regression 2 were calculated by the author. [5]

Using more sophisticated probit and logit models, Mishkin and Schmidt-Hebbel (2002) and Hu (2003) also find some evidence that high inflation increases the likelihood of adopting inflation targeting. [6]

An alternative and perhaps more obvious way to control for the potential endogeneity in the adoption of inflation targeting would be through the use of instrumental variable (IV) techniques. However, the use of an endogeneity correction in analysing the performance of inflation-targeting countries is not really feasible given the small sample available, and hence that approach is not attempted here. [7]

Indeed, the analysis in Section 4 shows that the inflation-targeting countries had higher inflation than other OECD countries ever since the early 1970s oil-price shock, which suggests that these ‘transitory factors’ lasted for over 15 years! [8]

As Friedman (1970, p 24) famously said, ‘inflation is always and everywhere a monetary phenomenon’. [9]