Is Delayed Disinflation More Costly?

• Author: Boschen, John F.
• Position: Statistical Data Included

John F. Boschen [*]

Charles L. Weise [++]

An often-stated piece of monetary policy wisdom is that postponing disinflation raises the ultimate cost of disinflation. However, there is little empirical work that directly tests the validity of this view. This paper is a study of the relation between the duration of inflation and the cost of subsequent disinflation using 67 disinflation episodes in OECD countries from 1960 through 1993. The estimates indicate that delaying disinflation raises the output loss per unit of inflation reduction, although the effect is not statistically significant in all models. The largest marginal effects of delay occur soon after inflation is allowed to get under way, a finding consistent with a rapid decline in the credibility of the inflating central bank.

1. Introduction

   The belief that postponing disinflation raises the eventual cost of disinflating is part of the received wisdom of many policy makers. The Board of Governors of the Federal Reserve System (1994, pp. 17-8) offers one statement of this view:

   Countering [the] threat of inflation with more restrictive monetary policy could risk small losses of output and employment in the near term but might make it possible to avoid larger losses later should expectations of higher inflation become embedded in the economy.

   There are some reasons why delaying disinflation might raise the cost of disinflation. One channel through which a delay could affect the cost of disinflation is by its effect on central bank credibility. The impact of credibility on the cost of disinflation has been widely discussed since Sargent's (1982) seminal work. In the present case, the longer inflation is allowed to rise, the more likely it is that private agents view inflation stabilization as having a low priority and the less credibility they would assign to an announced disinflation. A second channel through which a delay could adversely affect the cost of disinflation is by inducing changes in nominal wage-setting behavior. For example, the belief that sustained inflation is the ongoing economic policy could lead to an expansion of lagged indexation arrangements in labor contracts. The spread of such arrangements could prevent wages from adjusting quickly in response to a decline in inflation, and this could add to the cost of a disinflation. [1 ]

   There is no formal econometric evidence, of which we are aware, that directly supports or rejects the proposition that the duration of inflation raises the cost of disinflation on a per-unit-of-reduced-inflation basis. Given the policy implications, it is useful to present some empirical tests of this hypothesis. The framework for this analysis is the empirical literature on the determinants of the output loss associated with disinflation episodes. This literature has already examined a number of variables, notably nominal wage flexibility (Ball 1994), the speed of disinflation (Sargent 1983), and the level of inflation (Ball, Mankiw, and Romer 1988), with respect to their impact on the cost of disinflation.

   This paper extends our previous work on the cost of disinflation, Boschen and Weise (1999), by examining the relationship between the duration of inflation and the cost of disinflation in considerably more detail than was done in that paper. Here we use a less restrictive empirical structure to assess the robustness of the duration-sacrifice ratio relationship with respect to alternative control variables. We also address the reverse causation issue (i.e., high expected costs of disinflation cause policy makers to delay disinflation) using two procedures. First, we use a model of inflation duration and the cost of disinflation that controls for the major ways that reverse causation could arise. We estimate this model using instrumental variables. Second, we test whether the duration of inflation has nonlinear effects on the sacrifice ratio that are consistent with a rapid decline in the credibility of the central bank. Specifically, we examine whether the effect of inflation duration on the sacrifice ratio i ncreases most rapidly during the early stages of an inflation episode.

   Our study is based on a data set of 67 disinflation episodes drawn from 19 OECD countries over the years 1960-1993. Our analysis of these disinflations indicates that the longer a disinflation is postponed, the higher is the output cost per unit of inflation reduction. We also find that the largest marginal losses for delaying a disinflation occur early in the inflation regime. The statistical significance of these findings varies across model specification and estimation technique.

   The paper has three remaining sections. Section 2 provides an overview of the disinflation episodes used in the paper. Section 3 discusses empirical issues and presents the estimates of the models. Section 4 summarizes the findings.

2. Disinflation Episodes, the Sacrifice Ratio, and the Duration of Inflation

   The data set of disinflation episodes is based on the trend inflation measure used in Boschen and Weise (1999). We define the start date for a disinflation episode as the year in which our quarterly trend inflation series began a decline of two percentage points or more from the peak value. For example, if trend inflation peaked in 1974:2, then we use the year 1974 as the start date for the disinflation. Trend inflation for each country is a centered nine-quarter moving average of quarterly consumer price index (CPI) inflation. The 2% rule and the smoothing procedure filter out small temporary variations in inflation and focus on the larger policy-induced disinflations. [2] The duration of rising inflation prior to a disinflation (DURINF) is the number of years between the previous trough in trend inflation and the peak in trend inflation. As an example, trend inflation in the United States was at a trough in 1971, thereafter steadily rising to a peak in 1974. The value of DURINF for this disinflation episod e is three. The average value of DURINF across all countries in our sample is 4.3 years.

   Our measure of the output cost of disinflation is the sacrifice ratio, defined as the cumulative percentage deviation of actual log output from trend log output over the disinflation period divided by the decline in trend inflation. The construction of the sacrifice ratio follows Ball's (1994) method. Log output is assumed to be at its trend level at the peak of trend inflation and again at its trend level four quarters after the inflation trough. Trend output for each quarter during a disinflation is obtained by using the implied compound growth rate between log output at these two points. The output loss due to the disinflation is the cumulative difference between trend and actual output from the inflation peak to four quarters following the trough. The denominator for the sacrifice ratios is the peak in trend inflation less the trough in trend inflation.

   As Ball has pointed out, there are some clear advantages of this approach to measuring the cost of disinflation when working with a multicountry data set. First, we need not assume a fixed form for the Phillips curve (linear or nonlinear), which may or may not apply to all countries and all time periods. The trend-gap method also allows calculation of a separate sacrifice ratio for each disinflation. This advantage is the critical one, given that we want to pin down the impact of a delay in disinflation on the subsequent output loss for each disinflation. In addition, this calculation of the output loss treats all disinflation episodes in a consistent manner that does not rely on case-by-case assessments.

   The...