Testing for Credibility Effects
  • 1 https://isni.org/isni/0000000404811396, International Monetary Fund

This paper examines some recent techniques designed to draw inferences about the credibility of changes in macroeconomic policy regimes. An alternative two-step approach, based on the decomposition between permanent and transitory components of a "credibility variable" is proposed. The methodology is then used to test for the existence of a credibility effect in the Cruzado stabilization plan implemented in Brazil in 1986.

Abstract

This paper examines some recent techniques designed to draw inferences about the credibility of changes in macroeconomic policy regimes. An alternative two-step approach, based on the decomposition between permanent and transitory components of a "credibility variable" is proposed. The methodology is then used to test for the existence of a credibility effect in the Cruzado stabilization plan implemented in Brazil in 1986.

I. Introduction

The role of credibility factors in the design of a disinflation program has been a major focus of research in macroeconomics over the past decade. 1/ It has been argued that credibility about a future change in the course of policy may be able directly to reduce the costs of disinflation by changing inflationary expectations. To reduce inflationary expectations effectively in accordance with the disinflationary goal, a newly implemented policy must appear credible to the public. Since inflationary expectations have a significant influence on current wage and price decisions, a reduction in actual inflation may result. To the extent that it is quantitatively significant, this “expectations effect” suggests that there is a potential gain (in terms of output and unemployment) in establishing the credibility of a monetary disinflationary program. 2/ By contrast, when policies implemented lack credibility—because, for instance, they are inconsistent with each other—rational agents would eventually recognize that the announced program is not sustainable in the long run, and fears of the program’s collapse would become self-fulfilling.

In a world with imperfect information, however, the speed at which the economy will react to a major change in the policy regime may be slow. 3/ Even if a change in regime is sufficiently credible—in the sense that policies are consistent and are believed by agents—the economy may not move quickly to a new equilibrium with lower price expectations. Indeed, instant credibility is unlikely in a stochastic world, and rational individuals would typically learn gradually about the new environment (Taylor, 1975). Ascertaining how quickly economic agents perceive a change in their environment has important implications for the short-run impact of policy changes.

Despite the importance of this issue for the conduct of macroeconomic policy, there have been only limited attempts to examine empirically the evidence related to regime changes. 4/ The purpose of this paper is to analyze alternative procedures to estimate the size of, and the diffusion effect of, credibility effects in the context of stabilization policies. The paper focuses on an analysis of the time-series properties of the inflation rate in high-inflation developing economies, using as an example the Cruzado Plan implemented in Brazil during 1986.

The remainder of the paper is organized as follows. Section II examines some basic conceptual issues in analyzing the role of credibility factors in the context of stabilization policies. Section III presents a critical review of existing methods designed to test for credibility factors. Section IV presents an alternative two-step procedure, which is not affected by the drawbacks of some of these methods. Section V applies the methodology to test for the existence of a “credibility effect” in the first phase of the Brazilian Cruzado Plan of 1986. The final section summarizes the main results of the analysis and discusses some possible extensions.

II. Credibility and Disinflation Policies: Analytical Issues

It has long been recognized that when the public lacks confidence in the ability of politicians to carry out a newly-announced stabilization program, disinflation becomes more difficult to achieve. The lack of confidence has been shown to result from a variety of sources. First, the government’s stabilization program may be perceived as being inconsistent with other policies being pursued simultaneously, and be recognized as such by the public. For instance, a disinflationary program which does not include measures to limit the public sector budget deficit will lack credibility—irrespective of whether “orthodox” or “heterodox” policies are called for—because the public understands its inconsistent nature. 5/ As discussed below, one of the key reasons why the Brazilian Cruzado Plan implemented in 1986 lost credibility rapidly was because private agents became gradually aware that the public sector budget at the onset of the stabilization was not balanced. 6/

Secondly, the lack of credibility may result from a time-inconsistency dilemma for the government: its optimal ex-post strategy may differ from its ex-ante strategy. For instance, once nominal wages are set by the private sector, the authorities may find it tempting to disinflate less than they had promised to, in order to generate output gains (Barro and Gordon, 1983a). This arises in the context of a model in which a monetary authority is concerned about both inflation and unemployment in a world governed by an expectations-augmented Phillips curve. The policymaker wants everyone to expect low inflation, so that it will face a favorable trade-off between inflation and unemployment. But an announcement of a policy of low inflation is not credible. Once expectations are formed, the authority has an incentive to renege on its announcement in order to reduce unemployment. 7/ Private economic agents understand the incentive to renege and therefore do not believe the policy announcement in the first place. The general implication of this analysis is, therefore, that in circumstances in which a policymaker has an ex-post incentive to renege on its promises, rational agents will discount announcements of future policy actions—or assurances regarding the continuation of present policies.

A third source of credibility problems is incomplete or asymmetric information: private agents may not be able to assess how serious the government really is about stabilizing the economy. Imperfect information of this sort is particularly prevalent in developing countries, where policymakers tend to change rapidly, generating “confusion” about the true policy objectives. 8/

Finally, a fourth source of credibility problems results form the uncertainty regarding the predictability of policy reforms or measures. In a stochastic world, even if a program is coherently formulated and time-consistent—in the sense that policymakers have no incentive to depart from the announced policy measures—exogenous shocks may occur which may be large enough to throw the program “off track”. Such shocks may be external in nature (such as, for instance, the behavior of world oil prices or interest rates) but may also result from the policy environment itself, in particular when the authorities have imperfect control over policy instruments. For instance, the announcement of a fiscal target will not be fully credible if the government does not adequately control the level of government expenditure. Another example relates to a situation in which the rate of inflation depends on the rate of expansion of domesit credit, which in turn depends on both deterministic and stochastic factors. Private agents will in general be aware of this and will accordingly form a probability that the policy target will not be met. 9/ The lower the degree of precision over policy instruments, the more likely it is that agents will anticipate the possibility of a future collapse of the stabilization effort. 10/ The lack of policy predictability may create, therefore, doubts about the sustainability of the reform process, and will affect the degree of credibility of an otherwise consistent and “viable” program. 11/

An important distinction which has often been blurred in the recent literature is that between reputation and credibility. In general, reputation refers to a government, while credibility refers to a given policy. Clearly, these concepts are related: an “anti-inflationary” reputation can only be established by a track record of continuously low-inflation policies. This distinction is, however, important for empirical analysis and will be examined below.

The analysis in this paper does not focus on any particular source of credibility problems discussed above. It is, in practice, very difficult to test for the existence of one particular type of credibility problem, and few authors have indeed attempted to do so in the past few years. 12/ Rather, most researchers have worked on the general premise that a “credible” disinflationary program will translate into a change in the process driving prices, while a program which lacks credibility will have no discernible effect. We now examine how this general presumption has been implemented in practice.

III. Testing for Credibility Effects: An Overview

This section provides a critical review of recent techniques designed for testing for the presence of credibility effects. 13/ We emphasize both the conceptual and empirical limitations that various tests are faced with. 14/

In testing for credibility effects, it is important to distinguish the immediate effects occuring when a change in policy is announced, which we shall call the announcement effect, and the effects occuring after agents have experienced that the announced policy change really has been pursued, which we shall call the implementation effect. 15/ This is important, since a policy change may gain credibility only gradually. As shown below, some methods have implicitly assumed that credibility is gained instantaneously, and have focused on the announcement effect.

1. Prediction-error method and proxy variables

Most empirical analyses of policy credibility so far have made use of the prediction-error method. 16/ Essentially, the method consists in estimating, over the pre-reform period, a model of the inflationary process and forming predictions over the post-reform period. A disinflationary policy is deemed credible if such predictions overestimate actual values of the inflation rate during the “new” regime, simply because the model has proved unable to account for such a regime change. 17/ The precise nature of predictive failure tests used in these applications varies, of course, but quite often are variants of a Chow stability test (see Harvey, 1990).

However, this procedure provides in general an unsatisfactory test of the credibility hypothesis because such prediction errors might explain almost anything not considered explicitly in the model—just as dummy variables would. Only in the case in which the residuals are due solely to the absent credibility variable (that is, if the model is correctly specified) will the prediction error method prove useful in testing for credibility effects.

An alternative procedure which can allow proper identification of a regime break is to define and include in the regression model a “credibility variable” explicitly and to test for its significance and stability over time. Christensen (1987b, 1990), for instance, has used the variability of the exchange rate as a proxy variable for credibility of an exchange rate regime. He argues that a necessary condition to sustain high credibility is that the variability of the exchange rate becomes positively correlated with the nominal interest rate. In the context of a disinflationary program in developing countries, the spread between the parallel market exchange rate and the official rate has often been used as an indicator of the degree of confidence in macroeconomic policies (see, for instance, Dornbusch et al. 1990).

There are, however, difficulties associated with the use of proxy variables for credibility, either because of the largely arbitrary way in which they are defined, or because of the failure to recognize their endogenous nature. For instance, in Christensen’s study mentioned above, measuring the credibility variable by the standard deviation of the exchange rate and thereby restricting credibility to be affected equally by appreciations and depreciations of the domestic currency may be inappropriate. In the case of developing countries, the parallel market premium itself may be an inadequate indicator of credibility because it is an endogenous variable, which responds to both the behavior of “fundamentals” (money supply, foreign prices etc.) and the degree of confidence. In our procedure described below, a solution to this problem is proposed.

2. Bayesian learning procedure

An alternative procedure, which may be particularly appropriate when agents learn progressively about a change in regime, is to formulate a Bayesian learning procedure and derive explicitly a probability that a disinflationary program will collapse. Credibility, in this context, is high when the probability of failure is low.

Baxter (1985) has provided an application of these ideas in her attempt to measure the evolution of “credibility” after the introduction of a pre-announced exchange rate policy in Argentina and Chile, in the late seventies. In her model, agents observe the pattern of public deficits and domestic money creation, and then estimate, through a Bayesian procedure, whether or not that pattern is compatible with the announced exchange rate rule and thus sustainable in the sense of not violating any steady-state equilibrium conditions. Her findings show that the measure of credibility declines over time for the Argentinian policy but not for that of Chile—which nevertherless experienced the same ultimate fate. Furthermore, her empirical measure of credibility correlates well with various observed economic variables (for instance, interest and inflation rates) in Argentina, and not in Chile, thus suggesting that market participants also rely on sources of information other than domestic money creation and fiscal deficits. 18/

A Bayesian procedure for estimating reputational and credibility effects has also been developed by Weber (1991), who measures an anti-inflation reputation by the probability that policymakers consistently pursue low-inflation policies, that is, the government’s resolve to pursue its efforts. This probability is derived by learning over time from the actual behavior of the policymakers. Formally, this probability is estimated by a Bayesian procedure which consists in finding how inflation could be forecasted over a given period under various alternative assumptions about the degree to which inflation shocks are allowed by the central bank to become permanent. The relative success of these forecasts is then evaluated over time. In particular, the procedure determines the probability that a weighted average of the various forecasts predicts inflation better than each of them separately. The weight attached to the “low-inflation” forecast yields the required measure of anti-inflation reputation.

Following Cukierman and Meltzer (1986), Weber defines credibility as the extent to which beliefs concerning a policy conform to official announcements about this policy. To achieve credibility, policymakers must precommit themselves to a particular policy rule. Such a pre-commitment may or may not be binding. Credibility is thus viewed as a measure of the degree to which policymakers “tie their hands” on future policies by issuing policy announcements. Again, following Cukierman and Meltzer (1986), Weber defines two measures of credibility. The first one measures average credibility, and estimates the extent to which private agents expect policy outcomes to deviate from prior policy announcements. The second measure estimates marginal credibility, and focuses on the ability of policy announcements to influence the public’s expectations; it measures the impact of a change in the announcement on expectations and may be thought of as the weight placed on the announcement when the public forms its expectations. This weight is equal to unity if the policymaker always makes completely accurate (fully credible) announcements and tends towards zero as the announcements become non-credible.

Weber uses univariate time-series models to evaluate how expectations of the public are affected by policy announcements. 19/ In this approach, however, it is difficult to distinguish between changes in the goodness-of-fit of the time-series model generating expectations and changes in the credibility of policymakers, because possibly relevant variables are excluded from the model. This can be a serious problem if policy announcements are correlated with some of the omitted variables. The results can be even more misleading if the relationships between policy announcements and other relevant factors change during the sample period. A multivariate approach system can mitigate this problem, as discussed below. Moreover, Weber’s procedure does not allow an explicit examination of the effect of credibility on the persistence of the inflationary process.

3. Time-varying parameters and the Lucas critique

Closely related to the credibility hypothesis is the Lucas critique (see Lucas, 1976) which states that policy regime changes affect the parameters of reduced-form models. 20/ The usual procedure for testing the relevance of the Lucas critique has been to compare regression coefficients across different regime periods, but this method also seems unsatisfactory because parameter instability may be due to general mispecification rather than the Lucas critique. It is therefore important that one allows parameters to vary directly with the policy variable, which has been done by estimating a time-varying parameter model.

Time-varying procedures have been developed and used by several authors to examine the timing of the change in the stochastic process driving inflation and interest rates. 21/ A particularly interesting technique is the method used by Mankiw et al. (1987). 22/ The approach allows the determination of the most likely date for the change in regime, conditional on the assumption that the change occurred all at once. It also allows for the possibility of a gradual change in regime—a particularly useful case, when credibility only builds through time.

Suppose that the process for the inflation rate obeys

πt+1=α0+α1πt+εt+1,t=1,,Ts1,(1a)
πt+1=β0+β1πt+εt+1,t=Ts,,T(1b)

where Ts is the switch date (the first period of the new regime). Assuming normal errors, the log-likelihood function for this model is

logL=(T/2)log(2π)log(σo2)(TTs+1)log(σn2)(1/2)Σst=1T1(εt+12/σn2),(2)

where σo2 and σn2 are the error variances in the old and new regimes. The maximum likelihood value for Ts can be determined by computing the maximum likelihood estimates of the parameters for all possible Ts’s and then choosing the value of Ts with the maximum likelihood. This “switching regression” procedure was suggested by Goldfeld and Quandt (1973).

A second procedure for determining the timing of the change in the inflation process is to estimate a time-varying parameter model that allows the coefficients of the equation to change gradually over time, rather than moving instantaneously from the old to the new values as in the switching regression above. Mankiw et al. (1987), assume that the parameters in a model similar to (1) follow a logistic curve. 23/ Their approach allows the calculation of the rate of adjustment between changes in regime. Because the limit of the logistic curve is the step function, this procedure includes the earlier process as a special case.

A difficulty with the switching-regression technique is that, with a single switch point, it is only useful for examining regime shifts which are both credible and permanent within the data sample period. In many cases, stabilization programs may achieve credibility for a limited period before breaking down. Extending the above analysis to allow for such a situation would involve allowing for two or more switch points.

IV. An Alternative Approach

We now present an alternative technique, which is less vulnerable to the limitations of some of the procedures described earlier. Our analysis rests on two key assumptions. First, inflation is, because of inertial forces (wage contracts, financial indexation, etc.) a slow-moving process, which displays considerable persistence. 24/ Second, the transitory component of the parallel market premium and not its level is the appropriate proxy to measure the degree of credibility of a stabilization package. 25/ When a disinflationary program lacks credibility, agents will typically anticipate a resumption in inflation in the future. As a consequence, to avoid the inflation tax on real monetary balances, they switch to foreign-currency denominated assets. For given supply, the parallel market exchange rate depreciates, and the premium rises. The premium captures, therefore, the degree of confidence in existing policies (see Agénor and Bhandari, 1991). However, the premium is an endogenous variable, and its level will also reflect the behavior of “fundamentals”—domestic credit, foreign prices, etc. It is therefore important to “purge” the variable to account for this endogenous component, since these will typically be altered at the inception of a stabilization attempt.

To test for the existence of a credibility effect on the behavior of inflation, we then examine the relationship between the coefficients driving the inflationary process, and the transitory component of the premium. Such a relationship is estimated via a Kalman filter procedure.

1. Estimating the credibility variable

The first step in our procedure is to decompose the the parallel market premium, ρt, into a “systematic” (permanent) component and an “error” (transitory) component which will be taken as a measure of credibility. To do so, the procedure suggested here would start by estimating a static regression or a vector autoregression (VAR) system, which includes “fundamental factors”, captured in the vector of predetermined variables zt: 26/, 27/

ρt=ztδ+ut,(3a)
[ρtzt]=δ(L)[ρtzt]+ut,δ(L)=δ1L+(3b)

where δ and δk, k = 1,…, denote vectors of parameters and ut a white noise process.

A vector autoregression formulation is a natural one to estimate when ρt is known to respond to “fundamentals” according to some theory, but the exact form of the relationship is unknown. By focusing on a multivariate time-series representation in which lagged values of the premium as well as other variables affect its current value, we are assuming that agents use all available information in forecasting the premium. The negative of the transitory component can be taken as an indicator of expectations, or, more generally, confidence. In this procedure, therefore, it is not the premium itself but rather its transitory component that is interpreted as the credibility variable. By construction, the residuals from (3a) or (3b) are unbounded and have zero mean over the estimation period. 28/ It is therefore important to consider a sample large enough (relative to the period of interest) if one is to avoid imposing inadequate restrictions on the credibility variable. 30/

It is also important to note that any decomposition of an observable variable into a systematic and unsystematic components depends on the choice of information set, that is, on the choice of variables zt: white noise on one information set can be predictable using another. Thus, there is an inherent lack of uniqueness in using white noise residuals as a criterion for data coherency, although non-random residuals do indicate data “incoherency”. 31/ The VAR approach allows, however, to control for the environment to some extent without imposing “too many” a priori restrictions on parameters.

2. Credibility and inflation inertia

The second step in our procedure is to estimate a backward-looking process for inflation with parameters varying with the transitory component of the parallel market premium, using a Kalman filter approach. Assuming, for simplicity, that inflation, πt, is driven by a first-order autoregressive process (an assumption which is tested below), and denoting by ct the credibility variable, the system to be estimated is given by

πt=αtπt1+εt,(4a)
αt=αt1+γct+ηt,(4b)

where

[εtηt]N{[00],[σε200ση2]}(4c)

and πt is in mean-deviation form. Equation (4a) represents the measurement equation and equation (4b) the transition equation. The disturbances εt and ηt are serially uncorrelated disturbances with zero mean and constant variances, and are assumed uncorrelated with each other in all time periods. The Kalman filter allows us to update the estimate of a through time, using a two-step methodology. The first consists in forming the optimal predictor of the next observation, given all information currently available, by means of a prediction equation. The new observation is then incorporated into the estimator of the state vector using the updating equation. The resulting estimates of αt are “smoothed” over time, because they are based on all information available. 1/

The above discussion suggests that the resulting estimate of γ should be negative: the higher credibility is, the lower the “inertial” effect on inflation. Consequently, the coefficient αt should fall after the implementation of a credible disinflation program. Although the procedure does not allow to give a precise interpretation to the size of γ, its sign is what matters.

V. Credibility of the Brazilian Cruzado Plan, 1986-87

This Section applies the methodology developed above to a recent stabilization experiment, the Brazilian Cruzado Plan, implemented in 1986-87. We first provide a brief overview of the Plan, before considering the econometric results.

1. An overview of the Cruzado Plan

The Brazilian Cruzado Plan, introduced at the end of February 1986, consisted of a general freeze of prices, the exchange rate, and tariffs for public services. 33/ In addition, the government removed indexation, provided an upward adjustment of wages to their average real value over the previous six months, and established provisions for automatic adjustment of wages to be triggered when accumulated inflation exceeded 20 percent. However, key public sector prices (with the exception of electricity rates) and controlled private sector prices were not adjusted prior to implementation of the Plan, as originally planned.

The first months of the plan were highly successful. Monthly inflation rates were between 1 and 2 percent, industrial output increased by 40 percent between February and October of 1986 (while GDP expanded at an annual rate of nearly 8 percent during that period), and in November 1986 real wages were 9 percent higher than the average attained in 1985 (see Figure 1). 34/ But by late August 1986 the plan was becoming unsustainable. The price freeze was not consistent with the increase in domestic demand generated by the initial wage increase and the larger operational deficit of the public sector, which turned out to be 3.7 percent of GDP rather than the 0.5 percent projected early in the year. Excess demand developed in goods and labor markets. Imports rose sharply while exports fell, and the trade surplus diminished substantially in September 1986—turning into a deficit between October and December of that year. 35/ The combination of excess demand with misaligned relative prices (resulting from price controls) and large fiscal deficits led to unsustainable macroeconomic imbalances that finally undermined the stabilization attempt. 36/, 37/

FIGURE 1.
FIGURE 1.

BRAZIL: Prices, Output, and Net foreign Assets, 1985/87

Citation: IMF Working Papers 1991, 110; 10.5089/9781451853254.001.A001

When price controls were abandoned in December 1986, inflation rose to monthly rates of between 10 and 15 percent until March 1987 and accelerated even more from April 1987 to reach a monthly rate of 26 percent in May and June of that year. The acceleration of inflation led the government to adopt a new stabilization plan, known as the Bresser Plan. The plan tried to correct some of the errors of the Cruzado Plan, adjusting public sector tariffs and the exchange rate at the beginning of the price and wage freeze. This time the freeze was to be temporary, and wage indexation was to be restored three months later. This program too was ultimately a failure: inflation rose rapidly in the last quarter of 1987 (more than 10 percent a month), and by 1988 the annual inflation rate was approaching hyperinflation, at about 1,000 percent.

Table 1 indicates the results of opinion polls conducted immediately after the announcement and during the Cruzado Plan. The evidence suggests that the Cruzado Plan had strong initial popular support. However, support begun to dissipate once the initial success was replaced by a resumption of inflation. During March and June 1986, there was growing awareness that the public sector budget at the onset of the stabilization was not balanced. The situation grew worse as a result of the increase in expenditure of the federal government on the wage bill, subsidies, tax exemptions, and transfers to state enterprises and state and local authorities. The implementation of the plan suffered from three major shortcomings (Modiano, 1988). First, the public sector deficit turned out to be too large to finance without substantial recourse to the inflation tax. Second, consumption demand became excessive because of overly generous wage adjustment when the plan was announced, and because of the failure to realign some key public and private sector prices before the plan was announced and introduced. Third, the price freeze (including the exchange rate) was maintained too long given the excess aggregate demand which was apparent. In view of these increasingly perceived realities, credibility of the plan begun to erode, and inflationary expectations increased. The government was slow to take corrective action, further aggravating the situation. The inflationary spiral was reactivated because agents perceived that recourse to inflationary finance was likely to be renewed, considering the size of the fiscal deficit, and because excess demand developed.

Table 1.

Brazil: Public Opinion Polls, 1985-86 1/

article image
Source: Novaes (1990, p. 21).

Results refer only to Sao Paulo, but a similar pattern was observed in Rio de Janeiro.

The behavior of the parallel exchange rate and the premium are shown in Figure 2. The figure suggests that the premium fell substantially before the plan was implemented, but started to rise at the end of March 1986. Reflecting excessive monetization—the rate of growth of the broader monetary aggregate M4 reached 21 percent between the end of February and the end of June—the parallel market premium jumped between March and June 1986 from 26 percent to 50 percent, and reached 90 percent in October 1986. Deteriorating confidence as well as anticipations of a large devaluation of the Cruzado—which turned out to be unfounded—account for this sharp rise. After the announcement of new policy measures in November 1986, the premium fell substantially.

FIGURE 2.
FIGURE 2.

BRAZIL: Exchange Rates and Parallel Market Premia, 1985/87

Citation: IMF Working Papers 1991, 110; 10.5089/9781451853254.001.A001

2. Evaluating the credibllity of the Cruzado Plan

Following the methodology described in Section IV, we now examine whether statistical results can account for the “credibility pattern” suggested by survey evidence and casual observation—rapid gains in credibility at the inception of the Cruzado Plan, followed by a rapid erosion in the following months.

The first step of the procedure is to estimate either a static regression or a VAR system, which includes “fundamentals”. The variables considered here as “fundamentals” are the rate of growth of the money stock, changes in output, domestic and foreign inflation rates, and the rate of depreciation of the official exchange rate. 39/ Seasonal dummy variables were also included as independent variables in the regressions. Figure 3 shows the credibility measures—that is, the negative of the residuals series—derived from a first- and second-order VAR systems. 40/ For the period of interest—the period which covers the adoption to the abandonment of the Cruzado Plan—both measures display a fairly similar pattern: credibility rose sharply at the inception of the program and fell rapidly in the following months.

FIGURE 3.
FIGURE 3.

CREDIBILITY VARIABLES, 1982-89

Citation: IMF Working Papers 1991, 110; 10.5089/9781451853254.001.A001

Maximum likelihood estimates of the parameters of the model (4a)-(4c) using the credibility index derived from the first-order VAR system are as follows (asymptotic standard error in parentheses): 41/

γ=0.00035,(0.00007)σε2=0.0015,ση2=0.0029,(5)

while Figure 4 shows the behavior of the coefficient α. The coefficient γ is negative, as predicted by the model, and highly significant. Also, as suggested by the analysis, the persistence effect (which increased steadily from the end of 1985 to reach an all-time high in April 1986) started to fall dramatically until approximately June 1987—right after the Plan collapsed—and increased rapidly to its pre-program value, reflecting the explosive behavior of prices when policy credibility is low.

FIGURE 4.
FIGURE 4.

CREDIBILITY AND INFLATION PERSISTENCE, 1982-89

Citation: IMF Working Papers 1991, 110; 10.5089/9781451853254.001.A001

It is interesting to compare the results given in (5) with those obtained when credibility is measured by the inverse of the actual premium—so that ct = 1/ρt. Estimation of the model (4a)-(4c) yields in this case

γ=0.058,(0.067)σε2=0.0016,ση2=0.0031,(6)

The above results indicate that the parameter 7 has the wrong sign and is not statistically significant. Using the actual premium would therefore indicate no credibility effect on inflation persistence, in contrast to the results shown in equation (5) which allows for the effect of “fundamentals” on the behavior of the exchange rate spread.

Overall, the results suggest that although the Cruzado Plan seems to have gained credibility rapidly, its impact of the inflationary process has been less dramatic.

VI. Concluding Comments

The purpose of this paper has been to examine alternative procedures to estimate the size of, and the diffusion effect of, credibility factors in the context of disinflation programs. After a brief review of conceptual issues in evaluating the source of credibility problems and an overview of empirical techniques, a two-step methodology has been developed. The method, which may be viewed as particularly suitable for high-inflation developing economies, focuses on an analysis of the time-series properties of the parallel market premium and the inflation rate. It relies on a decomposition of the parallel market premium into a permanent and transitory component and tests for statistical correlations between the transitory component of the premium (which can be viewed as a “credibility variable”) and the coefficients characterizing the process driving the inflation rate. To demonstrate its usefulness in analyzing stabilization programs, the technique was applied to the Cruzado Plan implemented in Brazil during 1986.

There are several possible extensions of the analysis developed in this paper that may prove worthwhile. First, the robustness of the proposed two-step procedure should be tested by examining a larger group of case studies. A comparative analysis of the experience of Argentina, Bolivia, Israel and Peru, which have recently pursued disinflation strategies, would provide considerable insight in the general applicability of our technique. Second, a different approach to the first-step problem (generating a credibility variable) can be implemented. In many developing countries, public or private organizations conduct monthly opinion polls on economic predictions which can prove useful in determining the degree of confidence in the economy. Techniques used to derive quantitative series of expected inflation from qualitative survey data are well developed and could be applied to derive a “credibility index”. 42/ The major advantage of this approach is that it would allow one to evaluate the degree of credibility of a policy package ex ante, rather than ex post, as is done here. 43/ More generally, it would make it possible to estimate the extent to which regime changes have been anticipated—without recourse to “constructed” measures of expectations. Finally, even in the case where the credibility variable needs to be somehow “created”, it could be useful to consider variables other than the parallel market premium, such as stock prices, the degree of financial intermediation, or the extent of capital flight. Nevertheless, this paper demonstrates that quantification of credibility is essential to explain the empirical aspects of stabilization policies.

Testing for Credibility Effects
Author: Pierre-Richard Agénor and Mr. Mark P. Taylor