Models of Inflation and the Costs of Disinflation
Author:
Mr. Bankim Chadha
Search for other papers by Mr. Bankim Chadha in
Current site
Google Scholar
Close
,
Mr. Paul R Masson
Search for other papers by Mr. Paul R Masson in
Current site
Google Scholar
Close
, and
Mr. Guy M Meredith
Search for other papers by Mr. Guy M Meredith in
Current site
Google Scholar
Close

The focus of this analysis is on the output costs of disinflation. A model of inflation with both forward and backward elements seems to characterize reality. Such an inflation model is estimated using data for industrial countries, and the output costs of a disinflation path are calculated, first analytically in a simple theoretical model, then by a simulation of a global, multiregion empirical model. The credibility of a preannounced path for money consistent with the lowest output loss is considered. An alternative, more credible policy may be to announce an exchange rate peg to a low-inflation currency. [JEL E31, E52]

Abstract

The focus of this analysis is on the output costs of disinflation. A model of inflation with both forward and backward elements seems to characterize reality. Such an inflation model is estimated using data for industrial countries, and the output costs of a disinflation path are calculated, first analytically in a simple theoretical model, then by a simulation of a global, multiregion empirical model. The credibility of a preannounced path for money consistent with the lowest output loss is considered. An alternative, more credible policy may be to announce an exchange rate peg to a low-inflation currency. [JEL E31, E52]

The focus of this analysis is on the output costs of disinflation. A model of inflation with both forward and backward elements seems to characterize reality. Such an inflation model is estimated using data for industrial countries, and the output costs of a disinflation path are calculated, first analytically in a simple theoretical model, then by a simulation of a global, multiregion empirical model. The credibility of a preannounced path for money consistent with the lowest output loss is considered. An alternative, more credible policy may be to announce an exchange rate peg to a low-inflation currency. [JEL E31, E52]

It is now widely recognized that the primary objective of central banks is—or should be—the maintenance of price stability.1 Recent initiatives in several countries have attempted to formalize this objective: in the United States, a congressional resolution mandating a zero inflation target for the Federal Reserve has been proposed, while in Canada, the Bank of Canada has announced a near-term target for inflation and the ultimate goal of price stability.2 In Europe the statutes for a European System of Central Banks, agreed at Maastricht in December 1991, state that price stability will be the primary objective.3 In this respect, these statutes would echo the primary responsibility of the Deutsche Bundesbank, namely, to safeguard the stability of the currency.

Concern to reduce inflation has made disinflation a major macroeconomic issue of the past two decades.4 Reducing the rate of inflation through contractionary aggregate demand policies has in all countries been associated with a decline in output and employment relative to potential; for the United States, estimates of the “sacrifice ratio” of percentage points of cumulative output loss that must be incurred in order to lower the rate of inflation by 1 percentage point range from 3 to 18 (Sachs (1985)). Yet the causes of these output losses are imperfectly understood. If nominal rigidity is due to the stickiness of the levels of wages (as in the models of Fischer (1977), Phelps (1978), Taylor (1979, 1980), or Calvo (1983a, 1983b)), then a path that leads to a lower rate of money growth (as opposed to a lower level of the money supply) need not cause output losses, as long as price expectations correctly anticipate the monetary deceleration. Such a disinflation path was calculated for a staggered wage contracts model by Taylor (1983).5 In these circumstances, it is hard to understand why central banks could not convince the private sector that they were going to move to a money growth rate consistent with price stability, and when they did so, that wage increases would not decelerate smoothly, leaving employment unchanged. However, experience with disinflation suggests instead that there may be some fundamental stickiness in inflation rates that makes achievement of costless disinflation difficult, if not impossible.6

In this paper, we argue that modeling the process for inflation requires including both forward-looking and backward-looking elements. Inertia in inflation may be due to partially nonrational expectations; if expectations are adaptive, then the inflation process is purely backward looking, as in Ball (1990). In any case, a simple specification in which the two elements are present—in particular, expected future inflation and lagged inflation, with weights that sum to unity—rejects both restrictions of a unit coefficient or a zero coefficient on lagged inflation. Using this specification, we show that there is a critical value for the relative importance of the forward-looking part, above which a disinflation path with zero output losses is possible. Such a disinflation path, if it exists, implies a monotonic decline in inflation, which asymptotes to zero. The path for the money supply, however, is not monotonic.

The paper presents estimates of this inflation equation using pooled cross-section, time-series data for the industrial countries. Estimates of the forward-looking parameter are close to the threshold value mentioned above that permits costless disinflation. We then proceed to simulate various disinflation paths, using the estimated inflation equations in a global, multiregion model, MULTIMOD. These simulations imply sacrifice ratios that are positive, though below the range mentioned above. We explore the implications of changing the parameter on forward-looking behavior, by an amount suggested by the estimated standard error.

Another aspect of the disinflation problem is imperfect credibility of an announced deceleration of the money supply. Using model simulations, we illustrate how various degrees of credibility affect the output losses associated with a move to price stability. One particular feature of disinflation is that lower inflation is associated with higher money demand if money demand has a negative interest elasticity. To avoid a decline in the price level (that is, a period of negative inflation), the money supply has to be increased at some point through an acceleration of money growth. Such a path for the money supply may strain credibility, adding to output losses. An alternative available to some countries, for instance those in the European Monetary System (EMS), may be to use an exchange rate peg to a low-inflation currency as a focus for the disinflation policy, letting the path of the money supply be endogenous. We explore the implications of such a policy, compared to an imperfectly credible announced path for the money supply, accompanied by exchange rate flexibility.

A final section advances some conclusions on the basis of our results and sketches areas for further research.

I. Price Dynamics and the Phillips Curve

Many modern macroeconomic models, and most large macroeconometric models, embody some form of the expectations-augmented Phillips curve as the basic building block describing the adjustment over time of prices (wages) to movements in aggregate demand around potential, or capacity, output (the natural rate of unemployment). The specific form of the Phillips curve implemented can play a key role in determining the short-run response to changes in policy or exogenous shocks.7

The dynamics of price adjustment in Keynesian models has always been a subject of considerable debate. Much of the early literature on price adjustment in these models relied, either explicitly or implicitly, on mistaken expectations on the part of agents in goods or labor markets or on the ad hoc existence of asymmetrical rigidities between wages and prices. Since then, various attempts have been made to develop models of disequilibrium dynamics that do not rely on expectational errors.8 An important and influential development has been the staggered contracts models of Phelps (1978), Taylor (1979, 1980), and Calvo (1983a, 1983b). (See also Blanchard (1983) and Mussa (1981a, 1981b).) These models emphasize that wages tend to be set in nominal terms for a discrete period of time and are set by different agents at different points in time—that is, they are asynchronous—and, therefore, contracts overlap. Agents are assumed to contract a wage in accordance with their anticipations of future price and output levels for the expected duration of the contract. These models typically assume that prices are a constant markup over wages and focus on the persistence induced in the aggregate price (average wage) level due to the asynchronous and overlapping nature of wage contracts.9

The staggered contracts model is consistent with various traditional reasons put forward for the existence of nominal rigidities or the incomplete adjustment of nominal prices to their equilibrium levels. Much of the literature on optimal price (wage) adjustment and contracting assumes a fixed, lump-sum real cost of adjustment or negotiation.10 This assumption has been adopted by, among others, Barra (1972), Gray (1978), and Sheshinski and Weiss (1983). Given fixed, lump-sum costs of price adjustment, it will be optimal for firms in the presence of trend inflation, for example, to adjust their prices at discrete intervals of time.11 (Romer (1990) examined the microeconomic foundations for the staggered contracts model.)

This section explores the implications of alternative hypotheses about the wage-setting process and alternative assumptions on expectations formation, with a view to determining the restrictions implied for the form of the Phillips curve, in particular the dependence of current inflation on past and future expected inflation. The implications of two extreme benchmark cases are considered. First, a model in the tradition of Phelps and Friedman with adaptive expectations is shown to imply a weight of unity on past inflation in determining current inflation—that is, there is complete “inflation stickiness”—and a weight of zero on future expected inflation. Second, the Calvo (1983a, 1983b) model is shown to place a weight of unity on future expected inflation, and actual inflation is independent of the past rate of inflation—so that there is no inflation stickiness. An intermediate, alternative model and its implications for the form of the Phillips curve are then discussed: it is shown that current inflation is determined as a weighted average of past and future expected inflation.

The traditional expectations-augmented Phillips curve usually posits that the rate of inflation, ΔPt, equals the expected rate of inflation, Πte, plus a term representing excess demand:12

Δ P t = Π t e + β y t , ( 1 )

where expected inflation is computed as a weighted average of lagged inflation rates. Assuming a geometric lag distribution

Π t e = ( 1 α ) Σ i = 0 α i Δ P t 1 i , 0 < α < 1 , ( 2 )

which can be rewritten as

Π t e = α Π t 1 e + ( 1 α ) Δ P t 1 . ( 3 )

Then, substituting equation (3) into (1) and with some manipulation, current inflation is given by

Δ P t = Δ P t 1 + β ( 1 α ) y t + α β ( y t y t 1 ) , ( 4 )

so that actual inflation equals last period’s inflation, plus a function of current excess demand and the acceleration in excess demand. An important implication is that current inflation can be affected only by factors that alter current excess demand. In particular, disinflation must involve output losses.

We now turn to a version of the rational staggered contracts model in discrete time:13 Vt is defined as the log of the wage embodied in contracts or wage quotations initiated at time t by a representative agent, and is fixed during the length of the quotation. The quotation expiration date is assumed to be stochastic and to follow a geometric distribution. It is posited that the level of the new negotiated wage is determined as a weighted average of all future price levels, P, and excess demand, y, so that

V t = ( 1 b ) Σ s = t E t [ P s + β y s ] b s t , ( 5 )

or

V t = b E t V t + 1 + ( 1 b ) P t + ( 1 b ) β y t , ( 6 )

where each period in equation (5) is weighted by the probability that the contract currently being negotiated will survive to that period; Et denotes the expectations operator conditional on information available at time t. The (log of the) average price level (which equals the log of the average wage level), P, is then defined as a weighted average of all contract wages in existence:

P t = ( 1 b ) Σ s = t b t s V s , ( 7 )

where (1b)bts is the proportion of wages that were negotiated s periods ago. The general price level can be rewritten as a weighted average of last period’s price level and the new contract wage; that is

P t = b P t 1 + ( 1 b ) V t . ( 8 )

The one-period-ahead expected rate of inflation (using all information available at t) can then be written alternatively as

E t Δ P t + 1 = b Δ P t + ( 1 b ) [ E t V t + 1 V t ] , ( 9 )

or

E t Δ P t + 1 = ( 1 b ) [ E t V t + 1 P t ] . ( 10 )

Then, with substitution and rearrangement, the current rate of inflation can be written as

Δ P t = E t Δ P t + 1 + ( 1 b ) 2 b β y t . ( 11 )

Although equation (11) resembles the traditional expectations- augmented Phillips curve posited in equation (1), with the mathematical expectation today of the rate of inflation expected to prevail tomorrow. EtΔPt+1, replacing a distributed lag over past inflation, Πte, it is worth emphasizing that the difference has strong implications and will fundamentally alter the response of the economy to certain types of shocks. Note that equation (11) implies that regardless of the degree of price stickiness, as measured by the value of the parameter b, there is no stickiness in the rate of inflation. The rate of inflation is independent of the past rate of inflation and is free to jump to any value dictated by future expected inflation. The rate of inflation today will respond, therefore, to anticipated shocks that affect the rate of inflation tomorrow.14 This is in sharp contrast to the previous model where current inflation was tied to past inflation and could be influenced today only by contemporaneous movements in excess demand.

The two approaches represent extremes, with the traditional model of adaptive expectations implying complete inflation stickiness, and the rational staggered prices model implying no inflation stickiness. Whether or not inflation has backward- and for forward-looking elements is an empirical question that can only be settled by the data. Empirical testing is reported in Section III below. To allow for both backward- and forward-looking elements in the determination of inflation, we nest the two extremes in a more general model that includes both as special cases:

Δ P t = δ E t Δ P t + 1 + ( 1 δ ) Δ P t 1 + α y t + β Δ y t . ( 12 )

Such a generalization could be given a heuristic justification by assuming the existence of both backward- and forward-looking agents. Alternatively, and somewhat more formally, there are wage-setting characteristics that would lead to some inflation stickiness—and in particular, to current inflation being a weighted average of both past inflation and future expected inflation.

These characteristics may be classified as “incomplete forward looking.” The traditional adaptive expectations model discussed is, of course, an extreme example. We now present an example to illustrate the point more generally. Suppose that wage setters set contract wages according to the rule

V t = V t 1 + E t Δ P t + 1 + β y t , ( 13 )

so that those negotiating wages today set theirs to emulate the level of wages contracted last period, adjusted to compensate for inflation expected next period, and also as a function of prevailing excess demand. Wage setters are, therefore, partly backward looking and partly forward looking, but myopically so, in that they do not look at expected inflation or excess demand for the full duration of the contract. Then, again assuming that contract expiration dates are geometrically distributed, and substituting equation (13) into the first difference of equation (8) yields

Δ P t = b Δ P t 1 + ( 1 b ) E t Δ P t + 1 + ( 1 b ) β y t , ( 14 )

so that current inflation is a weighted average of past and expected future inflation.

II. The Dynamics of Disinflation

This section considers the dynamics of disinflation along a path with zero output losses. An inflation equation with backward- and forward- looking elements,15 as described above in equation (12), is embedded in a simple closed economy macromodel containing equations for aggregate demand and money demand. If perfect foresight prevails, then equation (12) can be written in terms of the acceleration of inflation ηt:

η t + 1 = [ ( 1 δ ) / δ ] η t ( α / δ ) y t ( β / δ ) Δ y t , ( 15 )

where ηtΔPtΔPt1. The root of this equation is equal to λ=(1δ)/δ, which is less than unity if and only if δ > 0.5.

We combine the inflation equation with a simple aggregate demand relationship that relates the output gap, y, to the real interest rate (where ɵ < 0) :

y t = θ ( i t Δ P t + 1 ) ( 16 )

and with a money demand equation

M t P t = ρ y t γ i t + Φ ( M t 1 P t 1 ) , ( 17 )

where it is the nominal interest rate, and Mt is the log of the nominal money stock.16 In this model let us consider what the dynamics of disinflation must be in order to avoid output losses completely. We will call this the “costiess disinflation path.” From equation (15), setting yt = 0 for all t ≥0 gives a path for the inflation rate. This yields

η t + 1 = λ η t , ( 18 )

so the rate of inflation in any future period (t ≥1) is given by

Δ P t = Δ P 0 + η 1 Σ i = 0 t 1 λ i = Δ P 0 + η 1 [ 1 λ t 1 λ ] . ( 19 )

Equation (19) expresses the path of inflation in terms of the initial deceleration of inflation, η1< 0, and the subsequent inflation dynamics, which depend on the weight given to future inflation in wage/price determination. It is clear from equation (19) that a necessary condition for convergence to zero inflation accompanied by zero output losses (yt = 0) is λ < 1, which will hold if and only if δ > 0.5.

Costless disinflation is thus possible only if δ is above 0.5, and provided the announced policy is credible.17 The intuition for this result is as follows: if inflation is expected to fall next period, it pulls down this period’s inflation, ΔPt. The key to costless disinflation is to decelerate money growth at a rate such that the fall in inflation today (relative to last period’s) is just offset by the further fall expected in the following period. The new steady-state rate of inflation is approached asymptotically—there is always an extra downward drag on today’s inflation rate from a further expected decline. However, provided that the forward term more than offsets the past (that is, δ > 0.5), and there is the expectation that inflation will decline in the future (that is, the policy is credible), then the whole process of deceleration from a positive inflation rate can be set in motion with a path of decelerating money growth, without incurring output losses.18

Assuming that costless disinflation is possible, we can calculate from equation (19) the initial deceleration of inflation that is needed to achieve price stability asymptotically, starting from an initial inflation rate, ΔP0. If

lim t Δ P t = Δ P 0 + η 1 / ( 1 λ ) = 0 , ( 20 )

then

η 1 = Δ P 0 ( 1 λ ) . ( 21 )

The closer λ is to 1—that is, the closer δ is to 0.5—the more gradual is the initial deceleration of inflation, η1 and also the slower is its subsequent decline, along a path with zero output losses. Of course, for estimates of δ slightly below 0.5, costs of disinflation, though positive, will be small; output losses will increase as δ declines further.

What does the path for the money supply look like along the costless disinflation path? Suppose we start from a position in which output is at potential, so y0 = 0, and ΔP0=ΔM0. Along the costless disinflation path. yt=Δyt=0, and from equation (16), therefore, it=ΔPt+1(fort1). If we take the first difference for the demand for money and substitute for Δyt and Δit

Δ M t Δ P t = γ ( Δ P t + 1 Δ P t ) + Φ ( Δ M t 1 Δ P t 1 ) . ( 22 )

Or, in terms of real balances, (mtMtPt), since ΔPt=ΔP0λt on the costless disinflation path

Δ m t = γ ( 1 λ ) Δ P 0 λ t + Φ Δ m t 1 . ( 23 )

In period 1

Δ m 1 = λ γ ( 1 λ ) Δ P 0 > 0 , ( 24 )

so real balances increase, as do nominal balances (since ΔP0 > 0). It is clear from equation (23) that in subsequent periods, (t > 1), real balances also increase.

It is also of interest to consider whether nominal balances accelerate or decelerate along the costless disinflation path; this depends on both the interest elasticity of money demand and the speed of adjustment of inflation. Since

Δ M 1 = Δ P 1 + Δ m 1 = λ Δ P 0 + λ γ ( 1 λ ) Δ P 0 = λ Δ P 0 [ 1 + γ ( 1 λ ) ] ( 25 )
Δ M 1 Δ M 0 = λ Δ P 0 [ 1 + γ ( 1 λ ) ] Δ P 0 = Δ P 0 ( 1 λ ) [ λ γ 1 ] . ( 26 )

In the first period. nominal balances accelerate if λγ>1. Thus, the Cagan (1963) condition emerges here, though in a modified form: the rate of growth of the nominal money supply increases upon implementation of a disinflation policy, if the product of the interest semi-elasticity and 1 minus the speed of adjustment of inflation (that is, λ) exceeds unity. The parallel with the Cagan analysis of hyperinflation using adaptive expectations becomes clear when the path of inflation is written as a first-order equation, as follows:

Δ P t Δ P t 1 = ( 1 λ ) ( Δ P ¯ Δ P t 1 ) ,

where ΔP¯=0, the long-run rate of inflation. If λγ>1, a small fall in inflation and, hence, interest rates, produces a large increase in the demand for real balances in period 1, which dominates the effect of a lower price level on nominal balances. In this case, the credibility of a costless disinflation path would be doubtful, since it involves an acceleration of money growth initially.

What about the subsequent paths for real and nominal balances? From equation (23), the solution for the change in real balances can be written

Δ m t = λ t γ ( 1 λ ) Δ P 0 Σ i = 0 t 1 ( Φ / λ ) i , ( 27 )

and the acceleration in real balances is

Δ m t Δ m t 1 = λ t 1 γ ( 1 λ ) Δ P 0 { ( λ 1 ) Σ i = 0 t 1 ( Φ / λ ) i + ( Φ / λ ) t 1 } . ( 28 )

Whether the acceleration of real balances is positive or negative depends on the sign of the term within the braces, { }. So, ΔmtΔmt10, if

( 1 λ ) Σ i = 0 t 2 ( Φ / λ ) i λ ( Φ / λ ) t 1 . ( 29 )

It is not in general possible to sign this expression; for some parameter values, money balances may accelerate for a time, and then decelerate. To calculate the implied path for the money supply requires parameter estimates; we now turn to estimation of the inflation equation.

III. Estimation Results

Reflecting the discussion in Section I, we now present estimates of a general form of the inflation equation where inflation is a function of both past and future expected inflation and the degree of capacity utilization. Variants of the equation are estimated that incorporate a nonlinear relationship between capacity utilization and inflation and country-specific parameters. Although little support for a nonlinear capacity utilization effect is found, there is some evidence of differences across countries in the degree of inflation stickiness. The polar cases of weights of unity on either lagged or led inflation are both strongly rejected.

Equation to Be Estimated

In addition to the lagged and led inflation terms, an absorption price term is included in estimation to allow for the potential desire of wage earners to be compensated for changes in the real consumption wage. The form of the equation that we estimate is the following:

Δ P t = δ E t Δ P t + 1 + ( 1 δ ) Δ P t 1 + α ( Δ P A t Δ P t ) + γ ( P A t P t ) + β f ( C U t ) , ( 30 )

where pt is the logarithm of the non-oil gross national product (GNP) deflator in period t; PAt is the log of the absorption deflator; CUt is the capacity utilization rate defined to equal 100 when output is at its capacity level; and α, β, γ and δ are parameters to be estimated. The function f(CU) is a (possibly nonlinear) function of the contemporaneous capacity utilization rate.19 The inclusion of either the level or the growth rate of the absorption price can be motivated by different theoretical models. The Calvo model, which specifies a fixed level for the wage over the life of a contract, implies that the relevant variable is the relative level of the absorption price. In other models, where contract wages grow over time, it is the growth rate that is relevant. Both terms have been included in equation (30) to nest these two possibilities in the initial specification.

The presence of ΔPt and Pt on the right-hand side of equation (30) is likely to generate a negative correlation between the relative price terms and the structural disturbance: to reduce this problem, the equations were reparameterized by adding (α+γ)ΔPt to each side.20 Dividing by (1+α+γ) then yielded the following equation to be estimated:

Δ P t = ( 1 α ¯ γ ¯ ) [ δ E t Δ P t + 1 + ( 1 δ ) Δ P t 1 ] + α ¯ Δ P A t + γ ¯ ( P A t P t 1 ) + β ¯ f ( C U t ) , ( 31 )

where α¯=α/(1+α+γ),γ¯=γ/(1+α+γ), and β¯=β/(1+α+γ).

Initial Parameter Estimates

The initial stage of estimation involved testing equation (31) with expected inflation replaced by led inflation and alternatives that had constraints imposed on the coefficients on future and lagged inflation and on the relative price terms.21 The limiting case of a zero weight on future inflation is consistent with the traditional Phelps-Friedman model with backward-looking expectations. At the other extreme, a weight of unity yields the Calvo model with fully forward-looking behavior and no inflation stickiness. A linear specification was initially used for the capacity utilization term, f(CU) = CU/100 - 1. This implies that capacity utilization exerts no pressure on inflation at a “normal” level of 100, consistent with the construction of the capacity utilization series as 100 times the ratio of actual to trend output.22 In the absence of a constant term in the price equation, the “natural” rate of capacity utilization then equals 100.

To control for the endogeneity of the right-hand-side variables dated period t and t + 1, they were first regressed on a set of country-specific instruments consisting of the lagged level of capacity utilization; the lagged ratio of government spending to capacity output: lagged growth in the non-oil GNP deflator; and lagged growth in real money balances. A Zellner-efficient systems estimator was then used to jointly estimate equation (31) for the Group of Seven industrial countries over 1966–88, with common parameter values imposed across those countries. Because preliminary results revealed large outliers in the residuals for the United Kingdom, it was excluded from the pooled sample.23

The unconstrained parameter estimates obtained from estimation of equation (31) are shown in Line 1 of Table 1.24 The weight on future inflation is estimated to be slightly less than one half; both of the terms in the absorption deflator have the expected (positive) sign; and the capacity utilization rate enters with the expected sign and its coefficient is statistically significant. Lines 2 and 3 present estimates with the parameter on future inflation constrained to zero and unity, respectively. The modified likelihood-ratio test suggested by Gallant and Jorgenson (1979) indicates either of these limiting values for a1 is strongly rejected by the data.25 This suggests that the inflation process is characterized by a degree of inertia intermediate between traditional backward-looking models, at one extreme, or the Calvo model, at the other extreme. Line 4 shows the results when the coefficient on growth in the absorption deflator is constrained to zero: a likelihood-ratio test indicates that this constraint is also strongly rejected by the data. In contrast, the constraint that the parameter on the level of the absorption deflator is zero cannot be rejected at conventional levels of significance, as shown in line 5. The latter estimates, then, represent the preferred price equation. They were used both as the starting point for the extended estimation results discussed below, and as the basis for the model simulations presented in the next section.

Nonlinear Capacity Utilization Effects

If there is an upper limit on the achievable level of output in the short run, then the output-inflation trade-off should become increasingly steep as this limit is approached. To test the empirical validity of this hypothesis, the equation was specified so that the degree of inflation pressure depends nonlinearly on the output gap. The functional form used was

Table 1.

Estimation Results for the Non-Oil GNP Deflator, Common Parameters

( Δ P = ( 1 a 2 a 3 ) [ a 1 Δ P + 1 + ( 1 a 1 ) Δ P 1 ] + a 2 Δ P A + a 3 ( P A P 1 ) + a 4 ( C U / 100 1 )

article image
Note: The estimation period is 1966–88; the estimation technique is iterative Zellner-efficient with instrumental variables. Absolute values of asymptotic t-ratios in parentheses.

Test of the null hypothesis that the constrained model is true; ɵ is asymptotically distributed x2(1): the critical value at the 2.5 percent significance level is 5.0.

f ( C U ) = a 3 [ a 4 2 a 4 ( C U / 100 1 ) a 4 ] . ( 32 )

The expression is parameterized so that a3 has the same interpretation here as in the linear relationship when CU equals 100: specifically, inflation pressure is zero at this point, and the slope of the price-output trade-off is equal to a3. The trade-off becomes vertical as CU/100 – 1 approaches a4: the latter parameter thus determines the limiting rate of capacity utilization. As a4 becomes large, the curvature of the function decreases; when a4 approaches infinity the function becomes linear.26 Because expression (32) is a nonlinear function of a4, a grid search was performed to identify the value that maximized the likelihood function.

Estimation results for this equation are shown in Table 2. The likelihood function reaches a maximum when a4 is about 0.08, implying that the short-run aggregate supply curve becomes vertical when output reaches 108 percent of long-run potential. The likelihood function, however, is quite flat for a range of values around this level. A test of the null hypothesis that the relationship is linear (that is, that a4 = ∞) yields a test statistic of 4.6, distributed asymptotically x2(1), thus the null hypothesis of a linear relationship cannot be rejected at the 2.5 percent level of significance (critical value 5.0), but it can at the 5 percent level (critical value 3.8). Although the historical data provide some evidence for a nonlinear trade-off between output and prices, it is not strong. For the purposes of further estimation and simulation, the null hypothesis of a linear trade-off was maintained.

Table 2.

Estimation Results for the Non-Oil GNP Deflator, Nonlinear Capacity Utilization Effect

( Δ P = ( 1 a 2 ) [ a 1 Δ P + 1 + ( 1 a 1 ) Δ P 1 ] + a 2 Δ P A + a 3 [ a 4 2 / ( a 4 ( C U / 100 1 ) ) a 4 ] )

article image

Linear specification.

Country-Specific Parameters

Finally, we tested for differences in parameters across countries. Because of a limited number of observations for each country, the six countries27 were grouped into three geographic regions within which the same price behavior was assumed to prevail. The regions were Europe, consisting of Germany, France, and Italy; North America, consisting of the United States and Canada; and Japan. The results are shown in Table 3 for both the unconstrained equation and several parameter constraints. In the case where all three parameters (a1, a2, and a3) differ across regions (line 1), results for Japan were implausible, with a negative (but insignificant) weight on future inflation and a very large parameter on capacity utilization. In what follows, therefore, we have constrained a1 to be the same across regions. A likelihood-ratio test could not reject the hypothesis that they were all the same (line 2).

Table 3.

Estimation Results for the Non-Oil GNP Deflator, Region-Specific Parameters

( Δ P = ( 1 a 2 ) [ a 1 Δ P + 1 + ( 1 a 1 ) Δ P 1 ] + a 2 Δ P A + a 3 ( C U / 100 1 ) )

article image
Note: Absolute values of asymptotic t-ratios in parentheses.

ɵ is asymptotically distributed x2(2); the critical value at the 5 percent significance level is 6.0. Test of constrained model against the unconstrained model of line 1.

Tests were then performed to see if either a2 or a3, or both, differed significantly across regions when a1 was constrained to he the same. When the parameter on absorption inflation (a2) varies across countries (lines 2 and 4), it is apparent that the coefficient is much larger for Japan than the other regions: however, the hypothesis that a2 is the same across regions cannot be rejected at conventional significance levels. The higher value for Japan may reflect two factors: a greater responsiveness of contract wages to consumption prices because, for instance, of bonus payments and indexation provisions in contracts; or a shorter average contract length than for the other regions.28 The results shown in line 3 indicate that the parameter on capacity utilization, a3, is also higher for Japan than the other regions, though differences are again not significant. The higher parameter on capacity utilization may reflect either greater sensitivity of new-contract wages to market conditions, or, as for a2, a shorter average contract length, implying that a higher percentage of overall wages is affected by current conditions.

To summarize, these results support an inflation equation in which current inflation depends on a weighted average of past and future expected inflation. The restrictions implied by the pure forward-looking, level models of Calvo (1983a, 1983b) and by the pure backward-looking, traditional Phillips curve models are both convincingly rejected. The data do not strongly support the hypothesis of a nonlinear Phillips curve. The evidence in favor of differing parameters across countries in the price equation is inconclusive, largely because the individual parameters are not estimated precisely. The point estimates, however, suggest that the response of prices to current market conditions and to relative price movements may be greater in Japan than in the other industrial countries.

IV. MULTIMOD Simulations of Disinflationary Policies

This section presents simulations of alternative disinflationary policies using MULTIMOD, a multiregion global macroeconomic model developed at the Fund (see Masson, Symansky, and Meredith (1990)). The focus is on the transitional costs in terms of lower output of reducing the inflation rate: the long-run benefits of lower inflation, such as possible reductions in relative price variability, inflation uncertainty, and distortions in tax systems, are not captured by the macroeconomic simulations. The results indicate that the output costs arc transitory, and vary according to factors such as the speed with which the disinflationary policy is phased in, the credibility of the authorities’ commitment to reducing inflation, the degree of forward-looking behavior in price and wage setting, and the responsiveness of prices to demand conditions.29 Sacrifice ratios are calculated that compare the cumulative output loss to the reduction in inflation: the implied costs of disinflation are lower than some others found in the literature. International aspects of disinflationary policies are also examined. In particular, we look at spillover effects on other countries of disinflationary policies in the United States, and also at the use of exchange rate versus money-supply targeting to reduce inflation.

Simulations of Disinflationary Policies in the United States

Alternative programs that reduce the target for U.S. money growth by 4 percentage points are described in Table 4a, and their cumulative effects on real output are shown in Table 4b.30 Figure 1 gives year-by-year paths for output, inflation, the short-term interest rate, and the real exchange rate. The reduction in target money growth was chosen to represent roughly the difference between the existing trend U.S. inflation rate and price stability: since money demand is homogeneous of degree one in prices in MULTIMOD, a permanent 4 percentage point decline in money growth leads over time to a 4 percentage point decline in the inflation rate. The price equation used for the simulations is shown in line 5 of Table 1 in the previous section, with common parameters across countries. The weight on future inflation is estimated to be 0.423. The simulations differ in terms of the speed with which the disinflationary policy is implemented, ranging from an immediate decline in money growth of 4 percentage points in 1990 (program 1), to a phase-in over an eight-year period starting in 1991 (program 4). In all cases except program 3, the policy is assumed to be fully credible, in the sense that future reductions in money growth arc anticipated at the time the policy is announced. For program 3, it is assumed that agents expect observed declines in money growth to continue indefinitely; as the program is phased in gradually, their expectations of future money growth are too high during the phase- in period.31

Figure 1.
Figure 1.

Effects of Alternative Disinflation Programs in the United States

Citation: IMF Staff Papers 1992, 002; 10.5089/9781451947106.024.A007

Note: The interest rate panel shows the percentage point deviation from baseline of the short-term nominal interest rate. The inflation panel shows the percentage point deviation from baseline of the rate of change of the absorption deflator. The other two panels show percent deviation from baseline. The programs are described in Table 4a.
Table 4a.

Alternative Disinflation Programs

article image
Table 4b.

Cumulative U.S. GDP Losses for Alternative Disinflation Programs

(Percent deviation from baseline)

article image

In the case of program 1, real GDP in the United States falls by 2.4 percent on impact and by a cumulative 7.9 percent over the first five years of the simulation. To understand the role played by price stickiness in the MULTIMOD results, it is useful to first consider those of a classical flexible price model where output is always at potential. In this model—which corresponds to having an infinite parameter on the capacity utilization rate in the MULTIMOD price equation—the price level jumps in each period to keep output at potential. There are two reasons why the price level changes: the first is to match the decline in nominal money balances associated with lower money growth; the second is to accommodate the increased demand for real money balances associated with a lower equilibrium nominal interest rate. The second effect—which is associated below with the “re-entry” problem—implies that the initial decline in the inflation rate exceeds its equilibrium decline. The MULTIMOD money demand function implies a total fall in the equilibrium price level in the first year of the simulation of about 14 percent: 4 percent to keep real money balances at their baseline level, and another 10 percent to accommodate the rise in real balances associated with a 4 percentage point decline in the nominal interest rate. Real interest rates are unchanged because both nominal interest rates and expected inflation beyond the initial period have fallen by 4 percentage points.32

The MULTIMOD results differ from those of the classical model because both the price level and the inflation rate are sticky. The inertia in the initial stages of a disinflationary process implies that current period inflation does not fall by the same amount as expected future inflation. As a result, the price level initially falls by less in MULTIMOD than in the long run. In the case of program 1, for instance, the GNP deflator drops by 3.9 percent on impact, compared to the 14 percent decline in the long run. This limits the decline in the short-term nominal interest rate to 60 basis points in the first year of the simulation, while expected inflation falls by about 5 percentage points. The resulting rise in real interest rates lowers domestic spending; in addition, the induced appreciation of the real exchange rate causes a deterioration in real net exports. It takes about five years for the negative effects on aggregate demand to wear off and for output to return to baseline.

The role of price stickiness in generating output losses can be reduced either by preannouncing the policy prior to reducing the money growth rate, or by phasing in the policy to achieve a more gradual reduction in inflation. Program la assumes a reduction in money growth that is (credibly) preannounced in year 1, but only implemented in year 2. The price level initially falls in anticipation of future declines in money growth; since actual money balances remain unchanged in year 1, the short-term interest rate drops by 160 basis points, as opposed to only 60 basis points under program 1 (Figure 1). The appreciation of the real effective exchange rate is also reduced, from almost 8 percent in program 1 to 5 percent in program 1a. Both of these factors moderate the decline in demand and output: the total output loss over the first five years is 6.0 percent, as opposed to 7.9 percent.

Program 2 has reductions in money growth of 1 percentage point in the first year of the simulation. rising linearly to 4 percentage points in the fourth year. The steady decline in money growth rates is assumed to be fully anticipated when the policy is introduced. The output loss in this case is 5.2 percent over the first five years, about two thirds that observed in program 1.

Program 2a specifies the same phased reduction in money growth, but starting in the second year as opposed to the first year; this further reduces the five-year output loss to 3.5 percent. Finally, program 4 specifies an eight-year phase-in period for the reduction in money growth, lowering the output loss to only 1.9 percent. The sensitivity of the output loss to the disinflationary program is moderately reduced when a ten-year horizon is considered, as the gradual programs tend to postpone some of the cost to the later years.

These simulations, then, suggest that the output cost of disinflationary policies can be reduced by phasing in or preannouncing the policy, as long as the policy is fully credible. This may be implausible when the program is either not implemented immediately (program la), or is phased in over time (programs 2 and 2a). To evaluate the sensitivity of the results to incomplete policy credibility, program 3 assumes the same phased-in path of monetary deceleration as program 2. Expectations, however, adjust only to observed declines in the money growth rate, as opposed to announced future declines.

Specifically, the decline in expected future money growth is initially limited to the observed decline of 1 percentage point in year 1. In year 2 agents are surprised, as money growth falls 2 percentage points below baseline. By year 4, money growth has stabilized at 4 percentage points below baseline and expectations have become consistent with the actual stance of policy. The initial output loss is about one fourth that in program 1, in line with the reduction in the size of the expected money growth shock. While the initial output cost is smaller, so is the adjustment of the price level. The sequence of surprises implied by declining money growth rates beyond the first year, then, requires further price adjustment. In the event, the cumulative output loss over the first five years is 7.2 percent, almost as large as when the cut in money growth is implemented immediately. This suggests that the potential benefits from a phasing-in of disinflationary policies can be negated if the credibility of the policy suffers as a result.

The costs of disinflation also depend on how much inertia there is in the inflationary process, which is inversely related to δ, the parameter on led inflation. As discussed in Section III, the historical evidence is consistent with a value of slightly under ½. At the same time, it is interesting to examine the sensitivity of the results to different values for this parameter, both to verify the theoretical results presented in Section II, and to examine the impact of more forward- or backward-looking behavior on output costs. Simulations were performed in which it varied from ¼ to 1, the latter representing the limiting case when there is no inflation stickiness.33

Table 5.

U.S. GDP Losses for Alternative Disinflation Programs with Different Degrees of Inflation Stickiness

(Percent deviation from baseline)

article image

The results are shown in Table 5 for programs 1 and 2. For both programs, the cumulative output loss tends to fall as the parameter on future inflation rises. It is also apparent that the reduction in the loss is not uniform across disinflation programs: for program 1, raising the parameter from ½ to 1 cuts the five-year loss in output from 7.1 percent to 5.8 percent, whereas for program 2 the loss falls from 4.0 percent to 0.4 percent. The reason for the different reductions in the output loss for the two programs involves the path followed by the inflation rate. When the weight on future inflation is 1, the inflation rate is not sticky and no output loss (or gain) is associated with “jumping” from an initial inflation rate to some arbitrary new rate, as long as it is expected to remain constant. When the decline in money growth is phased in gradually, inflation falls in the first year by 4.6 percentage points, close to the path consistent with a zero output loss when the parameter on future inflation is 1. When money growth is immediately reduced by 4 percentage points, the inflation rate overshoots its long-run level, falling by 6.9 percentage points in the first year of the simulation, then rising gradually back to its equilibrium level of 4 percentage points below baseline. This undershooting of the inflation rate reduces the benefits associated with having a higher weight on future inflation.

It is of interest to compare the output losses in these disinflation scenarios with estimates of what has occurred historically. A useful shorthand measure is the “sacrifice ratio” of percent output loss per percentage point reduction of inflation. Estimates of the sacrifice ratio for disinflations in the United States range from 3 to 18 (Sachs (1985)); for a typical estimate (a sacrifice ratio of 6), see Gordon and King (1982). In our scenarios, the fall in inflation is always 4 percentage points, so the cumulative output losses can be divided by 4 to obtain a sacrifice ratio. It can be seen from Table 4b. that our estimates do not exceed 2—at the bottom end of other estimates. However, it should he emphasized that our results imply that the sacrifice ratio is not a unique number; it depends also on the phase-in, on the extent of preannouncement, and on the credibility of the disinflation program. It could he that our estimate of output losses is low because the extent of inflation stickiness is underestimated: δ=0.25 would give a sacrifice ratio of 3 for program 1 (Table 5). A “cold turkey” disinflation without credibility (a combination of programs 1 and 3 in Table 4b) could generate even higher sacrifice ratios.34

Table 6.

MULTIMOD Simulation of the Effects on Other Industrial Countries of an Immediate Cut in U. S. Money Growth (Program I)

(Percent deviation from baseline)

article image

U.S. disinflationary policies also affect the economies of other industrialized countries. Three channels are at work: the impact of the reduction in U.S. economic activity on the demand for their exports of trading partners; the increased demand for their exports arising from the real appreciation of the U.S. dollar; and the transmission abroad of the rise in U.S. real interest rates. It turns out that the effects of these factors on the real output of the other industrialized countries are roughly offsetting, as shown in Table 6. Their output declines slightly on impact, and is roughly unchanged in subsequent years. Even for a country as closely tied to the U.S. economy as Canada, the net effect is relatively small; the principal impact being a jump in the rate of change of the absorption deflator in the first year in response to the depreciation of the Canadian dollar.

Another question related to international spillovers is whether the output costs of a disinflationary policy depend on the policies pursued by other countries. For instance, are the costs of disinflating in Canada independent of whether the United States disinflates or not? The answer is suggested by the results in Table 6, which indicate that Canada is little affected by disinflation in the United States. This was confirmed by simulating disinflation programs in Canada with and without similar policies being pursued in the United States. The output loss for Canada of program 1 amounted to 8.1 percent over the first five years of the shock in the absence of a disinflation program in the United States; the loss was raised to 8.3 percent when both countries pursued program 1. The difference is similar to the change in Canadian output when the United States alone disinflates. The implication is that the output cost of disinflationary policies is not significantly affected by the disinflationary objectives of other countries. One caveat is that the credibility of the domestic policy is assumed to be unaffected by monetary policies pursued in other countries. In this context, pursuing disinflationary policies jointly with other countries would reduce the output cost if their credibility were enhanced as a result—as indicated by the results for the United States reported in Table 4b.

Table 7.

Japanese GDP Losses with Different Degrees of Price Flexibility

(Percent deviation from baseline)

article image
Note: Equations: (1)ΔP=(10.306)(0.577ΔP1+0.423ΔP+1)+0.306ΔPA+0.285(CU/1001)(2)ΔP=(10.675)(0.570ΔP1+0.430ΔP+1)+0.675ΔPA+0.240(CU/1001).

Effects of More Flexible Labor Markets: The Case of Japan

The evidence in Section III suggested that Japan may have more flexible labor markets than the other industrial countries, as reflected in larger coefficients on both capacity utilization and relative prices. Specifically, the pooled estimates give coefficients of 0.306 on the contemporaneous rate of change of the absorption price and 0.285 on the capacity utilization rate; when the Japanese coefficients are allowed to differ from those for the other regions, they are 0.675 and 0.240, respectively. After transforming capacity utilization coefficients to give estimates of β in equation (30) (by dividing by 1 – a2), they equal 0.411 in the pooled case and 0.738 for Japan alone. One interpretation is that Japan has shorter contract lengths than the other countries, leading to a greater responsiveness of inflation to current economic conditions. This should reduce the output costs of a disinflationary policy by making Japan behave more like the purely classical economy discussed above. Indeed, as contracts become entirely contemporaneous, the parameter on capacity utilization becomes infinite.

The results of simulations for Japan using both the original and higher parameters are shown in Table 7. When programs 1 and 2 are simulated using the original equation, the output costs are slightly higher than those for the United States. Raising the parameters to their alternative values reduces the cumulative five-year output loss by about 30 percent for both programs 1 and 2. Greater price flexibility also leads to a sharper fall in inflation. Taking program 1 as an example, the rate of inflation initially drops by 5.8 percentage points with the higher parameters, as opposed to 4.6 with the lower parameters.

Exchange Rate Targeting Versus Monetary Targeting

The disinflation programs discussed above have all been implemented through lower targets for domestic money growth. Alternatively, the authorities could use an exchange rate target to reduce inflation. When the value of the domestic currency is fixed in terms of the currency of a trading partner with a lower inflation rate, the domestic rate of inflation will converge over time to that of the trading partner, absent an ongoing change in the real exchange rate. Under an exchange rate target, the domestic money supply would be endogenously adjusted to prevent the exchange rate from moving out of a target band.

There are three potential advantages to disinflating through exchange rate targeting as opposed to money targeting. The first involves the credibility of the policy. The private sector may view the institutional arrangements associated with pegging the exchange rate, such as the exchange rate mechanism (ERM) of the EMS, as being more binding on the monetary authorities than stated objectives for domestic money growth (Giavazzi and Pagano (1988) and Horn and Persson (1988)). The second advantage relates to the re-entry problem discussed above. The lower nominal interest rates associated with lower inflation raise the demand for real money balances (provided the demand for money depends negatively on the level of interest rates). With money targeting, this requires a once-and-for-all decline in the price level in addition to that associated with the reduction in money growth.35 The additional downward effect on prices increases the output costs of the disinflationary policy. With a credible exchange rate target, in contrast, the nominal money supply shifts upward to accommodate higher money demand, reducing the downward pressure on prices and the associated output cost.36

The third advantage of the exchange rate is that it is directly observable, unlike the money supply which is typically published only once a month and with some delay, and hence an exchange rate target conveys more information (Bruno (1990)). The complicated dynamics of money demand moreover may make interpretation of money targets difficult.

There are two disadvantages to exchange rate targeting. The first is the inability of the domestic monetary authorities to offset external shocks through changes in the nominal exchange rate as opposed to changes in domestic wages and prices.37 This problem becomes less important the more highly integrated is the domestic economy with that of the country to whose currency the domestic currency is pegged—in particular, the greater is the extent of factor mobility and bilateral trade between them.38 The second disadvantage, related to the first, is that for some countries, the trading partner with which they are most highly integrated may not have a lower inflation rate. For instance, the Canadian economy is strongly linked to the U.S. economy, but both countries have similar inflation rates; pegging the Canadian dollar to the U.S. dollar would not allow Canada to disinflate, unless the United States also chose to do so.

These considerations suggest that exchange rate targeting is more attractive when domestic inflation is initially higher than that in a major trading partner. This has been the case in the ERM. By establishing fixed parities versus the deutsche mark (or, more precisely, resisting downward realignments of those parities), non-German members, such as France, have successfully reduced their inflation rates to German levels. More recently, the entry of the United Kingdom into the ERM was designed to help achieve a reduction of high U.K. rates of inflation. The alternative for the United Kingdom would be to disinflate by tightening monetary conditions independently of the behavior of the exchange rate.

To examine the output costs of these alternative policies, simulations were run where a 4 percentage point reduction in the U.K. inflation rate was achieved by exchange rate versus money supply targeting (see also Bayoumi and Chadha (1991)). The exchange rate target was designed to be broadly consistent with a narrow-band ERM arrangement; a parity value for the pound sterling was established in terms of the deutsche mark that generated a 4 percent a year appreciation of the pound relative to its baseline path (which assumed a continual depreciation of the pound relative to the deutsche mark). Fluctuations in the exchange rate were restricted to a narrow range around this parity value.39

The results are shown in Table 8. For program 1, the output loss is cut almost in half under an exchange rate target compared to a money target.40 The cycle in output is somewhat longer, however, with the cumulative loss over ten years rising from 4.7 to 4.8 percent, as opposed to falling from 8.3 to 7.2 percent under a money target. The reduction in the output loss is consistent with the paths of the inflation rate in the initial years of the simulation: the average inflation rate in years 1 and 2 falls by 5 percentage points (relative to baseline) with a money target, as opposed to 2.8 percentage points with an exchange rate target. As discussed above, the difference results from the price level adjustment needed to accommodate the higher demand for real money balances with a money target. The difference in output costs is even more evident when disinflation is phased in, as in program 2. This is because the costs associated with the re-entry phenomenon for a money target are similar for programs 1 and 2, whereas they are not relevant for an exchange rate target.

Table 8.

U.K. GDP Losses with Exchange Rate Targeting Versus Monetary Targeting

(Percent deviation from baseline)

article image

These simulations underline the importance of the re–entry problem in determining the costs of disinflation under a money target. To some extent, this is an artifact of the path specified for the money supply, which makes no allowance for the one–time increase in money demand resulting from lower interest rates. If the monetary authorities could accommodate the initial rise in money demand without jeopardizing the credibility of the policy, then the output losses associated with the two targeting regimes would be closer. Indeed, if the authorities could credibly commit policy to the same path for the money supply as is implied under an exchange rate target, the two simulations would give identical results. However, the fact that disinflation would be accompanied by periods of accelerating money growth might be very hard to justify in a context of intermediate targeting of the money supply.

V. Conclusions

This paper has examined the transitional output costs of disinflationary policies. These costs depend critically on the form of the Phillips curve. The well-known staggered contracts models of Taylor (1979, 1980) and Calvo (1983a, 1983b) emphasize the predetermination of individual nominal wages and prices, which are revised at discrete intervals of time in an asynchronous manner. Agents are assumed to be forward looking and to set a wage in accordance with expectations of future price and output developments for the duration of the interval between revisions. The overlapping nature of the wage-setting process results in an aggregate price level that is sticky. There is, however, no inherent inertia in the rate of change of the aggregate price level—the rate of inflation. Hence, these models typically imply a rapid convergence of inflation rates to their new levels, with little or no output losses for relatively rapid decelerations in monetary growth. At the other extreme, if wage setters were purely backward looking (that is, expectations were adaptive), there would be complete inflation stickiness.

A general form of the inflation equation, with inflation determined in part by the past rate of inflation and in part by expectations of future inflation, was estimated using pooled cross-section, time-series data for the major industrial countries. The hypotheses that inflation was determined only (1) by expectations of future inflation as implied by the rational staggered contracts models where only the aggregate price level is sticky, or (2) by past rates of inflation, as implied by the traditional Phillips curve models with adaptive expectations, were both convincingly rejected by the data. The estimates support, therefore, an inflation equation in which inflation is determined as a weighted average of past and future expected inflation.

The output costs of disinflation in a model with such a Phillips curve were examined analytically. It was shown that there is a critical value for the relative importance of the forward-looking component above which a disinflation path with zero output losses is possible. Such a disinflation path, if it exists, was shown to imply a monotonic decline in the rate of inflation. The path for the money supply, however, was shown not to be monotonic, because the money stock needs to rise initially in order to accommodate increased money demand accompanying the new lower rate of inflation and avoid the re-entry problem.

The effects of alternative disinflation policies were illustrated by simulations using MULTIMOD, the Fund’s multiregion macroeconometric model. The results indicate that the output costs of a disinflationary policy are smaller (1) if the policy is announced in advance; (2) the more gradually the deceleration is phased in; (3) the more credible is the policy of disinflation; (4) (given credibility) the greater is the relative importance of expected future inflation in determining current inflation; and (5) the greater is the responsiveness of prices and wages to demand conditions. The international spillover effects of a unilateral disinflation in the United States were examined and were found to be largely offsetting and, on balance, small. Finally, the effects of disinflating with an exchange rate target rather than a money supply target were examined. It was shown that such a policy could avoid the re-entry problem, since the money supply would be endogenous at the targeted rate and any increase in money demand accompanying the lower inflation rate would be automatically accommodated without requiring a downward adjustment in the level of prices.

REFERENCES

  • Backus, David K., Patrick J. Kehoe, and Finn E. Kydland, “International Borrowing and World Business Cycles,” Working Paper 426R (Minneapolis, Minnesota: Federal Reserve Bank of Minneapolis, October 1989).

    • Search Google Scholar
    • Export Citation
  • Ball, Laurence M., “Credible Disinflation with Staggered Price Setting,” NBER Working Paper No. 3555 (Cambridge, Massachusetts: National Bureau of Economic Research, December 1990).

    • Search Google Scholar
    • Export Citation
  • Ball, Laurence M., “The Genesis of Inflation and the Costs of Disinflation,” NBER Working Paper No. 3621 (Cambridge, Massachusetts: National Bureau of Economic Research, February 1991).

    • Search Google Scholar
    • Export Citation
  • Barro, Robert J., “A Theory of Monopolistic Price Adjustment,” Review of Economic Studies, Vol. 39 (January 1972), pp. 1726.

  • Bayoumi, Tamim, and Bankim Chadha, “The Transition Effects of Entry into the ERM” (unpublished; Washington: International Monetary Fund, May 1991).

    • Search Google Scholar
    • Export Citation
  • Blanchard, Olivier, “Price Asynchronization and Price Level Inertia,” in Inflation, Debt and Indexation, ed. by Rudiger Dornbusch and Mario Simonsen (Cambridge, Massachusetts: MIT Press, 1983).

    • Search Google Scholar
    • Export Citation
  • Bruno, Michael, “High Inflation and the Nominal Anchors of an Open Economy,” NBER Working Paper No. 3518 (Cambridge, Massachusetts: National Bureau of Economic Research, November 1990).

    • Search Google Scholar
    • Export Citation
  • Bryant, Ralph C., Dale W. Henderson, Gerald Holtham, Peter Hooper, and Steve Symansky, eds., Empirical Macroeconomics for Interdependent Economies (Washington: The Brookings Institution, 1988).

    • Search Google Scholar
    • Export Citation
  • Cagan, Philip, “The Monetary Dynamics of Hyperinflation,” in Studies in the Quantity Theory of Money, ed. by Milton Friedman (Chicago: University of Chicago Press, 1963).

    • Search Google Scholar
    • Export Citation
  • Calvo, Guillermo A. (1983a), “Staggered Prices in a Utility-Maximizing Framework,” Journal of Monetary Economics, Vol. 12 (September), pp. 38398.

    • Search Google Scholar
    • Export Citation
  • Calvo, Guillermo A. (1983b), “Staggered Contracts and Exchange Rate Policy,” in Exchange Rates and International Macroeconomics, ed. by Jacob A. Frenkel (Chicago: University of Chicago Press).

    • Search Google Scholar
    • Export Citation
  • Calvo, Guillermo A. and Carlos Végh, “Credibility and the Dynamics of Stabilization: A Basic Framework,” IMF Working Paper 90/110 (Washington: International Monetary Fund, November 1990).

    • Search Google Scholar
    • Export Citation
  • Chadha, Bankim, “Is Increased Price Inflexibility Stabilizing?,” Journal of Money, Credit and Banking, Vol. 21 (November 1989), pp. 48197.

    • Search Google Scholar
    • Export Citation
  • Coe, David, “Nominal Wages, the NAIRU and Wage Flexibility,” OECD Economic Studies, No. 5 (1985), pp. 87126.

  • Commission of the European Communities, “One Market, One Money,” European Economy, No. 44 (Brussels: October 1990).

  • Cumby, Robert, John Huizinga, and Maurice Obstfeld, “Two-Step Two-Stage Least Squares Estimation in Models With Rational Expectations,” Journal of Econometrics, Vol. 21 (April 1983), pp. 33355.

    • Search Google Scholar
    • Export Citation
  • Fischer, Stanley, “Long-Term Contracts, Rational Expectations, and the Optimal Money Supply,” Journal of Political Economy, Vol. 85 (February 1977), pp. 191205.

    • Search Google Scholar
    • Export Citation
  • Fischer, Stanley, Indexing, Inflation, and Economic Policy (Cambridge, Massachusetts: MIT Press, 1986).

  • Gallant, A., and Dale Jorgenson, “Statistical Inference for a System of Simultaneous, Non-Linear, Implicit Equations in the Context of Instrumental Variable Estimation,” Journal of Econometrics, Vol. 11 (October/December 1979), pp. 275302.

    • Search Google Scholar
    • Export Citation
  • Giavazzi, Francesco, and Marco Pagano, “The Advantage of Tying One’s Hands: EMS Discipline and Central Bank Credibility,” European Economic Review, Vol. 32 (1988), pp. 105575.

    • Search Google Scholar
    • Export Citation
  • Gordon, Robert J., and S. R. King, “The Output Cost of Disinflation in Traditional and Vector Autoregressive Models,” Brookings Papers on Economic Activity: I (Washington: The Brookings Institution, 1982), pp. 20542.

    • Search Google Scholar
    • Export Citation
  • Gray, Jo-Anna, “On Indexation and Contract Length,” Journal of Political Economy, Vol. 86 (February 1978), pp. 118.

  • Grubb, Dennis, Richard Jackman, and Richard Layard, “Wage Rigidity and Unemployment in OECD Countries,” European Economic Review, Vol. 21 (March/April 1983), pp. 1139.

    • Search Google Scholar
    • Export Citation
  • Hodrick, Robert, and Edward C. Prescott, “Post-War U.S. Business Cycles: An Empirical Investigation,” Carnegie-Mellon University Discussion Paper No. 451 (Pittsburgh: Carnegie-Mellon University, November 1980).

    • Search Google Scholar
    • Export Citation
  • Horn, Henrik, and Torsten Persson, “Exchange Rate Policy, Wage Formation and Credibility,” European Economic Review, Vol. 32 (October 1988), pp. 162136.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, World Economic Outlook, World Economic and Financial Surveys (Washington: International Monetary Fund, May 1990).

    • Search Google Scholar
    • Export Citation
  • Lipsey, Richard G., ed., Zero Inflation: The Goal of Price Stability (Ottawa: C.D. Howe Institute, 1990).

  • Masson, Paul R., Steven Symansky, and Guy Meredith, MULTIMOD Mark II: A Revised and Extended Model, IMF Occasional Paper No. 71 (Washington: International Monetary Fund, July 1990).

    • Search Google Scholar
    • Export Citation
  • Masson, Paul R., and Mark P. Taylor, “Common Currency Areas and Currency Unions: An Analysis of the Issues” (unpublished; Washington: International Monetary Fund, September 1991); forthcoming in Journal of International and Comparative Economics.

    • Search Google Scholar
    • Export Citation
  • Mundell, Robert, “A Theory of Optimum Currency Areas,” American Economic Review, Vol. 51 (September 1961), pp. 65765.

  • Mussa, Michael (1981a), “Sticky Individual Prices and the Dynamics of the General Price Level,” in The Costs and Consequences of Inflation, ed. by Karl Brunner and Allan H. Meltzer, Carnegie-Rochester Conference Series on Public Policy, Vol. 15, pp. 26196.

    • Search Google Scholar
    • Export Citation
  • Mussa, Michael (1981b). “Sticky Prices and Disequilibrium Adjustment in a Rational Model of the Inflationary Process,” American Economic Review, Vol. 71 (December), pp. 102027.

    • Search Google Scholar
    • Export Citation
  • Obstfeld, Maurice, and Kenneth Rogoff, “Exchange Rate Dynamics with Sluggish Prices Under Alternative Price Adjustment Rules,” International Economic Review, Vol. 25 (February 1984), pp. 15974.

    • Search Google Scholar
    • Export Citation
  • Parkin, Michael, “The Output-Inflation Trade-Off When Prices Are Costly to Change,” Journal of Political Economy, Vol. 94 (February 1986), pp. 20024.

    • Search Google Scholar
    • Export Citation
  • Phelps, Edmund S., “Disinflation Without Recession: Adaptive Guideposts and Monetary Policy,” Weltwirtschaftliches Archiv, Vol. 114, No. 4 (1978), pp. 783809.

    • Search Google Scholar
    • Export Citation
  • Rodriguez, Carlos A., “The Argentine Stabilization Plan of December 20,” World Development, Vol. 10 (September 1982), pp. 80111.

  • Romer, David, “Staggered Price Setting with Endogenous Frequency of Adjustment,” NBER Working Paper No. 3134 (Cambridge, Massachusetts: National Bureau of Economic Research, 1990).

    • Search Google Scholar
    • Export Citation
  • Sachs, Jeffrey, “The Dollar and the Policy Mix: 1985,” Brookings Papers on Economic Activity: I (Washington: The Brookings Institution, 1985), pp. 11797.

    • Search Google Scholar
    • Export Citation
  • Selody, Jack, “The Goal of Price Stability: A Review of the Issues,” Technical Report No. 54 (Ottawa: Bank of Canada, May 1990).

  • Sheshinski, Eitan, and Yoram Weiss, “Optimum Pricing Policy Under Stochastic Inflation,” Review of Economic Studies, Vol. 50 (July 1983), pp. 51330.

    • Search Google Scholar
    • Export Citation
  • Taylor, John B., “Staggered Wage Setting in a Macro Model,” American Economic Review, Papers and Proceedings, Vol. 69 (May 1979), pp. 10813.

    • Search Google Scholar
    • Export Citation
  • Taylor, John B., “Aggregate Dynamics and Staggered Contracts,” Journal of Political Economy, Vol. 88 (February 1980), pp. 123.

  • Taylor, John B., “Union Wage Settlements During a Disinflation,” American Economic Review, Vol. 73 (December 1983), pp. 98193.

  • Végh, Carlos, “Stopping High Inflation: An Analytical Overview,” IMF Working Paper 91/107 (November 1991).

*

Bankim Chadha is an Economist in the Economic Modeling and External Adjustment Division of the Research Department. He holds a Ph.D. from Columbia University.

Paul R. Masson is the Chief of the Economic Modeling and External Adjustment Division and has a Ph.D. from the London School of Economics and Political Science.

Guy Meredith is Deputy Division Chief in the Asian Department; he did his graduate studies at the University of Western Ontario.

The authors are grateful to Ernesto Hernández-Catá for encouraging them to work in this area, to Joseph Gagnon and Ralph Tryon for discussions of the issues, and to Charles Adams, David Coe, Steven Symansky, and seminar participants at the Reserve Bank of Australia for their comments.

1

The goal of price stability was the focus of the policy discussion in Chapter II of the May 1990 issue of the World Economic Outlook (International Monetary Fund (1990)), which presented an analysis based on a preliminary version of this paper.

2

See Selody (1990) and Lipsey (1990) for a discussion of the issues in a Canadian context.

3

Commission of the European Communities (1990) discusses the goal of price stability in a European context.

4

We will not consider the costs of inflation here; see, for instance. Fischer (1986, Chaps. 1–4).

5

A reasonably rapid credible disinflation could, in fact, cause a boom in output, not output losses, in a model with sticky wages and prices. See Rodriguez (1982), Fischer (1986, p. 252), and Ball (1991).

6

Ball (1990) argued that output losses were the result of adaptive expectations.

7

For a comparison of alternative macroeconometric models see Bryant and others (1988). Obstfeld and Rogoff (1984) show how frequently used alternative price adjustment rules can alter even the qualitative response of the economy to the same shock.

8

This is not to imply that expectational errors cannot occur in these models, in that there are unanticipated shocks to the economy. The point is rather that agents do not systematically and continuously make mistakes in the same direction, as would occur, for example, in a model with adaptive expectations when the rate of inflation permanently increases.

9

There is no presumption in this approach that formal contracts are written; only that nominal prices (wages) are preset for some period of time.

10

Again, we do not distinguish between wages and prices, since this literature typically assumes that prices are a constant markup over wage costs.

11

For the price quotation or “contract” to be binding, it is necessary to make additional assumptions, since in a stochastic environment such contracts will typically be time inconsistent. An alternative approach was adopted by Barra (1972) and Sheshinski and Weiss (1983), who allowed the contract to he renegotiated when the benefits of such a renegotiation outweighed the costs. The optimal policy is an (S, s) policy; that is, the price is changed when it deviates by more than a critical magnitude from the optimal or equilibrium flexible price. Parkin (1986) examined the optimality of alternative wage-staggering rules.

12

And perhaps, also, a term representing the change in excess demand, due to a variable markup of prices over unit costs.

13

For a previous discrete-time version of the Calvo staggered contracts model, see Chadha (1989).

14

As Végh (1991) points out, in the rational staggered prices framework, the rate of inflation is a purely forward-looking variable.

15

That is, with weights that sum to unity on lagged and led inflation. However, the term “backward-looking” is not intended to imply “nonrational” or myopic expectations.

16

This model extends the model of the previous section because it makes money demand explicit and models the transmission of monetary policy through the channel of interest rates. For simplicity, we ignore the effect of trend growth in capacity output on money demand, and assume that equilibrium real interest rates are zero. These simplifications could be relaxed without changing the analysis.

17

Phelps (1978) and Taylor (1983) also examined disinflation paths with zero output loss.

18

If δ < 0.5, then zero output losses would require ever-accelerating declines in inflation, which, however, must eventually come to a halt to prevent ΔPt, from becoming unbounded in a downward direction. At this point (if not before), output losses would be incurred.

19

The first difference of capacity utilization did not enter significantly, and is ignored in what follows.

20

Instrumental variables were used in estimation to control for the endogeneity of the other regressors, as is discussed below.

21

In what follows, time subscripts are omitted on currently dated variables; Xt-1 is written X-1; and Xt+1 is written X +1.

22

The capacity utilization rate was constructed using a time-series filter to derive a capacity output series for each Group of Seven country from observed quarterly GNP data. The filter is the variant of the one developed by Hodrick and Prescott (1980), which has been used extensively to detrend economic data for the analysis of real business cycles (see, for example. Backus and others (1989)).

23

The problem with the U.K. data is associated with the surge in inflation from 7 percent in 1973 to over 27 percent in 1975, and the sharp decline thereafter. This surge of inflation was not related to strong economic activity—1975 was a recession year in the United Kingdom—but rather seems to reflect a wage-price spiral caused by attempts to recover real wage losses following the first oil price shock.

24

The (country-specific) constant terms were all small and insignificant in preliminary estimation: they have been constrained to zero. Estimation was also performed allowing for first-order autoregressive and moving average processes for the residuals: these coefficients were insignificant as a group and had little effect on the other parameter values and were thus excluded in further estimation. It should be noted that the estimate of the asymptotic variance-covariance matrix of the parameters used to construct the t-ratios may not be consistent for the reasons discussed in Cumby and others (1983).

25

The test statistic is asymptotically distributed x2(n), where n is the number of linear constraints. It will not necessarily be positive in finite samples, because the estimates of the variance-covariance matrix of the residuals in the restricted and unrestricted models are not identical.

26

This can be seen by calculating d2f(CU)/dCU2=a42/(a4(CU/1001))3. It is apparent that this expression—which indicates the change in the slope of the output-price trade-off as CU changes—goes to zero as a4 becomes large, regardless of the value of CU.

27

As noted above, the United Kingdom was excluded.

28

Other multicountry studies of price behavior also find a greater response in Japan to market conditions. See, for instance, Grubb and Jackman (1983) and Coe (1985).

29

Subject to the usual caveats of the Lucas critique; in particular, the relative weights of forward- and backward-looking elements in the inflation process may themselves depend on monetary policy.

30

Monetary policy is implemented in MULTIMOD in terms of an exogenous path for the target money supply. The actual money supply can differ in the short-run from the target, as the monetary authorities are assumed to smooth the interest rate changes that would be needed to keep money on target in each period.

31

There are, of couse, other ways to model credibility. See, for instance. Calvo and Végh (1990).

32

The real exchange rate would also be unchanged. The open interest parity condition implies that the real exchange rate depends on the differential between the domestic and foreign real interest rate; if real interest rates are unaffected, so is the real exchange rate.

33

Simulations were also run with a parameter of zero on future inflation, consistent with fully backward-looking models of price behavior. These results are not shown in Table 5, because, for some of the disinflation programs, the simulations exhibited potentially explosive cyclical behavior.

34

Fischer (1986, chap. 7), indicated sacrifice ratios of 1.9–3.7 (assuming perfect credibility) for a simple model with three-year contracts. He concluded that imperfect credibility is required to square the model with historical disinflation experience.

35

Fischer (1986, chap. 8) argued for the use of an exchange rate target in stabilization programs for this reason.

36

It may also alter the timing of output losses. Calvo and Végh (1990) contrasted money-based stabilizations (recession now) with those based on exchange rates (recession later). Végh (1991) surveyed the experience of several Latin American countries and Israel, finding that the stabilizations based on exchange rates produced initial booms.

37

The conditions under which the exchange rate instrument can be abandoned without incurring much of a cost are discussed in the optimum currency area literature, pioneered by Mundell (1961). This issue is also considered in Fischer (1986, chap. 8).

38

For a survey of issues involved in currency unions, see Masson and Taylor (1991).

39

The details of how the ERM is incorporated in MULTIMOD are discussed in Masson, Symansky, and Meredith (1990).

40

In contrast to Calvo and Végh (1990), however, the exchange-rate-based disinflation is not associated with a boom in output. In their model, currency substitution between home and foreign money implies an expansion in output, but that effect is not present in MULTIMOD, which contains conventional money demand functions.

  • Collapse
  • Expand
IMF Staff papers: Volume 39 No. 2
Author:
International Monetary Fund. Research Dept.