Managing Disinflation
“[We] are committed to bringing inflation back to our 2 percent goal. Over the past year, we have taken forceful actions to tighten the stance of monetary policy. […] Even so, we have more work to do.”
Federal Reserve Board Chair Jerome Powell, press conference, February 1, 2023.
Large, advanced economy central banks are working hard to lower inflation from 40-year highs. Policy rates are up sharply in Canada, the euro area, the United Kingdom and the United States. While disinflation has started, inflation remains far above policymakers’ common target of 2 percent.
Based on their latest projections published in December, most U.S. Federal Open Market Committee (FOMC) participants anticipate a largely benign return to price stability, without a decline in real GDP or a rise of the unemployment rate to much more than 4½ percent. Is this optimism justified? Pointing to the historical record, some prominent analysts wonder whether it is possible to engineer such a large disinflation at what would be such a low cost (see, for example, Lawrence Summers).
This is the setting for this year’s report for the U.S. Monetary Policy Forum that we wrote with Michael Feroli, Peter Hooper and Frederic Mishkin. In the report, we focus on the central challenge facing central banks today: how to minimize the costs of disinflation. To address this question, we employ two approaches: a historical analysis in which we assess the costs of sizable disinflations since the 1950s; and a model-based analysis in which we examine the degree to which policymakers might have been able to anticipate the recent surge of inflation, as well as the path of policy that is likely needed to achieve the desired disinflation.
In analyzing the historical record, we find no instance in which a central bank-induced disinflation occurred without a recession. Looking in more detail, we see that the cost (per percentage point of disinflation) was lower the more credible the central bank, the higher the initial level of inflation, and the faster the disinflation. In the modeling analysis, we identify a version of the Phillips curve that is better able to account for the behavior of U.S. inflation since 2020. Critically, our results suggest that a hot labor market raises inflation, but a cold labor market does not reduce inflation.
These findings have important implications for monetary policy. First, both the historical and modeling analyses cast doubt on the ability of the Fed to engineer a return to 2 percent inflation by 2025 without a mild recession. Second, the lack of a disinflationary impulse from a cold labor market highlights the importance of preemption in formulating an effective monetary policy strategy: the best way to manage a disinflation is to prevent a tight labor market from driving inflation above target in the first place. Finally, consistent with the opening citation from Chair Powell, to reach its inflation target, the FOMC likely needs to tighten policy further, and to maintain monetary restraint for some time.
In the remainder of this post, we expand on this summary of the USMPF report’s analysis and conclusions.
Historical Record
The following chart shows the seven U.S. disinflations since 1950. Looking at the chart, we can see that each one (disinflations are shaded in blue) is associated with a recession (shaded in gray). When we expand the data set to include Canada, Germany, and the United Kingdom since 1960, there are 10 additional disinflation episodes, all of which were associated with recessions. Of these 17 large disinflation episodes, all but one (the U.S. disinflation that started in 1951) were associated with a tightening of monetary policy.
U.S. Core and Trend Inflation, quarterly, 1950-2022
Notes: Core PCE inflation is annual percent change of the price index of personal consumption expenditures. Trend inflation is a nine-quarter centered moving average of seasonally adjusted quarterly changes of the core PCE price index at annual rates. Sources: BEA and authors’ calculations.
To continue, we measure the cost of these disinflations by computing the peak-to-trough increase in a trend measure of slack and then dividing by the scale of disinflation. Using the unemployment rate as a measure of slack, these U.S. “sacrifice ratios” have a median of 1.5—meaning that each percentage point of disinflation is associated with a 1.5-percentage-point increase of the unemployment rate. Unsurprisingly, our computed sacrifice ratios vary substantially across episodes—from only 0.5 in the disinflation that started in 1951 to 3.8 in the disinflation that began in 2007.
What can account for the variation in the costs of disinflation? We focus on three factors that we believe to be associated with lower sacrifice ratios: credibility, the starting point, and speed. Starting with credibility, in the absence of structural impediments (such as long-term labor contracts), theory suggests that a credible central bank is capable of an “immaculate disinflation.” That is, policymakers could reduce inflation without any rise of unemployment. In practice, all economies exhibit rigidities, so even a credible central bank faces at least some sacrifice when it needs to lower inflation substantially. On top of this, in our case study of the Volcker disinflation—the largest in the U.S. sample—we highlight the additional cost arising from premature easing. Put differently, credibility suffers if the central bank is not seen as “in it to win it.”
Second, looking at the entire sample of disinflations, we see that those starting with higher initial inflation are associated with a smaller per-unit sacrifice. One reason for this might be that the commitment of a central bank to act decisively is perceived as more credible when inflation is high. Another is that high inflation episodes often involve large, temporary global supply shocks that eventually allow inflation to drop back so long as expectations are stable. Whatever the reason, the cost of disinflation appears to rise as inflation approaches the central bank’s target. Third, as Laurence Ball observed nearly 30 years ago, faster disinflations seem to be less painful. This stylized fact buttresses the case for the Fed’s 2022 abandonment of policy gradualism.
Modelling Analysis
In our modeling analysis, we start with the formulation in the 2022 USMPF report (Carpenter et al.). Motivated by the recent work of Ball, Leigh and Misra, we then explore alternative versions of the U.S. Phillips curve, which links inflation to economic slack. We extend Ball et al.’s analysis in three ways. First, we include the earlier high-inflation period of the 1960s and 1970s. Second, we modify the vacancies to unemployment ratio (v/u) that Ball et al. find to be closely related to inflation by introducing a v/u gap (based on the difference between v/u and its noncyclical rate v/u*) measure that takes account of structural shifts in the relationship between v and u over more than 60 years (see our earlier post). Finally, we allow for the possibility that hot labor markets and cold labor markets have differential effects on inflation. This last innovation yields the potentially important result that a hot labor market (v/u higher than or u lower than their noncyclical rates) raises inflation, but a cold labor market (negative v/u gap or positive u gap) does not lower inflation.
The key results of this analysis are evident in the following chart, where we compare four-quarter core PCE inflation (labeled “Actual” in black) with model-based out-of-sample projections starting from 2020. The red lines show the results for the new v/u-gap Phillips curve, while the blue lines show the results for a conventional u-gap model (based on the difference between u and its noncyclical rate u*). The solid lines are based on estimating the Phillips curve over the entire pre-pandemic period (1962 to 2019), while the dotted lines use only the stable inflation period (1985 to 2019) for the estimation.
U.S. Out-of-Sample Inflation Simulation, 2020-2022
Notes: The simulations are based on estimating two Phillips curve specifications (u gap and v/u gap) over two sample periods: stable inflation (1985 to 2019) and pre-pandemic (1962 to 2019).
Sources: Barnichon, Bolhuis et al., BEA, BLS, and authors’ estimates.
Looking at the chart, we see that the v/u gap measure outperforms the u gap in matching the sharp rise in inflation over the past two years. However, while the measure of slack matters, the sample period of estimation matters more. The reason for this is that, over the stable inflation period from 1985 to 2019, unlike in the previous two decades, deviations of inflation from its 2% target were transitory. The implication is that any model estimated using that data alone always predicts that inflation will be low and stable. In our view, this result illustrates the impact of the FOMC’s 2020 decision to abandon preemptive monetary policy strategy: preemption was a key feature of the period from 1985 to 2019, but not of the earlier or subsequent years.
Finally, we use our preferred inflation model (based on the v/u gap with different inflation sensitivity to hot and cold labor markets) to project the path of inflation, unemployment and the policy rate out to 2025. The chart below again highlights how critical the sample period is for understanding inflation’s prospects. The projection based on the model estimated over the full historical period (1962 to 2022) indicates that inflation will fall only gradually to 3.7 percent by the end of 2025, while the projection based on the same model estimated over the 1985-2022 sample shows that inflation will fall quickly to the 2 percent target by mid-2024. Again, the difference reflects the variation in the persistence of inflation deviations from 2 percent across over the two samples.
Inflation model projections, 4Q 2022 to 2025
Notes: The solid lines the medians from 10,000 sequences of shocks from the estimated residuals of the mode for each sample period. The shaded areas depict the interquartile range (from the 25th to the 75th percentile) of each distribution.
Sources: Barnichon, Bolhuis et al., BEA, BLS, and authors’ estimates.
Interestingly, the unemployment and policy rate projections are less sensitive to the sample period, with the unemployment rate rising steadily to the 4½ to 5 percent range, and the policy rate peaking this year in the mid-5 percent range before declining in 2024 and 2025. As a result, when we base our projections on the full-sample model, the degree of inflation persistence implies a higher sacrifice ratio. Indeed, for the full-sample based simulation, with relatively high policy rates, unemployment rises above 5 percent and inflation generally remains above the Fed’s target of 2 percent at the end of 2025.
Finally, in the report we consider the alternative of raising the inflation target rather than lowering inflation to the current 2 percent target. However, raising the inflation target before reducing inflation to 2 percent would start the Fed down a slippery of slope in which inflation expectations could become unstable. As a result, the temporary costs of disinflation appear lower than the long-run costs of raising the inflation target (see our earlier post).
Conclusions
The analysis in this year’s report for the U.S. Monetary Policy Forum on managing disinflation leads to the following conclusions:
History
We find no precedent for achieving a major disinflation without a recession.
The cost of reducing inflation by one percentage point is lower when the central bank is credible, when disinflation starts at a high level of inflation, and when the disinflation is faster.
Based on the Volcker disinflation and on those that preceded it in the 1970s, premature policy easing raises the ultimate cost of disinflation.
Model
Policy-induced disinflations come with recessions.
Policy actions in 2022 that restored Fed credibility probably lowered the cost of disinflation. Even so, reducing inflation to the Fed’s 2 percent range likely requires a higher policy rate this year, and perhaps through 2025.
Despite the likely sacrifice, reducing inflation to the 2-percent target probably is less costly than raising the inflation target opportunistically when inflation exceeds 2 percent.