Literature review: Beveridge, Phillips, and the Concave Phillips

We review three new academic papers. The first paper (“Phillips Meets Beveridge”) by Régis Barnichon and Adam Hale Shapiro claim that V/U ratio and vacancy filling costs are superior to the unemployment rate in explaining and predicting inflation fluctuations due to changes in labor market matching efficiency. The second paper (“(Re-)Connecting Inflation and the Labor Market: A Tale of Two Curves”) by Hie Joo Ahn and Jeremy B. Rudd explains why a post-pandemic disinflation has occurred without large increases in unemployment (so far). The final paper (“The Concave Phillips Curve”) by Narayana R. Kocherlakota challenges the conventional belief that the Phillips curve is globally convex and instead argues for a globally concave curvature.

Paper #1: Barnichon and Shapiro (2024)

“Phillips Meets Beveridge” by Régis Barnichon and Adam Hale Shapiro (both San Francisco Fed), reexamines the relationship between economic slack and inflation within the Phillips curve framework by focusing on the vacancy-unemployment (V/U) ratio. The authors claim that V/U ratio and vacancy filling costs are superior to the unemployment rate in explaining and predicting inflation fluctuations. The authors attribute this finding to shifts in the Beveridge curve, particularly changes in labor market matching efficiency.

What the paper does and main results

The study assesses which variable is the most appropriate measure of economic slack in a Phillips curve framework. The authors answer this question using two approaches. First, they conduct an out-of-sample forecasting exercise. Based on a newly constructed panel dataset covering job openings, vacancy filling costs, and other labor market indicators from 1982 to 2022, the authors explore which variables best predict inflation at a one-year horizon. The V/U ratio outperforms other slack measures, including the unemployment rate, in both out-of-sample forecasting and structural Phillips curve estimation.

Second, the authors estimate the structural Phillips curve with different slack measures and evaluate their fit. Since many confounding factors such as inflation expectations and supply shocks confound standard OLS estimates, they use several approaches to address these endogeneity concerns: estimating a narrower measure of inflation less contaminated by supply shocks (“cyclical core”), employing Romer and Romer (2004) monetary shocks to identify movements in slack due to demand shocks, and analyzing cross-sectional data from U.S. metropolitan areas. Across these methods, the V/U ratio consistently provides the best fit of the Phillips curve to inflation, particularly during periods of significant inflation fluctuations, such as the post-COVID era.

The V/U ratio outperforms unemployment even though they are highly correlated, as captured by the Beveridge curve. This is because of shifts in the Beveridge curve. Figure 1 shows the curve for different decades using U.S. data. Shifts in the Beveridge curve may indicate changes in matching efficiency between job vacancies and unemployment, which can drop markedly in recessions.

Figure 1. The Beveridge curve in the US over time.

A critique of the paper

The gains in inflation forecasting ability are most likely moderate. While the V/U ratio performs as well as or better than the unemployment rate across forecasting exercises, the numerical differences of forecasting performance are small. Therefore, the out-of-sample real-time forecast performance probably depends more on the assumptions about the future evolution of V/U (and U), rather than the choice of the “slack” variable in the model. In other words, the paper is interesting, and reinforces the case to run models with different measures of “slack”. Having said that, as mentioned, ultimately the gains are probably limited.

Paper #2: Ahn and Rudd (2024)

“(Re-)Connecting Inflation and the Labor Market: A Tale of Two Curves” by Hie Joo Ahn and Jeremy B. Rudd (both at the Fed Board), investigates the dynamic relationship between inflation and labor market outcomes through the lens of the Beveridge and the Phillips curve. The authors argue that reallocation shocks were important drivers of the labor market during the Covid pandemic and explain why a post-pandemic disinflation has occurred without large increases in unemployment (so far).

What the paper does and main results

The analysis uses a vector autoregression that combines labor market and inflation dynamics by integrating two macro relations: the Beveridge curve, which reflects the relationship between unemployment and job vacancies, and the Phillips curve, which describes the relationship between unemployment and inflation. The analysis differentiates between two types of reallocation shocks: those resulting from job losses and those resulting from voluntary quits. The model additionally features shocks to aggregate activity, labor supply, and wage/price inflation. Estimation relies on 1967-2019 sample, and the estimated parameters are used to decompose the labor market and inflation dynamics during the pandemic.

Aggregate shocks and job losses explain most of the increase in unemployment during the pandemic, while wage and price inflation are largely caused by price-specific shocks. Figure 2 shows these results. Labor force participation (LFPR) declined sharply but recovered equally quickly due to job losses. The recovery in wage and price inflation is slow and not clearly associated with labor market dynamics. While certain factors such as job-loss reallocation shocks exert some influence on inflation, most of the dynamics appears to be due to price-specific factors. These can include energy shocks, markup changes, or changes in reservation wages.

Figure 2. Decomposition of labor market and inflation dynamics during and after the pandemic into six shocks.

Note : This figure displays the posterior medians of the historical decompositions of the variables in the authors’ VAR system during the Covid-19 pandemic (X-axis: calendar month; Y-axis: the magnitude of variations in the endogenous variable). LFPR: the labor force participation rate; UR (Job losers): the unemployment rate of job losers and entrants to the labor force; UR (Quits): the unemployment rate of job quitters; Wage inflation: Changes in average hourly earnings; Price inflation: Changes in adjusted market-based core PCE prices. “Aggregate” refers to aggregate activity shocks.

A critique of the paper

The results show that aggregate activity shocks and price-specific shocks capture most of the labor market and inflation dynamics, but it is unclear what those shocks represent. While great attention has been paid to separating different types of labor market shocks, those appear to play a minor role in driving inflation dynamics during the pandemic. Having said that, it is unclear the orthogonality of the shocks identified in the VAR. For instance,  reservation wages (which are unobserved) are endogenous to quits. Therefore, in the end, we are unsure to what extent the labor market has driven inflation through the lens of this analysis.

Paper #3: Kocherlakota (2024)

“The Concave Phillips Curve” by Narayana R. Kocherlakota (University of Rochester) studies the curvature of the short-run Phillips curve in a wide class of models, challenging the conventional belief that the curve is globally convex and instead arguing for a globally concave curve. This new perspective suggests that the inflationary cost of increasing economic activity is lower in booms than in recessions, which contrasts with traditional views where the curve steepens at high levels of economic activity.

What the paper does and main results

A common view is that the Phillips curve is convex, meaning that inflation rises more rapidly as economic activity increases. Kocherlakota argues that the perspective of a convex Phillips curve (PC) is flawed and demonstrates the concavity properties of the PC in a wide class of canonical models. Under a wide array of price-setting frictions, the paper finds that at high levels of economic activity, the curve becomes asymptotically horizontal, such that inflation increases very slowly as economic activity rises. Conversely, at low levels of economic activity, the curve is nearly vertical, indicating that inflation decreases sharply as economic activity falls. This is true even though firms’ real marginal costs may be convex in activity in such models.

The concave Phillips curve is driven by substitution effects. When economic activity is high, substitution effects lead consumers to spend more on goods with sticky prices, making inflation appear more stable. On the other hand, when economic activity is low, consumers shift their spending to goods with flexible prices, making inflation more responsive to economic activity.

Price indexes that understate the substitution patterns may lead to wrong conclusions about the Phillips curve in the data. Because the standard Laspeyres index used to calculate CPI indexes does not capture substitution patterns accurately, inflation as measured in this way is convex in economic activity for positive inflation rates, as commonly found in the data. Figure 3 illustrates the theoretical difference between the model-based (capturing substitution effects) Phillips curve and a traditional Laspeyres index.

Figure 3. True Phillips curve (accounting for substitution effects) against Laspeyres Phillips curve.

Note : “x” indicates “economic activity”.

A critique of the paper

The paper makes an important contribution to the ongoing debate about the shape of the Phillips curve. The theoretical results presented are useful in that they apply to a broad class of canonical models and can replicate the empirically found pattern of a convex Phillips curve using a Laspeyres price index.

The paper’s elegant derivations rely on simplifying assumptions that need not be innocuous. The assumption of a constant elasticity of substitution (CES) utility function is standard but not guaranteed to match substitution patterns accurately, especially for certain categories of goods. Similarly, firms’ pricing behavior may itself be endogenous to the level of economic activity and/or prices (see here and here), which pushes the slope of the Phillips curve toward convexity.

No evidence in PCE space. The most puzzling aspect of the paper is that the Phillips curve appears convex not only when looking at CPIs constructed using a Laspeyres aggregator but also at indexes which account for upper-level substitution. This is for instance the case of the US PCE price index (which uses an ideal Fisher index), the US CPI-C (which uses a Törnqvist formula), or the Swedish CPI (which uses a Walsh index). Unfortunately, the paper is completely silent about this evidence.

Contact

Want something more tailored?

We provide tailored consulting on ad-hoc projects.