The Case of the Reappearing Phillips Curve: A Discussion of Recent Findings

Abstract

The Phillips curve seems to have flattened over time. In this article, we use a simple New Keynesian model to analyze potential pitfalls in the estimation of the slope of the structural Phillips curve. Changes in the conduct of monetary policy or in the relative importance of supply and demand shocks may bias simple estimations of the slope of the Phillips curve. Recent proposals have favored estimations using regional or city data in an effort to overcome these issues. We use a simple model of a monetary union with a continuum of economies and find that some of the drawbacks of the aggregate model are still present in a cross-section of many regions in a monetary union. The relative importance of the demand and supply shocks largely determines the empirical relation between unemployment and inflation in both the aggregate and the cross-section of regions. Our analysis shows potential pitfalls in estimating the slope of the Phillips curve, even if using regional data.

Introduction

Central banks around the world intervene in financial markets by setting the short-term nominal interest rates to stimulate economic activity and control inflation. There are many theories that suggest that changes in interest rates due to monetary policy have effects on real activity: An increase in interest rates is associated with a tightening of the economy and a decrease in the real output, while a decrease in interest rates is associated with an increase in real output. Interest rates can affect output through several channels. For instance, an increase in interest rates implies an increase in the return on savings. In this case, individuals have an incentive to save more and forego current consumption, which translates to a decrease in current real output, all else equal. Clearly, the opposite is true in the case of a decrease in interest rates. Other possible mechanisms through which movements in interest rates affect real activity include borrowing costs and investment. As interest rates decline it becomes cheaper to borrow and invest, so business investment goes up, increasing total output.

While these theories shed light on the effects of monetary policy on the real economy, there is no consensus on the link between interest rates and inflation. A widespread view among policymakers is that there is a trade-off between real activity and inflation and that monetary policy decisions on interest rates, by affecting the real economy, ultimately affect inflation. This trade-off is known as the Phillips curve. 

The Phillips curve was popularized by A.W. Phillips in 1958, when he showed a statistically significant negative relation between the unemployment rate and the growth rate of nominal wages—that is, wage inflation. Based on this empirical relationship, Samuelson and Solow (1960) argued that a looser monetary policy could reduce the unemployment rate by allowing inflation to rise. This then implied that monetary authorities could exploit this trade-off. Since it was first discovered empirically, the Phillips curve has guided discussions of monetary policy and has shaped our understanding of the transmission of monetary policy to prices. More recently, several theories on price setting by firms can rationalize the existence of a Phillips curve in an economic model. 

Over the years, the Phillips curve has received several criticisms. Recent articles have argued that inflation can be approximated by statistical processes unrelated to the amount of slack in the economy (Atkeson and Ohanian, 2001; Cecchetti et al., 2017; and Stock and Watson, 2007). Moreover, the lack of a stable relationship between inflation and various measures of slack has led several articles to conclude that the Phillips curve has weakened over the years (Blanchard, Cerutti, and Summers, 2015; and Coibion and Gorodnichenko, 2015). 

Several articles have pushed back on this criticism and have attempted to "recover" the Phillips curve. Fitzgerald and Nicolini (2014) argue that aggregate data are uninformative about the true structural relationship between unemployment and inflation, and that in fact, under a specific definition of inflation targeting, the evolution of equilibrium inflation is a random walk. They then show that regional data can be used to identify the structural relationship between unemployment and inflation. The main intuition is that monetary policy typically reacts to the aggregate state of the economy, but not to regional conditions. Thus, it is possible to use the deviations of regional economic activity relative to the aggregate and the deviation of inflation relative to the aggregate to recover the relationship between unemployment and prices. 

A recent article by McLeay and Tenreyro (2019) also supports this view. The authors argue that it is difficult to identify the slope of the Phillips curve empirically, even if a negative relationship does hold true in the underlying model. This is because monetary policy will react to economic shocks in order to stimulate output when it is below potential and reduce inflation when it is above target. The actions of the monetary authority will typically affect the empirical slope of the Phillips curve. McLeay and Tenreyro (2019) then use a simple New Keynesian model to highlight this estimation bias due to the endogeneity of monetary policy. They propose several solutions including using regional Phillips curves to circumvent this identification problem. 

In our article, we follow the approach suggested by McLeay and Tenreyro (2019) closely and use a simple New Keynesian model to highlight the issues that arise with the identification of the empirical slope of the aggregate Phillips curve. We then use a simple New Keynesian model of a monetary union, as given by Gali and Monacelli (2008), to attempt to recover the Phillips curve at the regional level. However, we find that this approach is not sufficient to overcome the identification issues highlighted by McLeay and Tenreryo (2019), since even at the regional level our model fails to recover the slope of the Phillips curve. We argue that several factors, including the relative importance of the demand and cost-push shocks, affect the estimation of the slope of the Phillips curve, at both the aggregate and regional levels.