Shock-Dependent Phillips Curve: Evidence from the U.S.
One sentence summary: The negatively-sloped Phillips curve has flattened and switched sign over time mostly due to first lower positive and then negative oil price shocks.
The corresponding academic paper by Hakan Yilmazkuday is available as a working paper here.
Abstract
This paper estimates the slope of the (price) Phillips curve for the U.S. by using the implications of a structural vector autoregression model. The slope of the Phillips curve is measured by the ratio of cumulative impulses of inflation and unemployment, following alternative shocks. The results show that oil price shocks that are consistent with a demand-pull inflation due to higher global demand result in a negatively-sloped Phillips curve, whereas unemployment or inflation shocks that are consistent with a cost-push inflation result in a positively-sloped Phillips curve. These results are combined with historical estimated shocks to further show that the negatively-sloped Phillips curve has flattened and switched sign over time mostly due to first lower positive and then negative oil price shocks, although the negatively-sloped Phillips curve is still alive and well when the full sample is considered.
Non-technical Summary
The slope of the Phillips curve represents the relationship between economic slack and inflation. This relationship is important as it is commonly used by central banks to conduct monetary policy, where the interest rate and other policy instruments are used to affect the aggregate demand (and thus the slack) in an economy so that inflation can be stabilized. Within this context, several studies in the literature have shown that the Phillips curve has flattened or even switched sign over time, which suggests that the connection between economic slack and inflation is becoming weaker or even reversed (i.e., the so-called missing inflation or deflation puzzle), conducting monetary policy has become more delicate, especially under changing economic conditions.
This paper attempts to understand the reasons behind the flattening of the (price) Phillips curve over time. The key innovation is to focus on alternative shocks that can affect the relationship between economic slack and inflation, which we call as the shock-dependent Phillips curve. This approach is in line with studies such as by <cite>mcleay2020optimal</cite> who suggest to control for supply shocks that result in a cost-push inflation to recover the Phillips curve, although this paper takes one more step to identify shocks that result in both a cost-push and a demand-pull inflation. The investigation is based on a structural vector autoregression (SVAR) model, where U.S. monthly data on inflation, unemployment, federal funds (policy) rate and global oil prices are used. The shock-dependent slope of the Phillips curve is measured by the ratio of cumulative impulse responses of inflation and unemployment, both following a common shock.
The empirical results show that positive oil price shocks that are consistent with a demand-pull inflation due to higher global demand result in an increase in inflation and a reduction in unemployment, which corresponds to having a negatively-sloped Phillips curve. In contrast, positive inflation or unemployment shocks that are consistent with a cost-push inflation result in an increase in both inflation and unemployment, especially in the long run, suggesting a positively-sloped Phillips curve. Therefore, it is implied that the slope of the Phillips curve depends on the shock experienced by the economy.
After showing empirical evidence for the shock-dependent Phillips curve, we move to the slope of the Phillips curve over time. This investigation is achieved by using average (over time) historical shocks estimated by the SVAR model as they are considered as typical shocks representing a certain time period. Specifically, the slopes of the shock-dependent Phillips curve are combined with the average (over time) historical shocks during a specific time period to obtain the slope of the Phillips curve for that time period.
The corresponding results show that there is evidence for a negatively-sloped Phillips curve for the full sample representing the monthly period between 1990 and 2021. When subsamples of 1990s, 2000s, 2010s and the COVID-19 period are investigated, it is shown that the Phillips curve has been negatively sloped during 1990s, statistically insignificant during 2000s, and it has switched sign by having a positive slope in 2010s and the COVID-19 period. When we further investigate whether any specific shock is responsible for the flattening of the Phillips curve, we show that oil price shocks explain the lion's share of changes over time.
Important policy implications follow. Specifically, based on the results in this paper, central banks are supposed to determine the drivers of inflation and conduct their monetary policy accordingly. On one hand, if a higher inflation is driven by oil price shocks that are consistent with a demand-pull inflation, central banks would have control over inflation as higher interest rates and other policy instruments can be used to lower the aggregate demand in their economy without getting away from the potential output level. On the other hand, if a higher inflation is driven by inflation or unemployment shocks that are consistent with a cost-push inflation, although higher policy rates and other policy instruments would still reduce inflation, this would be at the cost of getting away from the potential output level.