The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
T HE C ASE OF THE D ISAPPEARING P HILLIPS C URVE James Bullard President and CEO 2018 ECB Forum on Central Banking Macroeconomics of Price- and Wage-Setting June 19, 2018 Sintra, Portugal Any opinions expressed here are my own and do not necessarily reflect those of the FOMC. I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY Introduction R EGRESSIONS R ELEVANCE I MPLICATIONS I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS I NTRODUCTION The slope of estimated Phillips curves in G-7 economies was negative in the 1980s but has been drifting toward zero in the inflation targeting era since 1995. This is an empirical phenomenon often referred to as a “flattening Phillips curve.” Monetary authorities have generally improved policy during the inflation targeting era—inflation has generally been lower, less volatile and closer to stated inflation targets. I will argue that the improved monetary policy has led to the flatter empirical Phillips curve. I will draw out the implications for monetary policy after making my core argument. I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS Empirical Evidence of a Flatter Phillips Curve R ELEVANCE I MPLICATIONS I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY E MPIRICAL EVIDENCE ON THE P HILLIPS R EGRESSIONS R ELEVANCE I MPLICATIONS CURVE In the past 30 years, the empirical Phillips curve has flattened in advanced economies. The following chart shows the coefficient on a measure of resource slack (unemployment) in a regression of price inflation on resource utilization. The analysis is contained in the latest BIS annual report. The data are for a panel of G-7 economies. The coefficient is estimated for rolling 15-year samples, from the 1980s to the present. The point estimate is a weighted average across economies. I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY F LATTENING OF THE P HILLIPS CURVE IN G-7 R EGRESSIONS R ELEVANCE ECONOMIES F IGURE : Time-varying Phillips curve slope. Source: Bank for International Settlements (2017). I MPLICATIONS I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS A Simple Model R ELEVANCE I MPLICATIONS I NTRODUCTION A F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE SIMPLE AND STANDARD MODEL I will use a simple and standard model to state the argument. This model is a version of more complicated models that underlie much of the analysis in modern central banking. I MPLICATIONS I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS T HE STANDARD N EW K EYNESIAN MODEL Dynamic IS equation: yt = Et ( yt + 1 ) − 1 [it − (ρ + et ) − Et (πt+1 )] σ (1) A structural, New Keynesian Phillips curve: πt = κyt + βEt (πt+1 ) + ut (2) Monetary policy conducted using a Taylor-type monetary policy rule: it = ρ + ϕπ πt + ϕy yt y, π, i, ρ + e: the output gap, inflation gap, short-term nominal interest rate and natural real rate of interest, respectively. e, u: the natural rate shock and the cost-push shock, respectively. σ, κ, β: structural parameters, all positive. ϕπ , ϕy : policy parameters, with ϕπ > 1 and ϕy > 0. (3) I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS M ODEL EQUILIBRIUM The equilibrium has the output gap and the inflation gap evolving as linear functions of the shocks: yt = πt = et − ϕπ ut , σ + ϕy + κ ϕ π κet + σ + ϕy ut . σ + ϕy + κ ϕ π (4) (5) I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS Monetary Policy R ELEVANCE I MPLICATIONS I NTRODUCTION F LATTENING C ONSTRAINED M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS OPTIMAL MONETARY POLICY We look for optimal monetary policy within the set of Taylor-type rules in the model. Fix ϕy to any positive value, and then choose the optimal value of ϕπ by minimizing a quadratic: ∞ (6) ϕπ = arg min (1 − β) ∑ βt απt2 + y2t , t=0 where α > 0 represents the relative weight on the desirability of inflation stabilization compared to output stabilization. Regardless of the value of α, the solution to this problem is to set a large coefficient on the inflation gap, technically, ϕπ → ∞. I NTRODUCTION F LATTENING I NTERPRETATION : M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS BETTER INFLATION TARGETING Interpretation of the solution: “The policymaker should promise to react aggressively to deviations of inflation from target in conducting monetary policy.” The idea that policymakers put more weight on inflation deviations during the post-1995 period could be related, in part, to quantitative easing and other unconventional policy measures during years when inflation has been below target. I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS Empirical Phillips Curves from Model Data R ELEVANCE I MPLICATIONS I NTRODUCTION F LATTENING T HE P HILLIPS M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS CURVE SLOPE IN THEORY Now let’s regress the inflation gap on the output gap inside the model and call the estimated coefficient “the slope of the empirical Phillips curve.” The slope can be calculated exactly as κσe2 − ϕπ σ + ϕy σu2 Cov (πt , yt ) γ= = . (7) Var (yt ) σe2 + ϕ2π σu2 σe2 , σu2 : variance of the natural rate shock and cost-push shock, respectively. Main result: Under the optimal monetary policy defined above, the empirical Phillips curve becomes flat, that is, lim γ = 0. ϕπ →∞ (8) I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS Empirical Relevance R ELEVANCE I MPLICATIONS I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS E MPIRICAL RELEVANCE Would this Lucas critique effect be large enough to importantly affect estimated Phillips curve coefficients? I consider a similar model, estimated by Lubik and Schorfheide (2004, AER). I use mean estimates for post-1982 data from their Table 3, p. 206, to generate artificial data and regress inflation on the output gap. I use Okun’s law with a coefficient of −2.3 to translate the Phillips curve slope in terms of unemployment. The following chart suggests that, at these parameter values, the slope of the estimated Phillips curve would attenuate significantly as ϕπ increases. I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS E MPIRICAL RELEVANCE R ELEVANCE Coefficient 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 -1.2 1 2 3 4 5 6 7 8 9 -1.4 10 F IGURE : Phillips curve slope as a function of the interest rate response to inflation. I MPLICATIONS I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS A DDITIONAL LITERATURE Boivin and Giannoni (2006, REStat) Monetary policy has been more effective in stabilizing the economy post-1980 by responding more aggressively to inflation expectations. Del Negro, Giannoni and Schorfheide (2015, AEJ Macro) During the Great Recession, did the Phillips curve (PC) break down (sharp decline in real activity, but only modest decline in inflation)? No: A standard DSGE model with a time-varying inflation target and financial frictions predicts a sharp contraction in economic activity and a modest and protracted decline in inflation in response to financial stress. McLeay and Tenreyro (2018, CEPR DP12981) The structural PC is a positive relationship between inflation and the output gap. Optimal policy induces a negative relationship between inflation and the output gap in response to cost-push shocks. Thus, the PC cannot be easily identified in the data. I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE Implications for Today’s Monetary Policymakers I MPLICATIONS I NTRODUCTION F LATTENING M ODEL M ONETARY POLICY R EGRESSIONS R ELEVANCE I MPLICATIONS L OOK FOR A DIFFERENT SIGNAL Ultimately, successful monetary policy can push the empirical Phillips curve slope all the way to zero. The model economy in this talk still has a structural Phillips curve; it is only the empirical Phillips curve that is “disappearing.” Today’s G-7 monetary policymakers are unlikely to glean a reliable signal for monetary policy based on empirical Phillips curve slope estimates—they have to look elsewhere.