Full Paper: SSRN
Journal of Macroeconomics, accepted subject to minor revisions
Working paper version: PDF, online appendix
Abstract: Remote work has split the labour market into two kinds of employers. Firms that hire remotely tap a national applicant pool, insulating their wages from local unemployment—a flattening force on the Phillips curve. On-site workers, meanwhile, can threaten to quit for remote jobs, forcing their employers to bid up wages when the market is tight—a steepening force. We identify both channels across 353 US metropolitan areas using instrumental variables and rationalise them in a New Keynesian model with search frictions and heterogeneous firms. Quantitatively, the steepening channel dominates: the aggregate Phillips curve steepened by 48% after 2020. The driver is not remote work itself but the tightness of the post-pandemic labour market—at 3.5% unemployment, firms must compete so aggressively for each hire that wages become far more responsive to demand. A central bank using the old, flatter slope underestimates inflationary pressure, generating 34% excess inflation volatility. Fiscal policy can reshape the slope too: a remote-hiring subsidy flattens the curve and raises the spending multiplier, while a job-retention subsidy steepens it and diverts a quarter of the stimulus into inflation. The broader lesson is that the Phillips curve slope moves with the labour market—how firms recruit and how workers search determines how sharply wages respond to demand.
Status: Draft available upon request
Demand, Supply, and Measurement Error: How to Conduct Monetary Policy When Shocks are Unclear (with Sydney Phiri)
Abstract: Optimal monetary policy requires distinguishing demand from supply shocks, yet the sign-based methods widely used for this classification rely on real-time data that are measured with substantial error. We develop a framework linking measurement error to misclassification probabilities and welfare losses, showing that noise in output data—the variable that distinguishes supply from demand—systematically biases classifications toward the demand label. The theory is validated on real-time data vintages from the Philadelphia Fed, where nearly half of quarterly classifications reverse upon revision, and on 120 disaggregated PCE categories, where over one-third of item-quarter observations lie close to the classification boundary. In an estimated Smets-Wouters model, misclassification erodes the welfare gain from shock-contingent policy by up to half under data-calibrated noise. We propose two factor-based methods—a FAVAR and a common-component VAR—that exploit the cross-section of disaggregated prices to filter out measurement error, restoring classification accuracy and recovering most of the welfare gain from correct shock identification
Status: Draft coming soon!!