Analyzing GDP and stock market dynamics

Abstract

This paper aims to examine whether the forces linking financial markets to real economic activity operate differently across business cycle phases, using quarterly US data from 1990 to 2024, spanning four recession episodes. Specifically, the author asks whether the mechanism that normally keeps equity markets anchored to corporate earnings and real output remains stable between expansions and recessions and what the accumulated output cost is when that mechanism breaks down. The author first estimates a vector error correction model among real gross domestic product (GDP), the S&P 500 Total Return Index and Earnings Per Share, using cointegration tests to identify the long-run equilibrium structure and controlling for monetary policy, consumer confidence, market uncertainty and real GDP expectations. The author then extends this to a Bayesian Markov-Switching vector error correction model that holds the cointegrating vectors constant while allowing adjustment dynamics and shock covariance structures to vary across regimes, with regime identification anchored to NBER recession dates. The author identifies two stable long-run equilibria anchored by earnings per share. This study finds that the stock market index self-corrects toward its earnings equilibrium in normal expansions, while in recessions, the adjustment coefficient linking the stock market index to earnings reverses sign, with the index moving further from earnings fundamentals; as the Granger causality tests detect predictive content from stock returns and earnings growth to GDP growth but not in the reverse direction, no offsetting predictive force is found within the estimated system. The accumulated output cost amounts to 1.78 percentage points of cumulative GDP growth deficit by quarter 20 following a recession onset. The author provides direct evidence that the corrective mechanism linking the stock market index to its long-run earnings equilibrium is regime-dependent, reversing during recessions in a way that has not previously been documented within a regime-switching cointegration framework.