Kim-Nelson (1999b) Has the U.S. Economy Become More Stable? A Bayesian Approach Based on a Markov-Switching Model of the Business Cycle

Summary

Kim and Nelson (1999b) ask whether postwar U.S. real GDP growth has undergone a structural break toward stabilization, when that break occurred, and what its nature is. Using a Bayesian approach that embeds a one-time permanent changepoint within Hamilton's (1989) Markov-switching business-cycle model, they identify a break at 1984:Q1 and find — by comparing Bayes factors across four nested models — that a narrowing gap between boom and recession growth rates is the primary source of stabilization, at least as important as the decline in shock volatility.

Key Claims

• Model II wins: Break only in regime-dependent mean growth rates (no variance break) achieves the highest log marginal likelihood (−247.01 vs. −253.70 for variance-only, −260.19 for both, −267.63 for no break). The narrowing boom–recession gap is the dominant source of the Great Moderation.
• Break date: Posterior mode of changepoint is 1984:Q1 under all three break specifications, consistent with McConnell-Quiros (1999).
• Pre/post break means (Model II): recession mean shifts from $\hat\mu_0=-1.025$ to $\hat\mu_0+\hat\mu_{00}=-0.195$; boom mean shifts from $\hat\mu_1=0.486$ to $\hat\mu_1+\hat\mu_{11}=0.137$. Recessions became shallower and booms more modest.
• Identification insight: Within a linear model a narrowing regime gap is observationally equivalent to declining unconditional variance. The Markov-switching framework is necessary to disentangle the two sources.
• Bayesian advantage: The unknown changepoint $\tau$ is not a nuisance parameter; it integrates out naturally via the absorbing-state Markov chain $D_t$, and its posterior distribution is a byproduct of the Gibbs sampler.
• Estimation: Seven-block Gibbs sampler extending Albert-Chib (1993); marginal likelihoods via Chib (1995, 1998) reduced-run decomposition; 12,000 draws (2,000 burn-in); three alternative prior specifications for robustness.

Concepts Introduced or Extended

• Great Moderation
• Markov-Switching VAR
• Marginal Data Density
• Gibbs Sampler
• Structural Break Testing

Entities Mentioned

• Chang-Jin Kim
• Charles R. Nelson
• James D. Hamilton
• Siddhartha Chib

Quotes

"a narrowing gap between growth rates during recessions and booms that is at least as important as any decline in the volatility of shocks"

"within the context of a linear model, one cannot distinguish between the two sources, and a narrowing gap between growth rates would show up as a decline in volatility"

My Take

The methodological contribution — modeling the unknown changepoint as an absorbing Markov chain and then using Chib's (1995) marginal likelihood algorithm to compare nested models — is clean and reusable. The economic finding is substantively important: the Great Moderation is not merely about smaller shocks but about the business cycle becoming less asymmetric. This is consistent with the plucking model (Kim-Nelson 1999a): improved monetary policy that shallows recessions shows up as a narrowing regime gap rather than a variance change, so the two papers' findings are complementary.