Overview
Chang-Jin Kim is an economist at the University of Washington (Seattle). He is best known for developing the Kim filter and smoother — an algorithm for inference in state-space models with Markov-switching dynamics — and for the graduate textbook State-Space Models with Regime Switching (with Charles R. Nelson, MIT Press 1999). His work bridges Bayesian computation, regime-switching models, and applied macroeconometrics.
Key Contributions / Features
- Kim filter and smoother (Kim 1994) — "Dynamic Linear Models with Markov Switching," Journal of Econometrics 60: 1-22. An approximate filter/smoother for state-space models where the measurement and/or transition equations depend on a Markov-switching variable. The filter collapses the growing mixture of Gaussians at each step to maintain tractability; the smoother runs backward to yield smoothed regime probabilities. Foundation for Gibbs-sampling estimation of MS state-space models.
- Long-run U.S./U.K. real exchange rate (Engel-Kim 1999) — Applied the Kim smoother within a Gibbs sampling framework to decompose the real exchange rate into a permanent (MS random walk) and transitory (MS AR(1)) component over 106 years; found a large permanent component inconsistent with PPP; showed ADF test size is distorted to 19% at the nominal 5% level under MS heteroskedasticity. See Purchasing Power Parity.
- UC-MS fad model (Kim-Kim 1996) — With Myung-Jig Kim, decomposed monthly real S&P 500 returns (1952:1–1992:12) into a permanent AR(2) and transient fad AR(1) component, with independent two-state Markov chains governing each component's innovation variance. Estimated via the Kim (1994) approximate filter; identified only two significant fad episodes (OPEC 1973–74, 1987 crash), both reflecting unwarranted pessimism. Log-likelihood −1383.29 vs. GARCH −1395.05. See Kim-Kim (1996).
- Business cycle turning points and coincident index (Kim-Nelson 1998) — With Charles R. Nelson, combined the Stock-Watson (1989, 1991) dynamic factor model with Hamilton's (1989) regime-switching model and estimated the synthesis fully Bayesian via multimove Gibbs sampling. Four monthly DOC coincident indicators (IP, personal income, manufacturing sales, nonfarm payrolls) are driven by a common factor whose mean growth rate switches between recession and boom states. The model generates a new coincident index and regime probabilities that align closely with NBER turning points; a Bayesian variable-selection extension tests for business cycle duration dependence and finds robust positive duration dependence for recessions. See Kim-Nelson (1998).
- Plucking model econometrics (Kim-Nelson 1999a) — With Charles R. Nelson, provided the first formal econometric model of Friedman's (1964) plucking hypothesis. Decomposes output into a stochastic trend ceiling and a transitory cycle with an asymmetric Markov-switching shock; estimated via the Kim (1994) approximate MLE filter on U.S. real GDP and unemployment (1951:1–1995:3). Finds strong support for plucking: π^=−0.011 (highly significant), symmetric shock essentially zero for GDP, apparent transitory persistence drops from 0.946 (linear UC) to 0.797 once asymmetry is modelled. Three business cycle phases implied: normal, recessionary, and high-growth recovery. See Plucking Model and Kim-Nelson (1999a).
- State-Space Models with Regime Switching (Kim and Nelson 1999) — Graduate textbook covering the Kim filter, Gibbs sampling for MS state-space models, and empirical applications to business cycles, exchange rates, and interest rates. Standard reference for Bayesian MS time-series analysis.
- Great Moderation and structural break in MS model (Kim-Nelson 1999b) — With Charles R. Nelson, embedded a one-time permanent changepoint (absorbing Markov state Dt) within Hamilton's (1989) Markov-switching business-cycle model. Applied to U.S. real GDP (1953:II–1997:I), Bayes factors favor a break in regime-dependent mean growth rates over a variance-only break (ln m=−247.01 vs. −253.70); break date posterior mode at 1984:Q1. Dominant source of the Great Moderation is a narrowing boom–recession gap: recession mean shifts from −1.025 to −0.195, boom mean from 0.486 to 0.137. See Great Moderation and Kim-Nelson (1999b).
Related