Overview
Christopher Sims is an economist (Princeton) and 2011 Nobel laureate who introduced Vector Autoregressions to macroeconomics in his 1980 paper "Macroeconomics and Reality." His work fundamentally changed how empirical macroeconomists approach dynamic modeling, replacing large structural models with flexible data-driven systems evaluated primarily on fit.
Key Contributions / Features
- Incredible Identification and VAR Introduction (Sims 1980) — "Macroeconomics and Reality," Econometrica 48(1): 1–48. Argued that exclusion restrictions in Cowles Commission models are "incredible" normalizations rather than genuine economic theory; Hatanaka (1975) dynamics problem (unknown lag lengths require exogenous instruments at every lag); rational expectations makes forward-looking identification (b+) orders of magnitude harder than backward-looking (b−). Proposed the VAR as a minimally-restricted alternative treating all variables as jointly endogenous. Empirical application: 6-variable quarterly system (M, Y/P, U, W, P, PM) for U.S. (1949–75) and West Germany (1958–76); (T−k)-corrected LR test selects p=4 (χ2(144)=166.09 U.S., 142.53 Germany); recursive Cholesky IRFs; variance decompositions show money dominates U.S. nominal variables at long horizons (64% of W, 60% of P at k=33); block exogeneity tests reject real-sector exogeneity to money in both countries (U.S. χ2(24)=64.63, p<0.001). See Structural Identification, Vector Autoregression, Granger Causality.
- Understanding Unit Rooters (Sims-Uhlig 1991) — "Understanding Unit Rooters: A Helicopter Tour," Econometrica 59(6): 1591–1599 (with Harald Uhlig). Demonstrated graphically that the likelihood function in AR(1) models is Gaussian-symmetric around ρ^ even at the unit root boundary — the asymmetry in Dickey-Fuller p-values is a property of the estimator's sampling distribution, not the likelihood. Monte Carlo "helicopter tour" of the 3D joint pdf of (ρ,ρ^): slicing along ρ=1 gives the asymmetric classical distribution; slicing along ρ^=1 gives a symmetric posterior. Showed that using DF p-values as posterior probabilities implicitly imposes a data-dependent prior favoring explosive ρ>1; for ρ^=0.95 the implicit prior is 2–3× higher at ρ=1 than ρ=0.9. Recommendation: report conventional t/F statistics alongside any unit-root p-values. See Unit Root Inference.
- Macroeconomics and Methodology (Sims 1996) — "Macroeconomics and Methodology," Journal of Economic Perspectives 10(1): 105–120. Epistemological defense of Bayesian inference in macroeconomics via a science-as-data-reduction argument (Kepler → Newton analogy): the Bayesian posterior is the formal analog of compressing data without sacrificing predictive content. Argues Bayesian inference is unavoidable when a single historically given time series is the object of inference, not repeated experiments. Critiques RBC calibration as "computations, not experiments," citing Watson (1993) showing RBC model dynamics are drastically dissimilar to U.S. data. Provides a historical narrative of monetary VAR identification from Sims (1972) through CEE/Sims-Zha (1994–95), and endorses Leeper-Sims (1994) — a DSGE model with fiscal sector that fits near a 3-variable reduced-form VAR — as the correct research frontier. See Marginal Data Density, Vector Autoregression.
- Sims (1982) — policy analysis with econometric models; modeled policy regime switches as sequences of random shocks.
- Sims-Zha (1998) "Bayesian Methods for Dynamic Multivariate Models" (IER 39(4): 949–968, with Zha) — unified A+∣A0 posterior derivation; Kronecker symmetry condition for 400× speedup in large VARs; three dummy-observation types (tightness, sums-of-coefficients μ5, initial-observation μ6); structural vs. reduced-form prior comparison. See Sims-Zha Prior.
- Sims-Zha (1999) — joint error bands for impulse responses (see Error Bands).
- Leeper-Sims-Zha (1996) — demonstrated feasibility of 18-variable Bayesian VARs.
- Sims-Zha (2005a, 2005b) — Markov-switching VARs and the Great Moderation debate (see Markov-Switching VAR).
Related