Sims-Zha (2006) Were There Regime Switches in U.S. Monetary Policy?

Summary

Sims and Zha confront a multivariate Markov regime-switching structural VAR with U.S. data to ask whether the apparent improvement in monetary policy between the 1970s and 1980s reflects genuine changes in the policy rule ("bad policy") or merely changes in the size of the shocks ("bad luck"). Comparing model variants by Bayesian posterior odds (which automatically penalize unneeded parameters), they find the best-fitting specification allows time variation only in the disturbance variances, not in coefficients. When coefficients are allowed to change, the best fit confines the change to the monetary-policy equation, yielding roughly three/four regimes that line up with the Burns, Volcker, and Greenspan eras — but the estimated differences among policy rules are not large enough to account for the rise and fall of inflation in the 1970s–80s. The Volcker reserve-targeting period stands out as one of high policy-shock variance. (American Economic Review 96(1): 54–81.)

Key Claims

• The question and the camps. Clarida–Galí–Gertler (2000) and Lubik–Schorfheide (2004) argue the 1970s policy rule was passive enough to admit sunspot (indeterminate) equilibria, explaining the Great Inflation; the later rule removed the indeterminacy. Sims–Zha (like Primiceri 2005, Cogley–Sargent 2005, Bernanke–Mihov 1998, Stock–Watson 2003) find only modest evidence that the policy rule changed drastically.
• Best fit: variances change, coefficients don't. Among all the switching specifications, the one that fits best (by posterior odds) allows only the structural disturbance variances to switch across regimes, with coefficients constant. Combining pre- and post-Volcker samples is therefore justified provided heteroskedasticity is modeled.
• If coefficients change, only the policy rule does. The best-fitting coefficient-switching model restricts changes to the monetary-policy equation (other equations constant). As in Cogley–Sargent and Primiceri, the point estimates of these changes are non-trivial but their magnitudes are left uncertain by the data.
• Four regimes, loosely the chairmen. The coefficient-switching model finds four regimes: a brief very-high-money-demand-variance regime (e.g. Sept–Oct 2001); the Volcker reserve/monetary-aggregate-targeting regime; a Greenspan regime (also spanning much of the 1960s); and a Burns regime (late 1960s–early 1970s) that leans more toward aggregate-targeting than the Greenspan regime but less than Volcker.
• Policy differences don't explain the Great Inflation. Counterfactual simulations swapping policy regimes show the differences among the estimated rules are too small to produce the observed rise then decline in inflation — so changing systematic policy is not the main driver; the volatility of shocks (the variance regimes) matters more. This is evidence on the "good luck vs. good policy" debate leaning toward luck.
• Method. A structural VAR with explicit identifying assumptions (isolating the policy block) plus Markov-switching in variances and/or coefficients; models (number of regimes, monotonic vs. recurring regimes, independent vs. synchronized timing of variance and coefficient changes) are compared by Bayesian posterior odds ratios.

Concepts Introduced or Extended

• Markov-Switching VAR
• Monetary Policy Shocks
• Great Moderation
• Structural Identification

Entities Mentioned

• Christopher Sims
• Tao Zha

Quotes

"The best fit allows time variation in disturbance variances only. With coefficients allowed to change, the best fit is with change only in the monetary policy rule … But the differences among regimes are not large enough to account for the rise, then decline, in inflation of the 1970s and 1980s."

"The Volcker reserves-targeting period emerges as a period of high variance in disturbances of the policy rule. This finding lends empirical support to the common practice … of combining the samples before and after the reserve-targeting period … as long as heteroskedasticity is properly taken into account."

My Take

This is the Markov-switching answer to the same "good luck vs. good policy" question that Primiceri's TVP-VAR poses with smooth drift, and it reaches a compatible verdict from the discrete-regime side: what changed most reliably across the postwar era is the variance of the shocks, not the systematic policy rule — a direct challenge to the Clarida–Galí–Gertler "the rule fixed indeterminacy" reading of the Great Inflation. Two features make it a methodological template for the MS-VAR literature: identifying the policy block structurally while letting only variances (or only the policy equation) switch, and adjudicating the many nested specifications by posterior odds rather than by fit alone. Its honest caveat — that the coefficient-change magnitudes are real but imprecisely estimated — is exactly why the debate with the "policy changed" camp never fully closed. For this wiki it is the empirical anchor connecting regime-switching VARs, monetary-policy identification, and the Great Moderation.