Assessing Changes in the Monetary Transmission Mechanism: A VAR Approach

Boivin and Giannoni (2002) ask whether the transmission of monetary policy shocks to U.S. output and inflation changed since the early 1980s, and whether any change reflects altered propagation, altered shock variances, or altered systematic policy conduct. They answer using a four-variable, four-lag recursive (Cholesky) VAR – detrended output, GDP-deflator inflation, commodity-price inflation, and the federal funds rate – estimated separately over pre-1980, post-1980, and post-1984 U.S. sub-samples, following the Christiano-Eichenbaum-Evans (1999) identification in which the funds rate is ordered last and does not affect the other variables contemporaneously.

The analysis uses quarterly U.S. data from 1963:Q1 to 1997:Q4 (detrended real GDP via a high-pass filter, GDP-deflator inflation, commodity-price inflation from the PSCCOM index, and the federal funds rate), split into a pre-1980 sample (1963:Q1-1979:Q3), a post-1980 sample (1980:Q1-1997:Q4), and a post-1984 sample (1984:Q1-1997:Q4). The 1979:Q4 break marks the Volcker shift in Federal Reserve operating procedure, and the 1984:Q1 break follows McConnell and Perez-Quiros’s (2000) estimated Great Moderation date; the post-1984 sample specifically excludes the 1979:Q4-1983:Q4 nonborrowed-reserves-targeting period, during which the funds rate is a poor indicator of policy stance.

Yes: Andrews (1993) sup-Wald tests (allowing an endogenous break) reject coefficient stability in 6 of 16 tests (38%) at the 5% level, and chi-squared tests reject stability of the innovation covariance matrix in 5 of 8 tests overall (63%, pooling both break dates) — rejecting in all 4 of 4 variables at the 1979:Q3 break but only 1 of 4 at the 1984:Q1 break. The authors conclude “instability is prevalent both in the systematic part of the estimated VAR and in the variance of shocks” (p. 98).

A reduced-form counterfactual decomposition finds that changed propagation coefficients (the A matrices) account for about 38% of the pre-1980-to-post-1980 decline in detrended-output variance (a fall of 2.85 percentage-points-squared, of which 1.07 is attributed to propagation) and about 58% of the inflation-variance decline (5.02 points-squared, of which 2.91 is propagation); for the pre-1980-to-post-1984 comparison the propagation shares are 22% for output and 56% for inflation. The authors’ own rounded headline figures – “roughly 40 percent” for output and “60 percent” for inflation – refer specifically to the pre- vs. post-1980 comparison, not the post-1984 one, where the output share is markedly lower.

Structural variance decompositions show the monetary policy shock’s contribution to output variance falling steadily, from 19% pre-1980 to 7% post-1980 to 3% post-1984, and its contribution to inflation variance falling from 14% to 10% to 6%. Notably, the estimated variance of the policy shock itself is higher in the post-1980 sample than pre-1980 (0.61 vs. 0.30), which the authors attribute to the anomalously volatile 1979:Q4-1983:Q4 nonborrowed-reserves-targeting episode included in that sample; once that episode is excluded in the post-1984 sample, shock variance falls to 0.14. Because output’s variance share still declined sharply from 19% to 7% even while the shock got bigger, the authors read this as evidence that the propagation of policy shocks – not just their size – changed.

Following an unexpected 1-percentage-point (100-basis-point) increase in the federal funds rate, the response of detrended output and inflation is “much less pronounced and persistent” in the post-1980 and post-1984 samples than pre-1980, with the trough of the output response “more than twice as large” pre-1980 as in either later sample (p. 104). The 95% confidence bands overlap substantially across sub-samples, so while the point-estimate differences are consistent across specifications, they are not sharply statistically distinguished; exact horizon-specific magnitudes are read from the paper’s chart rather than tabulated.

Splitting the SVAR’s parameters into a systematic-monetary-policy-rule block (Phi) and a rest-of-economy block (Omega), the authors find that replacing only the pre-1980 policy rule with the post-1980 (or post-1984) rule – while holding Omega fixed at its pre-1980 values – produces impulse responses that closely resemble the actual post-period responses, whereas replacing only Omega while holding Phi fixed at pre-1980 values “change[s] only little” (pp. 104-105). The authors conclude that “most of the reduction in the response of output and inflation in the post-1980 and post-1984 samples is accounted for by a change in the parameters characterizing systematic monetary policy” (p. 105), a result that holds in both the post-1980 and post-1984 comparisons.

Using a stylized intertemporal IS-curve model (y_t = E_t y_{t+1} minus sigmar_t + delta_t, with policy rule r_t = phiy_t + epsilon_t), the authors show the impact of a monetary policy shock on output is -sigma/(1+sigma*phi), which shrinks either if sigma (the interest-rate elasticity of output, i.e., genuine transmission strength) falls or if phi (the policy feedback coefficient) rises – both generate an identical, observationally equivalent attenuation of the impulse response. Because the Phi-block shift dominates in their counterfactual, the authors interpret their finding as more consistent with a more aggressive, stabilizing policy rule (higher phi) than with weaker underlying transmission (lower sigma), stating that “a reduced output response to shocks is exactly what one would expect from a successful monetary policy that aims to stabilize output” (p. 106) – but they caution that the VAR counterfactual alone cannot separately identify sigma from phi, a task left to a companion structural working paper.

The authors flag that their counterfactual is vulnerable to the Lucas critique: they acknowledge that “the description of the private sector behavior…corresponds to a reduced-form,” and that because firms and consumers care about their expectations of future states of the economy, “changes in the policy reaction function are thus likely to affect the way agents form their expectations of future variables, and therefore the coefficients of the equations in the system” (p. 105) – i.e., the private-sector equations (Omega), held fixed across policy regimes in their counterfactual, could themselves shift when the policy rule changes, an effect their reduced-form analysis cannot capture. They also note that the federal funds rate is a poor indicator of policy stance during the 1979:Q4-1982:Q4 nonborrowed-reserves-targeting period (motivating the post-1984 robustness sample), that their recursive identification depends on a specific zero restriction and variable ordering not tested against alternatives, and that the paper itself – appearing in a Federal Reserve policy review rather than a peer-reviewed journal – is shorter and less formal than a full journal treatment, though its VAR analysis and results are original.