[What Ends Recessions?]: Comment

Romer and Romer advance “a startling and intriguing proposition: Active, systematic monetary policy ended postwar recessions,” which Cochrane identifies as “the latest in a series of provocative papers in which Christina and David Romer have revived some of the methods and views of Friedman, and Friedman and Schwartz” (Section 1, p. 58). He immediately reaches for Friedman’s classic critics – Tobin’s (1970) “Post Hoc Ergo Propter Hoc” and Kareken and Solow’s (1963) “Lags in Monetary Policy” – which “outlined the issues that, formalized by Sims (1972) and others, today define the standard methodology for evaluating monetary policy,” framing his whole comment as a re-application of decades-old identification concerns to a paper that, in his view, does not engage them.

Cochrane accepts that the Romers’ collection of FOMC statements shows the Fed “perceives the onset of a recession quickly” and “seem[s] to advocate countercyclical policy,” useful evidence about the Fed’s information set and reaction function (Section 2, pp. 58-59). But he argues this documents only that the Fed reacted to output and inflation, “but doesn’t tell us if output reacted to the Fed.” Invoking Kareken and Solow’s parable – a central bank that perfectly stabilizes the economy would be observed doing nothing while the economy stays calm, and “an observer…would…conclude that monetary policy has no effects at all, which would be precisely the opposite of the truth” – Cochrane concludes that “predictable actions are precisely those actions whose correlations with other events have dubious causal interpretations,” quoting Sims’s (1992) question “does the cock’s crow cause the sunrise?” (p. 59).

Cochrane sets up a simple simultaneous system in which output depends on the monetary policy variable m and the Fed’s feedback rule depends on output, and shows that OLS cannot separate the structural output equation from the feedback rule unless contemporaneous m is assumed not to affect y within the quarter (Section 3.1.1, pp. 60-62). If that identifying assumption is wrong – if m can affect y within the same quarter – “OLS recovers mongrels, combinations of the structural output effects and the Fed feedback rule,” and in the polar case where the true output effect of m is zero, “OLS estimates recover the feedback rule and have nothing to do with the effects of m on y.” He notes the Romers’ regression is, in effect, “the first row of a bivariate vector autoregression” with a recursive orthogonalization ordering output first – the opposite assumption from most VAR studies using monetary aggregates, which assume the Fed cannot react within the quarter.

If other variables the Fed watches (and which help predict output) are left out of the regression of output on lagged output and the policy variable, “the estimated lag polynomials…are mongrels, combinations of all the lag polynomials in the system,” so the resulting “bias” – which Romer and Romer “acknowledge but belittle” – “can be upward or downward or both (at different lags)” (Section 3.1.2, p. 63). Cochrane argues this is not merely a theoretical possibility: “level” variables such as the consumption/output ratio, the term spread, and the default spread are known to significantly help predict output and monetary policy (citing his own 1994 “Shocks” paper), and the Romers’ own historical analysis “convinces one that the Fed obsessively watches an enormous number of economic variables when setting policy” – so omitted variables are, in his words, “a serious problem in Equation (1),” not just an in-principle worry.

Cochrane argues no: the Boschen-Mills index is “just another measure of the stance of monetary policy,” giving no reason to expect it is less correlated with the error term than the federal funds rate itself, and the Romer-Romer dates – meant to mark shifts to anti-inflation policy – require assuming “the change in regime is made without regard to the current state of output, anticipated future output, or other variables correlated with output,” a claim Cochrane finds hard to believe: “no disinflation event came in the depth of a depression!” (Section 3.2, pp. 63-64). He stresses this is “the crucial piece of evidence and it is not addressed by Romer and Romer’s historical analysis” – their own FOMC-minutes reading, in fact, seems to show the opposite, that the Fed does attend to output when shifting toward disinflation.

Cochrane sets up a structural model in which output responds separately to anticipated and unanticipated components of monetary policy, and shows in the Appendix that the coefficient governing the effect of anticipated policy on output and the coefficient governing the Fed’s response to output in its feedback rule “are not separately identified” – dividing the two estimated relationships to eliminate one unknown “leaves one equation in the two unknowns,” proving the proposition stated in the text (Section 3.3 and Appendix, pp. 64-65, 72-73). The practical upshot: “no regression can distinguish whether the true ‘policy multiplier’ is that estimated by Romer and Romer or zero. We must impose some theory or ‘identifying restriction’ to get an answer.” The Romers implicitly assume no distinction between anticipated and unanticipated money, which is exactly what delivers their large multiplier – but that assumption is a theoretical commitment, not a neutral default.

He points to counterexamples that most monetary economists accept: “the ends of hyperinflations, currency revaluations, and policies in countries with high and variable inflation” appear to have negligible real effects even though they are large monetary events, so “we need a view of money that explains why monetary policy does have effects in some circumstances and does not in others,” and “most monetary models that can explain both sets of observations give no role to systematic policy (beyond inflation-tax effects)” (Section 3.3, pp. 65-66). He is careful to frame this as a challenge rather than a refutation: “Perhaps Romer and Romer do want to assume there is no distinction. The point is that the assumption identifying what is policy-invariant is crucial, so it needs to be explicit and linked to a monetary theory.”

He objects on three grounds (Section 3.4, pp. 66-67): the responses are “permanent and delayed,” a pattern “no story for the effect of money on output that I know of produces”; the responses are “big” – a one-percentage-point decline in real rates raising output by up to 3%, which on a simple Keynesian investment channel “require[s] a 30% rise in investment for each percentage point decline in interest rates,” far larger than the investment literature supports even under expansive assumptions; and comparable two-variable VARs using the same federal-funds-output system produce similarly large, permanent, delayed responses only until other variables (notably commodity prices, to purge the “price puzzle”) are added, at which point “federal funds shocks…account for 10% or less of output variation.” He adds a reductio: if the estimated multipliers and the presumption that the Fed fully controls the real rate were taken literally, nothing would stop the Fed from setting the real rate at -4% and “permanently rais[ing] output by 20%” – since no one believes that is possible, the multipliers themselves must not be taken fully at face value.

He argues that most macroeconomics, including standard stochastic growth models, already predicts that the economy reverts after a shock without policy intervention – a calibrated growth model implies a 9.1-quarter half-life (5.8 quarters with higher depreciation) – and that the data agree: a simple autoregression of the nondurables-and-services consumption/output ratio implies a 5.07-quarter half-life, and Figure 1’s plot of consumption and output through four recessions shows “consumption is barely affected by the declines in output,” meaning “consumers expected the recessions to end promptly” on their own (Section 4, pp. 67-69). He preempts the rebuttal that consumers might be rationally anticipating Fed-engineered recoveries: to believe that, “one must again believe that completely anticipated, systematic policy can have real effects” while somehow not being anticipated in prices – an inconsistency he finds implausible. He concludes: “the ‘persistence’ of recessions that Romer and Romer seek to explain by persistent policy isn’t there. It couldn’t be. If it was, recessions wouldn’t have ’ended’!”

He charges that “Romer and Romer completely ignore” three strands of directly relevant work: the Tobin-Solow methodological critiques and the VAR literature built in response to them (which by his account is “converging on similar answers,” much smaller than the Romers’ multipliers); the theoretical literature on monetary and fiscal nonneutrality that is “the heart of macroeconomic training in every Ph.D. program”; and the applied literature on Fed decision-making and lags following Kareken, Solow, Poole, McCallum, and Meltzer (Section 5, pp. 69-71). He offers a “quite sympathetic” interpretation – that thirty years of difficult, inconclusive VAR and theoretical work make a return to narrative history tempting – but argues that once the Romers moved from narrative to quantitative multipliers (“Adam and Eve in the garden of Friedman, they have taken one bite of the forbidden econometric fruit”), they inherited exactly the identification problems that motivated the VAR literature in the first place, and “cannot be addressed by quotes from FOMC meetings” alone.

He does not dismiss history altogether: “Historical analysis should be able to help us figure out how monetary policy has nonneutral effects,” citing Sargent’s (1986) analysis of the ends of hyperinflations as an example that “brings home the potential neutrality of some large monetary events, the government’s intertemporal budget constraint, and the fact that inflation is often…a fiscal phenomenon, in a way that mountains of formal papers do not” (Section 5, pp. 71-72). He singles out the Romers’ reading of FOMC minutes as “very helpful in sorting out how the Fed reacts to the economy” – useful as an input to identifying the Fed’s reaction function – but insists that “a successful reading of history can’t ignore Tobin and Solow’s concerns,” and closes with the line that “if economic history simply ignores the history of economics, it is doomed to repeat it.”