Understanding the Price Puzzle

The price puzzle is the finding that, in standard reduced-form VARs, a positive innovation to the federal funds rate (an apparent monetary tightening) is followed by a rise rather than a fall in the price level. The paper asks (1) whether the puzzle’s severity has changed over time – specifically, whether it is stronger before 1980 than after 1982 – and (2) which of two candidate explanations, the Sims (1992) inflation-expectations story or a supply-shock story, better accounts for it.

The core VAR uses quarterly U.S. data on log real GDP (Y), the log GDP deflator (P), and the federal funds rate (f), estimated over 1960:1-1993:4 and over two subsamples, 1960:1-1979:3 and 1982:4-1993:4. The 1979:4-1982:3 non-borrowed-reserves period is dropped from the subsample analysis because the Fed was not targeting the funds rate during that operating-procedure regime. Additional specifications extend the base VAR with commodity prices (PSC) and/or the term spread (SP, the 10-year Treasury bond rate minus the 3-month Treasury bill rate). Identification is recursive (Cholesky): the base ordering is Y, P, f, with the funds rate placed last; when commodity prices are added the ordering is Y, P, PSC, f, and with both additional variables it is Y, P, PSC, SP, f – so PSC and SP are both ordered as more “contemporaneously exogenous” than the funds rate. All confidence bands reported are one-standard-error bands, narrower than the two-standard-error convention typical in academic papers, so the stated statistical precision should be read with that in mind. As a complementary, non-VAR check, the authors also run distributed-lag inflation regressions using a Romer and Romer (1989) narrative dummy for monetary contraction dates (1968:4, 1974:2, 1978:3, 1979:4, and 1988:3).

Yes – the price puzzle is present in the full 1960:1-1993:4 sample and is clearly present in the 1960:1-1979:3 subsample, where prices rise substantially for several years after a positive funds-rate innovation, but in the 1982:4-1993:4 subsample “a federal funds rate innovation does not cause prices to systematically rise; the effect on prices, though negative, is small and not statistically different from zero.” This pre/post-1980 contrast is robust to using the Romer-Romer narrative approach in place of the VAR shock: prices rise following Romer-Romer contraction dates in the full sample and in 1982-93, but fall six quarters after a contraction in the 1960-79 subsample.

Only partially by subsample: adding commodity prices eliminates the puzzle for the full sample and for the 1982:4-1993:4 subsample, but not for the 1960:1-1979:3 subsample, where prices still remain above their pre-shock level for nearly three years even with commodity prices included. This full-sample result replicates the finding of Christiano, Eichenbaum, and Evans (forthcoming at the time of writing, later published as their 1999 Handbook chapter), but the paper’s subsample split shows that resolution is specific to the post-1982 period rather than a general feature of the commodity-price fix.

Of a wide set of additional variables tried – oil prices, stock prices, unit labor costs, an index of leading indicators, capacity utilization, and individual short- and long-term interest rates – only the term spread (10-year minus 3-month Treasury rate) systematically helps resolve the price puzzle. Including the spread alone eliminates the puzzle in both the 1960-79 and 1982-93 subsamples, but not in the full 1960:1-1993:4 sample; the authors suggest this full-sample failure is “possibly” attributable to the extreme interest-rate volatility of the 1979-82 period rather than to a general shortcoming of the spread as a fix. When both commodity prices and the spread are included together, the puzzle is eliminated across all three samples (full, pre-1980, and post-1982).

When the spread rises, the federal funds rate falls rather than rises – a pattern the authors describe as “consistent with the explanation that the Fed partially extinguishes the inflationary effects of a temporary supply shock,” and inconsistent with the Sims (1992) inflation-expectations story. Under the inflation-expectations explanation, a rising spread should signal higher expected future inflation and should induce the Fed to tighten (raise the funds rate); the fact that the funds rate instead falls is the paper’s central piece of evidence against that explanation.

The authors conclude the supply-shock explanation is “consistent with those shocks reflecting supply-side shocks in the full and the 1960-79 samples,” based on the joint behavior of output, prices, and the funds rate following commodity-price and spread shocks. Under this story, a negative supply shock raises short-term real rates relative to long rates (so the spread falls) while raising commodity prices and, in general, the price level and lowering output; the Fed responds by raising the funds rate but does not fully offset the shock’s inflationary effect, mechanically generating a puzzle-like pattern in which both commodity prices and the funds rate rise before inflation does. Consistent with this, a positive spread shock is associated with output rising and prices falling (Fig. 9), and the funds rate’s negative response to spread shocks aligns with the Fed tightening in reaction to supply shocks (which lower the spread) rather than reacting to inflation expectations (which would require the opposite spread-funds-rate relationship). The muted post-1982 puzzle is attributed by the authors to either the Fed placing more weight on price stability after Volcker or to fewer severe supply shocks hitting the U.S. economy during the 1980s – offered as alternative, not mutually exclusive, explanations.

The main pre/post-1980 comparison is supported by formal structural stability tests that reject parameter equality across the two subsamples, and by triangulating VAR results against the independent Romer-Romer narrative approach. As a further check, when the Cholesky ordering is changed to place the funds rate before the spread (rather than after it), the price puzzle remains present in the 1960-79 sample – a result the authors read as consistent with the supply-shock interpretation. The authors also note that the large number of parameters in the seven-variable VAR relative to the number of observations in the post-1982 sample is itself a limitation affecting the joint Granger-causality tests in that specification.

The paper is published in a Federal Reserve Bank policy review, not a peer-reviewed journal, and reports one-standard-error rather than two-standard-error confidence bands throughout, so its stated statistical precision is narrower than the academic convention and should be treated cautiously. No lag-length selection procedure is reported for the VARs. The authors acknowledge that the post-1982 seven-variable specification has a large parameter count relative to its sample size, which limits confidence in the joint Granger-causality tests for that sample. The term spread’s failure to resolve the puzzle in the full sample may be a sample-specific artifact of the 1979-82 interest-rate volatility rather than a general limitation of the spread-based fix. No variance decompositions are reported; all results are impulse-response based.