The Cost Channel of Monetary Transmission

The paper is motivated by two long-standing empirical puzzles about monetary transmission – the “amplification puzzle” (small, transitory interest-rate movements appear to have large, persistent output effects) and the “price puzzle” (prices tend to rise, not fall, in the short run after a contractionary monetary shock in standard VARs) – and proposes a “cost channel” as a common explanation for both. The mechanism: firms must borrow to finance working capital (labor costs, inventories, and other inputs purchased before output is sold), so a Federal Reserve rate increase raises firms’ financing costs directly, shifting up the marginal cost of production. This pushes output down and prices up simultaneously, mimicking a contractionary supply (technology) shock rather than a contractionary demand shock. The authors are explicit that they are not denying demand-side effects or long-run monetary neutrality; they argue the cost channel is a short-run supply-side channel that operates alongside, and can amplify, the conventional demand channel.

Using a six-variable quarterly VAR (private-sector labor productivity, hours, real wages, the private-business price level, M2, and the federal funds rate; 1959:Q1-2000:Q3, 4 lags, estimated by GMM/Shapiro-Watson IV with 500 bootstrap draws), the authors identify a technology shock as the only shock permitted a permanent effect on productivity (following Gali 1999) and a monetary policy shock as the funds-rate innovation orthogonal to it. The resulting impulse responses show that a contractionary monetary shock produces falling productivity, falling hours, procyclically falling real wages, and a temporary price-level increase – a pattern that “really looks more like a technology shock than a demand shock,” and is inconsistent with a model in which real wages should rise counter-cyclically following a demand contraction.

As a second line of evidence, the authors compare the effects of Romer monetary-policy dates against Ramey-Shapiro military-buildup dates (reversed in sign for comparability) and find opposite real-wage responses: defense-driven output declines are accompanied by rising real wages, while monetary-driven output declines are accompanied by falling real wages. This is exactly what a standard neoclassical production function predicts for a demand shock (output and productivity move together, pulling wages with them) but is inconsistent with the countercyclical-markup story, under which demand contractions should also raise wages relative to the (falling) markup. The authors also document, in a supporting exercise on the aircraft-and-parts industry (SIC 372, 1977-1995), that defense spending is strongly procyclical with industry output (correlation 0.44) but strongly countercyclical with the real product wage (correlation -0.75), reinforcing that demand-driven episodes look nothing like the monetary-shock pattern.

Working capital is large relative to sales: averaged over 1959-2000, gross working capital (inventories plus trade receivables) equals about 17 months of final sales, and net working capital (netting out trade payables) equals about 11 months – comparable in magnitude to firms’ investment in fixed capital. Using a standard time-lag-in-production labor-demand equation (in which the marginal cost of labor is the real wage times the gross nominal interest rate), the authors show that with a capital cost share of 0.3, a 100-basis-point rise in the nominal rate lowers labor demand by about 3%, holding real wages fixed; because the average funds-rate increase across Romer tightening episodes is almost 400 basis points, the pure direct effect could be substantial. Layering on a credit-channel-style jump in external financing spreads (an illustrative 80-basis-point widening on top of the funds-rate move) raises the implied decline in labor demand to about 15% for a 400-basis-point tightening. The authors are careful to flag that this partial-equilibrium calculation holds real wages fixed and is meant to illustrate plausible magnitude, not to be read as a full general-equilibrium estimate.

The industry framework is a separate, complementary test: a seven-equation monthly VAR (1959-2000, 7 lags) with a five-variable macro block (industrial production, the PCE deflator, an index of sensitive commodity prices, M2, and the federal funds rate) and a two-variable industry block (industry output and the industry price-to-wage ratio, P/W) for each of 21 two- and three-digit manufacturing industries, with macro-block coefficients constrained to zero in the industry equations (a “pseudo-panel” block-restriction approach following Davis and Haltiwanger 1997). Where the aggregate VAR in Q2 asks whether monetary shocks look like technology shocks in the aggregate labor market, this framework asks a more direct cost-channel question at the industry level: after a contractionary funds-rate shock, does an industry’s price rise relative to its wage (P/W up) even as its output falls – the signature of a supply-side cost shock – or does output and P/W fall together, the signature of a demand shock? The two frameworks are thus testing the same hypothesis (a monetary contraction shifts supply, not just demand) from two different angles: aggregate labor-market comovements versus industry-level relative-price behavior.

Over the full 1959-2000 sample, output falls and P/W rises for 10 of the 21 industries (and for manufacturing as a whole), with 6 industries significant at the 10% level (textiles, apparel, industrial machinery, electrical machinery, transportation equipment, and motor vehicles); splitting the sample sharpens this considerably – 15 of 21 industries show significant P/W increases in the pre-Volcker period (February 1959-September 1979), versus only 3 of 21 in the Volcker-Greenspan period (January 1983-March 2000), even though 16 of 21 industries still show some rise in P/W in the later period. Only lumber and leather (and, to a lesser extent, hides and skins) show a dominant demand-channel pattern in the pre-Volcker subsample. The authors note that finding a supply-channel signature for a given industry does not rule out demand effects there too – the test only detects cases where the supply channel’s price signature clearly dominates.

Standardizing the shock to 25 basis points so magnitudes are comparable across subsamples, the authors find the industrial-production trough is almost four times as deep in the pre-Volcker period as in the Volcker-Greenspan period, with output still below its pre-shock level four years out (versus a rebound within about two years later); the price puzzle is “fully operational” pre-Volcker, with prices rising for over two years before falling (significant at 10% for more than three years), while post-1983 prices are mostly unresponsive after a brief positive spike in the first five months. The funds rate itself also normalizes much faster after 1983 (about nine months versus almost two years pre-Volcker). The authors attribute the weakening to institutional change: financial deregulation and innovation from the late 1970s onward gave firms more alternative funding sources; the Federal Reserve’s earlier use of direct “credit actions” restricting bank lending (per Romer and Romer 1993) had made pre-1979 credit crunches more severe; and the shift from fixed to floating exchange rates after the 1970s let currency appreciation offset some of the direct cost-side impact of a contraction. They also show the price-puzzle finding is robust to including commodity prices and oil-shock dummies in the Fed’s reaction function, and survives (albeit reduced) even under an economically implausible specification designed specifically to eliminate it.

Yes: using Quarterly Financial Reports balance-sheet data for 14 two-digit manufacturing industries, the authors construct an interest-expense-to-net-sales measure and find it correlates positively with each industry’s price-to-wage response – correlation of about 0.53 (peak response) and 0.52 (integral of the response) in an early-1970s cross-section (1974:Q1), falling somewhat but remaining strongly positive at about 0.43 and 0.40 in a 1990:Q4 cross-section. Industries carrying heavier interest burdens relative to sales are the ones whose relative prices respond most to a monetary tightening, which the authors read as suggestive (not structural) evidence that financing costs specifically, rather than demand conditions in general, drive the cross-industry pattern – and that the cost channel, while weaker after 1983, has not disappeared.

The authors test and argue against two alternatives – sticky prices with flexible wages (rejected because nominal prices themselves rise in virtually all pre-Volcker industries, while nominal wages are flat or falling, so “results are being driven primarily by rising nominal prices, not by falling nominal wages”; also, Christiano-Eichenbaum-Evans (1997) show profits fall after a contraction, contrary to the sticky-price prediction of rising profits) and a misspecified Federal Reserve reaction function (rejected because adding Romer-Romer Greenbook forecasts of output and inflation to the funds-rate equation does not eliminate the post-shock price rise) – but they explicitly concede that they cannot distinguish their cost-channel story from a countercyclical-markup story. Both mechanisms operate through liquidity constraints and are, in the authors’ words, “really just variations on a similar theme”: without direct data on marginal costs, a rise in price relative to wages could reflect either rising financing-driven marginal costs (cost channel) or a liquidity-constrained firm raising its markup (countercyclical markup). This observational-equivalence problem is left unresolved. The authors also note their industry framework’s block-restriction (pseudo-panel) approach assumes away cross-industry spillovers by construction, and that all reported significance levels use one-tailed bootstrap-based tests, which is a less conservative convention than a two-tailed test.