Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms

The paper asks whether small manufacturing firms account for a disproportionately large share of the aggregate manufacturing decline that follows monetary tightening, and whether that disproportion is driven by financial factors — a balance-sheet channel and a bank-lending (credit) channel — rather than by nonfinancial differences across size classes, such as industry composition or the flexibility of production technology. This distinction matters because a standard, frictionless view of monetary transmission (working purely through the cost of capital via a representative firm) does not predict that small and large firms should respond differently to the same policy-induced change in interest rates; a size-dependent, asymmetric response is instead the signature that capital-market imperfections — and specifically constraints that bind more tightly on small, less creditworthy borrowers — help propagate monetary policy into real activity. The paper is offered as “a first cut at sorting financial from nonfinancial factors” in explaining why small firms appear to bear a larger burden of monetary-policy-induced downturns.

The authors use quarterly, size-disaggregated aggregate time series for all U.S. manufacturing firms from the Census Bureau’s Quarterly Financial Report for Manufacturing Corporations (QFR), covering the main regression sample 1960:1-1991:4 (QFR data are available from 1958:4), broken into eight nominal asset size classes. The key variables are sales, book-value inventories, and short-term debt (maturity of one year or less, decomposed into bank loans, commercial paper, and other short-term debt), plus a balance-sheet proxy — the “coverage ratio,” cash flow divided by the short-term interest rate times short-term debt. Because nominal size-class cutoffs drift in real terms over time (small firms’ share of manufacturing sales measured this way falls mechanically from 52 percent in 1960 to 26 percent in 1990), the authors instead define “small” as a weighted average of the two cumulative asset classes whose combined sales share averages 30 percent of total manufacturing sales at each date, which corrects for this nominal drift.

The analysis proceeds in three stages: (1) descriptive plots of smoothed growth rates and log deviations of size-class variables around Romer-Romer narrative monetary-tightening dates; (2) reduced-form VARs — both a bivariate specification (each size-class variable on its own lags and 12 lags of a Romer dummy) and a multivariate specification that adds four lags each of real GNP growth, inflation, and the federal funds rate — estimated by seemingly unrelated regressions (SUR) across size classes so that standard errors on small-versus-large differences can be computed; and (3) structural GMM estimation of an inventory-adjustment (Euler-type) equation for each size class that includes the coverage ratio and the prime rate as balance-sheet and cost-of-capital controls. The reduced-form stage relies primarily on the Romer and Romer (1988, 1992) narrative dates (plus the 1966 credit crunch) as the monetary policy indicator, with the federal funds rate used as a robustness check; formal tests include an F-test for joint significance of the Romer dummy, a t-test on the sum of its coefficients, and a t-test on the small-minus-large difference.

Small firms’ sales, inventories, and short-term debt decline sharply and consistently after Romer dates, while large firms’ variables do not follow the same pattern — indeed large firms tend to accumulate inventories and initially increase short-term debt, consistent with “borrowing to smooth production” through the downturn. Small-firm short-term debt begins declining even before the associated recession and falls steadily thereafter, whereas large-firm short-term debt rises before falling only gradually. The authors note that “small firm sales drop more than large firm sales for a period of ten quarters after a Romer shock… there is no single episode where the reverse happens,” and that the divergence in inventory growth, if anything, is even more pronounced than in sales.

Romer dates are highly significant predictors of small-firm sales, inventories, and short-term debt (exclusion p-values of 0.00-0.04 across specifications) but have no significant marginal predictive power for the corresponding large-firm variables (p-values of 0.12-0.56), and the small-versus-large differences are themselves statistically significant. This holds in both the bivariate specification and the multivariate specification that conditions on real GNP growth, inflation, and the federal funds rate, so the authors describe the pattern as “robust to the inclusion of macroeconomic variables.” Quantitatively, small firms — who are by construction about 30 percent of total manufacturing sales — account for an estimated 55 to 60 percent of the drop in total manufacturing sales four, eight, and twelve quarters after a Romer episode; large firms’ inventories actually rise (by 2.4, 5.1, and 3.1 log points at four, eight, and twelve quarters), while small firms’ inventories fall sharply (by 2.2, 10.8, and 17.6 log points), pulling total manufacturing inventories negative by the twelfth quarter after the episode.

No — the response of small firms to a federal-funds-rate innovation switches significantly between high- and low-GNP-growth states (p=0.03 for sales, p=0.01 for inventories), while large firms show no significant evidence of such state-dependence (p=0.38 and p=0.68). In low-growth states the funds-rate coefficients are jointly significant for both small-firm sales and inventories, and also for large-firm sales (though not large-firm inventories); in high-growth states, small-firm sales and inventories are not significantly affected. Figure VI shows the small-firm inventory-to-sales ratio dropping significantly by the third quarter after a funds-rate shock specifically in low-growth states, with large firms showing no comparable asymmetry — evidence the authors read as consistent with credit constraints binding more broadly across small firms precisely when aggregate conditions, and hence balance sheets, are already weak.

In GMM estimation of a structural inventory-adjustment equation, the lagged coverage ratio (cash flow relative to the short-term interest burden) enters significantly for small firms — with a coefficient of about 1.80 (se 1.00) in the general specification and about 1.40 (se 0.40) once the interest-rate term is dropped, an exclusion restriction rejected at p=0.02 — while for large firms the coverage-ratio coefficient is insignificant and wrong-signed (-0.70, se 0.90), and does not enter significantly in any large-firm specification. The authors read this as more direct, structural evidence for the balance-sheet channel: the interest rate’s effect on small-firm inventory investment appears to operate substantially through its effect on cash flow relative to debt-service burden (the coverage ratio) rather than only through the cost of capital per se, whereas for large firms the coverage ratio adds essentially nothing once other controls are included. Splitting the small-firm sample further, the coverage-ratio coefficient is 1.21 (se 0.43) in high-growth periods versus 2.36 (se 0.64) in low-growth periods — “the coefficient on the coverage ratio doubles as the economy moves from the high to low growth state” — though the low-minus-high difference (1.14, se 0.77) is significant only at about the 20 percent level, so the authors describe this particular asymmetry as suggestive rather than “overwhelming.” Quantitatively, a one-standard-deviation fall in the coverage ratio is estimated to lower small firms’ annual inventory growth rate by roughly 2.25 percentage points, sizable relative to an annual inventory-growth standard deviation of about 8 percent for small firms (versus about 6 percent for large firms).

The authors address the “observational equivalence problem” — that industry composition or differences in production technology could generate the same size-dependent pattern without any financial mechanism — through several checks, though they stop short of claiming to rule out every alternative. They show that the durable-goods share of sales is nearly identical (about 0.52) across all size classes, so industry concentration cannot explain the differential response; they find no direct evidence that large firms subcontract production to small firms during downturns (only anecdotal evidence of the reverse); and they argue that a purely technological story would not naturally predict the cyclical asymmetry documented in Table VI and Figure VI (stronger small-firm sensitivity specifically in low-growth states), whereas a credit-constraint story does. The authors are explicit that these steps address the identification problem “to the maximum degree the data permit” rather than eliminating it entirely, since firm size itself is only a correlate — not a direct determinant — of capital-market access.

The authors flag several limits on interpretation: firm size is a proxy for, not a direct measure of, capital-market access; the QFR data are aggregated within size classes rather than observed at the individual-firm level, so within-class heterogeneity cannot be examined; and the cyclical asymmetry in the coverage-ratio coefficient, while economically large (roughly doubling from high- to low-growth states), is statistically significant only at the 20 percent level, which the authors themselves describe as “not overwhelming” evidence. They also note a subtler identification concern in the structural equations — the coverage ratio’s components (profits, interest rates) could have independent predictive power for future sales, which the authors address by including the coverage ratio in the information set used to forecast sales via two-step instrumental variables. Finally, the authors caution that their manufacturing-sector findings may not generalize without further study, since small firms (fewer than 500 employees) account for only about a third of manufacturing employment but roughly half of total U.S. employment.