Business Cycles: Real Facts and a Monetary Myth

The paper argues that the reporting of business cycle facts — without first assuming the data are generated by some probability model — is “an important scientific activity,” and that economics should be no exception to this practice found elsewhere in science (Introduction, p. 3-4). Kydland and Prescott agree with the first of Koopmans’s (1947) two criticisms of Burns and Mitchell’s Measuring Business Cycles — that theory is needed to decide which variables to report — but “strongly disagree” with his second criticism, that a study must specify a probability model and estimate/test it to be scientific; they say Koopmans “did economics a grave disservice” on this point. As a positive precedent, they cite the growth facts documented by Kuznets and formalized by Solow (1970) — constant growth rates of output per worker, a roughly constant capital/output ratio, a trendless profit rate — which were reported atheoretically yet “instrumental” in the development of Solow’s neoclassical growth model, for which both Kuznets and Solow received Nobel Prizes (p. 3-4).

Following Lucas (1977), the authors define business cycles as “deviations of aggregate real output from trend” — specifically, the comovements of the cyclical deviations of various aggregates with those of real GNP — rather than as Mitchell’s (1913, 1927) sequences of four distinct phases (prosperity, crisis, depression, revival) with emphasis on turning points (Sec. “Alternative Views of Business Cycles,” p. 4-6). They note Lucas’s definition is “incomplete” because he does not specify how to construct the trend, so their contribution is to “complete his definition by providing an explicit procedure for calculating a time series trend” (p. 4). They argue the field has moved in a “completely different direction” from Mitchell’s vision because deterministic cyclical laws of motion — the kind Mitchell’s four-phase view implicitly required — “have failed” to arise as equilibrium behavior for economies with empirically reasonable preferences and technologies, whereas random-shock mechanisms (Frisch 1933; Slutzky 1937; Adelman and Adelman 1959) can generate recurrent, Mitchell-like cyclical patterns from a stable, low-order stochastic difference equation (p. 5-6).

The trend component r_t is defined as the series that minimizes the sum of squared deviations from the data plus a penalty term, λ times the sum of squared second differences of the trend, with λ = 1600 for quarterly data (Sec. “Business Cycle Deviations Redefined,” p. 8-9, eq. 3). The authors state four criteria guiding this choice: the trend should approximate the curve “students of business cycles and growth would draw through a time plot”; the trend of a series should be a linear transformation of that series, with the same transformation applied to all series; lengthening the sample should not significantly change deviations at a given date (except near the sample end); and the scheme should be “well defined, judgment free, and cheaply reproducible” (p. 8). They note this smoothing method “has a long history” in actuarial science (1920s) and ballistics (von Neumann, early 1940s) and was earlier introduced to economics in an unpublished paper by Hodrick and Prescott (1980) (p. 9, fn. 5).

The authors state plainly that “whether measured by the implicit GNP deflator or by the consumer price index, the U.S. price level clearly has been countercyclical in the post-Korean War period” (Sec. “Nominal Facts,” p. 17), contradicting the widely held belief — which they call a myth — that the price level is always procyclical. They trace this myth to the interwar period, when the price level genuinely was procyclical, after which “no one bothered to ascertain” whether this held after World War II because of the “Koopmans taboo” against atheoretical fact-reporting; instead, business cycle theorists kept building equilibrium models with monetary policy or price surprises as the main source of fluctuations, a modeling choice the authors say was “largely” driven by this false perception (p. 7). They explicitly name Lucas (1977, p. 9), Bernanke (1986, p. 76), and Mankiw (1989, p. 88) as having stated or relied on procyclical prices as a business cycle regularity (p. 7). Their concluding caution is correspondingly strong: “any theory in which procyclical prices figure crucially in accounting for postwar business cycle fluctuations is doomed to failure” (Concluding Remarks, p. 17).

The authors find “no evidence that either the monetary base or M1 leads the cycle, although some economists still believe this monetary myth”; both series are “generally procyclical and, if anything, the monetary base lags the cycle slightly” (Sec. “Monetary Aggregates,” p. 16). They note one exception: during the long, steady 1980s expansion, M1 was unusually volatile and M2 “showed some evidence that it leads the cycle by a couple quarters” (p. 16-17). Because of this divergence between M1 and M2, they examine the difference M2 − M1 (mainly interest-bearing time deposits, about half of annual GNP versus M1’s one-sixth) and report that this component “leads the cycle by even more than M2, with the lead being about three quarters” (p. 17). They suggest in the concluding remarks that this pattern — M1 moving contemporaneously with the cycle while the much larger non-transaction component leads it — “suggests that credit arrangements could play a significant role in future business cycle theory,” which they call “an important open problem in economics” (p. 17).

The authors report that the human-capital-weighted real wage “behaves in a reasonably strong procyclical manner,” which they say is “counter to a widely held belief in the literature,” and they explicitly cite “the article by Lawrence Summers (1986, p. 25), which states that there is ’no apparent procyclicality of real wages’” as an example of that belief (Sec. “Production Inputs,” p. 13). Their argument runs through aggregate hours: because aggregate hours gives a brain surgeon’s hour the same weight as an orderly’s, and because less-skilled workers’ hours are “much more variable” (citing their own Kydland and Prescott 1989 study of nearly 5,000 workers, 1969-82), weighting hours by relative human capital lowers the measured cyclical variation of the labor input relative to raw aggregate hours — which correspondingly raises the implied procyclicality of the real wage (the ratio of total labor compensation to this smaller-varying input) above what is shown for average hourly compensation in their own Table 1 (p. 12-13). This is presented as a direct rebuttal of the “no apparent procyclicality” claim being debated in the adjoining Prescott-Summers exchange, though the present paper’s own scope is limited to this comovement finding and does not engage with Summers’s broader methodological objections to real business cycle theory.

Total hours worked (by either the household or establishment survey) is “strongly procyclical” with “cyclical variation which, in percentage terms, is almost as large as that of real GNP,” while “the capital stock, in contrast, varies smoothly over the cycle and is essentially uncorrelated with contemporaneous real GNP,” with a large correlation only once the capital stock is shifted back by about a year, meaning “business capital lags the cycle by at least a year” (Sec. “Real Facts — Production Inputs,” p. 11). Decomposing hours, they report employment lags the cycle while hours per worker is “nearly contemporaneous with it, with only a slight lead,” and that because the correlation between the two subseries is 0.86 rather than 1, “employment accounts for roughly two-thirds of the standard deviation in total hours while hours per worker accounts for about one-third” (p. 11-12). The inventory stock also lags the cycle, by about half a year, and is “nearly as volatile as quarterly real GNP” (p. 11).

“The bulk of the volatility in aggregate output is due to investment expenditures,” with consumption of nondurables and services — “by far the largest component (nearly two-thirds) of total output” — having “relatively little volatility” by comparison (Sec. “Real Facts — Output,” p. 13-14); their Table 2 reports a volatility (percent standard deviation) of 8.30 for investment expenditures against 1.71 for real GNP and 0.84 for nondurables-and-services consumption. Household and business investment in durables have similar amplitudes of fluctuation; consumer durables expenditure leads the cycle slightly while nonresidential fixed investment, especially structures, lags it (p. 14). Government purchases “has no consistent pro- or countercyclical pattern,” imports is procyclical with no phase shift, and exports is procyclical but “lags the cycle by from six months to a year” (p. 14). Measured as a share of GNP, the change in business inventories is procyclical and net exports is countercyclical, most strongly when exports are shifted back about a year (p. 14).

Labor income and capital income are both “strongly procyclical” in levels, and capital income is “highly volatile,” but measured as shares of national income the pattern flips: “labor income is countercyclical while capital income is procyclical” (Sec. “Factor Incomes,” p. 12-13). Proprietors’ income as a share of national income is “uncorrelated with the cycle” (p. 13). In the Concluding Remarks, the authors reiterate that, unlike Burns and Mitchell, they are “not claiming to measure business cycles,” and that the atheoretically-selected statistics they report are of interest precisely because they are (or were) “in apparent conflict” with established neoclassical growth theory; documenting such apparent deviations from theory “is a legitimate activity in the natural sciences and should be so in economics as well” (p. 17). They close by flagging the integration of money and credit into growth theory, in a way consistent with both the monetary and real facts reported, as “an important open problem in economics” (p. 17).