International Evidence of the Historical Properties of Business Cycles

Questions & answers

Q1. What is the paper’s data set, and why do the authors consider it more reliable, in places, than the standard prewar U.S. series?

The paper uses annual national-accounts data for ten countries with at least a century of coverage (Australia, Canada, Denmark, Germany, Italy, Japan, Norway, Sweden, the U.K., and the U.S.), filtered with the Hodrick-Prescott method, and the authors argue that “estimates for several of the countries we study are likely to be better than the American data examined by Romer (1989) and Balke and Gordon (1989)” because, for instance, the United Kingdom’s continuous income tax since 1842 and Australia’s extensive official statistics give those countries richer primary source material than the United States had before 1913, where Kuznets’s commodity-output-based estimation method is the subject of an active measurement-error debate the paper explicitly declines to resolve, instead bringing in nine additional countries’ evidence.

Q2. What regularities does the paper find in the comovement of real expenditure components with output, and how uniform are they across countries and periods?

Investment is “between two and five times more variable than output for every country and period” and strongly procyclical (the single exception being prewar Japan, where government accounted for roughly half of gross investment and its cyclical pattern was governed by factors such as natural disasters and military buildups rather than the business cycle); consumption is “uniformly procyclical” with variability “about the same as output’s” (cross-correlations with output ranging from 0.41 in prewar U.K. to 0.91 in prewar U.S.); net exports (measured as the ratio to output) are “for the most part, countercyclical,” negative in two-thirds of the cases studied; and government spending “exhibits little regularity,” with its relative variability ranging enormously (from 0.81 in interwar Sweden to 9.49 in prewar Japan) and its correlation with output roughly evenly split between positive and negative across the 18 country-period cases examined.

Q3. What does the paper find about output volatility across the prewar, interwar, and postwar periods, and how does it engage with the Romer/Balke-Gordon measurement-error debate?

Interwar output fluctuations are unambiguously larger than postwar fluctuations in nine of the ten countries (two to four times larger), a comparison the paper treats as settled because interwar national accounts are generally more reliable than prewar ones; the prewar-versus-postwar comparison is far less clear-cut, with “no consistent pattern” – six of the ten countries (including most of the European countries in Sheffrin’s 1988 study) show prewar volatility no more than 60% larger than postwar, while Australia, Canada, Sweden, and the United States show considerably larger prewar fluctuations, with prewar/postwar standard-deviation ratios across the full sample ranging from 0.8 to 3.3. For the U.S. specifically, the paper reproduces the Romer (1989) and Balke-Gordon (1989) debate over whether Kuznets’s reliance on commodity-output data overstates prewar GNP volatility, finding a prewar/postwar ratio of 1.9 with the traditional Kuznets-derived series, 1.29 using Romer’s revised series, and 1.77 using Balke and Gordon’s; for Sweden the paper similarly finds that roughly half of the excess prewar volatility reported by Sheffrin (1988) “disappears when revised estimates of prewar output are used” (Krantz and Nilsson 1975), while for countries with especially strong prewar source material – Australia and Canada – the authors conclude the greater measured volatility “probably indicates a change in the variability of real economic activity” rather than pure measurement error.

Q4. What is the paper’s central finding about how price-level cyclicality and inflation persistence have changed, and how widespread is this change across countries?

“Before World War II prices were predominantly procyclical; since then they have been consistently countercyclical” – the postwar correlation between Hodrick-Prescott-filtered price-level and output fluctuations averages -0.36 across the ten countries (strengthening from -0.27 in the earlier postwar half to -0.45 later), versus a predominantly positive correlation in both the prewar and interwar periods (the sole notable exception being prewar Japan, at -0.45, attributed to extreme early price instability as Japan established a national banking system and convertible currency). Separately, “in every country but Norway, the autocorrelation of inflation is higher in the postwar period than it was prior to the first world war, and in all but Japan and Sweden the increase is greater than 0.25” – a sharp rise in inflation persistence the paper shows is not an artifact of price measurement error, since applying the more accurate Balke-Gordon (1989) U.S. price series only modestly raises the prewar autocorrelation (from 0.17 to 0.31), “too small to account for the difference in persistence between the prewar and postwar periods.”

Q5. How does the behavior of monetary aggregates differ from that of prices across these periods?

Unlike prices, broad money-stock growth rates (typically M2 or M3, from Bordo and Jonung 1987) show “no obvious temporal pattern” in persistence – the autocorrelation of money growth fell by at least 0.25 in three countries, rose by at least 0.25 in three others, and changed little in the rest – while the correlation between money and output fluctuations shows only “a modest tendency toward lower correlation… in the postwar period,” with the mean cross-country correlation declining from 0.33 (prewar) and 0.31 (interwar) to 0.20 (postwar). The paper reads this contrast – persistent, structural change in price behavior versus comparatively stable money behavior – as itself informative: whatever explains the shift in prices evidently does not operate primarily through a parallel shift in the time-series properties of money.

Q6. What two competing explanations does the paper weigh for the postwar shift in price-level behavior, without choosing between them?

One hypothesis, which the paper attributes to Gordon (1980, 1983), explains the changing cyclical behavior of prices within an aggregate supply-and-demand framework as resulting from “a combination of increased persistence in inflation and greater variability of ‘supply shocks’ in the postwar period” – a change in economic structure (the propagation mechanism). The competing hypothesis, from Cagan (1986) and echoed by Bordo (1990), attributes increased inflation persistence to a postwar change in monetary-policy regime – “a general expectation that government policies will oppose severe price movements” showing up as greater autocorrelation, with the countercyclical-to-procyclical price shift reflecting that “the monetary authority did not accommodate price movements in the later period but did so in the earlier period,” consistent with the move away from the gold standard – a reading the paper notes is reinforced by prewar Japan’s exceptional price behavior, since Japan was “one of the few countries in our sample not on the gold standard during most of the prewar period.” The paper states plainly that it “leave[s] the resolution of this issue to others,” treating its own contribution as establishing that the price-behavior shift is a broad international phenomenon requiring an explanation that “cuts across national boundaries,” not as adjudicating between these accounts.

Q7. What scope conditions does the paper place on its own findings, particularly regarding data quality?

The authors repeatedly grade their underlying national-accounts series by source quality: the U.K. (Feinstein’s three independently constructed, then combined, “compromise” estimate), Australia (Butlin’s production-based data, which Butlin himself called possibly unmatched “in range and accuracy… over such a long period”), and Canada are treated as relatively strong; Sweden as good but “not as good as… the United Kingdom”; Norway and Denmark as weaker, with Danish data in the prewar period partly built from “smooth trends between dates,” which the authors note “is likely to understate short-term variability”; and Germany and Italy (sourced from Mitchell 1976 rather than country-specific compilations) as the least reliable, with German prewar data additionally compromised by shifting political boundaries and by basing estimates on incomplete income-tax records from only two of twenty-six states (Prussia and Saxony). The paper is explicit that this heterogeneity in source reliability is a first-order consideration in interpreting any prewar-versus-postwar comparison, and devotes its entire Section I and Appendix A to documenting it country by country so that “one could, at least in principle, replicate the entire study.”