as the evidence does not show a decline in the capital-labor ratio, although good annual data on this score are not available for this period.
Was the decline the result of “Keynesian” productivity movements? Specifically, did a decline in autonomous spending lead to falling output, while the number of workers did not fall as much? To examine that possibility with respect to the most important component of aggregate demand, consumption spending, we estimated a simple Keynesian consumption function for the period 1901 to 1928. The statistical fit was very good, with the coefficient of multiple determination (R2) approaching .98.
Deviations of actual consumption spending from what the consumption function relationship predicts are a measure of shifts in the propensity to consume from the long-run trend. Examination of those deviations (“residuals” to econometricians) shows that consumption in 1914 rose substantially above the predicted amount; there was an increase, not a decrease, in autonomous consumption. Indeed, the “overconsumption” (relative to the long-run trend) in 1914 was the greatest for any year in the entire 1901–1928 period. Consumption remained above trend (although less so) in 1915. Accordingly, the productivity drop cannot be attributed to underconsumption. As mentioned in chapter 3, based on longer-term evidence, there is no basis to believe that shifts in aggregate demand are systematically related to changes in labor productivity.
Much of the productivity decline may well represent what we previously termed “Schumpeterian” productivity change—exogenous, random occurrences that are reflected in rises and falls in the rate of innovative activity. This is also consistent with much of the new classical literature on the real business cycle. Of course, 1914 was the beginning of World War I, and some structural shifts in output probably were beginning to occur, shifts that may have rendered some of the capital stock less useful and contributed to the productivity decline.
Policymakers did relatively little to respond to the mounting unemployment in 1914–15; whatever progressivism Woodrow Wilson possessed did not translate into economic activism to deal with the problem. In his major message in December 1914, he did not even mention unemployment.13 Some talk was given to making it easier for the unemployed to go into farming; the Labor Department held a conference to promote intergovernmental cooperation to deal with the issue.14 But the problem seemed to solve itself before governmental efforts came to anything.
To be sure, there were some relatively isolated cries for more forceful governmental intervention. In a remarkable editorial that espoused a Keynesian approach some two decades before Keynes himself did, the New Republic proposed “to enrich the future through the transmuting of waste labor into permanent improvements and valuable stocks. … The only real obstacle to effective action … is a short-sighted reluctance on the part of the government to increase the national debt.…”15 This call for what later was termed countercyclical fiscal policy was ignored.
Following 1915, unemployment fell back to near-normal or equilibrium levels in 1916 and 1917, and then declined to 1.4 percent, well below the equilibrium level, in 1918 and 1919. Our analysis of that decline is somewhat handicapped by sharp movements in wages and prices, making the measurement of changes in real wages a somewhat hazardous process. Any inadequacies in the consumer price index were probably magnified during this period of intense inflation. One thing is clear: labor productivity rose sharply in the 1918–19 period (8 percent in 1918, 6.7 percent in 1919), and hence was the leading force in the fall in the adjusted real wage.
Despite the data problems, the model still does reasonably well in accounting for the unemployment decline. It indicates 1.9 percent unemployment for 1918, well below the normal rate although somewhat above the actual rate of 1.4 percent. As to 1919, the model dramatically underpredicts (a negative rate versus the actual rate of 1.4 percent), but captures the essential low-unemployment situation prevailing during that era.
The 1920–1922 Depression
THE EMPIRICAL EVIDENCE
By far the most important business cycle development of the first three decades of the twentieth century was the very sharp economic downturn of 1920 and 1921. Unemployment rose to the double-digit level in 1921. Since the annual rate of unemployment reached 11.7 percent, some months within that year witnessed even higher unemployment—possibly as much as 15 percent.
While the magnitude of the 1920–22 downturn was severe (and indeed exceeded that for the Great Depression of the following decade for several quarters), its duration was not. By 1922 recovery was already underway, and in the following year unemployment was actually less than its normal long-run rate. Despite movements in major components in the demand for labor, including sharp increases and then decreases in prices and money wages, the labor market adjusted reasonably quickly to disequilibrium conditions, in marked contrast to the 1929 downturn discussed in the next chapter.
The standard model with multiple lagged variables using annual data correctly predicts the rise and subsequent decline in unemployment. The model understates unemployment in 1920 rather considerably, but at least indicates a fairly sizable increase for the year (the actual unemployment rate rose 3.8 points.) Yet the model predicts very accurately the dramatic upsurge in unemployment in 1921 (calling for 12.1 percent instead of the actual rate of 11.7 percent), and the subsequent drop in 1922 and 1923 (the model somewhat underpredicts the magnitude of the decline, showing, for example, 8.6 percent unemployment for 1922 instead of the actual 6.7 percent).
Analysis of the 1920–22 downturn requires use of more detailed data than that employed in the basic model. While unemployment data are not available on a monthly or quarterly basis, factory employment data are, as are factory payroll, industrial production and wholesale price data. From these sources it is possible to discern changes in productivity in manufacturing by dividing industrial production by factory employment. This provides a measure of output per worker. To the extent that employers shortened the workweek in response to declining output, this productivity measure may misstate changing hourly output per worker. Fortunately, however, the evidence is that hourly output per worker fell only modestly in manufacturing. Data collected by the Bureau of the Census, and analyzed by pioneering statistician Willford I. King, reveal that the workweek, from peak to trough, declined slightly less than 4 percent.16 Accordingly, any bias introduced by the use of this productivity measure is relatively small.
Similarly, wages per worker are derived by dividing factory payrolls by the number of employees. Again, the relatively modest nature of changes in the workweek over time suggests that this very crude measure of employee compensation in fact is not too bad in this particular instance. For purposes of our analysis here, we averaged monthly figures to obtain quarterly estimates of the relevant variables.17
Table 4.3 reveals the trends in the major variables. Factory employment from the beginning of 1920 to the trough in the third quarter of 1921 fell slightly more than 30 percent on a seasonally adjusted basis, a sharp drop by any standard.18 Similarly, industrial production fell by a like proportion. An even steeper decline occurred with respect to wholesale prices. Between the second quarter of 1920, when they peaked, to the third quarter of 1921, a period of slightly over one year, wholesale prices fell nearly 44 percent, one of the steepest decreases recorded in American history.
The substantial fall in prices greatly exceeded the drop in money wages, so real wages rose markedly until the third quarter of 1921. It would be an overstatement, however, to characterize money wages as rigid. After all, they did fall over 19 percent from the summer of 1920 to the end of 1921. It is more accurate to say that wages proved less flexible than prices.
In no sense can the business cycle downturn of 1920–21 be attributed to a productivity decline. Only in two quarters did average
