1 See Stein, Herbert, The Fiscal Revolution in America (Chicago, 1969), p. 89;Krooss, Herman, Executive Opinion (New York, 1970), pp. 189–90 and 199.
2 Herman Krooss, Executive Opinion, pp. 161–62, 183.
3 Schlesinger, Arthur Jr, The Coming of the New Deal (New York, 1959), p. 569.
4 Allen, Frederick, Since Yesterday (New York, 1940), pp. 166–67.
5 Wolfskill, George and Hudson, John, All But the People (New York, 1969), pp. 144, 146, 157.
6 Leuchtenburg, William, Franklin D. Roosevelt and the New Deal (New York, 1963), pp. 176–77.
7 Walton, Gary and Robertson, Ross, History of the American Economy (New York, 1983), p. 926.
8 DeCanio, Stephen, “Expectations and Business Confidence During the Great Depression,” in Barry Siegel, Money in Crisis (Cambridge, Mass., 1984), pp. 157–76.
9 Schreiber, Harry, Vatter, Harold, and Faulkner, Harold, American Economic History (New York, 1976), p. 367.
10 Brownlee, W. Elliot, The Dynamics of Ascent: A History of the American Economy (New York, 1977), p. 300.
11 Friedman, Milton and Schwartz, Anna J., A Monetary History of the United States (Princeton, 1963), pp. 495–96.
12 Meltzer, Allan, “Comments on ‘Monetarist Interpretations of the Great Depression’” in Brunner, Karl, ed., The Great Depression Revisited (Boston, 1981), p. 151.
13 Moreover, unfilled orders also take into account the possibility that supplying firms, afraid that the political shock will cause the cancellation of previously-placed orders, speed up their shipments.
14 For electrical motors no data on unfilled orders are available; hence data on shipments were used instead. For machine tools only new orders could be used. Since as discussed below, in some regressions one lagged dependent variable and in other regressions three lagged dependent variables were used, the actual period for which the regressions were run starts either one quarter or three quarters later.
15 The data come from U.S. Department of Commerce, Commerce Yearbook, (Washington, D.C., 1932); and Survey of current Business; Supplement (Washington, D.C., 1936, 1938, 1942). There is, of course, a danger of too much aggregation. However given our autogressive approach monthly data are not practicable. To generate lagged variables covering one year, we would need 12 lagged (monthly) variables, and this could produce too much multicollinearity. Aggregation of quarterly data is simple way of reducing severe multicollinearity. Furthermore the use of quarterly data allows for short (3 months) lags in the effect of political shocks should they exist.
16 There is no reason to assume that a regression that explains price expectations in the postwar period also explains price expectations in the 1930s. See Temin's, Peter largely qualitative discussion of price expectations in his Did Monetary Forces Cause the Great Depression? (New York, 1976), pp. 160–64.
17 Andrews, Pauline, “Manufacturing Fixed Investment and the Great Depression” (Ph.D. diss., University of California, Berkeley, 1982).
18 The coefficients of the lagged dependent variables pick up the impacts of those variables that affect investment for more than one period. The impact of those that affect investment in only a single quarter becomes part of the error term.
19 The choice of third-order equations is arbitrary, but the results are similar if first-order equations are used instead.
20 An excellent discussion of the proper use of time-series models is in Anderson, O.D., Time Series Analysis and Forecasting (Borough, Green, Kent, England, 1977). More recently the use of time series methods has been given prominence in macroeconomic modeling. Thus Sims, Christopher in “Macroeconomics and Reality,” Econometrica, 48 (01 1980), pp. 1–49 says “… one can obtain macroeconomic models with useful descriptive characteristics, within which tests of econonucally meaningful hypotheses can be executed, without as much of a burden of maintained hypotheses as is usually imposed in such modeling.”
21 Since there were many shocks during the 1930s there is a danger that the political-shock hypothesis could be force-fitted by experimenting with enough sets of shocks until one set happens to fit. To avoid this the political shocks used were selected without looking at the data and before running any of the regressions. The only regressions that were run are those shown, except for some runs that contained errors.
23 Newsweek, Nov. 17, 1934, p. 7.
24 Crossley, Archibald, “Straw Polls in 1936,” Public Opinion Quarterly, 1 (01. 1937), pp. 24–35.
26 William Leuchtenburg, Franklin D. Roosevelt and the New Deal, p. 273.
27 Because the dummy variables were used only for the quarter in which the election occured, the shocks tested are the immediate shocks.
28 William Leuchtenburg, Franklin D. Roosevelt and the New Deal, p. 151.
29 Major, John, The New Deal (London, 1968), p. 196;Perkins, Dexter, The New Age of Franklin Roosevelt 1932–45 (Chicago, 1957), p. 39.
30 Herbert Stein, 1969, The Fiscal Revolution in America, p. 81; William Leuchtenburg, Franklin D. Roosevelt and the New Deal, p. 152.
31 Newsweek, June 8, 1935, p. 5.
32 William Leuchtenburg, Franklin D. Roosevelt and the New Deal, pp. 242–43.
33 In principle not all six shocks are independent. Suppose a negative shock in, say 1936-IV, reduces investment for that quarter only. In the following quarter, in the absence of new shocks, investment returns to its previous level but the autoregressive method used here predicts that investment will be lower. Suppose further that there is another negative shock in 1937–1, which does affect investment in that quarter. Since investment is underpredicted by the autoregressive terms, the coefficient of the dummy variable for the 1937–1 shock is then biased. Fortunately this is not a problem here since it turns out that none of the shocks—with the possible exception of the last one—had any effect on investment.
34 In a previous test of the political-shock hypothesis Smithies, Arthur, “The American Economy in the Thirties,” American Economic Review, 36 (05 1964), pp. 11–27, looked at the residuals from regressions for plant and equipment investment and from residential construction regressions. He too concluded that these residuals give no support to the shock hypothesis. However, such an approach is model-specific, and the models used by Smithies in 1946 are now outdated.
35 Both significant results are for nonresidential building contracts, and may be due to a special situation in the nonresidential building industry.
36 Since our results reject the null hypothesis, we are not subject to a standard criticism of econometric results; the improper use of significance tests whereby failure to reject the null hypothesis is seen as accepting the null hypothesis as true. Of course in alternative specifications the test outcome could have been different. However, our procedure is sufficiently general to encompass a wide variety of alternative specifications. This problem, known as Duhem's irrefutability thesis is unavoidable in scientific work. See Blaugh, Mark, Methodology in Economics (Cambridge, 1980), pp. 17–18.