The authors are indebted to Professors Arcadius Kahan, Robert Fogel, Richard Zecher and an anonymous referee who made helpful suggestions on an earlier draft. This research was made possible by a grant from the German Academic Exchange Service and aided by the Economic History Workshop of the University of Chicago and the Institute for Capital Market Research in Frankfurt. The authors are responsible for any errors or omissions.

¹ Among the better known works referred to are Riesser, J., The Great German Banks and their Concentration (Washington: U.S.G.P.O., 1911);Google ScholarJeidels, O., “Das Verháltnis der deutschen Grossbanken zur Industrie mit besonderer Berücksichtigung der Eisenindustrie,” in Staats- und sozialwissenschaftliche Forschungen, Band XXIV, Heft II, 1–271 (Leipzig: Duncker & Humblot, 1905);Google ScholarSombart, W., Das Wirtschaftsleben im Zeitalter des Hochkapitalismus (Munich: Duncker & Humblot, 1928);Google ScholarSchumpeter, J., Business Cycles, Vols. I & II, New York: McGraw-Hill Book Co., 1939;Google Scholar and Hilferding, R., Das Finanzkapital (Frankfurt: Europäische Verlaganstalt, 1968).Google Scholar

² Gerschenkron, Alexander, Economic Backwardness in Historical Perspective (Cambridge: Harvard University Press, 1962), p. 46.Google Scholar

³ Ibid., p. 51.

⁴ Eistert, Ekkehard, Die Beeinflussung des Wirtschaftswachstums in Deutschland von 1883 bis 1913 durch das Bankensystem (Berlin: Duncker & Humblot, 1970).Google Scholar

⁵ A Kreditbank performs the functions of a commercial bank, an investment bank, a development bank, and an investment trust.

⁶ Cameron, Rondo, France and the Economic Development of Europe (Chicago: Rand McNally, 1966), pp. 96–137.Google Scholar

⁷ It should be noted that Germany had neither a general anti-trust law nor any other statute restricting interlocking directorates. Furthermore, tax and securities market legislation appears to have been far more concerned with raising revenue and insuring honest dealing than with preserving competition in banking. Riesser cites this legislation as a significant cause of the concentration movement in German banking. See Riesser, Great German Banks, Section IV, Chapter 2.

⁸ Ibid., Appendix IV, pp. 897–920.

⁹ Buff, Siegfried, Das Kontokorrentgeschaft in deutschen Bankgewerbe (Berlin: Cotta'sche Buchhandlung, 1904), p. 47.Google Scholar

¹⁰ Eistert, Die Beeinflussung, pp. 89–91. Note that data used in this paper have been corrected to exclude current-account credit extended for other than industrial purposes.

¹¹ Ibid., p: 91.

¹² Ibid., p. 149.

¹³ Riesser, Great German Banks, pp. 259–260.

¹⁴ Also see Jeidels, “Das Verhältnis,” pp. 121–127.

¹⁵ For an explanation of the meaning of current-accounts from the banks' side, see Buff, Das Kontokorrentgeschäft, pp. 57–67.

¹⁶ This latter case is what was encountered in our empirical section where it was found that in our complete model the coefficients of labor and capital of the production function summed to more than one in contrast to our simple model where they summed to one. In the section “A Test of the Implications of our Findings” we subject our finding to an additional test suggested by Fogel and Engerman to filter out the scale effect from the credit-constant total effect.

¹⁷ Eistert, Die Beeinflussung, p. 33ff. It should be. noted that the use of Mittelbereitstellung as the denominator of this fraction indicates a concern only with the allocation of credit by the banks and not with the borrowing decisions of firms.

¹⁸ Gerschenkron, Economic Backwardness, p. 14.

¹⁹ Jeidels, “Das Verhältnis,” p. 270. For an account of government policy toward industry in defense related matters see Kehr, Eckart, Schlachtfiottenbau und Parteipolitik, 1894–1901 (Berlin, 1930).Google Scholar

²⁰ All of the above data come from Eistert, Die Beeinflussung. See Appendix I for a complete listing of data sources.

²¹ It is important to note that it is not being argued that some form of CA/MB is a factor of production. Instead it is being asserted that CA/MB measures the current-account segment of total bank credit extended to industry for productive purposes. By looking at the sign of the coefficient of this shift parameter, one may determine whether an increase or a decrease in CA/MB is associated with an increase or decrease in the efficiency with which the factors or production (i.e., labor and capital) are used.

²² To test whether the appropriate specification of the production function is Cobb-Douglas, a modification of the Kmenta test suggested by Nadiri (see Nadiri, M., “Some Approaches to the Theory and Measurement of Total Factor Productivity: A Survey,” Journal of Economic Literature, VIII [December 1970], 1137–1178), pp. 1150–1154Google Scholar has been used. What is required is the estimation of

where is a substitution parameter, γ a scale parameter, δ a distribution coefficient and μ the degree of returns to scale. If B is insignificant with μδ significant, this indicates that which means that the elasticity of substitution σ = 1 since . This would indicate that the Cobb-Douglas production function is the correct specification. The results of the test are

where R² = the coefficient of determination, SEE = the standard error of the regression, DW = the Durbin-Watson statistic, and t statistics are listed under the coefficients. Since the Durbin-Watson test indicated serial correlation of the residuals which bias the standard errors of the regression coefficients, generalized least squares has been used resulting in

which no longer shows serial correlation of the residuals. In both equations B is insignificant at the 95% level (since the t statistic is below 1.706), indicating that the Cobb-Douglas production function is the correct specification, μ is 1.491 which suggests increasing returns to scale. In subsequent empirical results (see Table 1) when time is added to the equation the indicated returns to scale fall substantially to a value of around unity. Data for K, L and Y are for non-agricultural output in the period 1883–1913. K and L have been corrected for utilization. For a more complete description see the data section of this paper and Appendix I on data sources.

²³ See Appendix I for a complete description of data sources which are all yearly averages.

²⁴ See Meinert, R., “Die Entwicklung der Arbeitszeit in der deutschen Industrie 1820–1956,” dissertation, Münster, 1958, pp. 110.Google Scholar

²⁵ Four observations for unemployment were missing in Kuczynski's, J. series. See “Germany 1800 to the Present Day,” A Short History of Labor Conditions Under Industrial Capitalism, Volume Three, Part 1 (London: Frederick Muller Ltd., 1945).Google Scholar These figures were generated by the use of a Phillips curve. See Phillips, A., “The Relationship Between Unemployment and the Rate of Change of Money Wage Rates in the United Kingdom, 1861–1957,” Econometrica, XXV (1958), 283–300.Google Scholar

²⁶ For a complete discussion of the generalized least squares technique see Johnston, J., Econometric Methods (New York: McGraw-Hill Book Company, 1963), pp. 179–200. This technique involves first estimating Y_t = a + B X_t + u_t to obtain a vector of errors (u₁ … u_t) to be used to estimate where . The original equation is lagged one period, multiplied by and this transformed equation subtracted from the original equation. The transformed data is then used to re-estimate the original equation which is now in the form . This procedure will not change the estimated coefficients. If it raises the Durbin-Watson test statistic, it will give better estimates of the standard errors of the coefficients. An alternative estimation procedure is to use first differences. This technique is merely a special case of generalized least squares where it is assumed that . When Y and X are in log form, such a procedure reduces to estimating the equation in percentage change rather than in level form.Google Scholar

²⁷ Both equation (11) and (12) have been estimated using generalized least squares with a first order lag scheme.

²⁸ The measured returns to scale can be calculated by adding the coefficients of labor and capital (α₁ and α₂).

²⁹ For a complete discussion of this test see Fogel, R. and Engerman, S., Time on the Cross: The Economics of American Negro Slavery (Boston: Little Brown and Company, 1974)Google Scholar, Appendix B. The essential idea is to rewrite:

In order to do this we must develop coefficients X₁ and X₂ such that

These reduce to .

We can now rewrite equation A as

We note that

Equation D can now be quickly reduced to equation B where

is the “adjusted” intercept that takes into account the effects of possible increasing (decreasing) returns to scale. The test reduces to computing ξ which is the ratio of the two “adjusted” intercepts. The general form of ξ is

which quickly reduces to equation (13) when the base equation has returns to scale ‘equal to one which implies B₀ = 0.

³⁰ The Fogel-Engerman test uses the mean values of all variables within the sample period. While this procedure is correct, the use of the technique does not answer the question of whether the scale effect ever will outweigh the static efficciency loss within the sample period. To answer this question, we have computed ξ for equations (7), (8), (9) and (10) in comparison to equation (6), where we have used end of sample period values for all variables. The values of ξ are .94, .97, .90 and .89 respectively, which are uniformly lower than those obtained for ξ when the mean values of these variables were used. The explanation is that during the sample period the rise in CA/MB was sufficient to outweigh the increase in production arising from the measured increasing returns to scale. If the level of CA/MB had remained constant, then the increasing returns might eventually have “canceled” the output reducing effect of the static loss associated with CA/MB.

³¹ In equation (16) the constant and CA/MB would not enter:

³² If the coefficient on time was the same in equation (8) and (15) this would suggest that there actually were constant returns to scale and that the estimates of α₁ and α₂ in equation (8) were biased upward. Our contrary finding suggests that in fact it is a misspecification to assume constant returns to scale and that α₁ and α₂ are not biased in equation (8).

³³ Gerschenkron, Economic Backwardness, p. 15 and Jeidels, “Das Verhältnis,” p. 270.

³⁴ it is interesting to speculate how much loss of non-agricultural output was actually associated with the rise of the use of the current-account by the banks. If we assume that — .1429 is the coefficient of (CA/MB)_t−1 (see equation 8) and that the index number on non-agricultural output in 1913 was 3.010300778, the rise in CA/MB from 50.8% to 72.8% in the period caused an apparent loss of real output of .094637835. The percentage loss was 3.14%.