The Time Structure of Investment Behavior in United States Manufacturing, 1947-1960

Abstract

T HE relationship between investment behavior and its underlying determinants is of critical importance for economic policy. Perhaps even more critical for the timing of economic policy is the time structure of this relationship.' A counter-cyclical policy predicated on the stimulation of investment expenditures requires a relatively short lag between changes in the policy instruments and actual investment. If the lag between policy changes and investment is long, a counter-cyclical policy based on stimulating investment expenditures may have an adverse effect on economic stability. The form of the lag between changes in policy and actual investment is also important for economic policy. If the effects of changes in policy instruments are highly concentrated in time, very precise control of the timing of policy measures is required. If, on the other hand, the effects of policy changes are distributed over very substantial periods of time, less precise control of the timing of economic policy may suffice. Empirical evidence on the average lag between investment and its underlying determinants has been slow in accumulating. Evidence based on sample surveys suggests an average lag between the determinants of investment and actual investment expenditures in United States manufacturing of seven quarters.2 Evidence based on regression studies of annual observations of investment expenditures suggests lags ranging from 25 to 50 quarters or more,3 while examination of quarterly observations suggests lags of six to ten quarters.4 The findings from quarterly data conflict with findings from annual data but coincide with survey results. Empirical evidence on the form of the lag distribution underlying investment behavior has accumulated even more slowly than evidence on the average lag. Most studies of investment begin with an a priori specification of the form of the lag distribution. This specification restricts the class of possible lag distributions severely Koyck [9] employs a geometric lag distribution, Solow [12] has proposed the Pascal lag distribution for this purpose, and deLeeuw [2] uses an inverted-V shaped distribution. Each of these classes of lag distributions has the undesirable property that it is impossible to approximate an arbitrary lag distribution by a member of the class. While evidence is available on the average lag resulting from the application of a given lag distribution, almost no evidence is available on the form of the distribution that should be employed. To obtain such evidence, a class of lag distributions that can approximate an arbitrary lag distribution must be used. Our first objective is to estimate the average lag between the determinants of investment behavior and actual investment expenditures. Our second objective is to characterize the form of the lag distribution. For these purposes we employ empirical results from a study of investment behavior in United States manufacturing.5 Quarterly data on investment exinvestment expenditures for 15 sub-industries penditures appear to be essential for a detailed study of this lag distribution. Our empirical results are based on quarterly observations on 1 A comprehensive discussion of the timing of economic policy is provided by Ando, Brown, Solow, and Kareken [1]. For a discussion of fixed investment, see especially pp. 2 5-38. 2See Mayer [11]. 'See Kuh [10], especially pp. 7-21 and 293-302. Kuh cites the results of Grunfeld [41 and Koyck [9], which are similar to his own. 4See Jorgenson [5 and 6]. Kareken and Solow [1] have studied lags between the determinants of investment and new orders and between new orders and production of investment goods using quarterly data. Their results are