T HEORETICAL models of inventory inl vestment including Belsley (1969), Holt et al. (1960), Whitin (1953) and others invariably suggest the opportunity cost of inventories should be included as a key explanatory variable in any empirical study of inventory behavior. In the absence of an opportunity cost (or other holding costs), the theoretically optimal inventory holding is infinitely large. Even so, it is rather rare any financial variable emerges as statistically significant in empirical studies of inventories.1 Michael Lovell, who has written extensively on inventory investment, goes so far as to comment that the probability of obtaining an interest-rate coefficient with negative sign is 50 percent (1976, p. 400). Even the MITPennsylvania-Social Science Research Council model, which represents an explicit attempt to specify in detail the channels of monetary policy, fails to include monetary variables in its inventory equation.2 The absence of empirical evidence in support of a cost of capital effect on inventory investment renders uncertain what many economists regard as a major channel of monetary policy. It is often commented roughly three quarters of the variance in GNP is accounted for by changes in inventory investment. Since monetary policy is commonly viewed as very powerful, it is truly remarkable so little econometric evidence exists to indicate a cost of capital effect on the most variable component of GNP. This study attempts to re-examine the size and significance of the theoretically important cost of capital effect on inventory investment by utilizing firm specific cost of capital measures, as suggested by the finance theory literature, in a pooled cross section econometric analysis of inventory behavior. The cost of capital measure is computed using the actual balance sheet capitalization particular to each firm in the sample for each point in time. Use of a firm specific cost of capital measure instead of a market interest rate avoids the measurement errors introduced into the analysis by the latter procedure. Risk differences among firms, such as between General Motors and Chrysler, imply substantial differences in capital costs. The errors in measurement problem introduced by a market interest rate will bias towards zero the cost of capital effect. Thus a firm specific cost of capital measure may serve as a more effective opportunity cost variable in an econometric analysis of inventory investment. Perhaps even more critical than the use of firm specific cost of capital measures, the econometric analysis is conducted using two samples of firms with each sample disaggregated by stage of fabrication. The first sample, which includes heavy machinery producing companies, attempts to explain inventory investment behavior for companies produce output in response to orders. The second sample consists of textile companies produce output predominantly to stock in anticipation of orders. Aggregation of firms produce to stock and produce to order-and, in addition, aggregation of inventories across stages of fabrication-may obscure the underlying behavioral characteristics operate, in fact, at the individual firm level. As suggested by theory, the findings of this study indicate the cost of capital is a highly sigReceived for publication May 30, 1978. Revision accepted for publication October 30, 1979. * Federal Reserve Bank of New York. A preliminary draft of this paper was presented at the August 1978 meetings of the Econometric Society in Chicago, Illinois. Financial support for the formative stages of this research was provided by the Computer Science Center of the University of Maryland and from the University Research Board in the form of a faculty research award. The data were provided by the College of Business and Management of the University of Maryland. The author thanks Clopper Almon, Robert Eisner, Irwin Friend, Robert J. Gordon and Joel Popkin for their comments and David Dossetter for very able research assistance. Neither they nor the Federal Reserve Bank of New York nor the Federal Reserve System are responsible for the errors or views contained in this paper. I Studies by Kuznets (1964) and Liu (1963) are among the very few report statistically significant interest rate effects. 2 The paucity of econometric evidence on behalf of cost of capital effects is nevertheless consistent with the prewar survey of Meade and Andrews (1938) (and others) and the postwar surveys of Crockett, Friend and Shavell (1967) and Shavell and Woodward (1971), which questioned managers on the degrees to which they adjusted inventories in response to changes in financial conditions.

THE critical nature of the demand for money in macroeconomic analysis has generated a considerable amount of theoretical and empirical research, although much controversy remains. The conventional theoretical frameworks have resulted in two basic money demand models: one an asset demand model and the other a transactions demand model. For the most part, empirical evidence has tended to favor the asset demand formulation.' This paper will report evidence that strongly suggests that the previous empirical studies of the transactions demand for money were misspecified, and that as a result, the conclusions drawn do not usefully discriminate between the asset and transactions theories. The misspecifications are twofold. First, either GNP or NNP has typically been used as a proxy for transactions although there are strong a priori reasons to believe that neither is an adequate measure of transactions.2 Second, as many authors have recognized, technological change could affect the demand for money and regression estimates might be biased if structural changes due to technological innovations are not taken into account. Even so, researchers have dismissed technological change from their studies, typically relegating their observations to footnotes. In this paper, debits to demand deposit accounts (adjusted for currency transactions) are used to replace income as the measure of transactions, and time, which provides a crude estimate of the mean rate of technological change over the sample period, is incorporated into both the theoretical and empirical model. The results demonstrate that the transactions theory of money demand outperforms the asset demand formulation, contrary to much of the currently available empirical evidence. This study also suggests that technological change reduces money demand, ceteeris paribus. by about 1.5% to 2.5% per year. In addition, other issues such as the speed of adjustment to changes in the desired level of money balances and economies of scale in holding money are also addressed. The empirical findings suggest that at least 75%, and as much as 100%, of the gap between desired and actual money holding is closed within one year and that there are substantial economies of scale in holding money, as suggested by monetary theory.

Although demand for money equations have traditionally been regarded as exceptionally stable, the literature abounds with empirical evidence of structural shifts and apparent trends in some of the coefficients over time. Subperiod estimates of long-term studies often yield different coefficients or suggest secular trends in the coefficients. Other studies report structural shifts or variables which are significant only during particular periods. And while numerous other instabilities have vanished along with revisions in the data, the 1974 instability is too large to be so obliging. Thus, the latest apparent instability is hardly without precedent. One possible source of these instabilities is the omission from the estimated equations of a measure of technological change which reduces over time the real cost of transactions in the management of money balances. The real cost of transactions plays a substantial role in explaining money holdings in the standard inventory theoretic models of money demand. The omission of a key explanatory variable from the estimation produces a misspecified equation which may result in biased coefficient estimates. While the sources of the most recent instability in money demand equations may be due to a structural shift, as suggested in several studies, it may also be due in part to an omitted technological change measure. This paper tests the usefulness of some technological change proxies and examines the forecasting ability of these modified models over the difficult post-1974 period. A Shiller lag specification is employed to obtain a better specification of the dynamic properties of money demand. Although the technological change proxies do improve the equations, the findings support the existence of another structural shift in money demand. The results nevertheless provide some hope that a fairly standard money demand equation, with only minor adjustments, may once again be capable of providing satisfactory forecasts, at least for the present.