Knowledge that Transforms

can take all the Sj and try to minimize the variance of vj via factor analysis. We could then find S and also see which Sj was most closely correlated with S. Unfortunately this technique requires that the Sj does not have common measurement error. Since none of the Sj are completely independent of all others (some common source of data is used), common error can creep in. In principle, if one data sourcebut not another -gave answers unacceptable in terms of the a priori considerations dictated by economics, we could eliminate the series. In the present instance, the only possibility would be the significance of NTW1 in the Sggc equations but not in the SOBi}B forms. The differences in cyclical behavior between series are disturbing. But none of the responses violate all saving theories especially since the more recent theoretical innovations, such as permanent income and the ratchet effect define saving to include purchases net of depreciation. Finally the relative quality of the data could be judged by a detailed examination of the primary data sources and subsequent manipulations. This cannot be done now since the last time the SEC and the OBE published detailed descriptions of their sources and manipulations was more than a decade ago and those descriptions in [3, 5] are out of date. Besides, the number of primary data sources used is quite large and diverse. Only a group of individuals familiar with the separate parts could hope to do a competent study. The conclusion, thus, is quite pessimistic. For the saving function, one of the most basic elements of macro-economics, the dynamic and cyclical characterization depends upon our choice of measurement of a given concept and we do not know which measurement is correct.

An intertemporal optimization condition follows from any model of optimal growth. Such an equation usually contains two parameters which cannot be directly observed: the discount rate of future utility and the elasticity of marginal utility. These two parameters can be empirically estimated if they remain constant over time and if the assumed criterion function is maximized in the real economic system. In this paper a method of estimating these two parameters is presented and it is applied to the United States, Japan, and Canada. Although the results are not conclusive, in view of the assumptions involved, they tend to confirm the hypothesis that they are fairly stable over time, at least during a sociologically well-defined short period.

Archen Minsol, Some Tests of the International Comparisons of Factor Efficiency with the CES Production Function: A Reply, The Review of Economics and Statistics, Vol. 50, No. 4 (Nov., 1968), pp. 477-479

T HE postwar growth performance of several European economies has been the cause of coincident feelings of awe, guilt, and envy. It has also been the reason for a number of attempts to explain what has allowed, say, the Germans, French, and Italians to go merrily on their way with only minor interruptions of remarkable growth records, while others, such as the British, have been faced with little other than stagnation. The studies have varied a good deal in the level of sophistication and abstraction, as well as in the relative weight given to empirical in contrast to theoretical considerations. Nevertheless, a common theme is the profound importance of capital formation as a source of growth. Neither of the two recently published studies is an exception, although the heavy weight given to capital formation in one is certainly inadvertent.' Each book is important in its own right, and together they provide an excellent opportunity to examine the main trends in an important and growing body of literature. Therefore, though emphasis will be on the two latest contributions in this area, we will include other studies for comparison and contrast. Together, they provide a sharp contrast to much current analysis which downgrades the importance of capital formation as a source of growth.

T HE model presented in this paper is a system of recursive linear regression equations, the parameters of which are esti-mated from monthly series of data covering the

EVERY adult consciously or unconsciously determines the extent to which he supplies labor for economic activities. Every business determines the extent to which it wishes to increase its work force by new hiring or the extent to which it wishes to reduce its work force by layoffs. The outcome of these decisions is recorded in a monthly tally of the employed (E), unemployed (U), new hires (h), layoffs and quits (s),' and the number of adults outside the labor force (N). As a first step towards understanding the supply and demand relationships that determine the flows of new hires and separations and the stock of unemployed, the authors have examined the Markov matrix A, that determines transitions between the state of Employment, Et, Unemployment of less than one month's duration Uot, Unemployment of more than one month's duration Ult, and non-labor force participation Nt. By definition

Z VI GRILICHES has written an interesting and provocative article on the first volume on the Brookings model. He does not report on later materials (cf. his footnote 2) that are pertinent to reviewing this continuing research project.1 In some places his own preconceptions and specializations caused some loss of perspective. In a few places too, he obtains seemingly contradictory or implausible results by making extreme assumptions about coefficients or by extrapolating to distant time points, inappropriate for testing a system designed for short-run business cycle analysis. This comment is designed to assure the reader a balanced view. At a general level, Griliches does not indicate the progress in model building associated with the Brookings project. It should be recalled that most of the prior models were small (on the order of 30 equations), had a limited degree of disaggregation of expenditure components and virtually no industry detail, lacked a financial sector, and did not explicitly include government policy parameters.2 The Brookings model has several noteworthy features which advance the state of the art of model building, solution, and simulation. 1. Scale: The model is substantially larger than its predecessors. Estimates of single equations do not necessarily give reasonable complete system results. We have solved a 200 equation condensed version of the model and obtained sensible cyclical and growth path predictions. Future large scale models may give better predictive results; but, at least we have shown that it is feasible to work with systems of several hundred equations. 2. Government Policy Parameters: Rather than simply dealing with variables such as tax yields and required reserves, the model treats the many instruments of government policy action explicitly. That is, tax yields are not controlled by the government but only tax rates. This makes for a more realistic description of the actual structure of the economy; it also permits more sensible simulation of policy changes. 3. Sector and Industry Detail: There is more extensive treatment of several sectors such as housing, financial, agricultural and foreign trade. At the industry level, there are now eight production sectors and an expansion to thirty-three is in progress. For each of these sectors, there are price, wage, employment, hours, investment, and so forth, equations. Such extensive treatment in a macro model has not been attempted previously. 4. Monetary Influences: With the exception of T. C. Liu's model, other formulations deemphasized the role of monetary factors.3 And, even in the Liu model, the monetary sector was quite limited. In the Brookings model, the full monetary sector comprises over thirty equations. 5. Input-output: This is the first attempt to integrate an input-output structure into a cyclical model. While input-output is only a limited approximation of the technical structure of the economy, it does permit the translation of GNP component demands into industry gross outputs and industry prices into GNP component prices. As has been shown elsewhere, within the same framework, it is possible to relax the input-output elasticity of substitution assumptions and apply more general CES production functions.4 Yet, an input-output type structure is still required because production from

One of the most fascinating aspects of Soviet economic development has been the remarkable pace in growth of aggregate output maintained over the substantial period of more than three decades. The pace has been remarkable, though not completely unprecedented, and there need be little doubt about its authenticity. Thanks largely to Professors Abram Bergson, Warren Eason, and Raymond Powell, to Dr. Richard Moorsteen, and to Nancy Nimitz there exists a carefully prepared and consistent record of Soviet Russia's gross national product, capital stock, and labor inputs for the period 1928 through 1961 [4] [13]. In table 1 a portion of this basic record on Soviet economic development is reproduced. The primary purpose of this paper is to explore the usefulness of the hypothesis of embodied technical change for providing insight into sources of the growth in output. II Alternative Aggregate Production Functions and Soviet Growth

originally proposed by Arrow, Chenery, A Minhas and Solow, the CES function was constrained to constant returns to scale. It has since been generalised to allow for any degree of homogeneity in the inputs. But the function is still constrained: if returns to scale are a when output is low, they are equally a when output is high. It is shown later in this paper that if this assumption is untrue, if what may be called point returns to scale are themselves functionally related to output, a common procedure for estimating the elasticity of substitution will generally be inconsistent, even if it would not otherwise have been so. To prove this, a modified CES function is derived in which point returns to scale are functionally related to output.