One of the leading hypotheses concerning the dynamics of production over time is the production smoothing hypothesis. Given a planning horizon which spans a number of production periods, the firm need not produce in each period an amount equal to expected sales. Rather, resorting to inventory accumulation and liquidation, the firm may follow a production plan temporally smoother than the path of demand. If firms faced with convex cost functions chose to smooth the rate of output in order to minimize costs, one would expect to observe that the rate of output would vary less than the rate of sales, with variations in inventory stocks absorbing some of the fluctuations in sales. Recently, work on the testing of the production smoothing hypothesis has cast doubt on its empirical validity. The evidence presented by Alan Blinder seems to indicate that the variance of production exceeds that of sales in seven out of eight two-digit retail industries (1981) and in eighteen out of twenty two-digit manufacturing industries (1983 and 1986). The purpose of this paper, then, is to examine the validity of such tests when seasonally adjusted aggregated data are used. The evidence presented show that the relative size of the variances of the seasonally adjusted production and sales does not provide valid tests of the production smoothing hypothesis. In addition, aggregating over firms where the seasonal patterns differ may also distort tests of production smoothing. Blinder realized that the use of seasonally adjusted data may not provide an adequate test of the hypothesis, stating Had they been available, I would have preferred to use data that were not seasonally adjusted since the production smoothing model presumably applies to seasonal fluctuations in sales. However, such data are not (1983, fn. 19). In this paper I focus on the cement industry because the unadjusted disaggregated data are available for the direct testing of the conjecture that aggregate seasonally adjusted data mask production smoothing phenomenon. Aggregate monthly data on five other industries will also be examined.

In his comment, Mark Ladenson contends that substantial differences in wages still exist between the North and the South even after accounting for differences in regional prices. We find his results objectionable on three grounds: a) his arbitrary exclusion of available data, b) his choice of a low budget cost of living as a deflator, and c) the possibility of heteroscedasticity which would render his testing procedure invalid. Ladenson objects to four of the five cities we used as our southern sample, all previous empirical work notwithstanding. These five cities were used in our analysis because they were the only five cities for which the necessary 1963 data were available. Similarly, the five cities included in our Northeast sample were the only Standard Metropolitan Statistical Areas (SMSA) for which complete 1963 data existed.' If we consider 1967 data, there are many more SMSAs in both regions than are utilized by Ladenson. (Our data sources are identical to Ladenson's: Census of Manufacturers 1967 and Handbook of Labor Statistics 1970.) In this study we use all the data available by: 1) examining wage differentials throughout the entire United States, 2) offering alternative definitions for the Northeast and South, and 3) using different deflators to measure differentials in the regional cost-of-living. The statistical techniques used are similar to those employed by Ladenson and in our previous paper, p. 934, except that we have added two additional dummy variables to account for the additional geographical regions, the North Central and western regions of the United States. Dummy variables, then, exist for the South, North Central, and West of the United States; consequently we are comparing wages in these regions to wages in the Northeast.2 Industry 21 of the Standard Industrial Classification