Labor Adjustment Patterns In The U.S. Steel Industry

Abstract

This study examines labor demand in the U.S. steel industry.  During the period of 1963-1988, the industry witnessed a tremendous decline in its output and employment.  This decline has been particularly severe in the 1980s.  The overall goal of this study is to estimate models of labor demand for the industry.

The study makes two main contributions.  First, the study constructs a new high-frequency monthly data set on steel output and factor prices.  Hamermesh (1993) notes that most studies of labor demand rely on annual data or quarterly data which is too intertemporally aggregated to model short-run labor adjustment.  Second, labor demand in the industry is modeled as a function of the unfilled orders variable.  This variable is included as a measure of future demand for the industry's output.  The main challenge of this study is to deal with a number of econometric modeling issues.  Recognizing that the industry's output demand is potentially an endogenous variable, the output is instrumented with a set of demand related variables including controls for various steel protection regimes.  This study also corrects for serial correlation in these data through differencing and Hatanaka’s autocorrelation procedures. 

The main results of the study are fourfold.  First, as in previous studies, the Instrumental Variable (IV) estimates show that the real wage rate and output are the key variables for explaining the fluctuations in employment.  Second, the IV results do suggest that endogeneity of output is a problem and that the OLS results are biased downward.  Third, the unfilled orders variable increases the demand for labor.  However, this result is somewhat sensitive across specifications.   Lastly, the study finds a fast short run employment adjustment period of 1.6 months.  A final note of caution concerns serial correlation.  Serial correlation is relatively severe in the data.  The study solves this problem through both differencing and autocorrelation corrections.  However, the success of these procedures proved to be mixed.  Thus, the estimates reported herein vary across different estimation approaches.