ESTIMATING THE TECHNOLOGY COEFFICIENTS IN LINEAR-PROGRAMMING MODELS

Estimation of a linear programming model's technology coefficients using data from a sample of firms is viewed as an application of random coefficient regression (RCR). An RCR estimator restricting predicted coefficient values to be nonnegative is proposed. The estimator's finite sample performance is examined in Monte Carlo experiments. The proposed estimator performs well compared with inequality-restricted least squares, despite its use of an estimated covariance matrix. In sampling a population of firms, a dependency may arise between coefficients and activity levels. Two tests for dependence are proposed and examined in Monte Carlo experiments. The tests' reliability varies with characteristics of the sampled population.