Applying Principal Components Regression Analysis to Time Series Demand Estimation

Demand functions for rice in Colombia and Venezuela, estimated by means of ordinary least square, were unsatisfactory because of problems with multicollinearity An alternative approach, principal components regression, was tried Results showed that principal components regression estimates were more consistent with theoretical expectations and were statistically more significant The cost of these gains was that the coefficients were biased However, the mean-square-error tests indicated that the reduction in variance outweighed the loss due to bias