Application of bootstrap techniques in econometrics: the example of cost estimation in the automotive industry

The application of bootstrap methods to regression models helps approximate the distribution of the coefficients and the distribution of the prediction errors. In this paper, we are concerned with the application of the bootstrap techniques to determine prediction intervals on econometric models when the regressors are known. We investigate problems associated with its application: determination of the number of replications, choice of the method to calculate the least squares estimator (pseudo-inverse or inverse) and sorting algorithm of the statistic of interest. These investigations stemmed from the need to predict costs in the automotive industry in the earliest phases of the development of a new vehicle. Generally, the sample size is small and the error term of the model has not necessarily a Gaussian distribution. Consequently, the use of bootstrap techniques strongly improved prediction intervals by reflecting the original distribution of the data. Two examples (car engine and fuel tank) illustrate the application of such techniques.