This support is for the continuation of a very interesting and important line of research in theoretical econometrics. Most applied econometric research involves using a statistical estimation technique known as "least squares" to analyze data. In essence this technique acknowledges that there are errors in the data, and these errors can derive from many sources. For instance, reported measurements of prices or incomes might be rounded to the nearest dollar rather than being exact. National economic data are highly aggregated, and survey data might include various types of clerical errors. An analyst, of course, knows that errors will exist in the data, but does not know their magnitudes. However, in order to draw consistent and unambiguous conclusions, any analysis of such data must incorporate the capability to account for the potential effects of the errors inherent in the data. Fortunately, the size of the errors can be estimated statistically, and the distribution of the errors can be assumed. Typically errors are assumed to follow the normal distribution. A proven analytical technique computes results by choosing parameters such that the sum of the squares of the estimated error values is at a minimum, hence the name "least squares". This technique forms the basis for virtually all economic analysis, thus knowing its inherent drawbacks is very important. Much research has shown the results of least squares estimation not to be robust if the true distribution of the errors is not normal. This project, and the line of work to which it belongs, presents an alternative to least squares estimation that is not so dependent on the errors being normally distributed, and so represents an important contribution to econometric analysis. Technically, this project develops a method of statistical estimation and inference based on regression quantiles. These are linear combinations of statistics based on data which have been ordered in some way. By studying these combinations, Professor Koenker derives estimation techniques robust to distributional errors. He also extends his analysis to estimation of summary statistics and confidence intervals.
