In using multiple regression methods for prediction, one often considers the linear combination of explanatory variables as an index. Seeking a single such index when here are multiple responses is rather more complicated. One classical approach is to use the coefficients from the leading canonical correlation. However, methods based on variances are unable to disaggregate responses by quantile effects, lack robustness, and rely on normal assumptions for inference. We develop here an alternative regression quantile approach and apply it to an empirical study of the performance of large publicly held companies and CEO compensation. The initial results are very promising.
翻译:在使用多重回归方法进行预测时,人们往往将解释性变量的线性组合作为一个指数。当这里是多重反应时,寻求单一的这种指数比较复杂。一种典型的方法是使用主要锥体相关性的系数。然而,基于差异的方法无法根据量化效应、缺乏稳健性以及通常的推理假设进行分解。我们在此开发一种替代的回归量化方法,并将其用于对大型公开控股公司和首席执行官报酬的实证研究。初步结果很有希望。