股票收益的可预测性是金融学中最重要的问题之一，也是当前计量经济学和金融学研究的热点问题之一。股票收益预测回归模型的一个重要数据特征是：预测变量为单位根或接近单位根过程然而被预测的股票收益是个接近白噪声的过程。这说明非线性的可积协整回归模型比文献中广泛使用的线性模型更适合用来对预测回归模型建模。本项目考察基于级数估计的单变量的非参数协整回归模型和多变量的半参数协整回归模型。这是对当前非线性非平稳时间序列计量理论的一个突破。同时我们将分析提出的模型是否在样本外预测表现优于传统的线性模型和历史均值（无预测性）模型。这将对非线性的股票收益预测研究具有重要的启发意义。

The predictability of stock returns is one of the most important problems in finance and is one of the current focus of the researches in econometrics and finance. The integrable nonlinear cointegrating regression model was shown to catch the characteristics of the data structure in stock return predictive regression where the predicting variables are unit roots or nearly unit roots while the predicted return is a nearly white noise process, thus is more suitable in modeling the predictive regression than the widely used linear model. This project examines univariate nonparametric cointegrating regression model and multivariate semiparametric cointegrating regression models based on series estimation which is a huge progress in the theoretical reseraches in nonlinear nonstationary time series. At the same time, we will examine whether the proposed models can out-perform the traditional linear model and historic mean (no predictability) model which will be of great interests for the nonlinear stock return predictability researches.