一、主題：Forecasting Stock Returns Under Model and Parameter Uncertainty: A Machine Learning Approach

二、主講人：姜富偉，bevictor伟德官网副教授，資産管理研究中心研究員。主要研究方向包括金融大數據與機器學習、行為金融、資産定價、投資管理等。曾在Journal of Financial Economics、Review of Financial Studies、Journal of International Money and Finance、Journal of Banking and Finance、Journal of Portfolio Management、《金融研究》等重要期刊發表多篇學術論文。曾獲得國際财務管理協會CFA最佳論文獎、中國金融評論國際研讨會Emerald優秀論文獎、《金融研究》優秀論文三等獎、全美華人金融協會最佳論文獎等學術獎項。

三、時間：2018年4月3日（周二），12:30-13:30

四、地點：學院南路校區主教學樓913會議室

五、主持人：黃志剛，bevictor伟德官网副教授

摘要：We propose a machine learning approach for combination forecasts of stock returns. When forecasting stock returns out of sample, sophisticated combinations, which average uni- and multivariate predictive regression forecasts, often fail to beat the historical average return, while simple mean combinations of univariate predictive regression forecasts often perform superior. In this paper, we apply the AdaBoost technique in machine learning to reduce parameter estimation risk and overfitting that impairs predictability of sophisticated combinations. Empirically, our new approach strongly beat historical average and simple mean combination with large out-of-sample $R^2_{OS}$. The predictability generates large utility gain for investors. In addition, the predictability is economically strong in both good times and bad times, and is linked to macroeconomic conditions.