| dc.description.abstract | This paper focuses on the quantiles (10%, 50%, 90%) of the equity return of listed companies in the United States. The sample period covers monthly data from January 1990 to December 2021. Inspired by the Feltham-Ohlson Model, this study collects a total of 414 variables with categories which include financial, operating, technical, supply, ESG, sentiment, and macroeconomics. Input the factors after extracting through dimension reduction into quantile regression model for prediction. Firstly, among the ten industries, it is difficult to predict the median return, which is consistent with the Market Efficiency Hypothesis. However, the consumer durables, manufacturing, energy, and technology industries exhibit predictability in extreme distribution (10%, 90%). Secondly, the structure of equity return undergoes continuous transformation over time, which is consistent with the Lucas Critique. Therefore, it becomes crucial to determine which variables and models to use for return rate prediction under different economic structures. Thirdly, this study decomposes important variables from each category factor, providing clear research directions for investors and financial researchers in stock valuation. Lastly, this paper creates stock selection strategy using a supervised learning partial quantile regression (PQR) method. The strategy effectively distinguishes individual stocks with higher and lower risk ratios. The long-short hedge strategy in the technology industry achieved a cumulative return of 400% from 2000 to 2021, outperforms the buy-and-hold market value-weighted technology stock index by nearly 200%. Compared to the mean, extreme distributions capture more important information and provides a comprehensive understanding of equity return. In the face of changing equity return, it enables more diverse strategies and more flexible adjustment of asset allocation. | en_US |