Core Insights - The article explores the underlying logic of stock selection ability Alpha, focusing on its purity, confidence, and heterogeneity, and proposes various improvement strategies to enhance its sustainability and predictive power [1][3]. Group 1: Characteristics of Traditional Time-Series Regression Alpha - Historical data shows that the proportion of equity funds with Alpha acquisition capability across different factor models fluctuates between 40% and 80%, significantly decreasing when requiring a significant p-value [3]. - Compared to cumulative return indicators, Alpha exhibits better sustainability [3]. - Long-term, constructing long positions with Alpha can yield returns exceeding market averages, but the presence of mixed components obscures the true fund capability, leading to unstable excess returns [3]. Group 2: Improving Alpha Purity through Regression Models - Cross-sectional regression is employed to reassess factor premiums, which helps mitigate information bias and omissions [5]. - Backtesting results indicate that cross-sectional regression Alpha shows significant improvements over time-series regression, with the IC mean for FF3 Alpha increasing from 4.52% to 6.30% [5]. - Key performance indicators such as annualized return and maximum drawdown for FF3 Alpha have improved, with tracking error decreasing from 4.8% to 2.5% [5]. Group 3: Incorporating Potential Factors to Purify Stock Selection Alpha - Incorporating different numbers of potential factors generally enhances the predictive performance of cross-sectional regression [6]. - For FF3, adding 1 to 3 potential factors increases the information ratio from 0.84 to 1.02, 1.00, and 1.24 respectively [6][8]. Group 4: Confidence of Alpha through P-Value Information - By integrating estimated standard error information, p-values can provide a more accurate assessment of estimation precision and stability [9][10]. - The annualized volatility decreases from 22.7% to 20.9% when using p-values to filter funds for constructing long positions, while tracking error and relative drawdown also improve significantly [10]. Group 5: Addressing Beta Anomalies - The average Alpha decreases significantly with increased exposure to SMB and HML Betas, indicating that traditional factor model-derived Alpha may not accurately reflect fund capabilities [15]. - Adjusting Alpha for Beta using various methods shows that fund regression Beta adjustments yield the best results, enhancing risk-adjusted returns [16][17].