Weak Factor Alpha Strategy 2(中文)
接續上一篇弱因子Alpha策略，此篇我們會將這些弱因子(特徵)放入CatBoost分類器中，並優化了一些過去論文中使用機器學習流程，同時加入財務理論的方法使回測更為合理。透過分類器預測買進相對最強和放空相對最弱的投資組合建構一個 Alpha策略，確實能在SPY500成分股中賺取較高的風險調整後報酬，也能有效控制市場的下行風險。
Continuing from the previous report, we will now introduce how to incorporate the filtered features into a classifier model for training and predicting future relative strength. Figure 1 illustrates the overall architecture of the entire process.
We first calculate the slope of the next three days and arrange them in ascending order. Then, we divide them into 10 equal classes, with each class containing 50 securities. These classes serve as the labels for training. After training the classifier, we will long the portfolio predicted to be the highest relative strength(9th) and short the portfolio predicted to be the lowest relative strength(0th). Figure 2 shows the cumulated return of each group.
In Figure 3 and Table 1, the performance of a strategy is presented where positions are entered on Monday and closed on Friday. The alpha denotes the return obtained from long the 9th group and short the 0th group, while alpha_2 represents the results after applying twice the leverage to the alpha strategy. It is observed that although trend returns may not be captured, the strategy still generate stable profits and higher risk-adjusted returns over the long term. Additionally, the strategy exhibits smaller maximum drawdowns during market downturns, indicating a low correlation with the market and reflecting the advantages of a long-short hedging strategy.
By combining the results from the previous report and the present study, we have optimized certain steps and incorporated financial theory into the backtesting process, making it more robust and rational. Through this process, we have been able to achieve higher risk-adjusted returns in the SPY500 constituent stocks while effectively managing market downside risk.
Davud, Wang
Quantitative Research Intern, Quantitative Finance, Gamma Paradigm
