結合 k-近鄰演算法模型解決 ESG 資料庫遺失值及其應用

Abstract

由於解決 ESG 資料庫數據嚴重缺失的相關研究還並不多，因此本論文嘗試\n將著名的分類模型 k-近鄰演算法（k-Nearest Neighbor, kNN）應用在 ESG 領域中，使沒有完全揭露 ESG 項目的公司，能夠依據該產業其他公司的表現來預測其遺失的值，再透過改良的計算 ESG 分數的方法，利用加大給分差距來加強好和壞公司在 ESG 表現上的差異，使得計算出來公司的 ESG 分數能對報酬率具有顯著解釋能力，也期望可以有效地區分出好公司與壞公司的投資組合，使好公司投組擁有較佳的績效報酬。\n實證結果指出，本研究展現了在 Bloomberg 資料庫原始公司 ESG 數據有缺失的情況下，使用 kNN 模型來插補遺失值比沒有使用 kNN 模型直接忽略遺失值，以及增加計算 ESG 分數級距的研究方法，都能夠讓 ESG 因子對報酬率的解釋能力有顯著提升。除此之外在特定的 ESG 因子篩選條件中，好公司投組具有顯著超額報酬，其歷史回測績效表現更贏過壞公司投組，且報酬率波動度也都較小；但相對地，在某些因子下則呈現相反的情況。總結來說公司 ESG 的表現對報酬率的影響，並非是 E、S 和 G 三個因子佔有相等權重，從實證分析來看，在使用 kNN 模型使得公司有完整資料時，單注重 S 因子，與同時考慮 S&E 和 S&G的情況，會有較優異的報酬表現與較低的報酬波動度。

Since there are not many studies to solve the serious lack of data of ESG database, this paper tries to apply the famous classification model k-Nearest Neighbor (kNN) to the ESG field, so that companies which do not fully disclose their ESG items can predict their missing values based on the performance of other companies in the same industry. Then the improvement of ESG score calculation method enhances the difference between the ESG performance of good and bad companies, so that the ESG scores of companies can have significant effect on the return rate, and is expected to identify the portfolio of ESG good and bad companies, then make ESG good portfolios have better performance returns.\nThe empirical results show that in the case of missing ESG data of Bloomberg, the method using the kNN model to impute missing values and expanding the calculation range of ESG score can significantly improve the explanatory ability of ESG factors on company returns more than ignoring missing values without using the kNN model. In addition, under some specific ESG factor criteria, the portfolios of ESG good companies have significant excess returns and outperform the portfolios of ESG bad companies, and also have less volatility in returns; however, the opposite is true for some factors. In conclusion, the effect of ESG performance on return is not equal weighting of the E, S and G factors. From the empirical analysis, when the kNN model is used to make the ESG data of companies complete, focusing on the unique S factor, as opposed to considering both S&E and S&G, results in better return performance and lower return volatility.