Article
OPTIMIZING FINANCIAL FORECASTS WITH BIDIRECTIONAL ANALYST MODEL FUSION
The primary audience for sell-side analysts' recommendations is institutional investors that are required to invest in a broad range of companies within client-mandated equity benchmarks, like the FTSE/JSE All-Share index. It might be difficult for portfolio managers to make unbiased investment judgements given the various sell-side recommendations for a single stock. Using random forest, extreme gradient boosting, deep neural networks, and logistic regression, this study investigates the utilisation of past sell-side recommendations to produce an impartial fusion of analyst forecasts that optimises bidirectional accuracy. While eliminating forward-looking biases, we incorporated 12-month rolling features derived from common sell-side recommendations, such as analyst coverage, point and directional accuracy. By combining forecast features from many analysts using machine learning techniques, we introduce a novel "AI analyst". By methodically producing objective and incrementally better prediction accuracy from publicly available sell-side recommendations, we were able to observe the additional benefits of using these features from multiple analysts. The Decision Tree algorithm, XGB Random forest, and KNeighbors algorithms demonstrated the highest relative performance. Machine learning algorithms perform better in resource-related industries with high volatility than in industries with low volatility, indicating the value of rolling features in bi-directional prediction under such conditions. We observe the incremental contribution of rolling features using feature significance, illuminating the connections between analyst coverage, volatility, and the accuracy of bidirectional forecasts. Furthermore, when modelling analysts' directional forecasts, factors using logistic regression highlight volatility features, initial and target price as some of the crucial aspects
Full Text Attachment
