機器學習在財務預測的應用： 月營收預測與超額報酬表現之比較研究

Abstract

本研究探討機器學習（ML）模型在營收預測與投資策略上的表現，特別聚焦於每月揭露的資料。我們評估六種 ML 模型，其中隨機森林（Random Forest）在預測準確度上表現最佳，且優於分析師預測。根據其預測建構的投資策略，在扣除交易成本後可產生年化超額報酬 51.29%。我們提出的大多數 ML 模型在報酬表現上優於大型語言模型（LLMs）與自我回歸整合移動平均（ARIMA）模型，顯示這些方法在提升投資績效方面具有明顯優勢。

This study examines the predictive performance of machine learning (ML) models in revenue forecasting and investment strategies, focusing on monthly disclosures. Six ML models are evaluated, with Random Forest achieving the highest accuracy and exceeding analyst forecasts. Strategies based on its predictions yield an annualized excess return of 51.29% after transaction costs. Most of the ML models we propose generate higher returns than large language models (LLMs) and Autoregressive Integrated Moving Average (ARIMA) models, demonstrating their effectiveness in improving investment performance.