- ニュース
- 2025年11月24日(月) に産研講演会「Waseda Organizational and Financial Economics Seminar : Adding Noise to Reduce Noise: A Counter-Intuitive Approach to Stock Return Prediction 」が開催されます。
2025年11月24日(月) に産研講演会「Waseda Organizational and Financial Economics Seminar : Adding Noise to Reduce Noise: A Counter-Intuitive Approach to Stock Return Prediction 」が開催されます。
Dates
カレンダーに追加1124
MON 2025- Place
- ハイフレックス開催
- Time
- 13:10～14:50
- Posted
- 2025年11月10日(月)
「Adding Noise to Reduce Noise: A Counter-Intuitive Approach to Stock Return Prediction」＊英語でのご報告
| 日時 | 2025年11月24日（月）13:10～14:50 |
|---|---|
| 開催方法 | ①対面 ＊11号館8階814教室にお越しください。 ②Zoom＊お申込み完了の自動返信メールにて、参加用URLをお知らせいたします。 |
| 対象 | 学生・教職員・一般 |
| 講演者 | 後藤 晋吾 氏 （Professor, College of Business, The University of Rhode Island） |
| 要旨 | This seminar introduces a new approach that, paradoxically, enhances the accuracy of cross-sectional stock-return predictions by deliberately adding random noise. In high-dimensional prediction settings, traditional regularization methods such as Ridge, Lasso, and PLS have been widely employed. However, these methods often produce negative out-of-sample R^2 values, making reliable prediction difficult in practice. To address this challenge, the study proposes two noise-based approaches—noise injection and noise augmentation—and demonstrates their effectiveness in stabilizing coefficient estimates and improving predictive performance. These methods are closely related to the phenomenon of “benign overfitting” recently highlighted in the machine-learning literature and are consistent with the emerging view that dense models, rather than sparse ones, may yield superior out-of-sample forecasting results. The central theme of the study is a seemingly paradoxical mechanism: strategically adding noise to a model induces implicit regularization and improves out-of-sample predictive accuracy. A related paper, “Does Noise Hurt Economic Forecasts?” by Liao et al. (2024), provides evidence for the usefulness of noise augmentation in economic forecasting, though the theoretical rationale remains complex. We extend this insight to the more complex and inherently noisy domain of stock-return prediction and confirm its empirical validity. Furthermore, we introduce a more intuitive noise-injection method, clarify its relation to noise augmentation, and discuss the role and implications of noise in an accessible manner. Overall, our results suggest that noise can function not merely as an error term but as a design element that improves the structure of predictive models. We hope this unconventional approach will open new possibilities for high-dimensional financial forecasting. |
| 世話人 | 宮島 英昭（早稲田大学商学学術院 教授） |
| 参加申し込み方法 | 参加はこちらからお申込みください。※11月20日（木）17:00締切 |
| 共催 | 早稲田大学商学部・産業経営研究所・谷川寧彦分科会 |
- Tags
- イベント