报告详情
RALE-Gate: Reliability-Aware Latent Analytics for Explainable Portfolio Risk Control
编号：101
访问权限：仅限参会人
更新：2026-11-10 16:46:02 浏览：8次
Online
报告开始：2026年11月21日 17:00（Asia/Kolkata）
报告时间：15min
所在会场：[S5] Data Science and Intelligent Analytics [S5-2] Data Science and Intelligent Analytics
暂无文件
摘要
Machine-learning models for portfolio management are usually trained to forecast returns, yet they rarely recognise when the market has moved outside the conditions they have learned. This paper proposes RALE-Gate, a reliability-aware latent analytics framework that replaces the question “what will the market return?” with “how well does the model currently understand the market?”. A standard autoencoder is trained to learn a representation of the thirteen causal market variables, from which three signals of reliability are extracted, namely the calibrated reconstruction-based unfamiliarity, the latent distribution shift and the latent drift, and added together in the form of an additive instability index. This trend-confirmed gate triggers equity selling only when high instability coincides with a negative risk-adjusted trend, ensuring that novelty isn't the sole trigger of a sell, and that the additive model provides a driver-level explanation for every allocation decision. It is further tested using a leakage-free walk-forward protocol and a prespecified design in which the choice of models is based only on the BSE SENSEX and the S&P 500 and the Nikkei 225 and the FTSE 100 are reserved for a single final test (2019–2026). Across all four markets RALE-Gate reduces the maximum drawdown compared to buy-and-hold-from 38.1% to 19.4% on SENSEX and from 35.0% to 23.2% on FTSE-with statistically indistinguishable Sharpe ratios. On the held-out markets it beats 200-day moving-average timing, the best rule in the development markets, by a combined Sharpe-ratio margin of 0.16 (one-sided p = 0.048), and Unreliable-Stress states are followed by volatility of 1.8 to 2.6 times higher.
关键词
intelligent analytics, financial analytics, autoencoder, model reliability, novelty detection, explainable AI, portfolio risk control, walk-forward validation
报告人
Tanuja Mathpal
Research scholar India；Department of Mathematics； Statistics and Computer Science； G.B. Pant University of Agriculture and Technology； Pantnagar稿件作者
全部评论
重要日期
-
会议日期
11月20日
2026
至11月22日
2026
-
10月31日 2026
初稿截稿日期
-
11月07日 2026
提前注册日期
-
11月22日 2026
注册截止日期
主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
IEEE Madras Section
联系方式
- ICon INDIA 2026
- ic******@usssociety.org
发表评论