MemGuard-Alpha: Limits of Membership Inference for Detecting and Filtering Memorization-Contaminated Signals in LLM-Based Financial Forecasting
“Large language models are increasingly used to generate financial alpha signals, but many have memorized the historical data in their training corpora, producing apparent accuracy that collapses out of sample.”
StrategyNews sentiment
TechniqueLarge language modellinear factor model
GeographyEurope
Sample window2019–2024
Also stated: Alpha, Excess return, Text and news, out-of-sample test, costs
Missing from abstract: benchmark, multiple testing, universe
Institutional-grade audit of this strategy.
£5 — Request the audit- Audit of the paper claims accounting for costs, luck, crises, originality, geography, etc.
- Reproduction code integratable into Bloomberg, IBKR, etc.
- Bias-free datasets for backtesting included
- Delivered in 2 working days
The paper's neighbourhoodNews sentiment and text signals by year
FamilyNews sentiment and text signals148Momentum and trend following289Factor investing240Machine learning return prediction15318 more · 854
AssetEquities97
The family is counted across the whole universe. Asset, method and authors are counted inside the family.