Levent Bulut releases study on the impact of conflicting historical data on return prediction by foundation models
Get up to speed
Levent Bulut released a new quantitative-finance paper on arXiv studying how, when foundation models make return predictions, old observations whose distribution differs from recent returns affect predictive performance. Across five foundation-model prediction pipelines, three simulation processes, and 1,024 independent replications each, conflicting histories reduced the probability that a model outperforms the zero benchmark in all 30 high-signal comparisons. An exploratory comparison across 56 assets showed that shortening the input sequence from 256 to 64 returns raised mean absolute error by 1.77–4.58 basis points for all models, and overall prediction error under both input lengths did not beat the zero benchmark.
AI-generated from reports · 3 hr agoUpdated
Coverage timeline
Follow the coverage to understand different sides of the event.
- arXiv q-finarXiv Paper: The Impact of Conflicting History on Return Prediction with Foundation Models
A new arXiv paper studies the predictive performance of foundation models for return prediction when historical observations are inconsistent with recent return distributions. Authored by Levent Bulut, the 10-page paper includes 5 figures and belongs to the statistical finance (q-fin.ST) field.
Trend of this event
Not enough consecutive observations yet; trend is not drawn.