Regime-Conditioned Information Routing in Institutional Quantitative Trading

Publication date: 2026-09-21

Contributors: Ogdn Ames, Ames Investment Systems

Abstract

This technical research paper develops Regime-Conditioned Information Routing (RCIR), a falsifiable theory for when transformer-like adaptive information routing may justify its statistical and economic complexity in institutional quantitative trading. It identifies state-dependent variation in predictive relationships, rather than architectural novelty, as the potential source of routing value. The paper sets out a formal conditional-linear result, a target-free Regime Reallocation Index, prespecified hypotheses, and a frozen forward-validation standard using point-in-time data, realistic costs, and multiple-comparison controls. It concludes that transformers should be treated as optional adaptive routing components that must demonstrate incremental net out-of-sample value before deployment.

Keywords: transformers, quantitative trading, financial time series, regime change, attention, foundation models, systematic macro, model selection, backtest overfitting, point-in-time data

Published by Ames Investment Systems

License: All rights reserved
Permission is required for reuse beyond applicable legal exceptions.