PAPER / ARXIV:2610.09048
Humzah Merchant, Alec Guthrie, Simon Mahns, Randall Balestriero, Bradford Levy
RESUMO
Building a world model requires a state representation useful for planning and decision-making, potentially over tasks unknown at training time. In financial markets, this may require a model to reason about market-wide conditions, asset-specific expected returns, liquidity, volatility, and cross-asset relationships, yet financial representation learning has largely been evaluated on individual predictive tasks over narrow datasets. We introduce Market-1T, a dataset of nearly one trillion observations across U.S. equities from 2008 to 2025 at 1 Hz resolution, develop a rigorous evaluation protocol, and conduct a systematic study comparing 18 encoder-training strategies across nearly two decades of market regimes. We find that encoders with similar predictive performance can organize market state very differently, establishing a foundation for training and evaluating financial market representations in support of world models.
NO MESMO MAPA
Resumo indisponível. Consulte o paper original.
Resumo indisponível. Consulte o paper original.