source: arxiv machine learning: algometrics: forecasting under algorithmic feedback
level: research
in algorithmic markets, predictive models become part of the data-generating process they aim to forecast. when their outputs are turned into trades, allocations, or risk controls, they change the future data on which they are evaluated. this paper introduces algometrics, a framework for time series whose evolution depends on the predictive algorithms forecasting them. the framework distinguishes historical risk, measured under passive forecasting, from deployment risk, measured when forecasts drive actions.
the author proves three results. first, deployment risk is not identifiable from passive historical data alone. even in a one-step linear feedback model, infinitely many algorithm-mediated environments induce the same historical law while implying different deployment risks for the same forecaster. second, historical model rankings can invert under crowding. a predictor with lower passive error can have higher deployment error when many agents use similar models, because their collective actions shift the data distribution.
third, the paper shows that deployment risk can be bounded by combining historical data with structural knowledge about the feedback mechanism. this provides a practical way to assess real-world performance without full knowledge of the deployment environment. the framework applies to any setting where forecasts influence outcomes, such as algorithmic trading, dynamic pricing, or automated decision systems.
why it matters: it warns that evaluating models on historical data alone can be dangerously misleading when those models will be used to make decisions that change future data, a common scenario in ai-driven systems.
source: arxiv machine learning: algometrics: forecasting under algorithmic feedback