Time-series models hide dynamics not reflected in forecasts
Dynamical Parameters: An Interpretability Framework for Time-Series Foundation Models
Artificial Intelligence
Summary
Sometimes computer models that predict data changing over time hold important information inside them but don’t show it in their predictions. The authors study how certain time patterns like trends and oscillations are stored inside these models’ inner workings but don’t always influence the output as expected. They find that while the hidden parts can reveal these dynamics very well, the models often don’t use that information to change their predictions correctly. This happens because changing the input doesn’t move the model’s inner state in the way needed to show the right output.
What this means in practice
- •For time-series model developers: Check if your model’s hidden states capture important time dynamics even when forecasts fail to reflect them, improving model interpretability and debugging.
- •For data engineers in finance: Evaluate whether predictive models used in financial time series capture underlying patterns internally to better assess forecast reliability.
Authors
Kang Yang, Gaofeng Dong, Liying Han, Mani Srivastava
Abstract
This work studies a central gap in interpreting time-series foundation models (TSFMs): a dynamical property may be accessible in a hidden state even when the forecast fails to respond correctly as that property changes. We formalize these properties as Dynamical Parameters, including trend slope, oscillation frequency, and autoregressive dependence. We compare their representation accessibility, measured by recovery from hidden states, with their forecast response, measured by agreement with the expected forecast change. Across nine frozen TSFMs and thirteen laws, 42 of 63 model-parameter cells achieve accessibility above 0.95, whereas their median reference-aligned response relative to the conditional reference is only 0.46. To explain this gap, causal geometry compares the hidden-state change required to produce the reference response with the change induced by the parameter intervention. Directly modifying the hidden state recovers the reference response, but the parameter intervention often moves the state in a different direction. These results show that accessible parameter information need not be expressed in forecasts when input changes miss the required hidden-state direction.