Driving · 2026-11-13

From Forecasting to World Modeling: Can Temporal Foundation Models Recover Causal Response?

Juwon Hong, Minha Hwang, Venkatesh Shankar

arXiv:2611.15009PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

The authors evaluate released temporal and causal foundation models for retail demand, asking whether good forecasting implies correct causal response to price changes. They report that forecast accuracy does not indicate accuracy of predicted price effects (on simulated data with ground truth and on real sales direction). Extra context improves forecasts but not counterfactual price response under confounding. Instrumental variables help only when supplied in a specific form (residuals).

Why it matters

For decision-making, a model must predict how outcomes change under interventions, not only how they co-vary in historical data. The paper shows that temporal foundation models can be miscalibrated for counterfactual response: a model can forecast well yet predict the wrong direction or magnitude of price effects. The results suggest evaluations of “world models” for control should include accuracy of predicted responses to interventions, not forecast error alone.

Method

  • Evaluate released tabular and time-series foundation models (TabPFN, TabICLv2, TabFM, Chronos-2) and causal foundation model (CCPFN) on retail demand, scoring counterfactual price-response accuracy.
  • Use CausalDemand simulated data with confounding/no-confounding variants and ground-truth sales under price interventions; also test direction-only on real retail scanner data.
  • Test instrumental variables delivery by comparing none vs control-function residuals vs raw instruments (Table 2).

Limitation

One limitation is that the models differ in training objective, capacity and pretraining; the comparison characterizes currently available models under their standard use rather than the capabilities of their architectures.

Abstract (from arXiv)

A model that forecasts sales accurately can still mislead a retailer about how sales respond to a price change. We evaluate released tabular and time-series foundation models, as well as a foundation model pretrained specifically to estimate causal effects. We report three findings. First, forecast accuracy does not indicate the accuracy of the predicted effect of price, both on simulated data in which the true effect is known and on real retail sales records. Second, for the tabular foundation models, additional context improves forecasts but not the predicted effect of price. Third, instrumental variables, the standard remedy in econometrics, reduce this error only when supplied in a specific form and increase it for every model under confounding when supplied as raw instruments.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First announced by arXiv: 2026-11-13 · Indexed by Pub-AI: 2026-11-13 UTC

preprint

From Forecasting to World Modeling: Can Temporal Foundation Models Recover Causal Response?

Juwon Hong, Minha Hwang, Venkatesh Shankar

First announced by arXiv: 2026-11-13 · Original record ↗

arxiv
2611.15009
Explicit source versions (1)
  • v1 · Date not supplied
  1. Preprint available

    Source date: 2026-11-13 · First detected: 2026-11-13 UTC · Historical baseline

    Source evidence (1)
    • Official evidence ↗ · RSS guid=oai:arXiv.org:2611.15009v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv

A model that forecasts sales accurately can still mislead a retailer about how sales respond to a price change. We evaluate released tabular and time-series foundation models, as well as a foundation model pretrained specifically to estimate causal effects. We report three findings. First, forecast accuracy does not indicate the accuracy of the predicted effect of price, both on simulated data in which the true effect is known and on real retail sales records. Second, for the tabular foundation models, additional context improves forecasts but not the predicted effect of price. Third, instrumental variables, the standard remedy in econometrics, reduce this error only when supplied in a specific form and increase it for every model under confounding when supplied as raw instruments.