Latent Dynamics · 2026-10-20

World Models, Trees and Foundation Models for Retail Forecasting: Evidence from Four Public Data Sets

Karthik Iyer, Madan Dabbeeru

arXiv:2610.25170PDF

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 compare a plan-aware world model (roll a carried latent item state forward under known price/promotion plan) against the same network without roll-forward, LightGBM, Temporal Fusion Transformer, three pretrained forecasters (TimesFM, Chronos-Bolt, Timestral), seasonal naive, and an equal-weight ensemble. Best model depends on the product-store-time grain: the average is best on item-by-store and item-across-stores, while cost/accuracy patterns differ at coarser grains.

Why it matters

Retail decisions are made at different hierarchical grains; using one forecasting model across grains can be suboptimal. This work tests plan-aware latent dynamics vs strong tree and neural baselines, pretrained foundation models given the plan, and assesses how forecast choice varies by grain using rolling-origin evaluation and model confidence sets on four public retail datasets.

Method

  • Evaluate plan-aware forecasting at multiple hierarchy levels (item-by-store up to category/chain totals) using rolling forecast origins, paired tests, and model confidence sets.
  • Train/compare models spanning: plan-aware world model with carried state roll-forward, same network without roll-forward, LightGBM, Temporal Fusion Transformer, pretrained TimesFM/Chronos-Bolt/Timestral, seasonal naive, and an equal-weight ensemble of world model, LightGBM, TimesFM with the plan.
  • Score forecasts with hierarchical WRMSSE (lower is better), including runs where removing plan inputs makes trained models less accurate and pretrained models require the plan inputs.

Limitation

These are results for these data sets, with the plan inputs and constraints they carry; a forecast solved across all four Ps, or a retailer’s own data, may give different results.

Abstract (from arXiv)

A retailer makes different decisions at different grains of the product, store and time hierarchy: store orders for an item in a store, supplier funding for an item across stores, category performance by store, and the financial plan for the chain. Which forecasting model to use depends on the grain. This paper compares models at each grain on four public data sets (M5, six Dominick's categories, Breakfast at the Frat, and Corporaci\'on Favorita, the last kept out of all development), over rolling forecast origins with paired tests and model confidence sets. It covers price and promotion through the plan inputs these data record, not product, assortment or place. The models are a world model that carries each item's state forward under the price and promotion plan, the same network without that state, LightGBM, the Temporal Fusion Transformer, TimesFM, Chronos-Bolt, Timestral, seasonal naive, and an equal-weight average of the world model, LightGBM and TimesFM given the plan. The answer changed with the grain. For an item by store and an item across stores, the average had the lowest error on seven and eight of nine data sets, and the 90% confidence set held a median of 1 and 2 models. For a category by store, a category and the chain total, the lowest-error model differed between data sets and a median of 4 of nine models were not separated by the four to six windows available. For category-by-store planning cost, the average or LightGBM was lowest on seven of eight data sets, and the world model was lowest only on M5; it was behind LightGBM item by store on six of eight. Every trained model used the plan, and on the weekly promotion-driven data pretrained models were less accurate without it. These results hold for these data sets; a forecast solved across all four Ps, or a retailer's own data, may give different results. Every model, test and change that lost is reported.

Related papers

Publication record

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

First public release: 2026-10-18 · Indexed by Pub-AI: 2026-10-20 UTC

preprint

World Models, Trees and Foundation Models for Retail Forecasting: Evidence from Four Public Data Sets

Karthik Iyer, Madan Dabbeeru

Preprint posted: 2026-10-18 · Original record ↗

arxiv
2610.25170
Explicit source versions (1)
  • v1 · 2026-10-18
  1. Preprint available

    Source date: 2026-10-18 · First detected: 2026-10-25 UTC · Historical baseline

    Source evidence (1)
Abstract · arxiv

A retailer makes different decisions at different grains of the product, store and time hierarchy: store orders for an item in a store, supplier funding for an item across stores, category performance by store, and the financial plan for the chain. Which forecasting model to use depends on the grain. This paper compares models at each grain on four public data sets (M5, six Dominick's categories, Breakfast at the Frat, and Corporación Favorita, the last kept out of all development), over rolling forecast origins with paired tests and model confidence sets. It covers price and promotion through the plan inputs these data record, not product, assortment or place. The models are a world model that carries each item's state forward under the price and promotion plan, the same network without that state, LightGBM, the Temporal Fusion Transformer, TimesFM, Chronos-Bolt, Timestral, seasonal naive, and an equal-weight average of the world model, LightGBM and TimesFM given the plan. The answer changed with the grain. For an item by store and an item across stores, the average had the lowest error on seven and eight of nine data sets, and the 90% confidence set held a median of 1 and 2 models. For a category by store, a category and the chain total, the lowest-error model differed between data sets and a median of 4 of nine models were not separated by the four to six windows available. For category-by-store planning cost, the average or LightGBM was lowest on seven of eight data sets, and the world model was lowest only on M5; it was behind LightGBM item by store on six of eight. Every trained model used the plan, and on the weekly promotion-driven data pretrained models were less accurate without it. These results hold for these data sets; a forecast solved across all four Ps, or a retailer's own data, may give different results. Every model, test and change that lost is reported.