跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2610.25170 · v3

Forecasting Retail Sales at Every Grain: Trees, Foundation Models, Transformers and World Models Compared on Four Public Data Sets

首次提交 2026-10-18 18:50
版本更新 2026-11-17 17:23
雷达观测 2026-11-18 19:42
arXiv ↗PDF ↗

摘要

A retailer makes decisions from the same demand at different grains: store orders by item and day, promotions by item and week, category reviews and supplier terms. This paper asks whether one forecast, built at the finest grain and summed to every coarser grain, can serve them all, and which model does that best. It compares LightGBM, pretrained foundation models (TimesFM, Chronos-Bolt, Timestral), the Temporal Fusion Transformer and a world model that carries each item's state forward under the price and promotion plan, on M5, six Dominick's categories, Breakfast at the Frat and Corporacion Favorita, with the world model and LightGBM tuned by a common search. When to use the average: the average of the tuned world model and LightGBM, summed to every grain, has the lowest error in 22 of 27 grain and data set cases and is never the least accurate on the final held-out periods. When to use the world model: on existing items kept out of training it loses at most 0.040 in scaled error, against up to 0.311 for LightGBM, and on lines with short history it is clearly more accurate on six of nine data sets. When to use LightGBM: on heavily promoted lines, fast sellers and lines with long history (six of nine data sets each) and in price-cut weeks. Inconclusive: lines never promoted and slow lines are close to even, and 52 weeks ahead the average is the most accurate on five of seven data sets, with the world model ahead of LightGBM on two of seven. The findings hold for these data sets and the price and promotion levers; anyone adopting them should rerun every comparison on their own data.

作者备注:38 pages. Version 3: title, abstract and code archive corrected; paper as version 2. Code and data: doi:10.5281/zenodo.24271672

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

仅对比实际观测的元数据,不代表 PDF 正文或实验变化。版本可能不连续;带删除线为移除,下划线为新增。

标题变化

WorldForecasting Models,Retail TreesSales andat Every Grain: Trees, Foundation Models, forTransformers Retailand Forecasting:World EvidenceModels fromCompared on Four Public Data Sets

摘要变化

A retailer makes different decisions at different grains offrom the product,same storedemand andat timedifferent hierarchygrains: store orders for anby item in aand storeday, supplier funding forpromotions anby item acrossand storesweek, category performance by store,reviews and the financial plan for the chain. Which forecasting model to use depends on thesupplier grainterms. This paper compares models at each grain on four public data sets (M5,asks sixwhether Dominick'sone categoriesforecast, Breakfastbuilt at the Frat, and Corporación Favorita,finest thegrain lastand keptsummed outto ofevery allcoarser development)grain, over rolling forecast originscan withserve pairedthem testsall, and which model confidencedoes setsthat best. It covers price and promotion through thecompares planLightGBM, inputspretrained thesefoundation datamodels record(TimesFM, not productChronos-Bolt, assortmentTimestral), orthe place.Temporal TheFusion modelsTransformer areand a world model that carries each item's state forward under the price and promotion plan, the same network without that state,on LightGBMM5, the Temporal Fusionsix Transformer,Dominick's TimesFMcategories, Chronos-Bolt,Breakfast Timestral,at seasonalthe naive,Frat and an equal-weightCorporacion averageFavorita, ofwith the world model, and LightGBM andtuned TimesFMby givena thecommon plansearch. TheWhen answerto changeduse withthe average: the grain.average Forof anthe itemtuned byworld storemodel and an itemLightGBM, acrosssummed stores,to theevery averagegrain, hadhas the lowest error onin seven22 andof eight27 ofgrain nineand data sets,set cases and is never the 90%least confidenceaccurate seton the final held-out aperiods. medianWhen ofto 1use andthe 2world models.model: Foron aexisting categoryitems bykept store,out aof categorytraining andit theloses chainat total,most the0.040 lowest-errorin modelscaled differederror, betweenagainst dataup setsto and0.311 afor medianLightGBM, ofand 4on oflines ninewith modelsshort werehistory notit separatedis byclearly themore fouraccurate toon six windowsof availablenine data sets. ForWhen category-by-storeto planninguse costLightGBM: on heavily promoted lines, thefast averagesellers orand LightGBMlines waswith lowestlong onhistory seven(six of eightnine data sets, and theeach) worldand modelin wasprice-cut lowestweeks. onlyInconclusive: onlines M5;never itpromoted wasand behindslow LightGBMlines itemare byclose storeto oneven, sixand of52 eight.weeks Everyahead trainedthe modelaverage usedis the plan,most andaccurate on thefive weeklyof promotion-drivenseven data pretrainedsets, modelswith werethe lessworld accuratemodel withoutahead itof LightGBM on two of seven. TheseThe resultsfindings hold for these data sets; a forecast solved across all four Ps, or aand retailer'sthe ownprice data,and maypromotion givelevers; differentanyone results.adopting Everythem model,should testrerun andevery changecomparison thaton losttheir isown reporteddata.

作者备注变化

3738 pages,. 18Version figures3: title, 26abstract tables.and Code:code https://githubarchive corrected; paper as version 2.com/karthik-iyer-au/retail-world-models, archivedCode atand data: doi:10.5281/zenodo.2344805624271672

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v3 · 2026-11-17 17:23 · 最新观测版本

Forecasting Retail Sales at Every Grain: Trees, Foundation Models, Transformers and World Models Compared on Four Public Data Sets

A retailer makes decisions from the same demand at different grains: store orders by item and day, promotions by item and week, category reviews and supplier terms. This paper asks whether one forecast, built at the finest grain and summed to every coarser grain, can serve them all, and which model does that best. It compares LightGBM, pretrained foundation models (TimesFM, Chronos-Bolt, Timestral), the Temporal Fusion Transformer and a world model that carries each item's state forward under the price and promotion plan, on M5, six Dominick's categories, Breakfast at the Frat and Corporacion Favorita, with the world model and LightGBM tuned by a common search. When to use the average: the average of the tuned world model and LightGBM, summed to every grain, has the lowest error in 22 of 27 grain and data set cases and is never the least accurate on the final held-out periods. When to use the world model: on existing items kept out of training it loses at most 0.040 in scaled error, against up to 0.311 for LightGBM, and on lines with short history it is clearly more accurate on six of nine data sets. When to use LightGBM: on heavily promoted lines, fast sellers and lines with long history (six of nine data sets each) and in price-cut weeks. Inconclusive: lines never promoted and slow lines are close to even, and 52 weeks ahead the average is the most accurate on five of seven data sets, with the world model ahead of LightGBM on two of seven. The findings hold for these data sets and the price and promotion levers; anyone adopting them should rerun every comparison on their own data.

查看此版本解读 →查看此版本原文 ↗
v2 · 2026-11-16 15:22

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

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.

查看此版本解读 →查看此版本原文 ↗