{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,25]],"date-time":"2026-10-25T16:34:53Z","timestamp":1792946093137,"version":"4.5.0"},"posted":{"date-parts":[[2026]]},"group-title":"SSRN","reference-count":7,"publisher":"Elsevier BV","license":[{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"unspecified","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"abstract":"<jats:p>&lt;div&gt;\n We evaluate Timestral-28M on daily returns, volume and range volatility using rolling annual adaptation and a locally reserved retrospective 2026 evaluation whose forecast origins precede model release. Thirty-six full-parameter trajectories require 101.6 minutes of training and validation on a 16 GB Apple M2 CPU. Among the evaluated combinations, the original model achieves 4.70% equal-asset relative-MAE skill for fourteen stocks' fifth-session log-volume against conditional OLS (descriptive 95% interval: 2.08-6.98%). The adapted ensemble's direct skill against the original is 0.14% [-1.15, 1.48]. Returns show no consistent benchmark advantage, and volatility results differ between stocks and equity funds. Unknown pretraining overlap limits generalization. The volume result supports a research baseline, with adaptation benefit and trading profitability unestablished.\n&lt;\/div&gt;\n&lt;div&gt;\n &lt;br&gt;\n&lt;\/div&gt;\n&lt;div&gt;\n AI assistance disclosure: AI tools assisted with implementation, verification, literature search and drafting. The human author directed and supervised the research and is responsible for the manuscript.\n&lt;\/div&gt;<\/jats:p>","DOI":"10.2139\/ssrn.7663019","type":"posted-content","created":{"date-parts":[[2026,10,24]],"date-time":"2026-10-24T11:34:35Z","timestamp":1792841675000},"source":"Crossref","is-referenced-by-count":0,"title":["Financial Forecasting with a Small Time-Series Foundation Model: A Local Evaluation of Timestral-28M"],"prefix":"10.2139","author":[{"ORCID":"https:\/\/orcid.org\/0009-0002-3126-6760","authenticated-orcid":true,"given":"Philipp D.","family":"Dubach","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"78","reference":[{"key":"ref1","author":"Miquel Noguer","year":"2026","journal-title":"Pretrained time-series foundation models for financial return forecasting"},{"key":"ref2","author":"Anubha Goel","year":"2025","journal-title":"Foundation time-series AI model for realized volatility forecasting"},{"key":"ref3","journal-title":"Mistral AI. Introducing Timestral. Official model release"},{"key":"ref4","author":"October","year":"2026"},{"issue":"1","key":"ref5","doi-asserted-by":"crossref","first-page":"61","DOI":"10.1086\/296071","article-title":"The extreme value method for estimating the variance of the rate of return","volume":"53","author":"Michael Parkinson","year":"1980","journal-title":"The Journal of Business"},{"issue":"2","key":"ref6","doi-asserted-by":"crossref","first-page":"174","DOI":"10.1093\/jjfinec\/nbp001","article-title":"A simple approximate long-memory model of realized volatility","volume":"7","author":"Fulvio Corsi","year":"2009","journal-title":"Journal of Financial Econometrics"},{"key":"ref7","author":"Marcel Meyer","year":"2026","journal-title":"Rethinking evaluation in the era of time series foundation models: (Un)known information leakage challenges"}],"container-title":[],"original-title":[],"deposited":{"date-parts":[[2026,10,25]],"date-time":"2026-10-25T16:24:57Z","timestamp":1792945497000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.ssrn.com\/abstract=7663019"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":7,"URL":"https:\/\/doi.org\/10.2139\/ssrn.7663019","relation":{},"subject":[],"published":{"date-parts":[[2026]]},"subtype":"preprint"}}