Transfer Learning and Locally Linear Regression for Locally Stationary Time Series
Jinwoo Park · 2025
Read Transfer Learning and Locally Linear Regression for Locally Stationary Time Series by Jinwoo Park (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-12948.txt.
Description
Jinwoo Park works out the theory for borrowing statistical strength across time series that keep changing. Locally stationary series drift slowly, so a regression fitted in one period may mislead in another. Locally linear smoothing handles the drift better than the classical local constant estimator. Park proves uniform convergence for the multivariate version under strong mixing, splitting the error into variation, smoothing bias, and a new term caused by the local stationarity itself. That extra term turns out smaller than in the older benchmark, which explains the method's edge.
On top of the theory, Park builds transfer learning estimators for the realistic case where the series of interest is sparsely observed but related series are dense. A smoothly varying bias function links them, with a local temporal adjustment for stability. Simulations and an analysis of international fuel prices support the predictions.
Catalog record
| CREATOR | Jinwoo Park |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 177 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-12948