A0506
Title: Transfer learning in conditional factor models
Authors: Yubo Tao - University of Macau (China) [presenting]
Abstract: The focus is on considering the estimation and prediction of a conditional latent factor model in the setting of transfer learning where, in addition to observations from the target model, auxiliary datasets are available. To effectively utilize the auxiliary datasets, a transfer learning algorithm is employed in conjunction with the instrumented principle component analysis (IPCA) to estimate the conditional latent factor models. Given the informativeness of the auxiliary datasets, a trans-IPCA algorithm is proposed, and its $\ell_1/\ell_2$-estimation error bounds are derived. It is proven that when the target and sources are sufficiently close to each other, these bounds could be improved over those of the classical IPCA estimator and its penalized variants using only target data under mild conditions. When the set of informative auxiliary data is unknown, a data-driven and algorithm-free procedure is introduced to detect transferable samples. Monte Carlo simulations confirm the superior performance of the proposed estimator compared to classical and penalized IPCA models, both in-sample and out-of-sample.
