A0638
Title: Conditional factor structures on large asset markets
Authors: Paul Schneider - USI Lugano and SFI (Switzerland) [presenting]
Damir Filipovic - EPFL and Swiss Finance Institute (Switzerland)
Michael Multerer - USI Lugano (Switzerland)
Abstract: The analysis in a prior study provides powerful implications stemming from the absence of arbitrage about approximate factor structures in models for large panels of asset returns. They establish expected returns and the cost of portfolios as sufficient inputs into mean-variance efficient allocations and seminal relations are established between certain return covariance matrices, attainable Sharpe ratios, and bounds for the errors approximating expectations produced by linear factor models. Accordingly, their foundational work serves as the basis for most econometric specifications and formulations in the literature on cross-sectional asset pricing. The analysis in CR is significantly extended to large panels of asset returns with nonlinear conditional factor models. This task is achieved by translating the CR framework to reproducing kernel Hilbert spaces (RKHS). Differently from CR, the return panel is directly modelled, rather than the cross-section of expected returns. The framework in particular also accommodates conditioning information. A model is termed within the setting a conditional approximative K-factor structure and conditions for its existence are provided. Importantly, these conditions are easily numerically tested, as a by-product of approximating the kernel matrices arising within the RKHS setting.