Nonlinear polymodels is a factor-based statistical analysis technique, which can be applied, in a broad range of fields. Inspired by pattern recognition methods used in DNA analysis or in hand writing recognition, it is particularly adapted to situations where space-dependent changes of regime modify the relation between dependent variables and their independent drivers, making it difficult for a single multi-factor model to fit all the possible situations that may potentially occur. We shall present in the particular context of financial modeling.
Traditional multi-factor analysis is essentially used in finance in a linear setting. Asset returns are replicated by a linear combination of factor returns. Not only it provides answers to questions related to the statistical behavior of assets with respect to the market, but it is intellectually comfortable, as a portfolio is naturally represented as a reduced "portfolio" of risk factors. However, this representation sadly lacks of any predictive value, especially when we need itthe most, that is, when a crisis is coming. We shall show how nonlinear polymodels provide a reliable solution to the main questions factor analysis aims at addressing:
1) finding the probability distribution of individual asset returns (risk measurement)
2) assessing the impact of a given shift of risk factors (stress testing)
3) estimating the joint probability distribution of family of assets (portfolio risk and optimization)
We shall show how the nonlinear polymodel-based "Dominant Factors™" methodology provides superior portfolio returns, simply thanks to a better control of the downside dynamics, without the procyclicality pitfalls of traditional Markowitz and Black-Litterman methods.
5:00 p.m. Tea, Keller Hall 3-176
5:30 p.m. - 6:30 p.m. Lecture, Keller Hall 3-180