A0473
Title: Model averaging marginal nonlinear logistic regressions for time series data
Authors: Rong Peng - University of Southampton (United Kingdom) [presenting]
Zudi Lu - University of Southampton (United Kingdom)
Abstract: A semi-parametric procedure named Model Averaging MArginal nonlinear LOgistic Regressions (MAMALOR) is proposed, which is flexible for forecasting of binary count time series data. It is motivated by applications in such scenarios of forecasting the price up/down direction in stock market and the default/non-default in credit scoring. Such binary time series exist in wide applications beyond finance though the considered financial application. The procedure can avoid the curse of dimensionality for high dimension d and be easily carried out by maximum likelihood methods. Our initial application shows that the MAMALOR procedure is promising in forecasting of the stock price direction, outperforming the linear logistic regression and the (logistic) generalised additive modelling. We comment that although this procedure is basically focused on binary time series data, the ideas and methodology as well as insights into applications, learned from this project, will help to further study other count time series data in a generalised setting.
