On development of novel hybrid and robust adaptive models for net asset value prediction

Abstract

Efficient prediction of net asset value (NAV) of various investment fund is crucial for both investors and fund management organisations. But prediction of such type of complex financial series is difficult because of uncertainty and influence by political and economical factors. In this paper a novel hybrid adaptive ensemble is developed and its performance is assessed both during training and testing phases using six different NAV data. Further a robust hybrid prediction model is proposed using minimisation of a robust norm and its prediction performance is evaluated and compared with its corresponding conventional ensemble model. Simulation based experimental results demonstrate superior prediction performance of proposed ensemble hybrid model compared to that of three individual component models. Statistical paired t-test is carried out to ensure the superiority of the proposed model in comparison to other three models. Further, it is observed that the proposed robust ensemble model outperforms its hybrid counterpart for all NAVs and for all percentage of outliers up to 10% present in the training samples. The same ensemble and robust models can also applied for efficient prediction of various NAVs for different day’s ahead prediction.

Keywords

• Net asset value (NAV) prediction
• Adaptive moving average (AMA)
• Adaptive auto regressive moving average (AARMA)
• Functional link artificial neural network (FLANN)
• Back propagation (BP) Algorithm and Wilcoxon norm