An Excursion towards Multi-frequency-Prediction Based Flexible Asset Allocation System

Abstract

The study probes the flexibility dimension underpinning the asset allocation strategy and develops a system which has the capability to adapt itself dynamically to the changing preferences of the investors and the uncertainties of the market and at the same time can leverage technology to make better foresight into the complex realities. The need of such a system has been accelerated by the ever-increasing competition to out-perform the market in an environment contaminated by fear from the trailing crisis and anxiety about the future sustainability. The authors engineered a multi-resolution convergence divergence indicator in the light of multi frequency trading behaviour of the heterogeneous agents and used the same in a recurrent neural network based non-linear autoregressive with exogenous input model to conditionally predict the future movements. The predictions are subsequently used in a flexible asset allocation system to manage dynamically both the exposure to the risky asset and the degree of risk tolerance. To ascertain the reliability of the model an out-of-the sample validation test is conducted and the model is compared with its peer and the market by back-testing the same on the testing data set. A multitude of relevant risk and performance measures are used for evaluating the proposed model. The outcome of the study reveals that the proposed system outperforms its peer and the market by a reasonable margin.