Neural network system for tactical asset allocation in the global bonds markets

This paper presents a neural network system for tactical asset allocation between seven major bonds markets. The system consists of several networks, each designed to optimise a local portfolio (local bonds plus US cash). These are subsequently integrated into a global portfolio management system which imposes financial constraints on asset allocation. The portfolio yields returns in excess of 100% for three years which compares favourably with industry benchmarks returning 34% for the same period. Intervals of values for the parameters that influence network performance over which this performance is persistent, are identified