Multi-Agent Model Based Proactive Risk Management For Equity Investment

This paper develops a new multi-agent model to create a novel warning signal for equity investment, which well replicates actual stock and bond futures price dynamics through estimating fund flows of typical trading agents. Particularly, our model contains four types of agents taking equity/bond strategies such as (i) trend-follow, (ii) contrarian, (iii) yield-curve, and (iv) valuation strategies, each of which is effectively modeled by fuzzy logic with technical indicators. Moreover, to reveal the effects of their trading activities on the price changes, we develop a state space model which represents how the agents' portfolio rebalances drive the price dynamics. Concretely, the model sets each agent's fund size as an unobservable state variable, and supposes its proportion to all agents' total fund size as a market impact of the agent's rebalance. As an estimation result, the price dynamics of popular stock index futures and bond futures in Japan and the United States are well explained by the trading activities.Finally, we provide a warning signal for proactive risk management utilizing the multi-agent model and show its validity on trading performance in Japanese and the United States equity markets. The results demonstrate that the estimated fund flow of each agent helps us to avoid a market crash, which reduces the risk and improves the return