Sequential Detection of Market Shocks With Risk-Averse CVaR Social Sensors

This paper considers a statistical signal processing problem involving agent-based models of financial markets, which at a microlevel are driven by socially aware and risk-averse agents. These agents trade (buy or sell) stocks at each trading instant by using the decisions of all previous agents (social learning) in addition to a private (noisy) signal they receive on the value of the stock. We are interested in the following: (1) modelling the dynamics of these risk averse agents and (2) sequential detection of a market shock based on the behaviour of these agents. Structural results that characterize social learning under a risk measure, conditional value-at-risk (CVaR), are presented and formulation of the Bayesian change point detection problem is provided. The structural results exhibit two interesting properties: (1) risk averse agents herd more often than risk neutral agents and (2) the stopping set in the sequential detection problem is nonconvex. The framework is validated on data from the Yahoo! Tech Buzz game dataset and it is revealed that (1) the model identifies the value changes based on agent's trading decisions. (2) Reasonable quickest detection performance is achieved when the agents are risk-averse.