Insuring Risk-Averse Agents
In this paper we explicitly model risk aversion in multiagent interactions. We propose an insurance mechanism that be can used by risk-averse agents to mitigate against risky outcomes and to improve their expected utility. Given a game, we show how to der
- PDF / 233,767 Bytes
- 12 Pages / 430 x 660 pts Page_size
- 47 Downloads / 173 Views
Abstract. In this paper we explicitly model risk aversion in multiagent interactions. We propose an insurance mechanism that be can used by risk-averse agents to mitigate against risky outcomes and to improve their expected utility. Given a game, we show how to derive Pareto-optimal insurance policies, and determine whether or not the proposed insurance policy will change the underlying dynamics of the game (i.e., the equilibrium). Experimental results indicate that our approach is both feasible and effective at reducing risk for agents.
1 Introduction In almost every decision people make, risk is a factor. When negotiating a business contract, there is the risk of either side being unable to fulfill its obligations. When bidding for multiple items in an auction, there is the risk of winning too many or too few items. Even when using the Internet, there is the risk of congestion depending on the routing policy used. In most of these cases people are risk averse. The importance of the influence of risk aversion on peoples’ decisions is reflected in the size of the insurance industry, a multi-trillion dollar business, [4] and the amount of research in economics relating to risk [17]. There is considerable research in multiagent systems on helping people make better decisions in settings such as those mentioned above. However, this research generally assumes that people are risk neutral [5,15,16]. Given the prevalence of risk aversion in the real world, we believe it is important to study how to manage the effects of risk and risk aversion in multiagent systems. In this paper, we study non-cooperative multiagent systems. Our main contribution is an insurance mechanism that can be used in games to reduce agents’ risk and increase their utility. We present a characterization of our mechanism that allows us to easily determine if the mechanism can be applied to any given game. Experimental results show that our mechanism is usable in many different situations and is scalable. The experimental results also examine how much risk aversion matters in different settings. We conclude with a discussion of related work and some promising areas for future work. F. Rossi and A. Tsoukis (Eds.): ADT 2009, LNAI 5783, pp. 294–305, 2009. c Springer-Verlag Berlin Heidelberg 2009
Insuring Risk-Averse Agents
295
2 The Model and Background In this section we introduce our model of risk aversion for multiagent systems, as well as define the key game-theoretic concepts used in this paper. For a more thorough introduction to game theory, we refer the reader to [8]. 2.1 A Model of Risk Aversion In this section we propose a model of risk aversion for a multiagent setting. The approach we take in modeling risk is motivated by models used in experimental economics [6]. If an agent is risk averse, then it dislikes uncertainty. For example, if given a choice between a lottery and a guaranteed payoff, a risk-averse agent will often prefer the guaranteed payoff, even when the expected payoff from the lottery is higher. In this paper, we model
Data Loading...
