Schedule - Parallel Session 2 - Experiments and Ambiguity 2

IMC Boardroom 2 - 14:00 - 15:30

Ambiguity Attitudes, Framing and Consistency

Alex Voorhoeve; Ken Binmore; Arnaldur Stefansson; Lisa Stewart


We use probability-matching variations on Ellsberg’s single-urn experiment to assess both the sensitivity of ambiguity attitudes to framing and the consistency with which subjects display these attitudes within a single frame. Contrary to most other studies, we find very little change in ambiguity attitudes due to a switch from a gain to a loss frame; we also find that making ambiguity easier to recognize has little effect. Regarding consistency, we find that 28% of subjects are highly inconsistent choosers; roughly the same share are highly consistent. Ambiguity attitudes depend on consistency: ambiguity seeking is much more frequent among inconsistent choosers; consistent choosers are much more likely to be ambiguity neutral.

Alex Voorhoeve

Associate Professor, London School of Economics

What You See is What You Bet

Dmitri Vinogradov; Angela Izah; Ceri Watkins


Decisions under uncertainty and risk depend on surrounding factors such as weather conditions, outcomes of sport competitions, TV shows (see, e.g. Baillon et al., 2014; de Martino et al., 2006; Edmans et al., 2007; Kamstra et al., 2003). Supposedly these factors affect people’s mood and/or emotional state, and through that the decision-making process. Can exposure to a particular computer interface produce similar effects? A typical interface of a stock trading software uses black background, colourful fonts and graphical charts. We show that interface and context (investment task) differently affect attitudes to risk and attitudes to uncertainty. Remarkably, mood does not appear to channel the impact of environment on decisions; instead, this impact is likely to be channelled by cognitive abilities. The results come from three online random assignment experiments, in which subjects perform tasks traditionally used to measure their attitudes to risk (willingness to pay and willingness to accept) and ambiguity (standard 2-colour Ellsberg task is used to classify subjects into ambiguity-averse, -neutral and -seeking). The control group faces neutral questions (referring to lotteries and balls in the two urns) in a black font on a white background (neutral interface). In the first experiment, the treatment group faces questions that are numerically identical to those in the control group, yet formulated as an investment task and presented in the interface that mimics the colours used in security trading platforms (aggressive interface). In the second and the third experiments, there are two treatment groups: one faces investment-styled questions in a neutral interface (investment treatment), and the other one faces neutral question in the aggressive interface (aggressive treatment). In all experiments we ask subjects to assess their mood in the beginning and in the end of the experiment, using a five-point scale (from bad via neutral to good). In experiment 3 we additionally run a cognitive reflection test (CRT, see Frederick, 2005, and extension by Sinayev and Peters, 2015). Exposing subjects to investment framing produces a decline in risk aversion: an average subject is going for riskier options; the fraction of risk-averse subjects becomes about one third smaller. Aggressive interface alone has no effect on risk attitudes, however it reduces the fraction of ambiguity-averse subjects, whilst investment framing reduces the fraction of ambiguity-seeking participants. Both types of treatment raise the fraction of EU-compatible subjects (all significant at p <0.05). In this sense what people see does affect what they bet. The observed impact is linked to cognitive abilities but not mood. Neither the aggressive interface nor the investment framing impact mood; instead, treatments impact cognitive abilities, as measured by both the time spent on the survey and the CRT scores.

Dmitri Vinogradov

Lecturer, University of Essex

Who to Hire, Optimists or Pessimists?: Optimal Contract Under Ambiguity

Qizhi Wang


Under ambiguity, where probabilities are unknown, Savage’s model is not applicable for describing a decision maker’s behaviour. Since no valid distribution can be obtained, ambiguity perception and ambiguity attitude are introduced to account for the decision maker’s choice. In this paper, we use the Choquet Expected Utility (CEU) model with neo-additive capacity to study the optimal contract in an employment relationship. We find that 1) optimistic employees could bring more profit to the employer because they are more easily motivated; 2) any optimal contract may fail if ambiguity factors are neglected and 3) ambiguous monitors can be introduced to reduce the negative impact from pessimists. In our model, there is a principal who actually knows the objective success probability of a project and wants to hire agents to exert efforts into it. However, the principal may obtain such objective probability by rule-of-thumb or it is indeed tacit knowledge, which makes her unable to provide solid evidence to convince her employees (agents). Therefore, the employees are not fully convinced by this prior distribution and only put limit confidence in it, which makes it an ambiguous situation for them. Knowing the employees’ ambiguity perception, the principal tries to maximize her expected profit by designing an optimal contract. In this paper, we analyse the consequences of the presence of ambiguity and provide some possible solutions to it. The intuition can be briefly summarised as follows. Firstly, since pessimists consider more about the worst situation which, in this case, is that their efforts do not increase the success probability very significantly, if outcome contingent wage scheme is used, they do not think their efforts can induce sufficient amount of increase in expected wages to cover the costs, so they would provide insufficient efforts. On the contrary, optimists consider more about the best situation and therefore more easily to be motivated. Secondly, as we have shown, personalities (e.g. ambiguity perception and ambiguity attitude) are too important to be omitted in optimal contract. If such factors are ignored by the employer, any contract will be vulnerable to some kinds of people. Thirdly, we propose a possible solution to motivate pessimists. If the employer initiates a regulator department which inspects employees irregularly and punish those who are found shirking, all employees will take such punishment as ambiguous as well. By creating such a threat, pessimists would try to avoid being caught shirking and be motivated consequently. This solution is interesting because more ambiguities are introduced to solve the problems brought by ambiguity. Based on these findings, further discussions are made. We explain why positive attitudes such as optimism are often praised in corporate cultures and provide a potential reason for why some start-ups have satisfying earnings in the beginning but then fail as they grow bigger.

Testing Biseparable Preferences

Veronica Roberta Cappelli; Fabio Maccheroni; Giorgia Romagnoli


A large body of empirical evidence shows violations of expected utility. In response, decision theory and behavioral economics have provided a large variety of non-expected utility theories. However, the existing evidence does not clearly discriminate among such theories. On the gains domain, many of the most well-known non-expected utility models belong to the class of biseparable preferences (Ghirardato and Marinacci, 2001), which includes Rank Dependent Utility models (Choquet Expected Utility, Prospect Theory and Choice-Acclimating Personal Equilibria), models of Disappointment Aversion, Maxmin Expected Utility and its Alpha-maxmin extension. Biseparable preferences satisfy the minimal behavioral restrictions that allow to separate tastes (as captured by a utility function on outcomes) and beliefs (as captured by the willingness to bet on events, often a distorted probability). In this paper we derive a non-parametric procedure for testing the biseparability of preferences hypothesis and we apply it to the results of a preliminary lab experiment. In the data we find little support for the separation of utility and beliefs as characterized by biseparable preferences. On the other hand, the observed behavior can be accommodated by alternative models, such as Smooth Ambiguity and Source Dependent Expected Utility.

Veronica Roberta Cappelli

PhD Student, HEC Paris