Working papers with abstracts

Preferences over Consumption and Status (2007)
download

Experimental evidence suggests that individual consumption has not only personal value but also enters the social part of the utility. Existing models of interdependent preferences make ad hoc parametric assumptions about the nature of this duality. This creates a problem of experimental identification of preferences since without such assumptions it is impossible to distinguish whether consumption or social concerns are driving the behavior. Given observed choice, the axiomatic model of preferences in this paper makes it possible to unambiguously determine personal and social utility without any assumptions about their relationship. The unique separation can be achieved only if the individual choices in different subgroups of other people are available. Preferences over consumption and status are used as an example to demonstrate how the utility is constructed. The model shows what kind of information about choice is needed to experimentally determine the nature of social preferences without making restrictive assumptions. This can help to estimate whether personal consumption or social value is more important in economic decisions.

Competition with Skill and Luck: Behavioral and fMRI Experiments (with Aldo Rustichini, 2007)

Many decisions that people make are driven by the concerns for social ranking. We study experimentally how strong the social ranking preferences are and what characteristics of others influence the ranking. In the behavioral experiment the subjects play two games against the computer: a game of skill and a game of luck. After each game the participants observe the winnings of everybody in the group. Each subject has a possibility to reduce the winnings of one other person at a cost to himself. We find that the majority of subjects use this costly option. Moreover, the decisions to subtract money depend on whether the game of skill or luck was played. The pattern of subtractions suggests that winnings made with skill are used as a proxy for social significance and are envied, whereas money won by luck do not convey such a signal. The same games are studied in the fMRI experiment. There are three participants: one person inside the scanner and two people outside. In the game of skill the activations in Orbitofrontal Cortex and Nucleus Accumbens, the areas known to represent reward, reflect the relative position of the subject: the higher the relative standing is the more rewarding it is felt. Relative position after the game of luck does not correlate with activation in the reward centers. This provides more evidence that skill is considered a socially salient factor whereas luck is not.

I'll Cross that Bridge When I Come to It:
Backward Induction as a Cognitive Process
(with Uri Gneezy and Aldo Rustichini, 2007)

We study experimentally how subjects plan ahead when they make sequential decisions. We use a simple combinatorial game of perfect information, with two players alternating, and with a winner and a loser. Subjects in the early stages follow the forward induction analysis, thinking about their possible moves, the subsequent possible moves of the other, and so on. Motivated by the experience of the early defeats they suddenly switch their mode of analysis to backward induction. This pattern is common to all subjects, although the change may occur at different times. This way of learning the solution of the game is different from a complete backward induction procedure: players are unable or unwilling to replace a sub-game with payoff defined by the optimal strategies in the sub-game. Our results indicate why people may find it difficult to plan their financial decisions effectively.

Backward Induction as a Cognitive Process:
fMRI Experiment
(with Aldo Rustichini and Ovidiu Lungu, 2007)

Backward Induction (BI) is a process of solving extensive form games of perfect information that substitutes sub-games with equilibrium payoffs starting from the end of the game. We check experimentally whether the process followed by subjects to discover the solution of a game is based on BI. We study behavior in a simple combinatorial game with two players with alternating moves, and with a winner and a loser. In this game the winning strategy does not depend on the actions of the other player, so we can separate the effect of the depth of reasoning of the subjects from that of their beliefs. The analysis of their moves shows that subjects switch from a forward induction to backward induction, so BI is the final mode of the learning process. The response times show that the BI is acquired one step at the time, rather than in a single step. Brain imaging analysis explains why this happens: we find, among other regions, Putamen, Insula and BA47 activation during the learning of the steps close to the end of the game. However, the activation during the early stages of the game suggests that the subjects do not replace the sub-games with payoff defined by the optimal strategy learned in the past. Rather they analyze the game forward each time to check whether the optimal strategy still works.

Nonprobabilistic Decision Making with Memory Constraints (2005) download

In the model of choice, studied in this paper, the decision maker chooses the actions non-probabilistically in each period (Sarin and Vahid, 1999; Sarin, 2000). The action is chosen if it yields the biggest payoff according to the decision maker’s subjective assessment. Decision maker knows nothing about the process that generates the payoffs. If the decision maker remembers only recent payoffs, she converges to the maximin action. If she remembers all past payoffs, the maximal expected payoff action is chosen. These results hold for any possible dynamics of weights and are robust against the mistakes. The estimates of the rate of convergence reveal that in some important cases the convergence to the asymptotic behavior can take extremely long time. The model suggests simple experimental test of the way people memorize past experiences: if any weighted procedure is actually involved, it can possibly generate only two distinct modes of behavior.