We study a single-queue system in which servers have discretion over the
effort with which to process orders that arrive stochastically and the
compensation is based on the total number of orders processed by the group.
We show theoretically that in the implied stochastic dynamic game, although
each individual server has a short-term incentive to free ride and provide
low effort, cooperation in the form of high effort can be sustained in
equilibrium if the interactions are long term and servers are patient enough.
In addition, we show that queue visibility plays an important role in the
type of strategies that can sustain cooperation. We conduct a controlled
lab experiment to test the theoretical predictions and find that when the
queue is visible, human subjects cooperate when it is long and defect when
it is short. We also find that, although possible theoretically, in practice,
cooperation is hard to achieve when the queue is not visible. We discuss
implications for managers and firms that are trying to improve service systems.
The indefinitely repeated prisoner's dilemma (IRPD) captures the trade-off between
the short-term payoff from exploiting economic partners and the long-term
gain from building successful relationships. We aim to understand how people form
and use beliefs about others in the IRPD. To do so, we elicit beliefs about the
supergame strategies chosen by others. We find that heterogeneity in beliefs and
changes in beliefs with experience are central to understanding behavior and learning
in the IRPD. Beliefs strongly predict cooperation, initial beliefs match behavior
quite well, most subjects choose strategies that perform well given their beliefs, and
beliefs respond to experience while becoming more accurate over time. Furthermore,
experience affects both transitions between strategies and cooperation. Finally, we
uncover a novel mechanism by which trusting subjects learn to cooperate through
their interaction with experience, which helps to explain how trust underpins successful
economic exchanges.
We investigate the role of performance feedback, in the form of a public leaderboard, in a sequential-sampling contest with costly observations.
The player whose sequential random sample contains the observation with the highest value wins the contest and obtains a prize with a fixed value.
We show theoretically that in the subgame perfect equilibrium of contests with a fixed ending date (i.e., finite horizon), providing public performance
feedback may result in fewer expected observations and a lower expected value of the winning observation. We conduct a controlled laboratory experiment
to test the theoretical predictions, and find that the experimental results largely support the theory. In addition, we investigate how individual
characteristics affect competitive sequential-sampling activity. We find that risk aversion is a significant predictor of behavior both with and
without leaderboard feedback, and that the direction of this effect is consistent with the theoretical predictions.
Social and political inequality among individuals is a common driving force behind the breakdown in cooperation. In this paper, we theoretically and experimentally study cooperation among individuals faced with a sequence of collective-action problems in which the benefits to cooperation are divided according to political power that is obtained through a contest. We have three main results. First, we find that cooperation predictably responds to the fundamental parameters of the collective-action problem. Specifically, it is increasing in the benefit to cooperation and how much benefit is gained from partial group cooperation, and decreasing in the number of players. Second, we find that when players are unrestricted in their expenditures in the contest, cooperation is much lower than when expenditures are set to a specific proportion of earnings. Finally, we find that individual norms and beliefs account for a substantial proportion of explained variance in individuals' decisions to cooperate.
Identifying the strategies that are played is critical to understanding behavior in repeated games.
This process is difficult because only choices (not strategies) are observable.
Recently, a debate has emerged regarding whether subjects play mixed strategies
in the indefinitely repeated prisoner's dilemma. We use an experimental approach
to elicit mixed strategies from human subjects, thereby providing direct empirical
evidence. We find that a majority of subjects use mixed strategies. However,
the data also suggest subjects' strategies are becoming less mixed over time,
and move toward three focal pure-strategies: Tit-For-Tat, Grim-Trigger, and Always
Defect. We perform an econometric analysis to provide support that the strategies
identified in our experiment are widely used.
Recent advances in technology have reduced frictions in various markets.
In this research, we specifically investigate the role of frictions in
determining the efficiency and biding behavior in a Generalized Second
Price Auction (GSP) -- the most preferred mechanism for sponsored
search advertisements. First, we simulate computational agents in
the GSP auction setting, and obtain predictions for the metrics of
interest. Second, we test these predictions by conducting a human
subject experiment. We find that, contrary to the theoretical prediction,
the lower valued advertisers (who do not win the auction) substantially
overbid. Moreover, we find that the presence of market frictions
moderates this phenomenon and results in higher allocative efficiency.
These results have implications for policymakers and auction platform
managers in designing incentives for more efficient auctions. Finally,
after establishing the validity of our computational model, we simulate
counterfactuals that provide additional insights into the role that
frictions play in the markets that are not feasible (or practical) to
investigate with human-subject experiments.
We run an experiment to compare belief formation and learning under compound
risk and under ambiguity at the individual level. We estimate a four-type
mixture model assuming that subjects may either follow Bayes Rule or behave
according to the multiple priors model of Epstein and Schneider (2007) for
each of type of uncertainty. Our results indicate that majority of subjects
are Bayesian, both under compound risk and under ambiguity, while the second
most frequent type are subjects that are Bayesian under compound risk but
who use multiple priors model of learning under ambiguity. In addition,
we find strong evidence against a common assumption that participants'
initial beliefs (and priors) are consistent with information provided
about the uncertain process.
We examine how individuals behave in the face of failure.
We investigate the willingness of individuals to persist at exploration
in the face of failure. Prior research suggests that a "tolerance
for failure" may motivate individuals to select more exploratory
courses of action. Little is known, however, about how individuals
persist at exploration when confronted by prolonged periods of negative
feedback. To examine this question, we design a two-dimensional maze
game to capture the essential trade-offs between exploration and
exploitation, develop predictions for the game using computational
models of reinforcement learning, and run a series of randomized
experiments with human subjects in the game. Our methods extend
beyond two-period models of decision-making under uncertainty
to account for repeated behavior in longer-running, dynamic
contexts. Our results suggest that individuals explore more when
they are reminded of the incremental cost of their actions,
a result that extends prior research on loss aversion and
prospect theory. The results also suggest that behavioral
factors may complicate innovation strategies that tolerate
failure. We discuss implications for future research and managers.
We study the evolution of cooperation and strategies in the indefinitely repeated
prisoner's dilemma when it is costly for players to change their strategies. In standard
repeated-game experiments, players directly choose an action in each period.
Our experimental interface allows subjects to design a comprehensive strategy that
then selects actions for them in every period. Subjects are also able to adjust their
strategies continuously within the supergames. We conduct lab experiments and
find that cooperation is lower when strategy adjustment is costly than when it is
not. The main difference is due to the evolution of cooperative strategies when it is
costless to adjust strategies within supergames and to the prevalence of less cooperative
strategies when it is costly to do so. These results highlight that within-game
experimentation and learning about strategies is critical to the rise of cooperative
behavior. We provide simulations based on an evolutionary algorithm to support
this result.
We propose a new approach for running lab experiments on indefinitely repeated games with high continuation probability.
The approach has two main advantages. First, it allows us to run multiple long repeated games per session. Second, it
allows us to incorporate the strategy method with minimal restrictions on the types of strategies that can be constructed.
This gives us insight into what happens in long repeated games and into the types of strategies that subjects use. We report
results obtained from the indefinitely repeated prisoner's dilemma with a continuation probability of δ = 0.99.
We find rates of cooperation that are lower than expected, given that the indefinitely repeated games are long and the
setting is approaching the continuous time form. However, when we analyze the constructed strategies our results are
largely similar to those found in the literature, specifically that the most common strategies are memory-1 strategies
such as Tit-For-Tat, Grim Trigger, and Always Defect.
Using laboratory experiments, we show that subjective valuations of projects with multiple risks
are highly sensitive to the format of risk specification. In the experiments, participants
considered a project that had two risks and would be considered a success
when favorable outcomes occurred on both risks. Participants were provided with
the probabilities of success for each risk in two different specifications. In the
reduced specification, each probability was directly specified. In the compound
specification, each probability was specified as a two-point distribution. The data showed that
under the reduced specification, decision makers' perceived value of the project was higher than
its true value, due to conjunctive probability bias in which decision makers overestimated the
conjunctive probabilities. Under compound specification, however, judgmental valuations were
subject to two biases that acted in opposite directions and, as a net outcome, managerial
valuations were closer to risk-neutral valuations.
Using behavioral experiments, we study the impact of queue design on worker productivity in service systems that involve human servers. Specifically, we consider two queue design features: queue structure, which can either be parallel queues (multiple queues with a dedicated server per queue) or a single queue (a pooled queue served by multiple servers); and queue-length visibility, which can provide either full or blocked visibility. We find that 1) the single-queue structure slows down the servers, illustrating a drawback of pooling; and 2) poor visibility of the queue length slows down the servers; however, this effect may be mitigated, or even reversed, by pay schemes that incentivize the servers for fast performance. We provide additional managerial insights by isolating two behavioral drivers behind these results: task interdependence and saliency of feedback.
We design and conduct an economic experiment to investigate the learning process of the agents
under compound risk and under ambiguity. We gather data for subjects choosing between
lotteries involving risky and ambiguous urns. Decisions are made in conjunction with a sequence
of random draws with replacement, allowing us to estimate the beliefs of the agents at different
moments in time. For each of the urn types we estimate the initial prior and a general updating
model for which the standard Bayesian updating model is a particular case. Our findings suggest
an important difference in updating behavior between risky and ambiguous environments.
Specifically, after controlling for the initial prior, when updating under ambiguity subjects
significantly overweight the new signal, while when updating under compound risk subjects are
essentially Bayesian.
Working Papers
On the Emergence of International Currencies: An Experimental Approach
(with Marcos Cardozo, and Cathy Zhang)
R&R at
Journal of Economic Behavior and Organization
We integrate theory and experimental evidence to study the emergence of different international monetary arrangements based on the circulation of two intrinsically worthless fiat currencies as media of exchange. Our framework is based on a two-country, two-currency search model where the value of each currency is jointly determined by private agents' decisions and monetary policy formalized as changes in a country's money growth rate. Results from the experiments indicate subjects coordinate on a regime where both currencies are accepted even when other regimes are theoretical possibilities. At the same time, we find the emergence of international currency depends on relative inflation rates where sellers tend to reject payment with a more inflationary foreign currency.
Cooperation under the Shadow of Political Inequality
(with Xinxin Lyu, Denis Tverskoi, and Sergey Gavrilets)
accepted at
Journal of Economic Dynamics and Control
We study cooperation among individuals and groups facing a dynamic social dilemma in which the benefits of cooperation are divided according to political power obtained in a contest. The main theoretical and experimental results focus on the role of the incumbency advantage. Specifically, an incumbency advantage in the political contest leads to a rapid breakdown of cooperation in the social dilemma. In addition, we provide simulations based on the individual evolutionary learning model of Arifovic and Ledyard (2012) to shed light on the difference between the behavior of individuals and groups.
Research Grants
USDA/NIFA AFRI Foundational and Applied Science Program: Markets and Trade. 2023 -- 2026.
"Behavioral Economics in Food and Environmental Policy: A Principal-Agent, Machine Learning, and Experimental Economics Approach" (with Steven Y. Wu)
The Blake Family Fund for Ethics, Leadership and Governance. 2021 -- 2023.
"Fairness in Machine Learning -- An Experimental Analysis"
(with Matthew Hashim, Karthik Kannan, Warut Khern-am-nuai, and Hajime Shimao)
The Department of Defense DEPSCoR Award. 2021 -- 2024.
"Dynamics of beliefs, power, and inequality in within- and between-group cooperation and conflict"
(with Sergey Gavrilets)
The Blake Family Fund for Ethics, Leadership and Governance. 2018 -- 2020.
"Collusion, beliefs, and personal characteristics"
(with David Gill)