Yan Xu

This paper introduces two simple betting mechanisms—top-flop and threshold betting—to elicit unverifiable information from crowds. Agents are offered bets on the rating of an item about which they received a private signal versus that of a random item. We characterize conditions for the chosen bet to reveal the agents’ private signal even if the underlying ratings are biased. We further provide microeconomic foundations of the ratings, which are endogenously determined by the actions of other agents in a game setting. Our mechanisms relax standard assumptions of the literature, such as common prior, and homogeneous and risk neutral agents.

We propose an instrument to measure individuals’ social preferences regarding equity and efficiency behind a veil of ignorance while controlling for idiosyncratic risk preferences. We construct a battery of portfolio and wealth distribution choice problems sharing a common budget set. A given bundle induces the same distribution over an individual’s wealth in both problems. The portfolio choice solely reflects an individual’s risk attitude, providing a benchmark to evaluate whether their wealth distribution choice exhibits equity or efficiency preferring tastes. Experiments show clusters of social preference types, which are unexpectedly independent of risk preferences.

In many economic scenarios, people resort to different tests to obtain information about the payoff-relevant states of the world. This paper studies how people evaluate and choose between useless tests (quacks) and genuinely useful ones (experts). Using a novel graphic experiment, I recover individual preferences over tests. I find people fail to distinguish between experts and quacks. Their quack choices are not driven by standard explanations, including belief updating bias, best-responding bias, and an intrinsic preference for information. The main culprit is the failure in reasoning the contingency value of tests for their decision problems.

This paper studies how reputation schemes build trust and promote cooperative behaviors in an open innovation interaction. An initiator develops a risky project internally or together with a partner. Cooperation is socially beneficial but entails post-cooperation exploitation from the partner; therefore, the initiator turns down the cooperation in the equilibrium. We find the partner’s reputation records, both partially and fully, boost cooperation. Our theoretical investigations show that a reputation scheme serves as (1) an informational device for the initiator to infer the partner’s hidden type (in the reputation consumption stage); and (2) a monitoring device to deter the partner’s undesirable hidden actions (in the reputation production stage). Using a mixed experimental design with two stages and three reputation intensities, we separate the reputation’s production and consumption stages and distinguish its informational and monitoring role. We find the reputation scheme is more effective in breeding cooperation as a monitoring device.

This paper tests the performance of the Bayesian market (Baillon (2017)), a new mechanism that incentivizes truth-telling from crowds. A participant in a Bayesian market trades a belief asset whose value is determined by other participants’ trading positions. In the truth-telling Bayesian Nash equilibrium, a participant will reveal her private signal through buying or short selling an asset when she believes others are also truthful. I create three Bayesian markets in the lab, varying in the belief uncertainties regarding participants’ truthfulness in the market. I find Bayesian markets effective when participants reasonably believe others are truthful. They are less effective when participants’ beliefs are subject to noise and updating biases. A further investigation of participants’ bids and ask prices demonstrates how bubbles are formed and impede the performance of Bayesian markets. With belief uncertainty, participants exhibit under-inference bias or ignorance in processing their private signals. The speculators in the market further amplify the updating bias, creating a speculative trend about the belief asset. This finding provides new insights into the role of belief uncertainty and updating biases in the evolvement of market bubbles.

We study Bayesian persuasion with communication. A sender (she) has partially conflicting interests with a receiver (he) and attempts to influence his action. She commits to an information structure about the state, privately learns a realization of the signal, and sends the receiver a cheap talk or partially verifiable message. The considered communication protocols impose different limits on the sender’s commitment power and the overall information transmission. Our equilibrium analyses predict a negative relationship between persuasiveness and communication credibility, while information transmission is invariant to communication protocols. We designed a new graphic Bayesian persuasion experiment and tested our model in the lab. More than half of our subjects follow the partially revealing mechanics of Bayesian persuasion in the benchmark. They react to different communication protocols qualitatively in line with the theoretical predictions. In addition, we find the sender’s informativeness in persuasion is negatively correlated with her truthfulness in communication, implying a signaling role of information structure. The identified signaling mechanism is new to the literature.

This paper studies people’s preferences for compound lotteries and examines their correlations with risk attitudes and loss aversion. We developed a new graphic experimental tool wherein a subject allocates a budget of state-dependent probabilities and designs his most preferred lottery. The constructed lottery is theoretically equivalent to a mixture of two fixed lotteries, a safer one and a riskier one. We recovered individual preferences over lottery mixtures with a large set of probability budgets. We also varied the lottery outcomes with all gains, avoidable losses, and unavoidable losses. Around half of our participants’ choices revealed a preference for lottery mixtures of gains, violating a broad class of models of risk (such as expected utility or disappoint aversion model) that require the betweenness axiom. A further investigation of the internal structure of mixture choices over the budget sets showed that 70 percent of the lottery mixtures were not monotonic, violating another major class of models that characterize convex preferences for risk (such as rank-dependent, cumulative prospect theory, and cautious expected utility models). Participants were found to be more reluctant towards mixtures involving losses, opting for the safer lottery to avoid losses but choosing the non-mixed risky lottery when losses were unavoidable. Individual loss aversion estimates explain the treatment patterns of non-mixing choices, but not lottery mixtures, nor the estimated risk attitudes towards gains and losses.

This study explores the impact of endogenous information acquisition in the context of public good provision. We examine a two-person threshold public goods game where the contribution threshold can be either high or low (baseline: no information acquisition). Prior to the contribution stage, players have the choice to acquire a 90 percent accurate signal regarding the threshold. Additionally, we manipulate whether players can observe each other’s information acquisition decisions, establishing private and public information acquisition treatments. Our experimental results reveal that the opportunity to acquire private information alone increases the successful provision of public goods by 13 percentage points compared to the baseline treatment. Furthermore, when information acquisition choices are observable, the provision increases by an additional 13 percentage points compared to the private treatment. Further analysis of participants’ strategies shows that players contribute effectively after receiving favorable signals (indicating a low threshold), regardless of information acquisition privacy. However, when players can publicly observe each other’s information acquisition choices, they further self-sort and also contribute more effectively after receiving unfavorable signals (indicating a high threshold).