Bettors' Curse

How to cite

Sarwar, Rubayat; Bayer, Ralph-Christopher. 2026. "Bettors' Curse." AEA RCT Registry. AEARCTR-0018433. https://doi.org/10.1257/rct.18433-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Rubayat Sarwar* , Ralph-Christopher Bayer

* Primary investigator

Registry ID
AEARCTR-0018433
DOI
https://doi.org/10.1257/rct.18433-1.0
First registered
Last updated
Status
In development
Trial dates
2026-04-26 – 2026-12-31
Countries
Australia ()
JEL codes
D81 (Information, Knowledge, and Uncertainty: Criteria for Decision-Making under Risk and Uncertainty), C91 (Design of Experiments: Laboratory, Individual Behavior)
Registry record
http://www.socialscienceregistry.org/trials/18433

Abstract

We define the bettor’s curse as the tendency for individual bettors to overvalue their private signals about the likelihood of events relative to the public information contained in market odds. In this study, we examine the extent to which betting behaviour is driven by the bettors’ curse.

Under Bayesian updating, private signals mechanically update posterior beliefs. Given correct priors and signal precision, posterior beliefs are unbiased. If market odds fully aggregate the private information of all bettors, then all relevant information is summarised by the posterior probability. Consequently, conditional on the posterior, betting behaviour should be identical regardless of whether a bettor observes the private signal or not. While private signals may be correlated with the true probability, they are not perfectly correlated and therefore not fully informative. In contrast, market odds—if efficient—perfectly reflect the true probability (up to the bookmaker’s margin).

In our experimental setting, individuals are not required to acquire private information themselves. Instead, we exogenously provide them with a private signal with known precision. Posterior beliefs are then computed based on all these signals and signal precision according to Bayesian rule and presented directly to participants in the form of market odds. This design removes the need for participants to perform belief updating themselves. Under the Bayesian benchmark, if individuals solely depend on the posterior probability, irrespective of the signal observed, their behaviour should be identical.

H1 (Bettors’ Curse): We hypothesise that when individuals observe private signals, they overweight these signals and perceive posterior probabilities as higher or lower than they objectively are, which lead their betting behaviour deviating from the Bayesian benchmark. On average, they place higher bets when they observe a private signal than when no signal is observed.

H2 (Directional Betting): If bettors overweight their private signals, betting behaviour should move in the direction indicated by the signal. In our experiment, odds are fair, so the expected value of betting on either outcome is zero, and there is no ex ante reason to favour one outcome over the other. However, if private signals are overweighted, participants will systematically bet in favour of the outcome suggested by their signal. Such directional behaviour indicates that bettors treat their private information as more informative than the information already incorporated into market odds.

H3 (Sole Dependence on Signals): If bettors fully disregard the information contained in market odds and believe that probabilities depend solely on their private signals, then, conditional on observing the same signal, they will place higher bets when the odds are more favourable (i.e., when potential payouts are higher).

Experimental design

We will conduct an incentivised laboratory experiment in which participants make betting decisions in 15 rounds of card-drawing games. For each round, one card was pre-drawn at random from the deck of cards. The drawing process was recorded, showing a dealer shuffling the cards and a person drawing one card randomly. In each round, participants receive a $10 endowment and choose how much to bet on an outcome, with stakes ranging from $0 to $10.

Treatments vary whether a participant observes only public market odds (baseline treatment) or also receives a private signal with known precision.

The baseline treatment is a no-signal treatment, where participants observe only the publicly available betting information: the market odds associated with the possible outcomes. No private signal is provided. Since participants do not observe additional private information, their betting decisions should depend only on the public information embodied in market odds and their individual risk preferences.

In the treatment with private signals, participants observe the same public market information (market odds) as in the no-signal treatment, but they also receive a private signal regarding the likely outcome of the card-drawing game. The signal is exogenously provided by the experimenter and has a known precision. Recall that posterior beliefs are then computed based on the signals of all bettors involved and signal precision according to Bayesian rule and presented directly to participants in the form of market odds. The purpose of these treatments is to test whether participants respond solely to the information contained in market odds—which aggregate private signals of all bettors—or whether they place excessive weight on their own private signal.

In two treatments (baseline and with private signals), the prior probability of the two mutually exclusive outcomes is set at 50–50. In two additional treatments, the prior probabilities are set at 75–25. This variation introduces additional heterogeneity in the betting environment and allows us to examine behaviour under different priors.

The experimental design allows us to compare betting behaviour across treatments that vary the availability of private signals while holding constant the underlying structure of the betting environment. By controlling priors, signal precision, and market odds, the design provides a clear benchmark for Bayesian updating and enables us to identify deviations consistent with the bettors’ curse.

All four treatments will present the payout of each dollar bet (odds) explicitly. In our experiment, as the card-drawing games are objective, the value of probability information given in the experiment is free from human judgment. Each participant will undergo only one of these four treatments, making this experiment a cross-subject design. For each round, participants will be given a virtual endowment of 10 dollars in the programmed experiment in Z-tree, from which they can allocate their stakes between 0 and 10 dollars. The difference between their endowment and the betting stake will be part of their earnings from the experiment. The other part will be the payout from the bet, received only if they win.

Design details

Randomization method
We randomly assign each experimental session to one of the 4 treatments. Participants voluntarily sign up to attend in our sessions. But at the time of registration, they do not know the treatment they will receive.
Randomization unit
Session
Clusters
As one independent observation will be a vector of 15 betting decisions by an individual participant, we will do clustering at individual level. The planned number of clusters is 200 and 15 observations per cluster. This approach captures the similarities in betting behavior across participants.
Observations
200 participants x 15 betting decisions = 3,000 betting decisions (observations)
Arms
Each treatment has 50 participants, and with 4 treatments, the total number of participants is 4 × 50 = 200.
Primary outcomes
The key outcome variable is bet size at the participant level. Each participant will be asked to decide how much of the provided endowment they would like to bet on the outcome of fifteen (15) card-drawing games. Thus, one independent observation within a treatment corresponds to a vector of 15 betting decisions made by a single participant, with each stake ranging from $0 to $10. We will then compare these decisions across treatment with and without private signals.

Cited by

No citation data available for this registration yet.

Keywords