When to Quit Gambling, if You Must!

Sang Hu, Jan Obłój, Xun Yu ZhouarXiv 2021 · DOI · arXiv

Abstract

We develop an approach to solve Barberis (2012)'s casino gambling model in which a gambler whose preferences are specified by the cumulative prospect theory (CPT) must decide when to stop gambling by a prescribed deadline. We assume that the gambler can assist their decision using an independent randomization, and explain why it is a reasonable assumption. The problem is inherently time-inconsistent due to the probability weighting in CPT, and we study both precommitted and naive stopping strategies. We turn the original problem into a computationally tractable mathematical program, based on which we derive an optimal precommitted rule which is randomized and Markovian. The analytical treatment enables us to make several predictions regarding a gambler's behavior, including that with randomization they may enter the casino even when allowed to play only once, that whether they will play longer once they are granted more bets depends on whether they are in a gain or at a loss, and that it is prevalent that a naivite never stops loss.

Introduction

In 2012, Nicholas Barberis proposed a model to explain the enduring mystery of casino gambling: why do individuals participate in games with negative or neutral expected values? Using Cumulative Prospect Theory (CPT), Barberis demonstrated that human risk perception—specifically the tendency to overweight small probabilities of large gains and an "S-shaped" utility function—can make a fair bet appear highly attractive.

However, CPT introduces a significant challenge: time inconsistency. Because our perception of probability and value shifts as we accumulate wins or losses, a strategy that appears optimal at the start of the night may no longer seem so halfway through. Solving this problem for a finite number of bets ($T$) has historically been a computational nightmare. Previous attempts relied on exhaustive searches that scaled exponentially; for instance, calculating an optimal strategy for just 8 bets could take nearly a year on a standard desktop. This paper introduces a framework that transforms this "difficult" problem into a manageable mathematical program.

Methodological Innovations

The authors overcome the computational barriers of the Barberis model by introducing two primary shifts in perspective:

1. The Role of Randomization

The framework allows the gambler to use an independent randomization (such as a coin flip) to assist in their decision to stop. While this serves as a technical mechanism to "convexify" the underlying mathematics, the authors argue it is a realistic behavioral assumption. In complex decision-making environments, individuals often delegate difficult choices to external randomizers, such as apps or traditional methods like divination sticks.

2. Distribution-First Optimization

Rather than attempting to find the optimal time to stop directly, the authors first search for the optimal probability distribution of final winnings. By utilizing a discrete-time version of the Skorokhod embedding theorem (specifically Root stopping times), they can work backward from an ideal distribution to find the randomized stopping rule that generates it. This reformulation allows the number of constraints in the optimization problem to scale linearly with $T$, rather than exponentially.

Key Insights and Behavioral Predictions

The analytical treatment of this model yields several striking predictions that contradict classical economic intuition:

  • The Single-Bet Entry: Without randomization, a CPT gambler might require a long horizon (e.g., $T \ge 26$) to justify entering a casino. However, with the ability to randomize, they can "design" a skewed outcome immediately, making them willing to enter even if they are only permitted a single round ($T=1$).
  • Precommitted vs. Naive Strategies:
    • Precommitted gamblers (those who can stick to an initial plan) typically follow a "loss-exit" strategy. They allow gains to run to build a "jackpot" distribution but quit once they hit a specific loss threshold.
  • Naive gamblers (those who re-optimize at every step) exhibit "gain-exit" behavior. They tend to cash out small wins too early while gambling "until the bitter end" on losses. This behavior is a direct mathematical manifestation of the disposition effect observed in financial markets.
  • The Paradox of Time: Interestingly, extending a gambler's deadline does not always result in longer play. If a gambler is currently at a loss, the fear of an even larger, overweighted loss often prompts them to quit immediately if they are granted more rounds.

Results and Computational Efficiency

The proposed algorithm represents a massive leap in efficiency. While an exhaustive search for $T=50$ would be impossible for modern computers, this framework solves the model in approximately 7 hours. Furthermore, the authors provide a theoretical bridge between finite and infinite horizons, proving that as $T$ increases, the value of the finite-horizon model converges to the infinite-horizon version.

Crucially, the paper confirms that for naive gamblers, it is mathematically optimal to stop when in a gain at the penultimate step ($T-1$) but to continue when in a loss. This reinforces the idea that "gambling until the bitter end" is a prevalent feature of naiveté in discrete-time models.

Conclusion and Future Work

This research provides a computationally tractable way to study time-inconsistent stopping problems under CPT. While the current model assumes "fair" bets to isolate psychological drivers, future work could incorporate the "house edge" found in real-world casinos. Additionally, the authors suggest that further exploration into "sophisticated" gamblers—those who recognize their own time-inconsistency and plan accordingly—remains a fertile ground for understanding the boundaries of rational and irrational gambling behavior.

Summary updated

Abstract

We develop an approach to solve Barberis (2012)'s casino gambling model in which a gambler whose preferences are specified by the cumulative prospect theory (CPT) must decide when to stop gambling by a prescribed deadline. We assume that the gambler can assist their decision using an independent randomization, and explain why it is a reasonable assumption. The problem is inherently time-inconsistent due to the probability weighting in CPT, and we study both precommitted and naive stopping strategies. We turn the original problem into a computationally tractable mathematical program, based on which we derive an optimal precommitted rule which is randomized and Markovian. The analytical treatment enables us to make several predictions regarding a gambler's behavior, including that with randomization they may enter the casino even when allowed to play only once, that whether they will play longer once they are granted more bets depends on whether they are in a gain or at a loss, and that it is prevalent that a naivite never stops loss.
BibTeX
@misc{hu2021when,
  author = {Sang Hu and Jan Obłój and Xun Yu Zhou},
  title = {{When to Quit Gambling, if You Must!}},
  year = {2021},
  eprint = {2102.03157},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2102.03157},
}
Figure