Abstract: We consider a specialized form of risk management for betting opportunities with low payout frequency, presented in particular for exotic horse race wagering. An optimization problem is developed which limits losing streaks with high probability to the given time horizon of a gambler, which is formulated as a globally solvable mixed integer non-linear program. A case study is conducted using one season of historical horse racing data.
Introduction
In parimutuel betting markets, "exotic" wagers—such as the superfecta, which requires predicting the top four finishers in exact order—represent a high-reward, low-probability frontier. Mathematically, these opportunities mirror "deep out-of-the-money" financial options: they are characterized by frequent small losses punctuated by rare, massive payouts.
For professional betting syndicates and individual gamblers alike, the primary obstacle to long-term success is not merely identifying value, but surviving the inevitable "dry spells." Extended losing streaks exert immense psychological pressure, often leading to "tilt"—a state where emotional frustration causes a bettor to abandon their statistical model in favor of irrational, high-risk decisions. While standard risk management frameworks like Markowitz mean-variance optimization are common in finance, they are ill-suited for this domain because they penalize the high positive skewness that exotic bettors specifically seek. This paper proposes a novel optimization framework designed to tame these losing streaks by incorporating a time-horizon-based risk constraint.
Methodology
The authors develop a risk-management model that maximizes long-term wealth growth while strictly bounding the probability of a losing streak exceeding a user-defined threshold.
1. The Objective Function
The model utilizes a fractional Kelly Criterion strategy. The Kelly Criterion is the industry standard for determining optimal bet sizes to maximize the expected logarithm of wealth, effectively balancing growth against the risk of ruin. By using a fractional approach, the authors provide a tunable parameter to further scale down risk exposure.
2. The Time Horizon Constraint
The core innovation of this research is the introduction of a Chance Constraint based on Value at Risk (VaR). The bettor defines a time horizon, $\tau$ (e.g., 20 races), and a confidence level, $1-\alpha$. The model then ensures that the probability of achieving at least one non-negative return within that $\tau$-race window is at least $1-\alpha$. This mathematically guarantees that losing streaks remain within manageable psychological and financial bounds.
3. Mathematical Transformation
The resulting optimization problem is initially a non-convex Mixed Integer Non-Linear Program (MINLP), which is notoriously difficult to solve globally. To overcome this, the authors employ a 1-to-1 mapping ($y_s = \log(x_s + Q_s)$) to transform the problem into a convex program. This transformation ensures that the model can be solved efficiently and globally, a requirement for real-time betting environments.
4. Probability Estimation
To accurately model the probability of specific race outcomes, the study utilizes the Lo & Bacon-Shone model. This approach is superior to the standard Harville model as it accounts for the "favorite-longshot bias"—a well-documented market inefficiency where bettors over-bet longshots and under-bet favorites.
Data and Empirical Analysis
The framework was tested using a real-world dataset from the 2013–2014 racing season at Flamboro Downs in Hamilton, Ontario. The dataset comprised 1,168 races, split into a 70% calibration set (to tune probability and payout models) and a 30% out-of-sample test set (350 races). Data points included handicapping factors, final win odds, and public pool sizes.
Key Results
The model was evaluated across different time horizons ($\tau$). The results demonstrate a clear trade-off between the stability of the strategy and its total profitability:
| Time Horizon ($\tau$) | Max Realized Losing Streak | Total Return (%) | Races Bet On |
|---|---|---|---|
| $\infty$ (Unconstrained) | 52 | 7.8% | 224 |
| 40 | 40 | 4.5% | 163 |
| 30 | 27 | 4.0% | 123 |
| 20 | 19 | 1.0% | 61 |
As the constraint becomes more stringent (lower $\tau$), the model becomes significantly more selective. At $\tau=20$, the model only placed bets on 61 races compared to 224 in the unconstrained version. However, the average bet size increased from $10.80 to $37.90. This suggests that to satisfy the risk constraint, the model waits for high-conviction opportunities and bets more aggressively on them to ensure a win offsets the preceding losses.
Discussion and Conclusion
The study demonstrates that incorporating a time-horizon constraint successfully "tames" the volatility of exotic wagering. By reducing the maximum losing streak from 52 races to just 19 (at $\tau=20$), the model provides a "psychological buffer," allowing the bettor to view dry spells as a planned component of the strategy rather than a failure.
While the total return decreases as the constraint tightens, the gain in financial and emotional stability is substantial. Future work could improve the model by moving from point estimates to stochastic programming to account for uncertainty in pool sizes and outcome probabilities. Furthermore, while applied here to horse racing, this methodology holds promise for any high-payout/low-probability investment field, including venture capital and specialized derivative trading.
Summary updated