A betting interpretation for probabilities and Dempster-Shafer degrees of belief

There are at least two ways to interpret numerical degrees of belief in terms of betting: (1) you can offer to bet at the odds defined by the degrees of belief, or (2) you can judge that a strategy for taking advantage of such betting offers will not multiply the capital it risks by a large factor. Both interpretations can be applied to ordinary additive probabilities and used to justify updating by conditioning. Only the second can be applied to Dempster-Shafer degrees of belief and used to justify Dempster’s rule of combination.

💡 Research Summary

The paper investigates two distinct ways of interpreting numerical degrees of belief—both ordinary additive probabilities and the belief functions of Dempster‑Shafer theory—through the lens of betting. The first interpretation is the classic “betting‑odds” view: a degree of belief p for an event A is taken as the odds at which one would be willing to bet, i.e., a fair bet pays p/(1‑p) to 1 if A occurs. Under this view, the coherence condition (no betting strategy can guarantee a positive expected gain) is equivalent to the Kolmogorov axioms, and the usual Bayesian conditioning rule follows directly because updating simply replaces the odds with conditional odds.

The second interpretation, introduced as “betting safety,” does not require the odds themselves to be offered. Instead it asks whether a given set of odds can be exploited by any admissible betting strategy to multiply the capital at risk by an arbitrarily large factor. If no such strategy exists, the odds are deemed safe; otherwise they are unsafe. This safety condition can be expressed mathematically as a bound on the supremum of the expected growth rate of any self‑financing portfolio built from the offered bets. Crucially, this notion does not rely on additivity and therefore can be applied to the interval‑valued belief/plausibility pair (Bel, Pl) of Dempster‑Shafer theory.

In Dempster‑Shafer theory a basic probability assignment m assigns mass to subsets of a frame Θ; Bel(A) = Σ_{B⊆A} m(B) is a lower bound on the support for A, while Pl(A) = Σ_{B∩A≠∅} m(B) is an upper bound. The paper interprets Bel(A) as the minimal safe payoff that a bettor can guarantee when betting on A, and Pl(A) as the maximal payoff that cannot be safely exploited. Thus the belief–plausibility interval itself is a “safe betting interval.”

With this interpretation the Dempster rule of combination acquires a natural betting‑safety justification. Suppose two independent bodies of evidence provide basic assignments m₁ and m₂, each defining its own safe interval for every proposition. The Dempster combination Bel⊕(A) is obtained by intersecting the two intervals and taking the lower endpoint; Pl⊕(A) is the upper endpoint. This operation guarantees that after combination no betting strategy can achieve unbounded capital growth, i.e., the safety condition is preserved. In other words, Dempster’s rule selects the most conservative (lowest) safe odds for belief and the most liberal (highest) safe odds for plausibility, ensuring that the combined evidence remains coherent in the betting‑safety sense.

The paper proves that when Bel = Pl (the special case where the belief function collapses to a precise probability), the two interpretations coincide: the betting‑odds view and the betting‑safety view both reduce to the standard coherence condition, and Dempster’s rule reduces to ordinary multiplication of probabilities. Hence the safety framework genuinely generalises the classical betting interpretation.

Beyond the theoretical equivalence, the authors discuss practical implications. The betting‑odds interpretation offers an intuitive justification for Bayesian updating in expert systems, while the safety interpretation provides a principled way to prevent over‑confident belief updates in Dempster‑Shafer based fusion systems. In fields such as sensor fusion, risk management, and AI uncertainty modelling, the safety criterion can be used to design fusion algorithms that automatically guard against evidence that would otherwise allow a gambler to exploit the system for unbounded profit.

In summary, the paper contributes a unified betting‑based perspective that bridges probability theory and Dempster‑Shafer belief functions. By distinguishing between offering odds (the classic view) and guaranteeing that no exploitable betting strategy exists (the safety view), it shows how conditioning and Dempster’s combination can each be justified on economic grounds. This dual interpretation enriches the philosophical foundations of uncertainty quantification and suggests concrete, safety‑oriented design principles for systems that reason under incomplete or ambiguous information.