No sure loss.
But is it coherent?
A lower prevision is your supremum acceptable buying price for a gamble. Compare one-sided buying commitments with precise, two-sided prices. Set four prices, then discover whether they fit together.
Set your buying prices
Three mutually exclusive outcomes: A, B, C. Each ticket pays €1 if its event occurs, and €0 otherwise.
Which distributions respect your prices?
Each point is a distribution (pA, pB, pC). The shaded region satisfies Ep[f] ≥ P̲(f) for every assessed gamble.
Try a portfolio
The search resets the sale to “Nothing” and normalizes total buying stakes to 1. A negative payoff in every state certifies sure loss.
Coherence means every price is tight
Natural extension is the smallest expectation over M. A gap reveals a buying price already implied by your other commitments.
Why these are different requirements
Avoiding sure loss: every nonnegative buying portfolio has a nonnegative payoff in at least one state.
for all λᵢ ≥ 0
Coherence: even when you sell one assessed gamble at its quoted price, buying other gambles cannot force a strictly negative payoff in every state.
On this finite domain, coherence is equivalent to P̲(f) = minp∈M Ep[f] for each assessed f. Avoiding sure loss only requires M ≠ ∅.
These prices are limiting buying prices. Strictly acceptable transactions can use prices just below them; strict sure-loss certificates persist under sufficiently small perturbations.
A two-minute experiment
Start with “Avoids sure loss only”: A and B each cost 0.30, but A ∪ B costs only 0.40. Buying A and B implies a price of 0.60 for their sum. Selling A ∪ B at 0.40 while buying A and B loses 0.20 in every state. This violates coherence, even though buying alone cannot force a sure loss.
Then choose “Sure loss” and find a portfolio. Finally, try “Complete ignorance”: coherent prices can leave the entire probability simplex open.