The SABR model is a widely used mathematical framework in finance for describing how asset prices (like interest rates or stock prices) and their volatility evolve together over time. A key feature of the model is a parameter that can force prices to be absorbed at zero, meaning they can hit zero and stay there. Understanding the "right wing" of the model means figuring out the probability of extremely large price movements, which matters for pricing options far out of the money. This paper works out precisely how rare those extreme outcomes are, specifically the rate at which the probability of a very large price move decays as the move gets larger and larger.
The main finding is that the answer depends critically on the correlation between price movements and volatility movements. When correlation is zero or positive, the decay rate matches what you would expect from the model's underlying geometry, and an earlier conjecture by a prominent researcher (Henry-Labordere) turns out to be correct. But when correlation is negative, something more subtle happens: the random path that would produce an extreme price move is additionally constrained to keep the asset price above zero, and satisfying that survival requirement imposes an extra cost. This extra cost changes the decay rate and means the implied volatility in the extreme wing is lower than previously conjectured. So the conjecture is confirmed for non-negative correlation and shown to be wrong for every negative correlation.
The paper also addresses a second conjecture about how to measure the "distance" to an extreme outcome in the model's geometric framework. For negative correlation, the naive distance measure fails, and the authors identify the correct replacement: you must measure distance while staying within the region where the price remains positive, using the natural curved geometry of the model. The mathematical proofs avoid sophisticated heat-kernel machinery and instead combine more elementary tools, including a clever sequence of transformations of the stochastic process, bounds using Gaussian moving barriers, and changes of probability measure.