The central question this paper addresses is deceptively simple: when is a certain fundamental mathematical object, called a stochastic exponential, a true martingale? Martingales are stochastic processes with no systematic drift, meaning the best prediction of their future value is always their current value. Stochastic exponentials appear constantly in financial mathematics, particularly when changing probability measures to price derivatives, so knowing exactly when they behave as true martingales versus merely local martingales has real practical stakes. The difficulty is that stochastic exponentials driven by multi-dimensional diffusions (think of several correlated random processes evolving together) can fail to be true martingales in subtle ways that are hard to detect, and prior results offered only sufficient conditions or handled only simple one-dimensional cases.
The paper's main contribution is a complete characterization, meaning conditions that are both necessary and sufficient, covering the full multi-dimensional setting. The authors analyze separately what happens near smooth boundary regions, near non-smooth boundaries, and at infinity. Near smooth boundaries, the key object is a modified drift term (the original drift shifted by the Girsanov transformation), and martingality turns out to be equivalent to this shifted drift pointing inward or tangentially at the boundary, with an additional local condition required at special points where the drift vanishes and the integrand blows up. At infinity, they use classical tools like Feller's test for explosions, extended carefully to the multi-dimensional setting through radial and projection-based criteria.
The paper then applies these abstract results to concrete financial models, most notably the inverse-CIR and 4/2 stochastic volatility models with correlated factors. These models are popular in quantitative finance because they can capture both mean-reversion and volatility-of-volatility effects, but their correlated, singular structure made prior martingale analysis incomplete. The authors pin down exact parameter thresholds, including critical boundary cases, and show that some failures of the martingale property are caused by the process hitting a finite boundary rather than exploding to infinity, something earlier infinity-focused tests could not detect. As a bonus, the martingale conditions also provide uniqueness results for the pricing equations associated with these models.