Boundary behaviour of the Volterra square-root process

Does Market Volatility Ever Truly Hit Zero?

Why do some financial models accurately track market turbulence while others collapse during periods of extreme stress? In mathematical finance, researchers use stochastic models to predict volatility—the rate at which asset prices change. While traditional models assume volatility moves in a relatively smooth fashion, modern "rough" models better capture the jagged, erratic nature of real-world markets. However, these rougher models introduce a mathematical crisis. They behave unpredictably when volatility approaches zero.

A new study by Martin Friesen, Stefan Gerhold, and Kristof Wiedermann explores this exact boundary behavior. They report that while standard, "regular" volatility models are mathematically guaranteed to stay above zero, the increasingly popular "rough" models actually hit zero with positive probability. This creates a mathematical "atom"—a single point that carries a concentrated chunk of probability. This finding fundamentally changes how these models must be used in professional trading environments.

The breakdown of classical smoothness

For decades, the gold standard for modeling variance has been the Cox-Ingersoll-Ross (CIR) process. In this framework, volatility is treated as a smooth, continuous flow. A "drift" term acts like a physical force, pushing the process away from zero. This avoidance is governed by the Feller condition. This is a specific ratio of parameters that ensures the process remains strictly positive.

The problem arises with the emergence of rough volatility models. These models incorporate fractional memory (a system where current states depend on past history). This "roughness" allows the model to fit the complex "skew" seen in market prices more accurately. However, it comes at a steep mathematical cost. Because the paths are so jagged, the classical safeguards that keep the process away from zero begin to fail. The authors note that moving from the smooth CIR regime toward the rough regime makes the process prone to hitting the zero boundary.

Analyzing the Riccati equation

To understand why these models crash into zero, the authors analyze the Volterra square-root process. This is a Volterra process (a process with memory) where the evolution of the variance is driven by a convolution of a kernel ($K$) with the history of the process. Think of the kernel as a weighted moving average. It determines how much influence yesterday's volatility has on today's movement.

The researchers tackle the complexity of this path-dependency by focusing on the Volterra Riccati equation. In simpler models, the characteristic function (a tool used to calculate probabilities) can be solved using standard differential equations. In the Volterra setting, the authors must solve an integral equation. Here, the solution itself is buried inside a convolution. Their approach involves three distinct stages:

Smoothing the solution

The authors use a "bootstrapping" argument to prove regularity. They show that even if the kernel is somewhat irregular, the resulting solution to the Riccati equation maintains a specific level of smoothness on the open interval $(0, \infty)$.

Using comparison principles

To handle difficult non-linearities, the authors employ comparison theorems for generalized Riemann-Liouville fractional equations. This allows them to sandwich the unknown solution between two known upper and lower bounds.

Matching asymptotics

The authors analyze how the kernel behaves at very small time scales versus very long time scales. This helps them predict if the process will settle into a stable distribution or get stuck at the zero boundary.

Evidence of the zero-hitting phenomenon

The study provides rigorous mathematical proof for the divergence between regular and rough regimes. For regular kernels, where the kernel is well-behaved at $t=0$, the authors establish a time-dependent version of the Feller condition. They report that if this condition is met, the process is guaranteed to stay positive on a given time interval $[0, T]$. Under these conditions, its negative moments also remain finite.

However, the findings change dramatically for the "rough" case. When the kernel is "regularly varying" at zero, the authors demonstrate that the process necessarily hits zero with positive probability. Specifically, for a pure fractional kernel, the paper provides an explicit lower bound for the probability $P[X_t = 0]$.

The authors show that in this rough regime, the law of the process possesses an "atom" at zero. This means there is a discrete, non-zero probability that the process sits exactly at the boundary. Instead of the process merely passing through zero, there is a measurable chance that it stays there. Interestingly, the authors find a nuance in the long-term behavior. While the process hits zero in the short term, its limit distribution can still have finite negative exponential moments.

Constraints on market pricing

These mathematical boundaries have dire consequences for the Volterra Heston model. This is a staple in quantitative finance used to price derivatives. Pricing relies on the ability to change "measures." This means shifting from real-world probability to a risk-neutral probability where assets are priced fairly. This shift is typically performed using a Girsanov transformation.

The authors report a striking limitation. In rough Volterra Heston models, standard affine changes of measure often fail. Because the rough process hits zero and stays there for a measurable amount of time, the mathematical requirements for a valid change of measure are violated. This failure means a valid risk-neutral measure might not even exist under standard assumptions.

Specifically, the authors demonstrate that an affine change of measure cannot alter the long-term mean of the variance process in the rough case. For practitioners, this means many common ways of adjusting models to account for risk-premia are mathematically invalid. If you attempt to use a standard Heston-style drift adjustment in a rough environment, you are building a model on a foundation that the math does not support.

The verdict for volatility modeling

Is the rough Volterra Heston model ready for production? The answer depends entirely on how you handle the boundary. If you are building a model that assumes volatility will never reach zero, the authors' work suggests you cannot use a rough kernel without changing your drift assumptions.

The research is a clear warning to quantitative engineers. The "roughness" that provides superior market fits also introduces structural instabilities at the boundary. While these models effectively capture short-term skew, they are mathematically fragile. For anyone implementing these in a high-stakes pricing engine, the "atom at zero" is a feature that necessitates a complete rethink of how risk-neutral measures are constructed.