Abstract:Guyon and Lekeufack recently proposed a path-dependent volatility model and documented its excellent performance in fitting market data and capturing stylized facts. The instantaneous volatility is modeled as a linear combination of two processes, one is an integral of weighted past price returns and the other is the square-root of an integral of weighted past squared volatility. Each of the weightings is built using two exponential kernels reflecting long and short memory. Mathematically, the model is a coupled system of four stochastic differential equations. Our main result is the wellposedness of this system: the model has a unique strong (non-explosive) solution for all parameter values. We also study the positivity of the resulting volatility process and the martingale property of the associated exponential price process.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the well - posedness problem of the Guyon - Lekeufack volatility model (Path - dependent volatility model, PDV), which specifically includes the following aspects:
1. **Existence and Uniqueness of Strong Solutions of the Model**:
- The paper mainly studies whether this path - dependent volatility model has a unique strong solution (non - explosive solution) for all parameter values. For the 2 - factor model and the 4 - factor model, the author respectively proves the existence and uniqueness of their strong solutions.
2. **Positivity of the Volatility Process**:
- For the 2 - factor model, the author gives a condition \(\lambda_2 < 2\lambda_1\) to ensure that the volatility process \(\sigma_t>0\). However, in the 4 - factor model, this property does not always hold, that is, the volatility process may become negative.
3. **Martingale Property of the Price Process of the Exponential Local Martingale**:
- The author also studies the martingale property of the price process \(X_t\) related to the volatility model. For the 2 - factor model, the author proves that under certain conditions, \(X_t\) is a true martingale. For the 4 - factor model, this problem remains open.
### Specific Formulas and Conclusions
#### 2 - factor model
The SDE system of the 2 - factor model is:
\[
\sigma_t=\beta_0+\beta_1R_{1,t}+\beta_2\sqrt{R_{2,t}}
\]
\[
dR_{1,t}=\lambda_1\sigma_t dW_t-\lambda_1R_{1,t}dt
\]
\[
dR_{2,t}=(\lambda_2\sigma_t^2 - \lambda_2R_{2,t})dt
\]
where the parameters satisfy:
\[
\beta_0,\beta_2,\lambda_1,\lambda_2\geq0\quad\text{and}\quad\beta_1\leq0
\]
The main results are:
- **Theorem 1.1**: The 2 - factor model (2 - PDV) has a unique strong solution.
- **Theorem 2.6**: If the initial condition satisfies \(\sigma_0 = \beta_0+\beta_1R_{1,0}+\beta_2\sqrt{R_{2,0}}>0\) and \(\lambda_2 < 2\lambda_1\), then \(\sigma_t>0\).
#### 4 - factor model
The SDE system of the 4 - factor model is:
\[
\sigma_t=\beta_0+\beta_1R_{1,t}+\beta_2\sqrt{R_{2,t}}
\]
\[
R_{1,t}=(1 - \theta_1)R_{1,0,t}+\theta_1R_{1,1,t}
\]
\[
R_{2,t}=(1 - \theta_2)R_{2,0,t}+\theta_2R_{2,1,t}
\]
\[
dR_{1,j,t}=\lambda_{1,j}\sigma_t dW_t-\lambda_{1,j}R_{1,j,t}dt,\quad j\in\{0,1\}
\]
\[
dR_{2,j,t}=(\lambda_{2,j}\sigma_t^2-\lambda_{2,j}R_{2,j,t})dt,\quad j\in\{0,1\}
\]
The main results are:
- **Theorem 1.2**: The 4 - factor model (4 - PDV) has a unique strong solution.
- **Proposition 3.4**: Even under the condition of \(\bar{\lambda}_2 < 2\bar{\lambda}_1\), there may still exist...