Abstract:The paper considers the problem of estimating the parameters in a continuous time regression model with a non-Gaussian noise of pulse type. The noise is specified by the Ornstein-Uhlenbeck process driven by the mixture of a Brownian motion and a compound Poisson process. Improved estimates for the unknown regression parameters, based on a special modification of the James-Stein procedure with smaller quadratic risk than the usual least squares estimates, are proposed. The developed estimation scheme is applied for the improved parameter estimation in the discrete time regression with the autoregressive noise depending on unknown nuisance parameters.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to estimate unknown parameters more accurately in continuous - time regression models in non - Gaussian noise environments. Specifically, the paper focuses on how to construct improved estimation methods when the noise is impulsive non - Gaussian noise, so that these estimation methods have higher precision than the traditional Least Squares Estimation (LSE).
### Background and Problem Description
In many practical applications, the noise of data does not always follow a Gaussian distribution. Especially in fields such as finance and biology, the noise may contain impulse components, and this kind of noise can be modeled by the Ornstein - Uhlenbeck process. The Ornstein - Uhlenbeck process is a stochastic process, which can be driven by a mixture of Brownian motion and compound Poisson process. In this context, the traditional least - squares estimation method may no longer be optimal because they assume that the noise is Gaussian - distributed.
### Main Contributions of the Paper
1. **Model Setup**:
- Considered a continuous - time regression model \( dy_t=\sum_{j = 1}^p\theta_j\varphi_j(t)dt + d\xi_t\), where \(\xi_t\) is an Ornstein - Uhlenbeck process driven by Brownian motion and compound Poisson process.
- The noise parameters \(a, \rho_1, \rho_2, \lambda\) are unknown.
2. **Improved Estimation Method**:
- Proposed an improved estimation method based on the James - Stein procedure. This method, by introducing a specific correction term, makes the risk (measured by the mean - square error) of the estimator smaller than that of the traditional least - squares estimation.
- The specific form is \(\theta^*=\left(1-\frac{c}{\|Y\|}\right)Y\), where \(c\) is a positive constant, which will be determined in subsequent theorems.
3. **Theoretical Results**:
- Proved that under certain conditions, the risk of the improved estimator \(\theta^*\) is indeed smaller than the risk of the least - squares estimator \(\hat{\theta}\).
- Theorem 2.1 gives an explicit upper bound, indicating that the risk of the improved estimator is smaller than that of the least - squares estimator.
4. **Applications**:
- Applied the improved estimation method to discrete - time regression models, especially in the case where the autoregressive noise depends on unknown perturbation parameters.
### Key Formulas
- **Regression Model**:
\[
dy_t=\sum_{j = 1}^p\theta_j\varphi_j(t)dt + d\xi_t, \quad 0\leq t\leq n
\]
where \(\theta = (\theta_1,\ldots,\theta_p)'\) is the unknown parameter vector, \(\varphi_j(t)\) is a periodic function, and \(\xi_t\) is an Ornstein - Uhlenbeck process.
- **Ornstein - Uhlenbeck Process**:
\[
d\xi_t=a\xi_t dt + du_t
\]
where \(u_t=\rho_1w_t+\rho_2z_t\), \(w_t\) is a standard Brownian motion, and \(z_t\) is a compound Poisson process.
- **Improved Estimator**:
\[
\theta^*=\left(1-\frac{c}{\|Y\|}\right)Y
\]
where \(c=(p - 1)\lambda^*\gamma_p\), \(\lambda^*\) and \(\gamma_p\) are specific constants.
### Conclusion
The paper successfully improves the accuracy of parameter estimation in non - Gaussian noise environments by introducing an improved James - Stein estimation method. This method has important application value in dealing with actual data.