Abstract
The aim of this paper is to develop estimation and inference methods for the drift parameters of multivariate Lévy-driven continuous-time autoregressive processes of order $p\in\mathbb{N}$. Starting from a continuous-time observation of the process, we develop consistent and asymptotically normal maximum likelihood estimators. We then relax the unrealistic assumption of continuous-time observation by considering natural discretizations based on a combination of Riemann-sum, finite difference, and thresholding approximations. The resulting estimators are also proven to be consistent and asymptotically normal under a general set of conditions, allowing for both finite and infinite jump activity in the driving Lévy process. When discretizing the estimators, allowing for irregularly spaced observations is of great practical importance. In this respect, CAR($p$) models are not just relevant for "true" continuous-time processes: a CAR($p$) specification provides a natural continuous-time interpolation for modeling irregularly spaced data - even if the observed process is inherently discrete. As a practically relevant application, we consider the setting where the multivariate observation is known to possess a graphical structure. We refer to such a process as GrCAR and discuss the corresponding drift estimators and their properties. The finite sample behavior of all theoretical asymptotic results is empirically assessed by extensive simulation experiments.
Introduction: Bridging Theory and Reality
Autoregressive models are the cornerstone of time series analysis, utilized extensively in fields ranging from finance to signal processing. While discrete-time (AR) models are common, Continuous-time Autoregressive (CAR) models offer a superior framework for handling irregularly spaced observations—a frequent challenge in high-frequency trading and medical monitoring.
Historically, two major hurdles have limited the application of multivariate CAR models: the presence of "jumps" (non-Gaussian noise) and the "observation gap" between continuous theory and discrete snapshots. This paper provides the mathematical framework to estimate drift parameters—the coefficients governing system evolution—under these realistic, difficult conditions.
Methodology: From Ideal MLE to Discrete Reality
The authors employ a three-step approach to construct robust estimators:
- The Ideal MLE: They first derive the Maximum Likelihood Estimator (MLE) assuming perfect continuous-time observation. This involves complex Radon-Nikodym derivatives to identify parameters that maximize the likelihood of the observed path.
- Discretization: To bridge the gap to real-world data, integrals are replaced with Riemann sums and derivatives with forward finite differences.
- Thresholding: To manage the "jumpy" nature of Lévy noise, the authors introduce a thresholding technique. By filtering out changes larger than a specific threshold, the algorithm separates sudden shocks from the underlying continuous drift. This allows the method to accommodate both finite and infinite jump activity.
The Derivative Difficulty
A significant challenge in estimating a $p$-order autoregressive process is the requirement of knowing the first $p-1$ derivatives. As shown in the figure above, while the process $Y$ may appear relatively smooth, its derivative $DY$ is significantly more volatile. The authors provide rigorous proof that finite difference approximations of these derivatives maintain estimator accuracy as sampling frequency increases.
Graphical CAR (GrCAR)
In high-dimensional systems, the number of parameters to estimate grows quadratically, often leading to overfitting or computational intractability. The authors introduce GrCAR (Graphical CAR) to address this. By assuming the variables interact according to a known network or "graph" structure, they significantly reduce the parameter space, creating a more parsimonious and interpretable model.
Key Results and Simulation
The paper validates its theoretical findings through extensive simulation experiments across three distinct regimes: Brownian Motion (purely continuous), Compound Poisson (discrete jumps), and Gamma Processes (infinite activity jumps).
Key findings include:
- Consistency and Normality: The authors prove that as observation time ($T$) increases and sampling intervals ($\Delta$) decrease, the estimators converge to true values (consistency) and errors follow a normal distribution (asymptotic normality).
- Robustness: The method remains effective even when the specific characteristics of the noise (covariance and jump intensity) are unknown and must be estimated from the data.
- Interpolation Utility: CAR models serve as a natural interpolation tool for discrete data that is messy or irregularly timed, proving their relevance beyond "true" continuous processes.
Conclusion and Future Directions
While this work provides a rigorous foundation for drift estimation, it assumes a known graph topology for GrCAR models. Future research may focus on simultaneously estimating both the graph structure and drift parameters. Additionally, extending this framework to Continuous-time Autoregressive Moving Average (CARMA) processes would allow for even more flexible modeling of complex temporal dependencies.
Summary updated