A1399
Title: A universal method for statistical inference of low-frequency time series
Authors: Alexis Derumigny - Delft University of Technology (Netherlands) [presenting]
Fernando De Diego Avila - Delft University of Technology (Netherlands)
Fang Fang - Delft University of Technology (Netherlands)
Abstract: Statistical inference for low-frequency time series is challenging due to limited non-overlapping observations, as commonly encountered in fields like financial risk management and weather forecasting. Traditional direct methods relying solely on low-frequency data often produce inaccurate estimates, while estimations based on overlapping observations have biases caused by autocorrelation. A novel simulation-based method is proposed for inferring low-frequency time series. It involves three steps: estimating the distribution of the corresponding higher-frequency process, simulating paths from this distribution, and generating a large dataset of aggregated low-frequency observations to enable accurate and robust statistical inference. Theoretical results are also given on the asymptotic properties of the estimators. The superiority of the proposed method over direct methods is verified via a comprehensive simulation study.
