A Curvelet based Approach to Time Series Forecasting

Abstract

In this paper we propose a new Curvelet based methodology for modeling the financial time series data, addressing and incorporating the diverse range of data characteristics. With the proposed methodology, we analyze and model the geometric multi scale and chaotic data characteristics. Empirical studies with some typical financial time series data show that more diverse heterogeneous data characteristics can be revealed and modeled in the projected time delayed domain. The proposed algorithm demonstrates the superior performance compared with the benchmark models.