Henghsiu Tsai Academia Sinica, Taipei, Republic of China.pdfVIP

Maximum likelihood estimation of linear continuous- time long-memory processes with discrete-time data Henghsiu Tsai? Academia Sinica, Taipei, Republic of China K. S. Chan University of Iowa, Iowa City, U.S.A. Revised, May, 2005 Summary. We develop a new class of Continuous-time Auto-Regressive Fractionally In- tegrated Moving-Average (CARFIMA) models which are useful for modelling regularly-spaced and irregularly-spaced discrete-time long-memory data. We derive the autocovariance func- tion of a stationary CARFIMA model, and study maximum likelihood estimation of a regression model with CARFIMA errors, based on discrete-time data and via the innovations algorithm. It is shown that the maximum likelihood estimator is asymptotically normal, and its finite-sample properties are studied through simulation. The efficacy of the proposed approach is demon- strated with a dataset from an environmental study. Keywords: CARFIMA models; fractional Brownian motion; innovations algorithm; irregularly- spaced data; polynomial trend. 1. Introduction It is well known that the long range dependence properties of time series data have found diverse applications in many fields including hydrology, economics and telecommunications; see Bloomfield (1992), Sowell (1992), Robinson (1993), Beran (1994), Baillie (1996) and Ray and Tsay (1997). A well-known class of discrete-time long-memory processes are the ?Address for correspondence: Henghsiu Tsai, Institute of Statistical Science, Academia Sinica, Taipei, Taiwan 115, R.O.C. E-mail: htsai@.tw autoregressive fractionally integrated moving average (ARFIMA) models; see Granger and Joyeux (1980) and Hosking (1981). Maximum likelihood estimation and forecasting of the ARFIMA models with missing values have been considered by Palma and Chan (1997) and Palma and Del Pino (1999). For irregularly-spaced time series data, it is often more convenient to analyze the data by assuming that they are sampled from an underlying continuous-time process.