The Distribution and Bias Property of Semi-parametric Estimates Using an ARFIMA (p, d, q) Process
This paper employs the Monte-Carlo simulation to explore the distribution and bias property of three semi-parametric estimates for long memory using an ARFIMA (p, d, q) process in finite samples. It demonstrates that the normality of estimators holds in most situations, hence a t statistics can be formed to test its significance when long-term and short-term dependences exist at the same time. However, in presence of the short-term parameter, the distributions are significantly biased so that the estimators and the following t test are less believable. In addition, when the true data process is near nonstationary or over-differencing, the distribution of estimators in finite sample is found non-convergent.
long memory semi-parametric estimator bias monte-carlo simulation
DENG Lu
School of Statistics, Central University of Finance and Economics, P.R.China, 100081
国际会议
威海
英文
244-252
2010-07-24（万方平台首次上网日期，不代表论文的发表时间）