Estimating the Mean Function of a Compound Cyclic Poisson Process

Abstract

A stochastic process has an important role in modeling various real phenomena. One special form of the stochastic process is a compound Poisson process. Many phenomena in different fields that have been modeled as a compound Poisson process, such as phenomena in the fields of insurance and finance, physics, demography, geology, and biology. A compound Poisson process model can be extended by generalizing the corresponding Poisson process. One of them is using cyclic Poisson process, so that the model becomes a compound cyclic Poisson process. This research has four objectives as follows: (1) to formulate an estimator of the mean function of a compound cyclic Poisson process; (2) to analyze the consistency of the estimator; (3) to analyze the bias, variance, and mean squared error (MSE) of the estimator; and (4) to observe the behavior of the estimator in the case that the length of the observation time interval is bounded. Let * ( ) + be a cyclic Poisson process with (unknown) locally integrable intensity function . We consider the case when the intensity function has a (known) period . We do not assume any (parametric) form of except that it is periodic, that is, the equality ( ) ( ) holds for all and . Let * ( ) + be a compound cyclic Poisson process that corresponds to the process * ( ) + where ( ) Σ ( ) with * + is a sequence of independent and identically distributed random variables having mean and variance , which is also independent of the process * ( ) +. The mean function of ( ) is given by ( ) , ( )- , ( )- , - . ( )/ where ⌊ ⌋ ( ) ∫ ( ) and ( ) Suppose that, for some , a single realization ( ) of the process * ( ) + defined on a probability space ( ) is observed, though only within a bounded interval , -. Furthermore, suppose that for each data point in the observed realization ( ) , -, say -th data point, 1, 2, , (, -), its corresponding random variable is also observed. The estimator of the mean function is given by ̂ ( ) . ̂ ̂ ( )/ ̂ where ̂ (, -) ̂ ( ) Σ (, - , -) and ̂ (, -)Σ (, -) with the understanding that ̂ when (, -) . The estimator of the mean function is both a weak and a strong consistent estimator, that is ̂ ( ) → ( ) and ̂ ( ) → ( ) as → . The rate of convergence of the bias, variance, and MSE of the estimator are respectively [ ̂ ( )] ( ) [ ̂ ( )] ( ) and [ ̂ ( )] ( ) as → . The behavior of the estimator is not good enough when the length of the time interval of observations is 20 period, but it is good enough when the length of the time interval of observations is 40 period, which is a bounded interval.