Truncated Poisson Regression for Time Series of Counts
A series of counts \(y_t\), \(t=1,2,\dots\), is considered such that the conditional distribution of \(y_t\) given the history of the process \(\mathfrak F_{t-1}\) is \[ f(y_t |{\mathfrak F}_{t-1})=\lambda_t^{y_t}/y_t!\psi(\lambda_t),\quad y_t\in\{c_1,\dots,c_2\}, \] where \(\psi(\lambda)=\sum_{x=c_1}^{c_2}\lambda^x/x!\) (this distribution is called doubly truncated Poisson distribution). It is assumed that \(\lambda_t=\exp(\beta^T z_{t-1})\), where \(\beta\) is an unknown parameter vector, and \(z_{t-1}\) is an \(\mathfrak F_{t-1}\) measurable predictor vector that can include past values of \(y\) and covariates. The author considers a partial maximum likelihood estimator of \(\beta\) and derives it's asymptotic normality. The likelihood ratio, the Wald, and the partial score statistics are proposed to test a general linear hypothesis about \(\beta\). It is shown that the distributions of all these statistics are asymptotically chi-square under the null hypothesis. Simulation results and analysis of poliomyelitis data (USA 1970-1983) are presented.
- Useful models for time series of counts or simply wrong ones?
- Poisson QMLE of count time series models
- scientific article; zbMATH DE number 1474822 (Why is no real title available?)
- Analysis of low count time series data by poisson autoregression
- Partial Likelihood Inference For Time Series Following Generalized Linear Models
- Robust estimation methods for a class of log-linear count time series models
- Some recent progress in count time series
- Right truncated homogeneous Poisson process
- Regression trees for analysis of count data with extra Poisson variation
This page was built for publication: Truncated Poisson Regression for Time Series of Counts
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4781093)
