Regression models for nonstationary categorical time series: Asymptotic estimation theory (Q1819873)

Let \(y_ t\), \(t=1,2,..\). be a time series with m possible categories for each observation. The t th observation is \(y_ t=(y_{t1},...,y_{t,m- 1})'\), where \(y_{tj}=1\) if the j th category is observed and 0 otherwise. Define \(\pi_{tj}=P(y_{tj}=1| y_{t-1},...,y_ 1)\) and \(\pi_ t=(\pi_{t1},...,\pi_{t,m-1})'\). The author proposes a model \(\pi_ t=h(Z_ t'\beta)\), where \(\beta\) is a vector of unknown parameters, \(Z_ t\) is a matrix which depends on past observations and on known exogenous variables, and h is a link function. It is proved that the maximum likelihood estimator for \(\beta\) is consistent, asymptotically normal and efficient.