An inference procedure for a simple epidemic model: A martingale approach (Q1186987)

The author considers an epidemic model in which a homogeneously mixing and closed population of susceptibles of size \(n\) is subject to an epidemic by one infective at \(t=0\). The rate of infection is \(\beta\) for a given susceptible due to one infective. It is assumed that there are no further infections from outside the population which is itself not subject to any birth, death or immigration. The author notices two zero mean martingales given by \[ M(t)=I(t)-\beta\int^ t_ 0 I(u-)S(u- )du,\quad M^*(t)=\int^ t_ 0 W(u)dM(u), \] where \(W(.)\) is any predictable process and \(I(t)\) and \(S(t)\) denote respectively the number of infections and susceptibles at time \(t\). If \(\tau\) is the stopping time given by \[ \tau=\inf\{t\geq 0;\;I(t)=n+1\text{ or }t=T\}, \] where \(T\) is the span of observation, an estimator \(\beta\) corresponding to \[ M^*(\tau)=E[M^*(\tau)]=0 \] can be defined. This gives rise to a wide class of estimates since \(W(.)\) is fairly general. The asymptotic properties of the estimator for large \(n\) leading to central tendency are studied. The maximum likelihood estimator is identified as a special member of this class and is used to study the efficiency of the estimators.