Autoregressive processes and first-hit probabilities for randomized random walks (Q3977673)

The author considers a random walk process in which the epochs of the jumps are regulated by a Poisson process, i.e. randomized random walk, and the jumps themselves are random variables assuming values \(\pm1,\dots,\pm k\) with probabilities \(p_ 1/2,\dots,p_ k/2\), independent of each other and of the Poisson process. The lifetime of this time-dependent process is regulated by an exponentially distributed random variable independent of the process. First-hit distributions of symmetric randomized random walk with exponential lifetime are found. The author uses a rather unusual method based on prediction of time series and on the theory of Dynkin, which relates Markov processes \(X\) to Gaussian processes \(\Phi\). Specifically, she uses the prediction formula of \textit{E. B. Dynkin} [Bull. Am. Math. Soc., New Ser. 3, 975-999 (1980; Zbl 0519.60046)] which expresses the coefficients of the best least squares predictor of \(\Phi\) in terms of the first-hit distribution of X. As it was noted previously, if \(X\) is a randomized symmetric random walk with exponential lifetime, the Gaussian process \(\Phi\) associated with \(X\) is an autoregressive sequence. This means that the coefficients of the least squares estimate of \(\Phi\) can be found through well-known time series algorithms; these coefficients are the first-hit probabilities of \(X\), according to Dynkin's prediction formula. In the paper there is firstly a review of Dynkin's theory of Gaussian processes associated with Markov processes, then there is a discussion on the \(AK(k)\) sequences which arise from association with randomized random walks and finally there is an explicit derivation of formulae for first- hit probabilities of random walks.