Estimating the adjustment coefficient in an ARMA(p, q) risk model