The traditional linear credibility formula provides an exact expression for
the predictive mean when the conditional loss distributions belong to the
exponential dispersion family (EDF) and when the prior distribution is conj
ugate. When the claim size follows a distribution outside this family, the
credibility formula offers an inaccurate estimation of the mean future clai
m. The established analogy between the credibility formula and stochastic a
pproximation [Landsman, Z., Makov, U., 1999. On Stochastic Approximation an
d Credibility. Scand. Actuarial J., to appear] offers a way of devising cre
dibility estimation for distributions other than members of the EDF. In thi
s paper we suggest sequential credibility estimators which are suited to de
al with distributions belonging to the symmetric location dispersion family
. (C) 1999 Elsevier Science B.V. All rights reserved.