Let XI, X2,. . . be independent random variables with common distribution Fo for which the mean, p = p(B), is a one-to-one function of the parameter 19 E 0 c (-oo. w ).Suppose that XI. X2,. . . a n to be taken one at a time up to stage f. Iff is chosen before observing X I , . ..,XI, then one may estimate p by the sample mean XI, which is an unbiased estimator of p . However, if f is determined according to a stopping rule, then may be biased for, p . Lctting r be a stopping time of the type proposed by Robbins (1959) and estimating p by a bias-reduction estimator, fi,, subject to the loss function L,, =&fit -p)' + f , a >O, we show that the asymptotic regret (as a + w ) of the sequential procedure (t,ji,) can be negative if the bias-reduction function is chosen properly.