In this paper, I examine whether the claim the more control variables, the better holds when matching in a dynamic context. I exhibit two situations where it is better not to control for past outcomes because DID matching is unbiased whereas matching on past outcomes is biased. I study the special case of evaluating a job training program, where both estimators are biased, borrowing a credible selection rule from Heckman, LaLonde, and Smith (1999) and relying on the parameters of the wage process estimated by MaCurdy (1982). I derive closed forms for the bias terms of the two estimators when the error terms are normally distributed, I show that DID matching performs better when used symetrically around the period of enrollment, as implemented by Heckman, Ichimura, Smith, and Todd (1998). Matching is less biased than DID-matching when considering to use the first pre-enrollment period, but symmetric DID-matching is less biased than the less biased matching estimator.