Firms that undertake corporate events often exhibit atypical financial characteristics, such that future performance might be expected to change even before the event is announced. I investigate five methods for generating control samples in various sampling situations. The results indicate that the methods sometimes produce severely biased test statistics. Overall, the method that generates control firms with similar prior levels and changes of performance and similar market-to-book ratios produces the most reliable test statistics. Furthermore, I find that it is more important to control for levels of performance than for changes in performance and market-to-book ratios.
