Peer Firm Selection for Discretionary Accruals Models

Taking as given the discretionary accruals models that exist in the literature, we examine how the selection of peer firms affects the researcher’s ability to detect earnings management. Researchers commonly estimate accruals models in cross section, and define the peer set (the firms used to estimate the cross section) as all firms in the same industry. We challenge this view by examining whether other factors – notably firm size – are at least as important as industry membership in selecting peers. Using U.S. data, we document that lagged asset peers perform as well as (or better than) industry peers at detecting earnings management both in simulations where we induce increasing levels of accruals (of between 2% and 100%) and in tests examining restatement data (where the existence of a restatement proxies for observed earnings management). Finally, we show that lagged asset peers continue to perform well for non-U.S. data, where it can be applied with significantly less sample loss than any industry peer definition.