Piecewise Linear Accrual Models: do they really control for the asymmetric recognition of gains and losses?

The asymmetric recognition of gains and losses underlying conservative accounting is not taken into account by Jones (1991)-type accrual models. Recently, Moreira (2002) and Ball and Shivakumar (2005a) have proposed piecewise linear accrual models designed to control for this asymmetric impact. Our paper first discusses the sign of the expected measurement error in discretionary accruals (DAC) estimates when models do not control for the asymmetry underlying conservatism. We find that DAC in firms with bad news (BN) are expected to be understated, while those in good news (GN) firms will be overstated. Based on this original result we empirically test, using graphical and statistical tools, whether piecewise linear accrual models correct such a measurement error. The empirical evidence shows mixed results. For GN firms the estimates are corrected downwards, as expected; for BN firms, unexpectedly, part of the estimates is also corrected downwards. The reason for this unexpected result seems to lie in a non-linear relationship between accruals and the proxy for BN that the models are unable to control for. Thus, DAC estimates under piecewise linear models are not deemed to be of better quality than those of traditional accrual models.