Omitted Variables and Tests of Dividend Displacement

Abstract

Theory suggests that, under certain assumptions, corporate distributions should reduce market value on a one-to-one basis. This result is sometimes known as dividend displacement. Nonetheless, dividend displacement tends to be rejected by empirical tests for total dividends and regular dividends, but not for share buybacks, in the UK. This is associated with regular dividends having a stronger relationship with future earnings than share buybacks (that is, more information content). We study a conjecture in the literature that the omission of lagged accounting variables in accounting-based market value models is the cause of the rejection of dividend displacement for regular or total dividends. Our results suggest that this is not so, although lagged accounting variables are value relevant. When various controls for omitted variables in models used in prior literature are employed, it is in only one case that the results of prior studies are overturned. This case is when lagged market value is added into the estimated equation. Theoretically, within a general study of controlling for omitted variables in accounting-based market valuation models, we argue that including lagged market value as an independent variable cannot control for the omission of lagged accounting variables, or other omitted variables, without producing additional model mis-specification problems which, in turn, introduces bias into statistical tests. Qualitative analysis of the various empirical results suggests that the results concerning dividend displacement when lagged market value is included in the model are induced by coefficient bias. As a consequence, we conclude that the results in previous literature are robust to the inclusion of lagged accounting variables, and the evidence against these results when lagged market value is included in the model is unreliable.