Do fair value adjustments influence dividend policy?

Abstract

We examine the impact of positive fair value adjustments on dividend policy. If fair value adjustments are transitory in nature and managers are able to assess their implications for future earnings, fair value adjustments in net income is expected to have no distribution consequences. However, positive fair value adjustments may lead to higher dividends when management incorrectly assesses their persistence. This can have a procyclical impact because higher dividends increase leverage, and thus risk. We use a Russian setting that mandates fair value accounting for financial instruments and requires disclosure of unrealised fair value adjustments in income. We find no empirical support for the concern that dividends increase in response to positive fair value adjustments. Rather, there is a negative relationship between positive fair value adjustments and dividend changes, which holds after controlling for dividend policy determinants and any endogenous nature of the revaluation decision. We discuss several possible explanations for this finding.

Keywords:

• fair value accounting
• dividend policy
• earnings persistence

Acknowledgements

We appreciate the useful comments of Allan Hodgson, Christopher Humphrey, Alexander Muravyev, Brendan O'Dwyer, Deborah Schanz, Thorsten Sellhorn, Hervé Stolowy, David Veenman, Dmitry Volkov, Derek Weaving, Jörg Werner, and seminar participants at WHU – Otto Beisheim School of Management, University of Amsterdam and the 32nd Annual Congress of the EAA, as well as the constructive remarks of Pauline Weetman (editor) and the reviewers.

Notes

Core (permanent, sustainable) earnings are defined as income after excluding items, which are transitory in nature. Collins et al. (1999) proxy for core earnings with income before extraordinary items, discontinued operations, cumulative effects of accounting changes and tax-adjusted special items. In our tests, we differentiate between earnings before revaluations – our raw estimate of core earnings – and fair value adjustments.

We exclude observations for which financial data or information on common dividends is missing and for which the primary outstanding security is not common stock. Furthermore, we drop financial companies and utilities, although keeping them in the sample does not change our inferences.

Worldscope covers mostly large firms and does not provide information on the amount of fair value adjustments reported in income.

Including these observations in the control (i.e. non-revaluation) sub-sample does not materially affect our results. Results of these tests, as well as of all other robustness checks, are available from the authors on request.

As in La Porta et al. (2000) we calculate the payout ratio after deleting firms with negative earnings before fair value adjustments. We have also replaced earnings with earnings changes in the denominator of this ratio and concentrated on the sample of firms reporting positive earnings changes. Results of this analysis are qualitatively similar and suggest that revaluers do not fully adjust their dividends to a change in core earnings.

In addition to differencing the dependent variable and earnings (indirectly, via using lagged earnings) to control for unobservable characteristics, we also run the model using a difference in fair value adjustments, with similar results.

By construction, our measure of net share issues also includes dirty surplus adjustments.

We also test whether selection bias affects our inferences by fitting the two-step treatment effects model (see Maddala 1983, Leuz and Verrecchia 2000), where the treatment variable REV_D _it is conditioned on size, leverage, growth, profitability and cash. In the first step, the probit regression (4) is used to produce a selection hazard (inverse Mills ratio). In the second step, the dividend policy regression (model (e) in Panel B of Table 5) is augmented with the inverse Mills ratio. Testing the null hypothesis that the coefficient on Mills ratio is zero represents a test for sample selectivity. This test indicates that self-selection is not an issue (z = 1.08, p = 0.281). As the two-stage selection procedure is expected to produce inferior results (Francis and Lennox 2008), we chose to report results based on the propensity score matching.

The Wald χ ² of Model (4) estimated for a matched sample equals 2.63 (p = 0.757). Furthermore, the t-tests show that revaluers cannot be distinguished from a control sample in terms of size, growth, leverage, (current and lagged) profitability, cash balance and lagged dividends.

Our inferences are based on standard errors robust to both serial correlation and heteroscedasticity (Huber/White standard errors). We do not cluster by firm, as most firms now have only one observation.

We thank an anonymous reviewer for suggesting this explanation.

The median annual growth in sales was 18% during our sample period (see Table 1).

Aggregate investments have substantially increased in each year during our sample period. Based on data of the Russian Federal State Statistic Service we estimate that the average growth in investments in fixed assets was 28% during our sample period.