The US financial crisis and corporate dividend reactions: for better or for worse?

Abstract

We examine how changes in dividend policy in 2008 as the financial crisis was unfolding influenced firm risk-adjusted returns in the following years. Our sample consists of NYSE- and NASDAQ-traded firms that paid dividends in 2007. We divide these firms into four groups based on their dividend policy in 2008. We find that firms that decreased or eliminated dividends in 2008 had higher risk-adjusted returns in 2009. The higher risk-adjusted return is consistent with the better corporate governance in 2007. This finding suggests that the firms that quickly reacted to the deteriorating economic conditions by cutting dividends and preserving cash were able to better weather the coming financial crisis.

Appendix: Modified GRS test

Gibbons et al. (1989) provide their GRS test on abnormal returns. The null hypothesis of GRS test is all α_s equal to zero. In our work, we want to compare abnormal returns between groups. Therefore, we make a modification to GRS test. Following Morrison (2005), we have

$$H_{0} : \alpha_{i} - \alpha_{j} = 0 \quad or\quad R^{{\prime }} \alpha = 0$$

$$\sqrt {\frac{T}{{1 + \hat{\theta }}}} \hat{\alpha } \sim N\left( {\sqrt {\frac{T}{{1 + \hat{\theta }}}} \alpha ,\Sigma } \right).$$

(14)

$$\sqrt {\frac{T}{{1 + \hat{\theta }}}} R^{\prime } \hat{\alpha } \sim N\left( {\sqrt {\frac{T}{{1 + \hat{\theta }}}} R^{\prime } \alpha ,R^{\prime }\Sigma R} \right).$$

(15)

$R^{{\prime }} \hat{\alpha }$ and $R^{\prime }\Sigma R$ are independent. $\left( {T - 2} \right)R^{\prime }\Sigma R$ follows Wishart distribution. Then apply GRS’s conclusion,

$$F = \frac{{T\left( {T - N - K} \right)}}{{N\left( {T - 2} \right)}}\frac{{\left( {R^{\prime } \hat{\alpha }} \right)^{\prime } \left( {R^{\prime } {\hat{\varSigma }}R} \right)^{ - 1} \left( {R^{\prime } \hat{\alpha }} \right)}}{{1 + \hat{\theta }^{2} }} \sim F_{N, T - N - K} .$$

(16)

N is the number of restriction, which is 1 in our tests. K is the number of factors, which is 4 since we use four-factor model. $\hat{\theta } = \hat{\mu }_{f}^{\prime } {\hat{\varOmega }}^{ - 1} \hat{\mu }_{f}$, where $\hat{\mu }_{f}$ is the sample mean of factor loadings and ${\hat{\varOmega }}$ is the max-likelihood estimation of covariance matrix of factor loadings.

We are interested whether alpha of one group is significantly larger than alpha of another group.

$$H_{1} : \alpha_{i} - \alpha_{j} > 0\quad or\quad R^{{\prime }} \alpha > 0.$$

Null is rejected when $F \, < F_{2a, N, T - N - K}$ and $\alpha_{i} - \alpha_{j} > 0$ (Follmann 1996) (Table 16).

Table 16 The persistence effect of institutional holding: results of first stage regression

We test the effect of management on the return effect of dividend policy changes controlling the persistence effect of institutional holding. We use institutional holding as a measure of managers’ effort in 2007. A two-stage regression is used to exclude the effect of institutional holding in 2009. Alphas and excess returns in 2009 are dependent variables in first stage regressions

$$Alpha_{i} = \alpha + \beta IH_{i}^{2009} + u_{i} .$$

Residual from first stage is the dependent variable in second stage.

$$\hat{u}_{i} = \alpha + \mathop \sum \limits_{k = 1}^{6} \beta_{k} Control_{ik}^{2009} + \gamma_{1} Group_{i} + \gamma_{2} IH_{i}^{2007} + \gamma_{3} Group_{i} IH_{i}^{2007} + \varepsilon_{i} .$$

The interested variable is Group and its interaction term with institutional holding. Results of the second stage regression are reported in Table 14.