Peer effects and dividend policy /

No Thumbnail Available
Yang, Fang
Journal Title
Journal ISSN
Volume Title
Middle Tennessee State University
This paper adds a variable capturing peer effects to the dividend regression models explaining a firm's dividend behavior. The existing literature on dividend policy primarily focuses on agency theory, tax effects, or investors' preferences to explain the observed trends in firms' dividend behavior. Peer effects are formulated by a spatial lag variable, which is constructed on the basis of equal 2-digit, 3-digit, or 4-digit Standard Industrial Classification (SIC) codes.
The estimation of peer effects has been a very difficult task because of the reflection problem as well as data availability problems (Manski, 1993). Manski indicates that it is impossible to separately identify peer effects from the other types of neighborhood effects in the linear model. This study directly confronts the reflection problem by using a lagged peer variable in the dividend model.
The hypothesis is that a firm is more likely to change its dividend policy when its peers are doing the same regardless of its own financial conditions. To test this hypothesis, peer effects are treated in analogy to spatial correlation in regional science and real estate economics. The empirical methodology uses spatial econometrics techniques as typically employed in these fields (e.g., Anselin, 1988). Specifically, to identify peer effects, a spatial lag variable is constructed and added to the dividend regression models with traditional control variables. The models are estimated on both cross-sectional and panel data. The cross-sectional regression models are estimated on the companies of the S&P 1500 Super Composite Index and the S&P 500 index using data from the years 2003 to 2006. The panel regressions employ S&P 1500 data for the seven years from 2000 to 2006.
The cross-sectional results from the S&P 1500 sample show strong evidence of peer effects. The spatial lag variable constructed for the 2-digit Standard Industrial Classification (SIC) data is highly significant and has the expected positive sign for all four years. The results are similar for the models that employ spatial lag variables for the 3-digit and the 4-digit industries. There is also strong evidence in support of the peer effects hypothesis for the S&P 500 sample. On average, the peer effects measures have a stronger impact on the amount of dividends paid than do size and profitability, which are the traditional explanatory variables emphasized in the dividend literature.
For the S&P 1500 panel data, the coefficient of the peer variable also tends to be positive and statistically significant. However, the peer effects results are not as consistent across alternative models as those for the cross-section regressions.