Sampling Error and the Joint Estimation of Imputation Credit Value and Cash Dividend Value

Abstract

Since dividend imputation was introduced to Australia 32 years ago, researchers and corporate finance practitioners have debated the extent to which imputation credits are incorporated into share prices. One reason for divergence of opinions is the selective interpretation of coefficient estimates from regression. Sample observations exhibit little dispersion of corporate tax rates and franking percentages. This means that if noise in a sample leads to the value of cash being understated, the same noise is likely to lead to the value of credits being overstated. Using simulation analysis we show that there is an inverse relationship between estimates of credit value and cash value due to random variation in samples. This problem is exacerbated by a lack of independence across observations. Regression analysis has merit, provided inference accounts for the inverse relationship between estimates of cash dividend value and imputation credit value. We consider five studies which reported estimates of 0.34 to 0.57 for imputation credit value and 0.73 to 0.88 for cash dividend value. The implication of our simulation analysis is that it would be incorrect to claim that cash is fully valued, but that imputation credit value lies within the range of 0.34 to 0.57. This would represent selective interpretation. A researcher cannot claim to have a reliable sample and research method which allows interpretation of one coefficient, but at the same time ignore the implications of other coefficients from the same sample and research method. The evidence suggests that, if cash is in fact fully valued by the market, then sampling error leads to the coefficient on cash (0.73 to 0.88) being understated and therefore the coefficient on credits (0.34 to 0.57) being overstated.