The Role of the Book-to-Price Effect in Predicting Stock Returns: International Evidence

Abstract

Recent literature show that leverage has a negative effect on stock returns, which is contradicting with influential finance theories and models. Based on the time-period 1966-2015, the five-factor model and an international dataset, this thesis sets the focus on the question what kind of effect leverage has on stock returns when the B/P factor is decomposed into two components: an operating business risk component, defined as the enterprise B/P and reflecting asset risk, and a financing risk component defined as ND/P and reflecting risk caused by leverage. Investigation was done by performing 15 regressions based on the Fama-Macbeth (1973) procedure, with some of these 15 models also taking into account the twofold leverage effect. Although standard finance theories state that operating risk should be rewarded by higher stock returns, the results in this thesis indicate that the operating business risk component, enterprise B/P, has a statistically significant negative effect on stock returns. However, the operating covariance risk premium, another (additional) measurement of operating risk, seems to have a statistically significant positive relationship with stock returns. This not only raises the question whether the enterprise B/P is an inaccurate measurement of operating risk due to probably failing to proxy for only operating risk, but this also creates a new puzzle to examine. The results of this thesis also show that the financing risk component has a statistically significant negative effect on stock returns, confirming the findings of recent literature. Moreover, when financial leverage gets to only purely account for only interest expenses of debt, the relationship between leverage and stock returns does not alter. Finally, this thesis shed light on the effectiveness of these factor models by using the GRS-test. It came to light that none of the factor models passes the GRS-test due to the fact that, for each factor model, all pricing errors are jointly statistically significant different from zero, implying that these 15 factor models are not as effective as wanted.