monash_171907.pdf (6.76 MB)
"Empirical studies of asset pricing models: an emerging market perspective for Indonesia"
thesisposted on 2017-10-11, 04:19 authored by Nurjannah, Nurjannah
The Capital Asset Pricing Model (CAPM) has been one of the most challenging topics in financial economics since its introduction in the early 1960s. However, several empirical studies have progressively questioned its ability to explain the risk-return relationship. First, some arguments confront the usefulness of beta as the single measure of risk of a security. Therefore it is important to consider extensions of the CAPM such as the Fama French 3-factor model and the Carhart 4-factor model. Second, the implementation of asset pricing models has traditionally relied on the assumption of constant betas and constant risk premia which contradict the mounting empirical evidence that both may vary over time. As the accuracy of beta is crucial in determining investment strategies and company valuation it is important to study its behaviour. Especially in an emerging markets such as the Indonesian stock market where international investors and fund managers are seeking global diversification for a better risk-return trade-off. This thesis aims to provide empirical evidence of the applicability of the asset pricing factor models to an emerging market by extending the CAPM to the Fama French 3-factor model and the Carhart 4-factor model that incorporate the value, size and momentum effects. Further, we investigate the extent of risk (and risk premium) instability and its time-varying properties by using several approaches including quantile regression and nonparametric techniques. The empirical performance of the CAPM, Fama French 3-factor model and Carhart 4-factor model at firm level as well as portfolio level using both full-sample period and Fama MacBeth testing methods reveal inconclusive results. and yield ambiguous conclusions regarding the impact of risks (betas) on returns. Further, there is an indication of asymmetry in the return distribution which needs further attention. Empirical results obtained with quantile regression indicate that the relation between stock returns and the risk premia are not uniform and, occasionally, the relation is not even monotonic. The general conclusion that can be drawn is that there exist disparities in behavior between underperforming and overperforming stocks (portfolios) that may be receiving negative or positive idiosyncratic shocks regarding beta risk, size risk, value risk and occasionally, momentum risk. Furthermore, it can also be concluded that the Kalman filter estimates generally outperforms the rolling regression estimates. The nonparametric method using a cubic spline function to estimate the risks with knot 8 reveal that for each method, the estimated betas are not stable over time and most of them are significantly different from zero. Various methods generate different betas in their temporal evolution and nonparametric risk premia are generally smoother compared to the rolling and Kalman filter estimates. Overall, the Kalman filter estimates is the best compared to the nonparametric and the rolling regression methods. The last chapter investigated the functional concurrent linear model also known as functional regression to estimate the time-varying beta. The results suggest that the functional return series of Indonesian stocks typically exhibit upward and downward movement of beta and show high level of heterogeneity. The dynamics of average returns through velocity and acceleration reveals high fluctuation during the monetary crisis periods. In conclusion, the main finding of this research confirms that the beta risk changes through time with the changing economic environment. There is no clear pattern of association between the market risk betas estimated from four different methods and size or industry portfolio returns. There is weak evidence that the functional regression estimator outperforms the other methods in both full-sample and time-varying risk premium derived from the cross-sectional regression.