TY: THES
T1 - Application of alternative regression models to deal with proportions as dependent variables
A1 - Marques, José Lourenço Pires
N2 - The main purpose of this thesis is to consider different approaches to deal with proportions as dependent variables in regression models.
The Classical Linear Regression Model (CLRM) is the approach that most researchers apply to their data. However, the CLRM is inappropriate to deal with bounded variables whose response is restricted into the interval (0, 1) as dependent variables since it may possibly yield fitted values for the variable of interest that surpass its lower and upper limits.
Due to the CLRM weaknesses, in this thesis we will consider some alternative parametric regression models that include the additive logistic normal distribution, the censored normal distribution, the Beta distribution and the normal distribution with nonlinear response function. A quasi-parametric regression approach will also be considered.
In the empirical case we consider a dataset with financial information from US firms. The dependent variable of the models we intend to estimate is the debt to maturity, which is measured as a proportion of the total debt of the firm that has a maturity larger than three years. The explanatory variables are the abnormal earnings, the asset maturity and the size of the firm.
To compare the above models will be used the Akaike?s information criterion (AIC) and Schwarz criterion (SBC). The distribution that displays the lowest values on both criteria is the best to study proportions as dependent variables. We will also study the adjusted value of each model.
UR - https://repositorio.iscte-iul.pt/handle/10071/3355
Y1 - 2012
PB - No publisher defined