.png){kind=icon}
.png){kind=icon}
Return
COPULA REGRESSION AND GENERALIZED LINEAR MODELS: A COMPARATIVE STUDY WITH AN APPLICATION
.png){kind=icon}
PRE
AI
DOI：10.17654/0972361726020.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Classical linear regression, despite its widespread use and simplicity, often proves inadequate for various applications. The presence of nonlinear relationships and non-normal error distributions necessitates the adoption of alternative modeling approaches. Generalized linear models (GLMs) are a prevalent alternative method, yet they require the response variable to belong to the exponential dispersion models (EDMs) family of distributions. In certain practical scenarios, the response variable may follow a distribution outside of this family, highlighting the significance of copula regression, which does not impose stringent conditions on the probability distribution. Consequently, this paper conducts a comparative study between GLMs and copula regression using real world data. To strictly evaluate the predictive performance and generalizability of the models, a 5-fold cross-validation approach was used. The findings demonstrate that the t-copula regression model yielded superior performance compared to the gamma regression with a log-link function.
Keywords:
t-copula
copula regression
gamma regression
elliptical
Journal
A
IF:
0.2
Papers:
37
Citations:
0