-
Notifications
You must be signed in to change notification settings - Fork 6
Factor analysis based on higher order moments
Several factor extraction methods exist based on the covariance matrix. The most well known one being PCA.
Also higher order moments can be used for factor extraction. The most well know approach being ICA.
Several other approaches have been proposed for extracting factors from higher order moments. There is no unified way to do this factor extraction allowing to easily compare results. The goal of this GSOC project is to extend the hofa package and include such a systematic approach both in testing for the number of factors and extraction of them.
There are two stages:
- Selection of number of factors (function already available in hofa is ger.sel)
- Estimation of the factors (function already available in hofa is hofa.als)
The package is complementary to the PerformanceAnalytics package which uses factor models to estimate higher order comoments (structured approach) but does not provide functionality for a detailed analysis of the factor model results.
The hofa package is available at https://github.com/GuanglinHuang/hofa
The package is too specific and does not allow an easy comparison across methods.
What is needed is a general function for factor selection (methods: Bai and Ng, Anh and Horenstein, Lu et al., ...) and a general function to estimate the factors.
Simulated and empirical datasets are needed.
Mentors, please explain how this project will produce a useful package for the R community.
Kris Boudt, Nathan Lassance
- EVALUATING MENTOR: Nathan Lassance is an expert in higher order moment modelling.
- Kris Boudt is an expert in higher order moment modelling and coauthor of various R packages. He has extensive experience with mentoring GSOC projects.
/
The potential student is the author of the hofa package