-
Notifications
You must be signed in to change notification settings - Fork 31
Bayesian Hierarchical Models in Finance
Multivariate Bayesian Hierarchical Models are being increasingly investigated to solve problems in finance.
These models offer potential advantages over many other model types in that they will adapt to new data
automatically over time, and that the researcher does not necessarily need to identify all the factors or
features which are important to the model before model development begins.
R has strong support for the STAN modeling language via the rstan package. We do not feel that there are deficiencies in this package, and intend to use it for this project.
This project will be a research replication project where the student will plan to replicate three or more papers using Bayesian Hierarchical Models in finance using rstan. The replication should follow, to the degree possible, the process outlined here.
The primary goal is to produce a documented, rigorous, replication of three or more papers in this area, including such things as literature review, hypothesis summary, hypothesis tests, and reproducible code (likely as rmarkdown vignettes.
It is unclear whether this reproducible research should live in its own R package, or be integrated into another existing package such as PortfolioAnalytics.
Reproducible research has its own value, of course. The primary additional value to the R (finance) community will be a set of fully documented examples of reproduced research backed up with code which can be used to build further models, or fit on other data. Eventually, this could become the template for a more standardized workflow for creating multivariate Bayesian Hierarchical models, but I do not feel that a GSoC student will be able to get that far in three months.
Brian Peterson (brian@braverock.com) is the primary author of PortfolioAnalytics.
Michael Weylandt
- provide code to demonstrate a multivariate Bayesian Hierarchical model using R and STAN. It does not need to be on of the target papers for the project, and can be a different model flavor, but must demonstrate skill in R, STAN, and rmarkdown or LaTeX.
Students, please post a link to your test results here.
- Abhinav Gupta: Test Solution
- Yun Yan: Solution webpage (see here) | Source Codes at repo (see here)