Bayesian Hierarchical Hidden Markov Models applied to financial time series.
- Mentors
- Brian Peterson, Michael Weylandt
- Organization
- R project for statistical computing
The goal of this project is to replicate research in Hierarchical Hidden Markov Models (HHMM) applied to financial data. This model is a generalization of Hidden Markov Models (HMM), which in turn is part of the Dynamic Bayesian Networks (DBN) family. I identified three academic works with interesting ideas and application that do not provide data and code. Replication will allow future readers to assess the credibility of the results and program improvements on top of fully working code. Ultimately, published code can be integrated into current research code (for example, for backtesting) and will be proposed as a Case Study to the Stan Development Team.
The concept of hidden states could enrich many trading strategies. A detailed replication that provides literature review, reproducible code and solid documentation will allow future readers to implement HHMM logic into existing trading frameworks (for example for covariates, signals and benchmarks). Published code for this GSoC 2017 may be leveraged in future work to provide a generic implementation of hidden state models to already existing R Packages for trading analysis.