gcosta151 / rs-portfolio-opt

Regime switching portfolio optimization

Portfolio Optimization under a Markov Regime-Switching Framework

This repository contains the source code to conduct numerical experiments similar to those presented in the following two papers:

• Costa, G. and Kwon, R. H. (2019). Risk parity portfolio optimization under a Markov regime-switching framework. Quantitative Finance, 19(3), 453-47
  • Link to PDF
• Costa, G. and Kwon, R. H. (2020). A regime-switching factor model for mean–variance optimization. Journal of Risk, 22(4), 31-59
  • Link to PDF

The work in these two papers pertains to a Markov regime-switching factor model that captures the cyclical nature of asset returns in modern financial markets. Maintaining a factor model structure allows us to easily derive the first two moments of the asset return distribution: the expected returns and covariance matrix. By design, these two parameters are calibrated under the assumption of having distinct market regimes. In turn, these regime-dependent parameters serve as the inputs during portfolio optimization, thereby constructing portfolios adapted to the current market environment. The proposed framework leads to a computationally tractable portfolio optimization problem, meaning we can construct large, realistic portfolios.

Dependencies

• Julia v1.x
• JuMP.jl v1.x
• Ipopt.jl v1.x
• TimeSeries.jl v0.23

Usage

This repository contains all the files used to necessary to run the numerical experiments of the regime-switching portfolio optimization framework. To run the experiments. please refer to the main.jl file. Anyone wishing to make any changes to the models can do so by tinkering with a copy of the code base. The code base is made up of the following files:

• dataload/DataLoad.jl: Module to download data from Kenneth French's data library. Use this module to download returns of the Fama-French three-factor model, as well as the Industry Portfolios to serve as the historical asset returns.
• optimization/PortfolioOptimization.jl: Module to construct optimal nominal and regime-switching portfolios. Six portfolio optimization models are currently available for use:
  • mvo: Nominal mean-variance optimization
  • rsmvo: Regime-switching mean variance optimization
  • minvar: Nominal minimum variance optimization
  • rsminvar: Regime-switching minimum variance optimization
  • rp: Risk parity portfolio optimization
  • rsrp: Regime-switching risk parity portfolio optimization.
• optimization/Optimization.jl: Supporting script called by the PortfolioOptimization.jl module. This script contains the JuMP-based optimization models.
• optimization/HiddenMarkovModel.jl: Supporting script called by the PortfolioOptimization.jl module. This script contains an implementation of the Baum-Welch algorithm to fit a hidden Markov model to the factor returns.
• optimization/FactorModels.jl: Supporting script called by the PortfolioOptimization.jl module. This script contains an implementation of linear regression under a single regime, as well as under the assumption of multiple regimes.

Licensing

Unless otherwise stated, the source code is copyright of Giorgio Costa and licensed under the Apache 2.0 License.