Moment set selection for the SMM using simple machine learning

This repository contains the official implementation of Moment set selection for the SMM using simple machine learning by Eric Zila and Jiri Kukacka.

If you find our work useful, we encourage you to use the following citation:

@article{zilakukacka2023,
    title = {Moment set selection for the {SMM} using simple machine learning},
    author = {Eric Zila and Jiri Kukacka},
    journal = {Journal of Economic Behavior & Organization},
    volume = {212},
    pages = {366-391},
    year = {2023},
    issn = {0167-2681},
    doi = {https://doi.org/10.1016/j.jebo.2023.05.040},
    url = {https://www.sciencedirect.com/science/article/pii/S0167268123001944}
}

Requirements

To download the content of the repository and move to the root directory, run:

git clone git@github.com:jirikukacka/Zila_Kukacka_Moment_Selection.git
cd Zila_Kukacka_Moment_Selection/

To install DrWatson, run Julia and execute:

julia> using Pkg
julia> Pkg.add("DrWatson")

To install the required packages, run Julia (>= 1.6.1) in the root directory and execute:

julia> using Pkg
julia> Pkg.activate(pwd())
julia> Pkg.instantiate()

Data

Empirical data used for empirical estimation are located in the data/ folder.

Reproduction

The repository contains all the code needed to reproduce experiments from the paper for the Markov-switching multifractal (MSM) model. The necessary scripts are located in the scripts/ folder.

To estimate the MSM model using the three benchmark moment sets employed in the paper, execute:

julia scripts/main_markov256state_bench.jl

To perform the backward stepwise moment elimination (BSME) algorithm as proposed in the paper, execute:

julia scripts/main_markov256state_bsw.jl

To perform the forward stepwise moment selection (FSMS) algorithm as proposed in the paper, execute:

julia scripts/main_markov256state_fsw.jl

Results

For the Markov-switching multifractal model, the full set of results is located in the results/ folder. The corresponding visualizations are located in the plots/ folder. In order to reproduce them, please, refer to the notebooks/markov256state.ipynb Jupyter notebook file executable with a Julia kernel.

License

All content in the repository is licensed under the MIT license. More information can be found in the license file.