/DTVP-VAR

Neural Network SVAR

Primary LanguageHTML

Deep Time-Varying Parameters (DTVP) Structural VAR model (SVAR)

By Rodrigo Chang (rrcp@banguat.gob.gt) at Banco de Guatemala @r2cp

Check out the introductory presentation

This is a toy example of a prototype model that combines a Multilayer Perceptron to estimate the time-varying coefficients of a SVAR.

This is work-in-progress. Inspiration for this work comes from Prof. Phillippe Coulombe's idea of a Macro Neural Network.

flowchart LR
    A[Time variaton sources S] --> |Neural network G| B(VAR parameters β)
    B --> C(Macro variables y)
    D[Lagged values X] --> C
Loading

I will upload here the following work:

  • scripts/nn-svar.jl: The toy example network estimated with simulated data.
  • scripts/poly-example.jl: This is an example I worked for understanding how to estimate a neural network with the Lux.jl package in Julia.
  • notebooks\intro.qmd: This is a Quarto notebook with my elevator-pitch presentation given on Sept. 2.
  • notebooks\final.qmd: This is the Quarto presentation with my progress at the end of the course.

For this project, I used Julia. Feel free to try out the project by instantiating it:

julia> ] 
(DTVP-VAR) pkg>