cavriends / tvp_vars_vb Goto Github PK

This repository contains all the code for the thesis that I have written to complete the MSc program in Econometrics & Management Science at the Erasmus University Rotterdam. The title of the thesis is: "Prior sensitivity in time-varying parameter vector autoregressions".

A rough outline of the file structure is as follows:

• data - contains all the empricial data for the the thesis
• excel - contains all the cross-table analysis of the results
• notebooks - contains all the notebooks that were used in the thesis
  • Runtimes.ipynb - computes the runtimes for each of the VI-based TVP-BVARs
  • Sensitivity analysis.ipynb - runs in parallel the sensitivity analysis of the hyperparameters for several priors
  • Simulation analysis.ipynb - analyses the results for the simulation study and compares the results of the VI-based TVP-BVAR to the MCMC-based TVP-BVAR
  • Simulation datasets.ipynb - generates all the datasets necessary for the simulation study according to several DGPs
  • Simulation study.ipynb - runs in parallel the simulation study for each of the priors
  • Visualisations.ipynb - to visualise the plots for the sensitivity analysis
• rcode - contains all the rcode that was used
  • GelmanRubin.R - calculates the Gelman-Rubin statistic for the MCMC-based TVP-BVAR and BVAR
  • functions.R - all the functions that are necessary to conduct the sensitivity analysis
  • results.R - analyses the results of the simulation study and compares MCMC-based TVP-BVAR to the BVAR and VAR
  • runtimes.R - calculates the average runtimes for the models that were programmed in R
  • simple_lr.stan - a simple OLS in Stan syntax to experiment with ADVI
  • simulation_study.R - runs the simulation study in parallel
  • stan_lr.R - used to experiment with ADVI and compare viability to MCM
• sensitivity - contains all .pkl files that were created in the sensitivity analysis
• simulations - contains the simulated datasets and results of the simulation study
  • datasets - contains the simulated datasets, mind you that this folder is aroud 1GB.
  • results - contains all the .pkl files with the results of the simulation study
• utils - the additional functions that are used in Python
  • data_utils.py - contains the standardisation, transformation and DGP functions
  • lstm_models.py - contains the code for an experiment with an LSTM
  • lstm_utils.py - contains the extra utilities necessary for the LSTM
  • tvp_models.py - contains the implementation of the VI-based TVP-BVAR for the three different priors
• visualisations - contains all the visualisations that are in the thesis

If you're not familiar with Git a good place to start is here.