#HMC.jl

This repo contains code to estimate the implementation of the Hidden Markov Chain model. Code was written against julia 1.0

Getting Started

To download the needed packages do the following

1. Start julia
2. Enter the Pkg manager prompt by pressing ]
3. Enter activate . to activate project
4. Enter instantiate to download needed packages

From now one start julia with the project flag from the root directory, julia --project=. to load the needed packages.

File Structure

• code contains scripts to run stuff. Some of these are softlinked into the project root directory to make it easier to run on the cluster

• src contains the code for the module Hmc.jl which contains all the functions used in the estimation

• test contains unit test code

• data contains the data for the project -- raw contains the raw inflation data -- output contains output files from the estimation scripts

• thoughts contains latex code outline various procedures

• slurmscripts contains the slurm shell files to run the commands on the cluster

Data

The estimation code outputs the results from each Gibbs draw for the various parameters in the model. If 1000 samples are drawn in the estimation for each date then the csv files will have 1000 lines for the samples drawn for date range estimated. Each line starts with the end date of the estimation. For each of these csv's there is also a second csv with the mean of the samples for each date. In this csv there is exactly one line for each date. These csv's will have _summary as suffix. The directory structure should make clear what's being estimated

• filtered_forecasts_x.csv are the 1-12 month ahead forecasts for inflation for the estimation ending on the given date
• filtered_means_x.csv are the means for the three states for the estimation ending on the given date
• filtered_variances_x.csv are the variances for the three states for the estimation ending on the given date
• filtered_state_probs_x.csv are the state probabilities for the three states for the last date of the estimation ending on the given date
• filtered_trans_probs_x.csv are the transition probabilities for the three states for the estimation ending on the given date. These are outputted by A[:] to transform the matrix to a vector and need to be reshaped in column major order (default in Julia and Matlab but not Python)
• smoothed_state_probs_x.csv are the state probabilities for the three states for each date of the sample from the last estimation

Code

run_hmm.jl is the main script which estimates the HMM model. It takes a command line argument to determine which inflation series to use in the estimation. Call the script as either julia --project=. -p <number of cores to use> run_hmm.jl official or julia --project=. -p <number of cores to use> run_hmm.jl alter

aggregate.jl is a secondary script that calculates aggregate statistics