Bottom-up leading macroeconomic indicators: An application to non-financial corporate defaults using machine learning

This repository holds the replication files for experiments from:

Tyler Pike, Horacio Sapriza, and Tom Zimmermann. Bottom-up leading macroeconomic indicators: An application to non-financial corporate defaults using machine learning September 2019. Working Paper

Abstract: This paper constructs a leading macroeconomic indicator from microeconomic data using recent machine learning techniques. Using tree-based methods, we estimate probabilities of default for publicly traded non-financial firms in the United States. We then use the cross-section of out-of-sample predicted default probabilities to construct a leading indicator of non-financial corporate health. The index predicts real economic outcomes such as GDP growth and employment up to eight quarters ahead. Impulse responses validate the interpretation of the index as a measure of financial stress.

Workflow

1. Prepare data

   1. create firm-level data
      script: corp_default_data_creation
      output: corp_default_data_fisd.RDS

   2. calculate summary statistics and latex tables
      script: corp_default_data_tables
      output: latex table in terminal

   3. create summary charts of firm-level data
      script: corp_default_data_charts
      output: /Figures/observed_data.pdf

2. Create NFCH

   1. estimate firm-level defaults
      script: corp_default_estimation
      input: corp_default_data_fisd.RDS
      output: /Estimation/[predicted values], /Estimation/[tuning AUC charts][ROC curves][variable importance]

   2. calculate cross-sectional moments and aggregate firm-level probabilities into macro index
      script: corp_default_index_creation
      input: /Estimatin/[predicted values]
      output: /Data/Output/indexes_aggregate.csv

   3. chart aggregate index
      script: corp_default_index_charts
      input: /Data/Output/indexes_aggregate.csv
      output: /Evaluation/Index/[charts]

3. Validate index

   1. compare in- and out-of-sample AUCs of machine learning methods
      script: corp_default_auc
      input: /Estimation/[predicted values]
      output: /Evaluation/AUC/auc_plot.pdf

   2. estimate impulse response functions
      script: corp_default_IRF
      input: /Data/Output/indexes_aggregate.csv
      output: /Evaluation/JordaIRF/[charts]

   3. perform forecasting exercises
      scripts: corp_default_forecast_insample, corp_default_forecast_outsample
      input: /Data/Output/indexes_aggregate.csv
      output: latex tables in terminal

Notes

Set up

1. The required output folders are included in this repo, however, if they need to be reinstantiated, then running the initialize.sh script will create the needed directories to run the replication.
2. One will need access to FRB data sources to fully run this replication, but publicly available data may be substituted in as a proxy in several instances.

Run-time

1. File paths assume a unix environment and that the working directory is the project's root.
2. The recursive index takes days to run, if constructed from scratch; parallel processing is necessary for reasonable execution times.
3. The user's WRDS credentials will have to be supplied in corp_default_data_creation.R and corp_default_index_creation.R
4. Files may be run one-at-a-time independently or all together via replicate.sh