SGDinference

The replication_qr_sgd repo provides the replication files for the following paper:

• Lee, S., Liao, Y., Seo, M.H. and Shin, Y., 2023. Fast inference for quantile regression with tens of millions of observations. Forthcoming in Journal of Econometrics. https://doi.org/10.1016/j.jeconom.2024.105673.

Applications

Data Files

Because of the large file sizes, the Github repo cannot hold the data files. To replicate the application results, please download a zip file: data.zip (1.6GB).

You need to extract it under the working directory. All data files should be located at application/data/

We use the following 6 files for estimating the model:

• data_ipums_1980.csv
• data_ipums_1990.csv
• data_ipums_2000.csv
• data_ipums_2005.csv
• data_ipums_2010.csv
• data_ipums_2015.csv

These files are contained in the above data.zip file. You can also generate those 6 files from the scratch by running data_cleaning.R file over

• PCEPI.csv
• usa_00005.cbk
• usa_00005.csv.gz
• usa_00005.xml.

All of these files are also contained in the data.zip file.

Source Files

Four R files are located in the application/src/ folder:

• fn_estimate_model.R: contains an R function to estimate the model.
• make_table_04.R: generates Table 4 (summary statistics) in the paper.
• make_tables_and_charts.R: generates all other Tables and graphs in the paper.
• run_qr_sgd_app.R: is the main source code to estimate the model.

All results are saved under the application/results/ folder

Simulations

All simulation results and the example running code is contained in the simulation folder.

• all_results_from_clusters: This folder contains all simulation results conducted in the larger cluster system.
• example_running_code: This folder contains running code for a small-scale simulation design.
  • The filename is sim_coverage.R.
  • We set $n=10^5$, $p=10$ with 10 replications.
  • The simulation results are saved under simulation/example_running_code/p10/n1e05/.