Code for "A Macroeconomic Model with Financially Constrained Intermediaries and Producers"

Vadim Elenev, Tim Landvoigt, Stijn Van Nieuwerburgh¹

Fast Steps to Reproduce Benchmark Model

• In Matlab, execute the script main_create_env.

  • If you have the Matlab Symbolic Math Toolbox and use version 2018b or earlier, leave the flag useJacobian in line 43 on. Otherwise set to false to use numerical differentiation.

  • Note: generating the analytic Jacobian for the benchmark model takes approximately 5 minutes with version 2018b, and can take longer for the other experiments.

  • main_create_env will create a file env_bench_ini0 that contains the experiment definition for the benchmark economy on the initial coarse grid.

• Execute script main_run_exper to run the benchmark on the coarse grid for up to 100 iterations.

  • You can set the number of parallel workers to be started in line 12.

  • Set to zero if you want to run it with a single process.

  • On a computer with sixteen cores (and 16 parallel workers) this should take about 30 minutes.

  • main_run_exper creates a results file named res_[current_date_time] that contains the converged policy functions.

  • Rename this file to res_20200910_bench_i100.mat.

• Execute main_create_env again using results file from previous step as input.

  • Configure main_create_env as follows (leave other variables as is):

    • xpername = 'econ';

    • guess_mode = 'guess';

    • guess_path = 'res_20200910_bench_i100';

  • main_create_env will create a file env_bench that contains the experiment definition for the benchmark economy on the fine grid, using the resulting policy functions from the coarse grid iterations in res_20200910_bench_i100.mat as initial guess.

• Execute script main_run_exper to run the benchmark on the fine grid for up to 30 iterations.

  • Configure main_run_exper as follows (leave other variables as is):

    • exper_path = 'env_bench.mat';

    • maxit = 30;

    • price_zns = true;

  • Set to zero if you want to run it with a single process.

  • On a computer with sixteen cores (and 16 parallel workers) this should take about 45 minutes.

  • main_run_exper creates a results file named "res_[current_date_time]" that contains the converged policy functions.

  • Rename this file to res_20200910_bench_s130.mat.

• Simulate the model using sim_stationary and sim_trans_cluster.

  • sim_stationary simulates the model policies contained in res_20200910_bench_s130.mat for 10,000 periods and writes out the resulting time-series and several statistics. The main output is a file named sim_res_20200910_bench_s130.mat.

  • sim_trans_cluster reads both res_20200910_bench_s130.mat and sim_res_20200910_bench_s130.mat, and simulates generalized IRFs.

  • To plot IRFs, run plot_trans.

For More Details See readme_ECTA.pdf and readme_replication.txt

¹: Elenev: Johns Hopkins University, Carey Business School; email: velenev@jhu.edu. Landvoigt: University of Pennsylvania Wharton School, NBER, and CEPR; email: timland@wharton.upenn.edu. Van Nieuwerburgh: Columbia University Graduate School of Business, NBER, and CEPR; email: svnieuwe@gsb.columbia.edu.