This file provides step by step instructions to replicate all results in the paper, as submitted for publication to Review of Economic Studies.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
        Part 1: Data Availability Statements
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
This paper does not involve analysis of external data (i.e., no data are used or the only data are generated by the authors via simulation in their code). The code generating the Figures and Tables are in this repository.


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
        Part 2: Solve and Simulate the Model
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

This part can be done locally or on a HPCC cluster. Instructions below are for local execution. If recomputing all 40 economies (=paramater combinations) required for the paper, it is much faster to do it on the cluster.

------------------------------
Part 2a: Computing locally
------------------------------

Repeat the steps below for each economy in experdef_20201117.txt. The definitions of these economies are in experdef_20201117.m. Let the placeholder "econ" represent the name of the economy from this text file.

Run the economy for up to 300 iterations.
1.  Configure mainSB_create_env.m as follows (leave other variables as is):
	- expername = 'econ';
	- guess_mode = 'no_guess';
2.  Run mainSB_create_env.m and it will produce a file called env_econ.mat
3.  Configure main_run_exper.m as follows (leave other variables as is):
	- no_par_processes = XX; % where XX is the number of processors on the machine you're running it on
	- exper_path = 'env_econ.mat';
	- maxit = 300;
4.  Run main_run_exper.m. On a machine with 16 cores, this should take about 1 hour. It will create a file named res_TIMESTAMP.mat, where TIMESTAMP is the time at which the calcuation is finished expressed as YYYY_MM_DD_hh_mm
5.  Rename the res_TIMESTAMP.mat file to res_2020117_econ.mat.

Next, simulate. The model must be solved and res* file must exist.
6. Configure sim_stationary.m as follows (leave other variables as is):
	- resfile = 'res_2020117_econ';
7. Run sim_stationary.m. If you are running sim_stationary having run it before during the current MATLAB session, run "clear" beforehand. This operation will create the following files:
	- sim_res_20201117_econ.mat: all simulation results incl. full series, statistics, errors, and parameters
	- Results/statsexog_res_20201117_econ.xls: statistics using exogenous subsampling to define crises (one sample per worksheet)
	- Results/errstats_res_20201117_econ.xls: statistics of EE errors

Next, compute impulse response functions. Model must be solved and simulated. Both res* and sim_res* files must exist.
8. Configure sim_trans.m as follows (leave other variables as is):
	- resfile_list = {'res_20201117_econ'};	
9. Run sim_trans.m. On a machine with 16 cores, this should take about 10 min. It will create the following files:
	- GR_res_20201117_econ.mat: mean, median, and sd of IRF paths for each of 4 shocks (no shock, recession, run recession)
	
------------------------------------
Part 2b: Computing on the cluster
------------------------------------

The following instructions work for a HPCC cluster that uses a job scheduling engine for batch jobs. The code folder contains a sample batch script 'hpcc_script_aws_array.sh' used to run all 40 economies as part of a single batch job on Wharton's HPCC using dedicated AWS nodes. 

The configuration needed to run the numerical experiments as batch job on a cluster will vary substantially based on the specific cluster configuation. However, the basic structure should be transferable.  



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
			Part 3: Compute and Save Results
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

This part pre-supposes that the res* and sim_res* for each economy created by Part 1 exist. NOTE: the .mat files for all economies are available as zip archive "res_sim_files.zip" with the supplemental materials via the Review of Economic Studies publication.

The steps below re-create all the results used in the paper.
	
1.  Run writeStats.m. This will create Results/simres.mat, which contains a variable called "mapDict," a hash-map of every simulation and IRF statistic and model parameter. They can be retrieved by running mapDict("key"), where "key" has the following format:
	[command] - [econ] - [subsmaple] - [variable] - [statistic]
	
	command:
		- sim: simulation statistic
		- comp: simulation statistic relative to bench (either level or percent change)
		
	econ: name of the economy (as in experdef_20200910.txt)
	
	subsample: if command is "sim" or "comp"
		- u: unconditional
		- e: expansions 
		- r: recessions
		- c: crises (ie run recessions)
		
	variable: all variables reported by sim_stationary (as well as welfare "VHcons" for unconditional mean) 
	
	statistic: for command = "sim" and "comp", all statistics reported by sim_stationary e.g. "mean," "std, "AC". 
			
	
2.	After simres.mat has been created in step 1, use build.m script in subfolder "paper_tables" to fill in placeholders in the file "BLrev_v2021_tables.tex". To do so, open Matlab to this folder and execute: 
	
	build('BLrev_v2021_tables','../Results/simres.mat',[])
	
	This will create a new file "BLrev_v2021_tables_filled.tex", in which placeholders are replaced by the numbers from model simulations as in the paper. This file can be compiled in a standard Latex environment.	
		
3.  Compute base economy IRF graphs, Figure 1 in the paper.
		a. Configure sim_trans.m as follows:
			- resfile_list = {'res_2020117_base'};
			- no_par_processes = XX;  % where XX is the number of processors on the machine you're running it on
		b. Run sim_trans.m. This will create GTR_res_20201117_base.mat.
		c. Configure plot_trans.m as follows:
			- resfile = 'res_20201117_base';
		   Running plot_trans.m will create Figure 1.(pdf|eps).

4.  Compute financial crisis simulations graph, Figure 2 in the paper.
		a. Configure sim_trans_GFC.m as follows:
			- resfile_start = 'res_20201117_precrisis';
			- resfile_end = 'res_20201117_postcrisis';
			- resfile_steps={'res_20201117_postcrisis085',...
               'res_20201117_postcrisis085',...
               'res_20201117_postcrisis09',...
               'res_20201117_postcrisis09',...
               'res_20201117_postcrisis095',...
               'res_20201117_postcrisis095',...
               'res_20201117_postcrisis10',...
               'res_20201117_postcrisis10',...
               'res_20201117_postcrisis105',...
               'res_20201117_postcrisis105'};
			- no_par_processes = XX;  % where XX is the number of processors on the machine you're running it on
		b. Run sim_trans_GFC.m. This will create GFC_res_20201117_postcrisis.mat.
		c. Configure sim_trans_GFC.m as follows:
			- resfile_start = 'res_20201117_precrisis';
			- resfile_end = 'res_20201117_postcrisis08';
			- resfile_steps={'res_20201117_postcrisisc1',...
               'res_20201117_postcrisisc1',...
               'res_20201117_postcrisisc2',...
               'res_20201117_postcrisisc2',...
               'res_20201117_postcrisisc3',...
               'res_20201117_postcrisisc3',...
               'res_20201117_postcrisisc4',...
               'res_20201117_postcrisisc4',...
               'res_20201117_postcrisisc5',...
               'res_20201117_postcrisisc5'};
			- no_par_processes = XX;  % where XX is the number of processors on the machine you're running it on
		d. Run sim_trans_GFC.m. This will create GFC_res_20201117_postcrisis08.mat.
		e. Run plot_trans_GFC.m. This scripts reads the contents of the two GFC_ files generated above and also "Post Crisis.xlsx" that contains the time series for the data lines in the plot. It will create Figure 2.(pdf|eps).