This file describes the replication package for:
"AIDS, Human Capital and Development", by Rodolfo E. Manuelli and Emircan Yurdagul, prepared for Review of Economic Dynamics.

The replication folder has two blocks: Data and Model. Data part uses Stata to construct the statistics. Model part uses Fortran to solve the model and uses Stata to construct figures/tables from model outputs.

All run times refer to a computation with Intel I5 processor with 2.5Ghz. 


-----------------------------------

Data:
First, we need to run the codes in the Data folder, which gives the empirical input of the paper. The do files below should be run in the following order (unless noted otherwise).

- main.do: This is the parent do file that calls the sub-codes for each separate data set used in this paper. It first calls the do files that take the raw data downloaded from each source and converts them to the format merge-able with the rest of the data. It then calls a do file that puts these together to produce the output to be used in the calibration, and the data entries of figures/tables in the paper. Below is the description of these sub-codes. (2 minutes)

- WB2020_Together.do: This file takes the excel downloads from World Databank provided by the World Bank and puts them together. The details on the raw file such as the download info are provided on top of the do file. The output of this file is called later in All_data.do.

- UNAIDS_Build.do: This file uses three excel sheets downloaded from AIDSinfo provided by the UNAIDS. It converts them into compatible format with the rest of the datasets and produces a dta file for each variable to be called later in All_data.do.

- HDD.do: This file uses one csv file downloaded from the Human Development Data and converts into a dta file to be called later in All_data.do.

- All_data.do: This file constructs the main data that we use in the figures/tables of the paper corresponding to empirical results. 

- Construct_IES.do: (This file is self-standing; it is irrelevant at what order it is run.) This file uses the downloads from Income and Expenditure Survey provided by the Statistics South Africa. It computes the two wage ratios that we use in our calibration.

In addition to the results from the raw data, we include in the Data folder the explanation of how we derived the costs of intervention (ART treatment and condom distribution) that we use in the Section 5 of our paper. The descriptions are given in:

- Costs\Treatment: This gives an excel file that uses the information in Dutta et al. (2015) to construct the treatment costs we use in the paper. There also an accompanying a word file that explains the computation.
- Costs\Condom: This gives an excel file that uses the information in a supplementary file for Stover et al. (2017) and computes the costs of condom intervention (costs per infection averted). It includes an accompanying word file that explains the computation. 

Note 1: In constructing the data folder, we included a subfolder for each separate data source, to highlight better the separation across sources. Do files for each source is included in the corresponding subfolder. For sources that directly provide the dta file the folder is empty. The researcher should download the raw data files into these folders. 
Note 2: Except some in-text numbers, the above do files do not produce themselves any figure or table in the paper. The do files for the model (Main_model.do) merges the data output with the model output to generate the figures/tables. In-text numbers from the data are printed to the excel file “Data\In-text numbers.xls”.

----------------------------------
Model:
We solve each model or experiment with two fortran codes. One is for the demographics (age-health distribution, population growth) and one is to find the policy functions and get the aggregates using the demographics. We first solve the benchmark, then experiments, and then the robustness checks. Here we first go over the fortran codes, and then describe the stata do files that are run to construct the tables/figures. For the fortran codes, the run times are conditional on starting with the guesses included in corresponding folders.

Benchmark model: We first run the demographic solution, and then use the output of it in finding the policy functions and aggregate objects.

- distribution.f95: This file computes the distribution of health-age groups in each country-year, and calibrates the demographic parameters of the model. Its inputs are guesses for parameters, and the calibration targets. (3 minutes)
- solve_oldy.f95: This one computes the policy functions, equilibrium paths of human capital across health-age groups, and then aggregates to compute the output and other macroeconomic variables for each country-year. It calibrates for each country year the TFP level to match its output per worker relative to South Africa. Its inputs are the distribution of health-age groups from distribution.f95, guesses of TFPs for each country-year. (2 minutes)
Experiments: Once we run the benchmark, we run 18 experiments with the model. First we run the code for demographic distributions in these experiments then solve the rest of the model using its output. The experiments here the twists of the model we do for the accounting exercises of Section 4 and the policy interventions of Section 5. The codes explain what each experiment corresponds to.  
- distribution_experiment.f95: This is the counterpart of the distribution.f95 code of the benchmark. It loops over the experiments, solves each of them and calibrates their demographic parameters. (35 minutes)
- solve_experiment.f95: This the counterpart of the solve_oldy.f95 of the benchmark. It loops over the 18 experiments and solves the non-demographic parts of each model, calibrating country-year-specific TFP for each experiment. (35 minutes)

Robustness: We have 5 robustness checks (See Online Appendix). For each of them, we have a separate folder in this replication package. These folders are independent of each other. We explain in the beginning of each code how it differs from the benchmark codes.

- Robustness_hb: This sets the initial human capital (h_b) at 0.5. We first run the fortran code “Robustness_hb/distribution.f95” and then “Robustness_hb/solve_oldy_robustness_hb.f95”. (3+2=5 minutes)

- Robustness_nu: This sets the effective labor of infected not-treated people (nu_A) at 0.85. We first run the fortran code “Robustness_nu/distribution.f95” and then “Robustness_nu/solve_oldy_robustness_nu.f95”. (3+2=5 minutes)

- Robustness_r: This sets the interest rate (r ) at 0.07. We first run the fortran code “Robustness_r/distribution.f95” and then “Robustness_nu/solve_oldy_robustness_r.f95”. (3+2=5 minutes)

- Robustness_retirement: This sets the retirement age (R) at 70. We first run the fortran code “Robustness_retirement/distribution_robustness_retirement.f95” and then “Robustness_retirement/solve_oldy_robustness_retirement.f95”. (3+2=5 minutes)

- Robustness_pstar: This sets the length of non-susceptible period (p*) at 17. We first run the fortran code “Robustness_pstar/distribution_robustness_pstar.f95” and then “Robustness_pstar/solve_oldy_robustness_pstar.f95”. (3+2=5 minutes)

Do files for figures/tables:

- Main_model.do: This is the parent do file for all model results. It calls the text outputs of the fortran files to construct the figures tables. (2 minutes)
Below we go over each sub-do file called by Main_model.do. The main file specifies which figure or table each of these correspond to.
- Tables.do: Merges the benchmark model output with data summaries, produces the main figures and tables. (Figures 1,2, Appendix Tables 1,2,3,4)
- additionalfigures.do: Produces the figures and tables for the policy interventions of Section 5. (Appendix Figure 1)
- Figures_experiments.do: Combines the output of experiments with the benchmark model and the data.
- Figures_experiments_2020_intervention.do: Produces the figures and tables for the policy interventions of Section 5. (Figures 6,7, Table 2)
- Figures_experiments_2020_accounting.do: Produces the figures and tables for the accounting exercises of Section 4. (Figures 3,4,5)
- Figures_robustness_X, where X={r, nu, hb, pstar, retirement}. These codes take the outputs of the robustness check model solutions, merges them with benchmark results, can plots the comparison of implied average schooling in each robustness check against those in the model. (Appendix Figure 2)