By: Adam Michael Bauer -- adammb4 [at] illinois [dot] edu

To cite our working paper that uses these codes: Bauer, A. M., F. McIsaac, S. Hallegatte. How Delayed Learning about Climate Uncertainty Impacts Decarbonization Investment Strategies. World Bank Policy Research Working Paper No. WPS10743, World Bank Group, Washington DC, 2024.

This set of codes reproduces all of the figures and analysis carried out in How delayed learning about climate uncertainty impacts decarbonization investment strategies. This package uses Gurobi, a commerical nonlinear programming solver that is free for academics, but may not be free for everyone. (It is unclear to me if it's freely available for all researchers or just researchers at universities. I imagine Gurobi will handle this on a case-by-case basis, so just shoot their customer support staff an email and they can help.)

Each code is assigned a number corresponding to the figure it creates. Note an si before the script name indicates that figure is in the Supplementary Information. So code 01_xxx.py makes Figure 1 from the main text, which shows our calibration of the marginal abatement cost curves, while code si01_xxx.py makes the Figure 1 from the Supplementary Information. Here is the full table for both versions:

If you're an academic, you can email Gurobi customer support to get a free academic license. It's easy to install, and once it's installed, I believe you'll be good to go to run the codes.

A final note is that you should consider using the .yml file provided in this directory to establish a virtual python environment that should include all of the necessary dependencies for the code to run smoothly. I recommend using conda to do this.

To run the codes, simply navigate to the codes directory and run the numbered code to recreate the desired figure. If you want to run the program script_name, you may need to execute:

    chmod +x script_name

to grant execution permissions (hence the +x) to the script you want to run.

As an example, if you want to recreate Figure 1 which shows our calibration of the marginal abatement cost curves, you would simply run:

./01_mac_calibration.sh

Notice the first bit of the above program name, 01_mac_calibration.sh, matches the figure number we wanted to create, Figure 1.

All figures will be deposited into the codes/figs folder. To run indiviudal simulations, you can run any of the files in simulation_mains, and to make individual figures, you can run any file in the figure_mains folder. Note: You should run all scripts from the codes directory. As an example, let's say you want to run the invBase_cvxpy_main.py file in the ar6_15 calibration, but not save the output. Then in your command line, you'd use:

python simulation_mains/invBase_cvxpy_main.py ar6_15 1 0

Note: You should be operating in the Python environment provided at the head directory. Without it, I make no guarantees any of this will run on your machine (and even then, well, mileage may vary...).

The only figures that were not able to be reproduced on a member of the World Bank Group's Reproducibility team's computer were scripts si07 and si08. These figures were verified virtually, with a team member joining via video call and watching the code run on the author's laptop.

The hypothesized reason is that both si07 and si08 have highly convex objective functions, which requires more powerful hardware to solve precisely than what was available to the reproducibility team member. The team member got an optimal_inaccurate solution during optimization. The code will throw an error when anything other than an optimal solution is found. It was verified that the original author of the code gets an optimal solution when the si07 and si08 codes are run.

If a user gets an optimal_inaccurate solution for either of these codes, one possible course of action is to edit the scale parameter found in the codes/simulation_mains/invRec_RiskPrem_cvxpy_main.py file. This can be found on lines 34 through 42.

The hardware of the original author is a 2023 MacBook Pro with an M2 Pro Chip and 16 GB of RAM.

Last edited: 7 June, 2024.