Code README for: Managing Climate Change Under Uncertainty: Recursive Integrated Assessment at an Inflection Point
The model is written in MATLAB and requires the COMPECON toolbox by Miranda and Fackler. It can be found at: http://www4.ncsu.edu/~pfackler/compecon/toolbox.html. As is, the model will replicate results found in the paper. However, users can change the code to explore different scenarios (e.g. different damage functions).
There is a main folder and four subfolders. One subfolder, smolyak, contains the Smolyak collocation codes. Some of these codes are adapted from ones graciously provided online by Lilia and Serguei Maliar. Others are partially adapted from Python code provided at EconForge: https://github.com/EconForge. Three scripts within the smolyak folder,
ind2subVect, are from the MATLAB file exchange, while
uniqueperms is from MATLAB central.
The second subfolder, results, contains the solutions (basis function coefficients) to the problem.
The third subfolder, compecon, is the folder for the COMPECON toolbox codes which can be downloaded at: http://www4.ncsu.edu/~pfackler/compecon/toolbox.html.
The fourth subfolder, terminal_workspaces, contains the structure for the terminal value function.
How to run the model:
main_dice.m - the main script to run the entire model. This will generate a workspace.mat in a folder under the results subfolder, with indicators in the folder name for what type of run it was.
time_paths.m - The subroutine to simulate paths conditional on the solution to the finite horizon problem. Load the workspace.mat from the folder that corresponds to the model you wish to simulate. channels_out contains the optimal carbon tax channels found in the main text.
- The controls for the model code are effective consumption and abatement cost. Using abatement cost as a control instead of abatement allows for a linearly constrained maximization problem. Investment is the residual from the resource constraint.
- The analytic components of the SCC/tax calculation in time_paths will not work for
approximation_level= 1 due to the low order approximation to the value function.