dice2016

python implementation of DICE2016 using pyomo and ipopt

installation

1. clone from GitHub
2. install required dependencies (requires conda, see Installing Conda) with conda env create -f environment.yml

run

1. activate environment with conda activate dice2016
2. to edit parameters that control the model behaviour, edit the dice2016_parameters.csv file (note that some combinations of parameters may cause infeasibilities or weird results)
3. then python dice2016.py.

notes

for my template I used the DICE2016R3-112018-v8 implementation. This was adjusted for 2018 prices. There are several versions of DICE2016 knocking around. Colours may vary if you compare to a different version. However I think all the versions of DICE2016 are structurally the same.

this implementation doesn't quite reproduce the GAMS version of Nordhaus (you can do your own comparison with the data at https://github.com/chrisroadmap/loaded-dice/blob/main/dice2016/DICER3-opt.csv), but it gets pretty damn close. I don't know if this is down to using a different solver (ipopt versus CONOPT), differences between GAMS and Pyomo, or me mucking up the translation of the GAMS code. In the third case do raise an issue if you find a clear and obvious error (i.e. one that VAR would correct). Results are sensitive to assumptions and parameter values anyway so I claim that they are close enough to be useful.

ipopt version

versions of ipopt that run on different platforms vary. On Windows, newer versions of ipopt are not found by the conda interpreter. The workaround is to downgrade to v3.11.1 which is found. In the environment.yml file, change the line under dependencies to ipopt==3.11.1 (https://stackoverflow.com/questions/64912995/applicationerror-no-executable-found-for-solver-ipopt-in-pyomo). On Mac Silicon and Ubuntu Linux, the version available is 3.14.5 (#1), and this is picked up by the interpreter and should run out of the box.

acknowledgements

this code was inspired by the DICE2013 implementation by Optimized Financial Systems available from https://github.com/moptimization/pythondice2013implementation. Without their skeleton I would not know where to start. They produced this model several years ago and it no longer runs out of the box, so I also brought their DICE2013 implementation code up to date to work with python 3 and newer versions of pyomo, at https://github.com/chrisroadmap/pythondice2013implementation. Also thanks to Bill Nordhaus for making the original DICE2016 GAMS code available.