This is a periodic newsletter of the interesting things we’ve seen and what we are thinking about in open source policy analysis.
Tax-Calculator 3.0.0 released. Policy Simulation Library has released version 3.0.0 of the Tax-Calculator microsimulation model. The latest release adopts ParamTools for parameter processing and validation, hosts new documentation on Jupyter Books, creates a new parameter for adjusting inflation-indexed parameters, and includes various data updates. Release notes
Research using Tax-Calculator presented at the Tax Economists Forum. Max Ghenis of the UBI Center presented results from a coauthored paper comparing the effects of extended unemployment relief, payroll tax cuts, and a universal basic income. Presentation Slides
Salesforce open sources the AI Economist. Researchers from Salesforce and Harvard University have collaborated to create the AI Economist, a machine learning project based on reinforcement learning that analyzes individuals’ behavior to produce an optimal tax policy. While using reinforcement learning to assess various tax policies from the government or planner’s perspective has been implemented before, the application of reinforcement learning to individuals is an important feature that reveals behavioral responses to tax policy not previously estimated in the literature. The project was developed by Richard Socher from Salesforce Research and David Parkes from Harvard University. Link
The Economist magazine open sources its election prediction model. The Economist magazine has released the source code for its 2020 presidential election forecasting model on Github. The open-source Bayesian model is written in R, and Stan and has been updated from previous iterations to incorporate support for nonresponse bias and data from state-level elections earlier in the year. The project was developed by the data science team at The Economist in collaboration with political scientists Andrew Gelman and Merlin Heidemanns from Columbia University. Link
Edited by Matt Jensen, Peter Metz, and Jacob Chuslo