Open UBI Optimizing UBI policy with machine learning
Open. UBI Optimizing UBI policy with machine learning NABIG 2019 Max Ghenis, Founder, Open. UBI
Open. UBI has come a long way LIFT+
Open. UBI But important questions remain. . . How will poverty change? How will work incentives change? How will the deficit change? How many people win or lose? How will inequality change?
Open. UBI . . . including the most important question HOW WILL UBI AFFECT ME?
Open. UBI Work's been done on some UBI-like policies Citizens' Climate Lobby (carbon dividend advocacy) https: //citizensclimatelobby. org/household-impact-study/
Open. UBI Work's been done on some UBI-like policies Citizens' Climate Lobby (carbon dividend advocacy) https: //citizensclimatelobby. org/calculator/
Open. UBI This is what I do for UBI policy overall Demo ubiplans. org (Short for www. open-ubi. org/plans)
Open. UBI Lots of metrics can quantify UBI policies Inequality impact Poverty impact Official, extreme, SPM; child, senior, etc. Gini, top 1% income share, etc. Income distribution Median, bottom 10%, etc. Deficit impact Work incentives Average marginal tax rate Household disruption Share better off, share with incomes decreasing more than 1%, etc.
Open. UBI What if we didn't just use metrics to assess plans. . . but also to design them?
Open. UBI We just saw a simple version of this Inequality impact Poverty impact Official, extreme, SPM; child, senior, etc. Gini, top 1% income share, etc. Income distribution Median, bottom 10%, etc. Deficit impact Work incentives Average marginal tax rate Household disruption Share better off, share with incomes decreasing more than 1%, etc.
Open. UBI Optimal taxation theory also addresses this Inequality impact Poverty impact Official, extreme, SPM; child, senior, etc. Gini, top 1% income share, etc. Income distribution Median, bottom 10%, etc. Deficit impact Work incentives Average marginal tax rate Household disruption Share better off, share with incomes decreasing more than 1%, etc.
Open. UBI For example, Saez and Gruber (2000) "The Elasticity of Taxable Income: Evidence and Implications" Given: ● ● ● Labor response to taxes Income distribution Social weights by income group Calculate: ● ● Optimal tax rates Optimal UBI
Open. UBI Kasy (2019) is applying optimal taxation to UBI https: //maxkasy. github. io/home/files/slides/UBI-experimental-design-Kasy. pdf
Open. UBI Let's apply this to a new type of UBI plan REPLACE ● All tax deductions and credits ● Payroll & AMT tax ● All benefits except Medicare/aid Includes Social Security and the standard deduction WITH 50% flat tax on all income Extra revenue funds UBI Modeled with Tax-Calculator 2. 2. 0
Open. UBI $5. 2 trillion Raised by this plan
Open. UBI $15, 731 UBI person (children, too) financed by this plan
Open. UBI A flat UBI would have dramatic effects Poverty falls 99. 6% 34. 5% come out behind Inequality falls 19% 29. 4% lose 5% or more disposable income Median income rises 31%
Open. UBI $x per child Under 18 What if we vary the amount by age group? $y per adult 18 to 64 $z per senior 65 or older
Open. UBI What if we vary the amount by age group? Optimizing for: ● Share left worse off ● Inequality
Open. UBI Suppose we want to minimize the losers 34. 5% come out behind today. To minimize would you increase kids, adults, or seniors? How much is too much?
Open. UBI Suppose we want to minimize the losers There are three variables to tweak to optimize: 1. Child UBI 2. Adult UBI 3. Senior UBI But this is actually only two, since the budget forces one of the UBI amounts to go up when others go down, and vice versa Choose child/senior, infer adult
Open. UBI Enter: machine learning Machine learning optimization techniques can explore the multivariate space. Steps: 1. Try a combination of {child UBI / senior UBI} (infer adult) 2. Calculate % losers 3. Try new combination (smartly) 4. Repeat 100 times
Open. UBI How much better do we do? 34. 5% losers to 31. 0% With UBIs of: $10, 710 per child $21, 250 per adult $2, 295 per senior (!)
Open. UBI How does this optimal reform compare? Child UBI Adult UBI Senior UBI Flat $15, 731 Min. losers $10, 710 $21, 250 Median income rise Loser share 5%+ loser share Poverty reduction Inequality reduction $15, 731 99. 6% 19% 31% 34. 5% 29. 4% $2, 295 22. 0% 16% 33% 31. 0% 26. 8% Less poverty and inequality reduction, higher median income, fewer losers
Open. UBI What if we optimize for inequality? From 19% inequality cut (Gini index) to 22% cut With UBIs of: $6, 141 per child $17, 481 per adult $25, 310 per senior (!!) Likely due to differing household sizes
Open. UBI Reducing inequality creates more losers Child UBI Adult UBI Senior UBI Flat $15, 731 Min. losers $10, 710 Min. ineq. $6, 141 Median income rise Loser share 5%+ loser share Poverty reduction Inequality reduction $15, 731 99. 6% 19% 31% 34. 5% 29. 4% $21, 250 $2, 295 22. 0% 16% 33% 31. 0% 26. 8% $17, 481 $25, 310 94. 0% 22% 21% 35. 2% 27. 9%
Open. UBI Values sometimes compete
Open. UBI These are political/philosophical questions How much disruption are we willing to give for a more equal, lower-poverty income distribution ?
Open. UBI Given those values, modeling can help This was a simple "search space" of only two parameters. What about expanding the problem to: ● ● Changing tax rates? Choosing programs to replace? Keeping certain deductions? Modeling dynamic effects like labor response?
Open. UBI How we can move forward What's nonnegotiable? What is negotiable? By what metrics do you choose? Philosophers, activists, and economists together
Open. UBI Thank you! Questions? max@open-ubi. org
- Slides: 31