+
+
+
+
+
+
+ |
+
+ |
+
+
+
+
+ |
+
+ New research · Tax year 2027
+
+
+ The IRS’s inflation adjustments for tax year 2027
+
+
+ On November 5, the Internal Revenue Service released its annual adjustments for tax year 2027 to almost 70 tax provisions that Congress indexes to inflation. The IRS raised the standard deduction for single filers from $16,100 to $16,600 and the maximum Child Tax Credit from $2,200 to $2,300 per child. It also raised the credit’s refundable portion from $1,700 to $1,800 per child. A family whose credit exceeds its income tax can receive up to that amount per child as a refund.
+
+
+ We project these amounts before the IRS publishes them, and for 2027 the IRS raised most of them by more than we projected. The IRS’s increases exceeded our projected increases by about 9% at the median across the income tax bracket thresholds and the standard deduction, Earned Income Tax Credit, capital gains and alternative minimum tax amounts. For example, we projected that the standard deduction for single filers would rise $450, to $16,550, and the IRS raised it $500, 11% more than we projected. Our projections matched both Child Tax Credit amounts, $2,300 and $1,800.
+
+
+ We added the IRS’s 2027 amounts to the latest release of policyengine-us, our open-source US model for Python. Our web app and parameter explorer run an earlier version of the model, so for 2027 they show our projections, not the IRS’s amounts.
+
+ |
+
+
+
+
+
+
+
+ |
+ Standard deduction
+
+
+ | Filing status |
+ 2026 |
+ 2027, IRS |
+ 2027, our projection |
+
+
+ | Single |
+ $16,100 |
+ $16,600 |
+ $16,550 |
+
+
+ | Married filing jointly |
+ $32,200 |
+ $33,200 |
+ $33,150 |
+
+
+ | Head of household |
+ $24,150 |
+ $24,925 |
+ $24,850 |
+
+
+ | Married filing separately |
+ $16,100 |
+ $16,600 |
+ $16,550 |
+
+
+ |
+
+
+
+
+ Also in this issue: Military retired pay · Ballot measure results · Prop 40 webinar recording · PolicyBench AI benchmark · Model updates · Talks
+
+ |
+
+
+
+
+ |
+
+ Military retirement · State income tax
+ How states tax military retired pay in 2026
+
+ Twenty-eight states exempt all military retired pay from income tax in 2026, including Utah, which gives a credit that cancels its tax on the pay. Another 13 states exempt all, part or none of it, depending on the amount of pay, the retiree’s age, income, earnings and residency, and how long and when the retiree served. Nine more states levy no broad income tax and don’t tax the pay, and the District of Columbia taxes all of it.
+
+
+ For example, take a married couple in which one spouse, age 45, draws $30,000 of military retired pay and earns $60,000 in a civilian job. Because of the pay, the couple owes $3,600 more in federal income tax. The couple also owes more state income tax in nine states and DC, from $60 in Maine to $2,310 in Oregon. Alabama and Missouri exempt the pay and let filers deduct federal income tax. There, the extra $3,600 of federal tax lowers the couple’s state income tax, by $180 in Alabama and $25 in Missouri. After the extra federal and state income tax that the pay causes, the couple keeps $24,090 to $26,580 of the $30,000, depending on the state.
+
+
+ Or take a retired couple aged 67 and 65 who live on $45,000 of military retired pay and $30,000 of Social Security. Because of the pay, the couple owes $1,710 more in federal income tax, or $2,080 more in Alaska, where it receives the Permanent Fund Dividend. The couple also owes more state income tax in ten states, up to $440 in Hawaii. After the extra federal and state income tax that the pay causes, the couple keeps $42,850 to $43,290 of the $45,000, depending on the state. Of those ten states, Delaware and Montana tax the pay itself, and in seven, including Hawaii, the couple loses a credit. In the tenth, Idaho, the couple has to file a return because of the pay, and Idaho charges $10, its permanent building fund tax, on each return it requires. Read the military retired pay analysis.
+
+ |
+
+
+
+
+ |
+
+ November 3 ballot · Results
+ What voters decided on the seven measures we analyzed
+
+ Voters approved three of the seven tax and benefit measures whose effects we estimated on our 2026 ballot measures dashboard, and rejected four. State election offices reported these unofficial counts by November 13.
+
+
+
+ | • |
+ California Proposition 40. Voters rejected a one-time tax on the net worth of California’s billionaires, 47.6% yes to 52.4% no. On Forbes data as of October 29, we estimated $117.6 billion in one-time revenue, assuming no behavioral response, such as billionaires leaving the state. Applying the assumptions of the Berkeley estimate, by Galle, Gamage, Saez and Shanske, we estimated a net present value of $103.1 billion. Applying those of the Hoover estimate, by Jaros and Rauh, we estimated a net loss of $34.1 billion in present value. Two Hoover assumptions account for most of the difference: counting the future California income tax that billionaires who leave would no longer pay, and the chance that legal challenges stop collection. |
+
+
+ | • |
+ Colorado Amendment 87. Voters approved replacing Colorado’s 4.4% flat income tax rate with six rates, from 3.7% to 8.4%, starting in tax year 2027, 51.6% yes to 48.4% no. We estimate that the state collects $1.43 billion more in individual income tax in 2027 than under the flat rate. Colorado’s nonpartisan Legislative Council Staff estimates $635.7 million more from individual taxpayers on 2023 incomes. |
+
+
+ | • |
+ Colorado Proposition NN. Voters approved permanently raising the revenue limit in Colorado’s Taxpayer’s Bill of Rights (TABOR), 57.0% yes to 43.0% no. Under TABOR, the state must refund revenue above the limit. With the revenue it keeps under the higher limit, the state first reimburses local governments for property tax exemptions and puts the rest in a new Children’s Account. The state pays part of its TABOR refunds through a sales tax refund to filers. We estimate that filers receive $276 million less from it in 2027 than they would have without the measure, and Legislative Council Staff projects $259.8 million less. |
+
+
+ | • |
+ California Proposition 3. Voters approved making California’s 10.3%, 11.3% and 12.3% income tax rates permanent, 60.7% yes to 39.3% no. At 2025 thresholds, California applies the rates to taxable income above about $371,000 for single filers and $743,000 for joint filers. Under prior law, the rates would have ended after 2030. We estimate that they raise $10.1 billion in 2031, and California’s nonpartisan Legislative Analyst’s Office expects $5 billion to $15 billion a year. |
+
+
+ | • |
+ Washington Initiative 645. Voters rejected repealing the 9.9% tax on individual income over $1 million and enacting a state law that bars state and local taxes on individual income, 42.1% yes to 57.9% no. The Legislature enacted the tax in 2026, and it starts in 2028. We estimated that repeal would have cut revenue by $3.66 billion in tax year 2028. Washington’s Office of Financial Management, the executive branch’s budget office, estimated $3.11 billion for fiscal year 2029, from July 2028 to June 2029. |
+
+
+ | • |
+ North Dakota Initiated Constitutional Measure No. 2. By 2,889 votes, voters rejected requiring every public school to serve free breakfast and lunch to all its students, 49.4% yes to 50.6% no. The State Canvassing Board meets on November 18 to canvass the votes and make the results official. For kindergarten through grade 12, we estimated a cost of $114 million per two-year budget. The Secretary of State’s guide, which also counts pre-kindergarten, gave $124.3 million to $134.3 million for 2027–29. |
+
+
+ | • |
+ South Dakota Amendment I. Voters rejected tying the state’s constitutional Medicaid expansion to a 90% federal share, 44.2% yes to 55.8% no. Under current federal law, the federal government pays 90% of the cost of expansion and the state pays 10%. The Attorney General’s explanation said the amendment would end expansion if the federal share fell below 90%. The sponsors’ statement said it would end the constitutional requirement to provide expansion, not expansion itself. We modeled a scenario that follows the Attorney General’s reading, in which the federal share fell below 90% and expansion ended on July 1, 2027. In that scenario, at the state’s 10% share, we estimated that the state would have spent $32.6 million less on Medicaid in 2028. The 29,055 people enrolled through expansion in an average month from July 2024 to June 2025 would have lost Medicaid. We estimated that 20,694 of them would have had incomes below the poverty line, too low to qualify for the premium tax credits that help pay for Affordable Care Act marketplace coverage. |
+
+
+ |
+
+
+
+
+ |
+
+ Webinar recording · October 30
+ Why the Prop 40 revenue estimates differ
+
+ In our October 30 webinar, Fernando Hoces de la Guardia, a researcher who works on transparency in policy analysis, joined David Trimmer and Max Ghenis of PolicyEngine. David covered Prop 40, which voters rejected on November 3, and the Berkeley and Hoover estimates of its revenue. In his open policy analysis, Fernando put the two on one definition to reconcile them and published the materials to reproduce his work. Max then ran both sets of assumptions in our Prop 40 calculator, and David walked through the 2026 ballot measures dashboard. Watch the recording (62 minutes).
+
+ |
+
+
+
+
+ |
+
+ PolicyBench
+ GPT-6.1 Terra and Claude Fable 5.2 join the leaderboard
+
+ We run PolicyBench, an open benchmark of how accurately language models calculate household taxes and benefits. We count a dollar amount as right when it falls within $1 of PolicyEngine’s answer, and an eligibility answer as right when it matches PolicyEngine’s. On the November 11 leaderboard of 52 models, OpenAI’s GPT-6.1 Terra ranks third with 93.1% of answers right, and Anthropic’s Claude Fable 5.2 ranks fifth at 92.0%. OpenAI’s GPT-6 Sol leads at 95.2%, ahead of Anthropic’s Claude Opus 5.5 at 94.1%.
+
+
+ We also tested America.gov, a General Services Administration site that answers questions with AI and information from government websites. In our October test, it scored 87.9%, matching OpenAI’s GPT-5.5 from April, and we run every leaderboard model, GPT-5.5 included, with AI alone and no web access. Read the PolicyBench notes on these results.
+
+ |
+
+
+
+
+ |
+
+ Model updates · Since October 29
+
+ We added these changes, among others, to the latest release of policyengine-us:
+
+
+
+ | • |
+ We added regular heating assistance from the Low Income Home Energy Assistance Program (LIHEAP) and state energy programs in 18 more states. In Hawaii, one of the 18, we model eligibility and the flat benefit for subsidized housing but not the dollar awards for private housing. We also extended our model of California’s LIHEAP utility and wood heating assistance from Riverside County to the whole state, and added its propane and fuel-oil assistance in Madera County. |
+
+
+ | • |
+ From the health policy research organization KFF, we added 2027 Affordable Care Act marketplace premiums in every rating area for the lowest-cost bronze plan and the benchmark plan, the second-lowest-cost silver plan. Under federal law, each household’s premium tax credit depends on the benchmark premium in the rating area where it lives. KFF puts the national average benchmark premium for a 40-year-old at $703 a month. |
+
+
+ | • |
+ DC’s temporary 100% match of the federal Earned Income Tax Credit for filers with qualifying children expired after 2025. DC matches 85% for 2026 through 2028 and 100% from 2029, and we apply those rates. |
+
+
+ | • |
+ Under Senate Bill 164, California lowers the asset limit in Medi-Cal, its Medicaid program, to $21,000 for one person and $31,000 for two on July 1, 2027. California applies the limit in Medi-Cal programs that don’t use modified adjusted gross income (MAGI) rules, such as coverage for aged, blind or disabled people, and we added the new amounts. |
+
+
+ |
+
+
+
+
+ |
+
+ Talks
+
+ On November 5, Max joined Tara Watson of the Brookings Institution and Linda Giannarelli of the Urban Institute on a roundtable, “State Policy Impacts on Family Resources: Tools to Improve Policy Analysis.” Cynthia Osborne of Vanderbilt University moderated the session at the Association for Public Policy Analysis and Management’s Fall Research Conference in Boston. See Max’s slides on PolicyEngine for state policy research.
+
+
+ On the Tax Chats podcast, Max talked with accounting professors Scott Dyreng of Duke University and Jeff Hoopes of the University of North Carolina about whether AI can correctly calculate a tax liability. Listen to the 32-minute episode.
+
+ |
+
+
+
+
+ |
+ Best wishes, The PolicyEngine Team
+ |
+
+
+
+
+
+
+
+ |
+
+