Estimating Income Without a Job

Every dollar figure in this guide that depends on your income depends on the income you predict, not the income you end up with. For a person with a salary, prediction is trivial. For a person three weeks into a business, it is the hardest number in the file, and since 2026 it is also the most dangerous one to get wrong. This lesson is how to build the estimate honestly and manage it as the year reveals itself.

The number the marketplace wants

The application asks for your household's modified adjusted gross income for the full calendar year. Start from last year's tax return as a floor, then rebuild it piece by piece for this year: wages through your last day, any severance or payout, unemployment benefits, interest and dividends, a spouse's salary if you file jointly, and net profit from the business. Net, always: self-employment income counts after business expenses, so the $30,000 invoice year with $12,000 of costs is $18,000 of income.

Two rules do most of the tripping. First, the whole calendar counts: a salary through July pushes the number up even if you earn nothing after. Second, the household is your tax household: a spouse's income joins yours in the estimate even though only you lost coverage.

What changed in 2026: the caps are gone

Through 2025, the system forgave estimation errors. If your advance subsidy came in too high and your actual income outran the estimate, repayment was capped by income band, $375 for a household under twice the poverty line, scaling up to about fifteen hundred dollars. A law passed in 2025, Public Law 119-21, eliminated the caps for 2026 coverage: every excess dollar of advance subsidy is now repaid at tax time. The marketplace's own answer to this is matter-of-fact: there is no limit on the repayment. One exception survives in the fine print: if your actual income would have made your household eligible for Medicaid or CHIP, you do not repay the excess.

The flip side is worth saying out loud: overestimating is safe in direction and expensive in monthly cash. A founder who projects $70,000, takes no subsidy, and finishes at $45,000 gets the correct subsidy on the tax return as a refundable credit. The safe error, in other words, is high. A worked reconciliation from the tax-preparer literature: a household estimated at $80,375, took its credit monthly, finished $2,000 over, and repaid $325. Finished $2,000 under, and it collected $349 more. Small misses in the middle of the scale stay small. Misses that cross the cliff line do not.

How I would build the estimate, concretely

Start with committed money: contracted revenue, retainers, the severance, the spouse's salary. Add a conservative haircut of pipeline you can actually see, not pipeline you hope for. Subtract documented expenses you know you will incur. That is the estimate. Write it down with its pieces, because you will update it, and updates need the original. If your launch year began with a layoff, you have built pieces of this projection once already; estimating launch-year income is the severance-and-unemployment version, and its numbers plug straight into this one.

Then set two triggers in your calendar. If a big contract lands and the year is trending over the cliff line, update the marketplace application that week: raising your estimate reduces or stops the advance subsidy and shrinks the eventual repayment to zero. If the business stalls and the year is trending toward the Medicaid line from "Medicaid and Your State's Line," update it too: eligibility moves with the fact, and the marketplace will route you correctly if you tell it.

The December move

The one structural advantage of the cliff is that it is measured on a calendar year, and by November you usually know. If you are within a few thousand dollars of the line, the last weeks of the year decide which side you land on, and you still have open enrollment in front of you: if the year is landing over, you can choose a cheaper metal, or price the spouse plan, or hold the COBRA option differently. The system gives you one correction window per year, and it lands exactly when your information is best. Use it that way.

The estimate is the last piece of machinery. One subject remains before the plan: the products that look like doors and are not.