Skip to content
wheremyrefund

How the forecast works.

Time-to-refund statistics on anonymous timelines, with the waiting people counted correctly.

Waiting on a refund?

Check my refund

Free. Read in your browser.

Timelines

A shared timeline holds transaction codes and the number of days between them, counted from the day the return posted (code 150) or from the e-file acceptance date. It never holds names, SSNs, calendar dates or amounts. The clock for a forecast starts at the event that matters for you: the return posting, a code 570 hold, a code 571 release, a code 810 freeze or a code 977 amended return.

Counting people who are still waiting

Many timelines end before the refund arrives. Dropping them would make the wait look shorter than it is, so each one counts as "not paid by day N" (right-censored). The curve of the share paid by each day is a Kaplan-Meier estimate (Kaplan and Meier, 1958), weighted by each timeline's anti-abuse trust score; self-reported dates count half.

You have already waited

Dates shown to you are conditional on not having been paid yet: the median is the first day by which half of the people still waiting on your day were paid. Unconditional figures would put the date in the past for anyone who has waited a while.

Groups and small samples

Groups go from the most specific (season, situation, week, credits, processing cadence) to the broadest. The first group with at least 8 timelines is used and blended with the next group of 30 or more, with weight n / (n + 50) (empirical Bayes, after Efron and Morris, 1975). Under 30 timelines, quantiles come from a log-logistic model fitted to the censored data (Nelder-Mead).

Season pace

Last season's timelines are scaled by how fast this season pays compared with last season at the same week, read from the IRS's weekly filing season statistics (limited to 0.7 to 1.6). In 2026 the IRS paid faster than in 2025: the scale was 0.971 at the end of February, 0.922 at the end of March and 0.842 in mid-April.

When there is not enough data

The page shows the IRS's own published window instead, with its source, and says so.