Visa Bulletin forecast: method, accuracy and current numbers

Dates in the Visa Bulletin are set by hand each month, so any forecast is a rough guide. This page explains the simple method used on this site, shows how well it did on past data, and lists the current projections. No false precision: the ranges below are wide because the history is.

The method

1. Pace. For a category and country, take the average change per bulletin of the Final Action Date over the last 12 bulletins, counting only months where both the month and the previous month have a real date (not C or U).
2. Damped extrapolation. Forecast for h months ahead = latest date + λ × pace × h. The damping factor λ (between 0 and 1) shrinks the extrapolation because fast months are usually followed by slow ones. “λ = 0” means no movement.
3. Uncertainty band. The range is the 10th to 90th percentile of the errors this method made in the past, pooled over all 40 category-country series, so it covers roughly 80% of past outcomes.
4. Choice of λ. λ and the band were chosen using forecasts made up to December 2021 and then tested on forecasts made afterwards, so the test numbers below are out-of-sample.

Backtest: how wrong was it?

For every month since October 2013, every category and country in the main grid (EB-1/2/3, F1, F2A, F2B, F3, F4 × five columns), the method forecasted 3, 6, 12 months ahead and was compared with what the bulletin later published. Months where the starting or target value was C or U are skipped. “MAE” is the mean absolute error in days.

Reading it honestly. At 3 months, the damped method is only slightly better than assuming nothing moves (λ = 0). At 12 months it helps more, but the typical error is still about 10 months, and the band is wide. For individual series with stalls, sudden jumps or Unavailable months the error can be much larger than these pooled averages; for instance, a retrogression is not something this method can anticipate.

Current projections (Final Action Dates)

Central estimate and 80% range of the cut-off date for the bulletin month 6 and 12 months after October 2026. “Pace” is days per month over the last 12 bulletins.

No projection (Current, Unavailable or too little date-to-date movement): EB-1 Rest of World (Current); EB-2 Rest of World (1 Jan 2025).

What this does not model

• Pending demand in USCIS or NVC queues (not public in detail).
• Policy changes, court orders, legislation or processing pauses.
• Spillover of unused numbers between categories and fiscal years.
• Your own case: priority date, category and country, and whether you can file.

Code and data are reproducible from the repository (scripts/model.py). Use the calculator to apply the same pace to your own priority date.