Methodology & sources
How forecasts are collected, matched to actual data, and graded.
1. Which forecasts count
- Only published, dated, attributable forecasts with a public source link. Every row records forecaster, publication date, target period, value and source URL. If we cannot source a number, it is not on the site.
- For scoring we use year-ahead calls: the forecast published in roughly November–January for the end (or average) of the following calendar year. Mid-year revisions are shown as "current" but not scored, so each forecaster gets one scored call per target year.
- Forecasters: Fannie Mae Economic & Strategic Research (ESR), Mortgage Bankers Association (MBA), National Association of Realtors (NAR), Realtor.com, Wells Fargo Economics, Redfin and Zillow. NAHB and others are not yet included because we could not find a dated, numeric year-ahead call in a public source.
2. Actuals
- 30-year rate: Freddie Mac Primary Mortgage Market Survey, weekly (FRED: MORTGAGE30US). Target definitions follow each forecaster: Q4 average (Fannie Mae, MBA quarterly tables) = mean of weekly readings in Oct–Dec; year-end (Realtor.com, NAR, Redfin, Wells Fargo quarter-end figures) = mean of weekly readings in December; calendar-year average = mean of all weekly readings in the year.
- Home prices: FHFA purchase-only house price index, seasonally adjusted, monthly (FRED: HPIPONM226S). Q4/Q4 targets compare Oct–Dec averages; Dec/Dec compare December levels; annual targets compare calendar-year averages. The S&P CoreLogic Case-Shiller national index (CSUSHPINSA) is charted for context only.
- A period is scored only when its data are complete (rates: data through the last week of December; prices: December index published).
3. Scoring math
Error = forecast − actual (percentage points). Mean absolute error (MAE) = average of |error| over a forecaster's scored rows. Bias = average signed error (negative = expected lower than reality). Sample size n is always shown.
Head-to-head MAE uses only the target years covered by every forecaster with at least two scored rows for that metric. This stops a forecaster from looking good just because its record starts after a hard year (e.g. 2022). Forecasters without full coverage of that window are marked "prov." (provisional) and unranked.
| Grade | Rate MAE (pts) | Home-price MAE (pts) |
|---|---|---|
| A | ≤ 0.25 | ≤ 1.5 |
| B | ≤ 0.50 | ≤ 3.0 |
| C | ≤ 0.75 | ≤ 4.5 |
| D | ≤ 1.00 | ≤ 6.0 |
| F | > 1.00 | > 6.0 |
Small samples (n = 2–4) mean one lucky or unlucky year moves a grade a lot. Grades describe past accuracy only.
4. Known limitations
- Home-price forecasters target different indexes (existing-home median price, ZHVI, Redfin median sale price, Fannie Mae HPI, FHFA). We score all against FHFA for comparability; this penalizes or flatters some forecasts for index differences rather than skill.
- Some rows come from reputable news roundups quoting a forecaster (marked by source domain) rather than the forecaster's own document; the publication date is then the article date.
- Wells Fargo's tables report quarter-end rates; we compare them with December averages. NAR's 2024–2026 calls are calendar-year averages.
5. Sources
Loading source list…
Disclaimer
This site is for informational and educational purposes only and is not financial, investment, or lending advice. Forecasts belong to their respective publishers; we are not affiliated with any of them. Data may contain errors — always check the linked source. Past forecast accuracy does not predict future accuracy. Talk to a licensed professional about your own situation.
Questions or a sourced forecast we missed? We review additions monthly. — The Forecast Scorecard team