Trend and Forecast Tools · Status
The tools are finished. What is not finished is the proof that the forecast works, and only time can supply it: each forecast is scored when its months have passed.
Section 1 · Table 1 · What is live
| Tool | Status | Where | Detail |
|---|---|---|---|
| Market trend line | Live | Every market statistics page | County repeat-sales index, the same one valuations use to bring comparable sales to today. Williamson: −3.7% over the 12 months to September 2026. |
| Forecast model v2 | Running | jensen, monthly on the 15th | Writes one forecast a month to a hash-chained ledger that is never edited. Last run October 15, 2026: completed; ledger verified, 2 rows, newest origin September 2026. |
| Price outlook | Live | Every new valuation | Per-home 6- and 12-month forecast, calibrated: its 80% range held 80% at 6 months and at least 80% at 12 months on 14,415 sales (back-test, September 2025 – August 2026). |
| Forecast on market pages | Off | Since October 22, 2026 | The county-level range is too narrow (Section 3). Market pages show the trend line only. |
Section 2 · Table 2 · The ledger · every forecast issued
| Origin | Issued | Row hash | 6-month score due | 12-month score due |
|---|---|---|---|---|
| August 2026 | Oct 8, 2026 8:54 PM | c23a13cc6591 | May 2027 | Nov 2027 |
| September 2026 | Oct 8, 2026 9:09 PM | e79670256f6f | Jun 2027 | Dec 2027 |
A forecast is scored once the market truth for its horizon is frozen, 3 months after the forecast month, so late-reported sales are counted. The ledger chain was verified by the newest monthly run.
Section 3 · What the evidence says so far
County range too narrow. In replay over 2005–2022 the market finished inside the 80% range only 62–69% of the time.
No gain at 12 months. Nothing tested has beaten “prices stay where they are” at 12 months; on the 2023–2026 holdout the bootstrap p-value was 0.498.
Per home, calibrated. The price outlook in valuations adds the valuation’s own error and is scaled to hit its range: 80% held in back-test.
Why it stays on in valuations. It is calibrated and recorded; the county forecast is neither calibrated nor better than no change.
Section 4 · Table 3 · When the proof arrives
| Milestone | Rule | Earliest |
|---|---|---|
| First score | August 2026 forecast, 6 months, truth frozen at T + 6 + 3 | May 2027 |
| Interim readout, 6 months | 12 matured 6-month forecasts | April 2028 |
| Interim readout, 12 months | 12 matured 12-month forecasts | October 2028 |
| Formal verdict | 60 matured 12-month forecasts (pre-registered) | October 2032 |
Dates assume one forecast is issued every month from August 2026 on, as the schedule does. A missed month moves every later date by a month.
Section 5 · Nothing to do
The monthly run, the ledger and the scoring run on their own. The next forecast, origin October 2026, is issued on November 15, 2026. The method is set out in The ALEX Market Trend Model (alex-companies.com/market-data/).
Status, not a valuation. Ledger rows, run results and schedule are read from jensen when this page is built. Replay, holdout and back-test figures are the recorded results of the v2 specification and its calibration. Forecasts are model outputs, not appraisals and not promises of future prices. ALEX Companies · Licensed by the Texas Real Estate Commission · Broker: Alexander Tiffany Southwest, LLC, #565194-BB.