---
title: "NAHB/Wells Fargo Housing Market Index (HMI) — methodology and forecast use"
authors: "National Association of Home Builders (NAHB)"
source: "NAHB.org"
source_url: "https://www.nahb.org/news-and-economics/housing-economics/indices/housing-market-index"
paper_id: "NAHB HMI (monthly survey, 40+ years)"
harvested: 2026-08-22
year: 2026
keywords: [NAHB, HMI, builder sentiment, survey, housing starts, leading indicator]
---

## Abstract / Summary

The NAHB/Wells Fargo **Housing Market Index (HMI)** is a monthly survey of single-family
builders, conducted for 40+ years, gauging builder perception of three conditions: current
single-family home sales, sales expectations for the next six months, and traffic of prospective
buyers — each rated "good", "fair", or "poor." These are scored into a seasonally-adjusted index
on a 0-100 scale where **over 50 indicates more builders view conditions as good than poor**. It
is a widely-cited gauge of builder sentiment and a directional predictor of single-family housing
starts (NAHB chief economists explicitly reference HMI when forecasting single-family starts).

**Relevance to StratMac RE-sector work — documented exclusion (do NOT wire in):**
- **Not on FRED's public API/CSV** under HMI/NAHB/NAHBINDEX (documented in
  `re-predictive-modeling.md`; full history is NAHB member/commercial data).
- **Structural reason it cannot improve the long-horizon walk-forward:** HMI history begins 1985
  (~41 years to present), which is < the ~15y-train + 10y-test walk-forward requirement
  (MIN_TRAIN + TEST_SPAN ≈ 25y) → the aligned sample collapses to < 2 full folds = the spectral
  skill's **thin-window trap**. A 1985+ builder-sentiment series cannot be walk-forward-validated
  inside the current long-horizon design.
- This is a *defensible exclusion, not an access problem*: builder-level sentiment leaders most
  likely to lower the HOUST/PERMIT error ceiling cannot meet the walk-forward sample bar. An SM
  data-access outreach letter was drafted (`strategy/outreach-nahb-research-data.md`, unsent).