---
title: "Forecasting Using Long-Order Autoregressive Processes: An Example Using Housing Starts"
authors: "Michael A. Sklarz; Norman G. Miller; Will Gersch"
source: "AREUEA Journal (Wiley) / IDEAS RePEc"
source_url: "https://onlinelibrary.wiley.com/doi/abs/10.1111/1540-6229.00438"
paper_id: "AREUEA Journal 15(4):374-388 (1987)"
harvested: 2026-08-22
year: 1987
keywords: [housing starts, autoregressive, AR, forecasting, metro]
---

## Abstract

Michael Sklarz, Norman G. Miller, and Will Gersch applied **long-order autoregressive (AR)
processes** to forecast housing starts — the canonical early univariate / time-series treatment
of starts specifically. As documented in the *Counselors of Real Estate* review of house-price
forecasting methods, this was an early example of forecasting at the **metropolitan level**: the
model was run on 316 metropolitan areas so the forecasts could catch local effects (of high
practical interest to practitioners). The authors found **strong positive serial correlation**
in housing starts, so a long-AR specification could track the series well.

**Effectiveness / limitation:** autoregressive methods extrapolate the cycle the series is
already in, so they track short-horizon movement well but are structurally **flat at turning
points** — they contain no independent structural or leading-indicator content.

**Relevance to StratMac RE-sector work:** this is the classic counterpoint to StratMac's
screen-then-model approach. Our Tier-1 finding that *prices, credit, and broad activity*
(CSUSHPINSA, REALLN, TOTALSL) are the load-bearing predictors — rather than the target's own
history — argues structurally against a pure long-AR model for level/amplitude. Cite as the
historical root of univariate starts forecasting, setting up why a leading-indicator panel is
the improvement.

*Note: AREUEA Journal is paywalled via Wiley; this entry records metadata + the review's
characterization. Full text requires institutional access.*