Rational expectations in housing markets: the case of survey forecasts

This article examines whether survey forecasts of U.S. housing activity satisfy the rational-expectations hypothesis. Using monthly data for building permits, housing starts, construction spending, and home sales, we test both long-run cointegration and short-run efficiency in Mincer – Zarnowitz and supersaturated Autometrics regressions. Forecast errors are systematically predictable, indicating departures from full rationality. The Autometrics results reveal time-varying and episodic bias – most pronounced around the 2007–09 financial crisis and the 2020 pandemic – consistent with adaptive learning, structural change, or shifting behavioural anchors rather than persistent inefficiency. Anchoring-bias regressions further show that consensus forecasts adjust sluggishly to new information. Together, these findings suggest that housing forecasts reflect evolving expectations and regime shifts, not random noise. Recognizing these dynamics can improve applied forecasting: bias-adjusted or regime-sensitive consensus forecasts may offer more reliable guidance for analysts, investors, and policymakers.