Think national, forecast local: a case study of 71 German urban housing markets

ABSTRACT

In this article, we examine whether the local indicators are able to predict the city-level housing prices and rents better than national indicators. For this purpose, we assess the forecasting ability of 126 indicators and 21 types of forecast combinations using a sample of 71 large German cities. There are several predictors that are especially useful, namely price-to-rent ratios, national-level business confidence, and consumer surveys. We also find that combinations of individual forecasts are among the top forecasting models. On average, the forecast improvements attain about 20%, measured by a reduction in root mean square error, compared to the naive models.

KEYWORDS:

• Housing prices and rents
• forecast combinations
• spatial dependence
• Germany

JEL CLASSIFICATION:

• C21
• C23
• C53

Acknowledgement

This article was presented at the DIW Berlin Macroeconometric Workshop (Berlin, 2013), the Panel Survey Data and Business Cycle Analysis Workshop (German University in Cairo (Berlin campus), 2014), the 54th ERSA Conference (St. Petersburg, 2014), and the 32nd CIRET Conference (Hangzhou, 2014). The authors thank the participants of these conferences as well as the anonymous referees for their useful comments. The standard disclaimer applies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ In its study, Deutsche Bundesbank (2013) stated that house prices might be overvalued by 5–10% in several cities.

² For more details, see the coalition treaty (CDU-CSU and SPD, 2013).

³ Gesetz zur Ergänzung des Finanzdienstleistungsaufsichtsrechts im Bereich der Darlehensvergabe zum Bau oder zum Erwerb von Wohnimmobilien zur Stärkung der Finanzstabilität (Referententwurf) as of 31 October 2016.

⁴ http://www.empirica-institut.de/empi2007/startseite.html

⁵ http://www.bundesbank.de/Navigation/DE/Statistiken/statistiken.html?nsc=true.

⁶ Four additional observations were lost due to incorporating the fourth-order lag of the dependent variables in the forecasting equation.

⁷ As reported in Tables 2–5, the forecasting performance of the benchmark autoregressive (AR) benchmark model was only in few cases substantially better than that of the RW model. This is unsurprising, given that the AR term enters with the fourth-order lag and the housing price indices displays no seasonality. As a result, in an overwhelming number of cases, the RW model provides a benchmark that is difficult to improve upon. This is the primary reason why we focus on comparing the forecasting performance of the indicator-augmented models with the benchmark RW model.

⁸ The medians are negative since higher value of $\hspace{0.17em}RMSFE\hspace{0.17em}/\hspace{0.17em}RMSFE{\hspace{0.17em}}_{AR}$ than $\hspace{0.17em}RMSFE\hspace{0.17em}/\hspace{0.17em}RMSFE{\hspace{0.17em}}_{RW}$ indicate that the AR model produces more accurate forecasts than the RW model for a given city.