On the internal consistency of the term structure of forecasts of housing starts

Abstract

We use the term structure of forecasts of housing starts to test for rationality of forecasts. Our test is based on the idea that short-term and long-term forecasts should be internally consistent. We test the internal consistency of forecasts using data for Australia, Canada, Japan and the United States. Using a simple model of forecast formation, we find that forecasts are not internally consistent, leading to a rejection of forecast rationality.

Keywords:

• forecasts
• housing starts
• internal consistency
• panel data

JEL Classification:

• R30
• D84
• C33

Acknowledgements

We would like to thank the Euro Area Macroeconomic Developments Division of the European Central Bank (ECB) for providing the data. We thank the Fritz-Thyssen-Stiftung for financial support (AZ.10.11.1.167). The usual disclaimer applies.

Notes

¹ MacDonald (2000) surveys the vast literature on unbiasedness and orthogonality tests.

² Pierdzioch et al. (2012c) provide details on the derivation of the restrictions. Here, we focus on the economic interpretation of and .

³ Formally, the coefficients have the same sign if . We refer to the interval , however, because this interval covers our empirical estimates and both the case of contrarian and the case of bandwagon forecasts.

⁴ As an alternative model of forecast formation, we study a distributed lag model featuring two lags. The dynamics implied by such a distributed lag model can capture a potential ‘twisting’ behaviour of forecasts (Frankel and Froot 1987). Forecasts twist if forecasters form short-term ‘bandwagon’ and long-term ‘contrarian’ forecasts. While evidence results (not reported, but available upon request) do not support internal consistency of forecasts. Evidence of internal consistency is strongest in the case of Japan, for Japan the distributed lag model fits the data very poorly.