Significant revisions in estimating housing starts (Hebrew)
- The data on housing starts serve as an important input in formulating policy in the housing market, as well as in formulating monetary policy. A Bank of Israel Research Department analysis shows that between 2004 and 2014, the initial quarterly estimate of housing starts was biased downward by an average of about 10 percent. The figure is revised upwards mainly during the first year after publication, and stabilizes after two years (8 quarterly publications).
- Policy makers in the housing field are exposed to the initial estimate—and the broad public is exposed to it since in recent years it has gained wide media exposure. They may therefore develop a misleading concept regarding activity in the construction industry. Therefore, caution must be exercised when using the initial estimate, and resources should be invested in improving it.
- Data collection depends on many entities dealing with construction initiation, approval and actual start, and some of these entities provide the Central Bureau of Statistics with only partial data. This complex process makes it difficult to publish quality data in real time.
- A statistical model for forecasting the revisions makes it possible to narrow the size of the average revision by at least one-third, and to formulate a forecast that is not biased in a particular direction—an estimate that is sometimes revised upward and sometimes downward. This model relies on data that is known in real time (ex ante), and can be used to improve the estimate at the time of publication.
Housing starts, the construction stage in which the digging of the building’s foundations begins, are a critical link in the production chain of a housing unit. The estimate of the number of housing starts is a very important figure, which has attracted broad public interest in recent years in view of the increase in home prices, and is published on a quarterly basis by the Central Bureau of Statistics (CBS). The high number of entities involved in the construction process (the Ministry of Construction and Housing, more than 100 local planning and building committees and hundreds of developers surveyed by phone) makes it difficult to gather data in real time, and leads to retroactive revisions in the quarterly figure. The CBS makes great effort to cleanse the administrative data that it receives from the Ministry of Construction and Housing and from the developer surveys in real time. In addition, it uses a statistical model to estimate missing data that it expects to receive in the future.
An analysis conducted by Adi Finkelstein of the Bank of Israel Research Department examined the development of the quarterly estimates of housing starts since 2004. The examination showed that while typical initial economic estimates are generally revised at random (we cannot know in advance whether they will be revised upwards or downwards), the estimate of housing starts is consistently biased downwards to a significant extent. This figure is revised upwards mainly in the first year after publication, stabilizing after two years at a value that is about 10 percent higher on average. A breakdown by district shows that the problem exists in all districts, but is particularly marked in the data on the Tel Aviv and Haifa districts. A breakdown by developer shows that between 2009 and 2014, the Ministry of Construction and Housing revised the data on public development starts upward by about 28 percent on average. In contrast, the figures on private development starts were revised upward by about 10 percent on average. This shows that all the data are downward biased, but the administrative data from the Ministry of Construction and Housing are typically of low quality in relation to the figures from the private sector.
The analysis also proposes a statistical model based on the revision patterns between 2004 and 2015. The model uses the initial housing start estimates and their revisions to forecast the expected revisions after four and eight quarters. The results of the real time forecasting exercise show that less conservative models can be proposed to estimate the missing data, which will make it possible to reduce the size of the average error by at least one-third and to formulate a forecast that is not biased in a particular direction.
The figure on the number of housing starts serves various government authorities in estimating supply in the housing market and in formulating policy regarding the need for government intervention in the construction industry. The figure also serves the Bank of Israel in the formulation of monetary policy. Since in most cases the estimate is revised upward retroactively, and in some cases, estimates that showed a decline in the volume of housing starts were changed to indicate an increase, it may create an impression that activity in the construction industry is slowing down or not expanding, while in actuality the number of new homes has been in a moderate upward trend over time. This paper strengthens the warning that the CBS attaches to its publications on housing starts and provides the consumers of the data with the magnitude of the expected revisions of the data.