Thinking about local housing market volatility

by Ken Gibb

A recent paper in Bank Underground (the Bank of England’s new blog site) by Arzu Uluc uses an interesting local-level data set (325 local authorities in England covering 1997-2009) with which to examine local housing booms and busts. The underlying model draws on data including real house prices, real gross disposable household income per capita, dwelling stock per capita and various mortgage market variables like loan to income, loan to value and the share of interest-only mortgages.

The author concludes that the local research allows us to infer (and reasonably so) that volatile housing markets can ‘threaten financial and macroeconomic stability’. Credit conditions in terms of the key ratios and types of mortgage products play an important role through the ‘1997-2009 housing cycle’.

Uluc presents his empirical work by generating six stylized facts. These are:

  1. Real house prices rose on average by 150% between 1997 and 2007 and then fell 12% by the end of 2009.
  2. Changes to the proportion of high loan to income mortgages were positively correlated with local housing booms and busts.
  3. There is a negative relationship between the changes to the proportion of high loan to value mortgages and the size of local booms and busts.
  4. Changes in the share of interest only mortgages were pro-cyclical (similar to 1. above).
  5. Housing booms and busts were also associated with real drivers like real income and dwelling stock growth.
  6. The bigger the local boom, the larger the subsequent bust.

A few caveats are in order – it is not clear why local authority data is particularly appropriate rather than broader more functional geographies; nor does it appear from the blog that there is spatial dependency accounted for in the analysis. I also wondered if the London effect needed to be explicitly modelled (or indeed some sense of North and South)? On the other hand, the author is clearly concerned about attribution and possible reverse causality or confounding factors in the models deployed. These are all things that can no doubt be explored in a longer paper.

What about the bigger messages?  The model is best summarised by a quote from the post: ‘the econometric analysis…suggests that booms and busts ere associated with both real factors and credit loosening. Higher real income growth [was] associated with larger booms in [the] 1997-2004 period, and in 2007-09 areas with higher growth in the dwelling stock per capita tended to see larger price falls”.

Thus, and despite the fall in endowment mortgage volume shares after 1997, changes in the share of interest only mortgages were associated with local house price booms. Similarly, higher loan to income ratio shares of new mortgages were associated with local booms. The interesting negative relationship between high loan to value mortgages as a share of the total and local house price growth – may be explained by the growth of the share of home movers as opposed to first time buyers among total transactions and the increased financial support utilised by remaining successful first time buyers from family and other savings (which also drove down LTV ratios).

As Uluc stresses, there are reverse causality explanations and possibly confounding omitted variables that may instead better explain what is going on – but the associations found are striking. It is a reminder of how difficult it can be to disentangle the relationship between the real housing market and monetary transmission through credit variables – something well known in the wider housing and economy literature but just as striking here.

Stylised facts are useful, but as the author points out, they are the starting point that we then develop models and theories from and look at fresh data to test the ideas that originally flow from these facts. It would therefore be interesting to extend this data forward beyond 2009 and also to aggregate a sub-sample of the local authorities into clusters that approximate for sub-regional housing market areas. If we did so, would we also be able to detect the influence of the changing regulation and practice of mortgage lenders in more recent years?