How can Regional Differences in the Risk‐of‐Foreclosure be Explained? Evidence from Swedish Single Family Housing Markets

Abstract

In Sweden, quite large differences in the risk‐of‐foreclosure for single‐family housing exist between regions. The aim of this paper is to explain such differences, using data on foreclosures for all Swedish regions. In an option‐based model, the risk‐of‐foreclosure is a function of such things as housing prices and incomes, as well as interest rate and housing price volatility. Instead of using housing prices and incomes to explain the risk of foreclosure, we use explaining variables in the labor market model. The main results indicate that the option‐based model explains the variation in foreclosure rates. Specifically, interest rate – together with price volatility, price changes, price and rent level, income, and employment – explains around one‐third of the total regional variation. Our extended option‐based model explains slightly more. Specialization within the industrial sector seems to have a positive effect on foreclosure risk in that it, together with the educational level of the workforce, reduces the risk. Specifically, mortgage lenders and banks can reduce their risk by concentrating their business on dense regions with a high degree of employment within the manufacturing industry and with a higher educational level of the workforce.

Keywords:

• Foreclosure risk
• default
• spatial distribution
• housing market

Acknowledgements

Funding from the Sweden‐America Foundation and the Wenner‐Gren Foundation is highly appreciated. We also thank the Haas School of Business and the Goldman School of Public Policy (UC, Berkeley) for their hospitality. Finally, we are grateful to Sandy Nixon for her editorial help with the manuscript.

Notes

1. Normally, if one tries to estimate prepayments without default, or default without prepayment, the results will be biased. However, prepayment risk in Sweden is zero, as the borrower has to pay a penalty fee if the mortgage is prepaid in advance. Hence, prepayment is not included in the model.

2. The traditional foreclosure rate metric relates the total number of foreclosure relative to the number of mortgages outstanding. Mortgage outstanding should have little short run fluctuation. As we lack information of the mortgages outstanding we are using an alternative definition with sales as a denominator. Sales, however, are highly cyclical series. Thus time series movements in our ratio may not be dominated by movement in foreclosures, like the conventional series would be. Hence, we may not know the extent to which foreclosure behavior or housing sales are driving the results in our empirical study. However, we have carried out some robustness test in the empirical section.

3. We do not have the information about mortgage balance; on the other hand, we do have the sale price of the house. Hence, we do not define loss severity as in Lekkas et al. (1993) where it is defined as the difference between mortgage loan balance and the value of the house. Our measure can be seen as an implicit assumption that the mortgage balance is equal to zero, which is not likely, suggesting that the loss is equal to the total value of the foreclosed sales.

4. Rosen and Smith (1983) introduced the concept of natural vacancy as dependent on the issue of matching at the same time the demand for and supply of rental properties.

5. In Sweden, there exist no (or a very small) secondary rental market for single‐family houses, which makes it difficult to obtain information about the rent level. Therefore, we have used rent from the rental market as a proxy for the rent level on the single‐family housing market.

6. As Capozza (1997) we are using income as a proxy for transaction cost. His argument is ‘as personal income rises, the costs of default are expected to increase since the financial consequences of a negative credit rating and the threat of deficiency judgment will increase’.

7. The data comes primarily from NAI Svefa, Statistics Sweden (SCB) and Föreningssparbanken ‐ SPINTAB.

8. , where S is the total number of industries (24 in our case) within region i and e is the numbers employed.

9. We are using a shorter period due to lack of data concerning some of the independent variables, that is, we are investigating the period 1994–2001 and not 1993–2006.

10. See Anselin (1988) for an in‐depth discussion about spatial econometrics. Using longitude and latitude coordinates (squared and cross products) [suggested by Galster et al. (2004)] do not reduce the spatial dependency in our data set.

11. We have also used the inverse of the distance. However, the results are not altered, and the same conclusions can be drawn.

12. The question about what drives the results have been investigated by using number of foreclosures per housing stock instead of foreclosures per sale. The reason is that the housing stock has little short run fluctuation. Most of our parameter estimates (not shown in the table) have the same sign and are significant. The only exception is the variables Price and Sigma which are not significant different from zero.

13. All spatial models are available upon request.

14. It may be more reasonable to expect that the relationship is temporal lagged the first examined years and not lagged at all the last years, as the legal process has gradually improved. The empirical tests that we have performed weakly indicate that (not presented).