among top 10 papers at 2020 International Real Estate Society (IRES) Doctoral Symposium  

(previously named as "Spatial-temporal asymmetry, shock and memory: housing transaction prices in Sweden")

 [PDF] 

Abstract:

This research, using spatial-temporal framework, analyzes if segmentation of housing market by object type (houses vs apartments), affects the features of idiosyncratic volatility pattern: volatility clustering, volatility mean reversion and asymmetry. Volatility features are tested at transaction level by novel spatial - temporal ARCH and GARCH model, applied to unbalanced non-repetitive sales data from Jönköping, Sweden, from 2012 through 2019.  Results show significant difference in the price volatility patterns between the market segment of apartments and houses. Houses have higher constant price variance. With wider geographical distances between observations, constant price variance of houses increases, as more heterogeneous areas are included. Estimated parameters for the segment of apartments show positive shock response and positive memory effect, which  confirm the assumption of spatial-temporal persistence in the variance. Apartments show significantly higher response to negative price shocks than to the positive price shocks.  This asymmetry is clearly observed in the spatial context compared to the  spatial-temporal one.  Spatial-temporal models have better goodness-of-fit and better statistical adequacy compared with the benchmark spatial ARCH model. 

Keywords: spatial-temporal, ARCH, GARCH, transactions, housing price volatility, idiosyncratic risk, persistence, asymmetry, memory, housing market segments, Sweden

JEL Classification: C33, R30, R32 

Abstract:

In this research it was shown that, in general, spatial filter enhances the fit and moderately improves the prediction of the logit credit risk model. It was observed that the fit and prediction results depend on the created weight matrix when using spatial filtering. With the increase of the neighbor links, the prediction by the spatial model slightly outperform the base model. The detected positive auto-correlation indicates the existence of clusters of defaults within a geographical area, which could confirm the need for use of spatial filter or other spatial techniques. Hence, existence of positive spatial pattern in the credit risk assessment could be taken in consideration by the national banking regulators (central banks) and appropriately treated in the regulation, so that estimated credit risk parameters reflect the true risk condition of the companies and their microeconomic surrounding.