This paper examines the housing-output growth nexus in South Africa by accounting for the time variation in the causal link with a bootstrapped rolling Granger non-causality test. We use quarterly data on real gross domestic product, real house prices, real gross fixed capital formation and a number of building plans passed. Our data span 1971Q2-2012Q2. Using full sample bootstrap Granger causality tests, we find a unidirectional causality from output to a number of building plans passed.

The unidirectional causality from real house price to output and a bi-directional causal link between residential investment and output. However, using parameter stability tests, we show that estimated VARs are unstable, thus full-sample Granger causality inference may be invalid. Hence, we use a bootstrap rolling window estimation to evaluate Granger causality between the housing variables and the growth rate. In general, we find that the causality from housing to output and, vice versa, differ across different sample periods due to structural changes. Specifically speaking, house price is found to have the strongest causal relationship with output compared to residential investment and the number of building plans passed, with real house price showing predictive ability in all but one downward phase of the business cycle during this period.