Abstract
The present study applies GARCH models to explain and forecast the variation in monthly IPD total returns for office, retail and industrial space. With the exception of retail returns, the forecasting performance of the estimated GARCH models is very good. Over the 16βmonth period from January 2000 to April 2001 total office returns were 17 per cent and industrial returns 16.2 per cent. The GARCH models would have predicted 21 per cent for offices and 18.3 per cent for industrial units. Improvements in the forecast performance are sought with the inclusion of information based on the yield curve, a wellβknown indicator of future economic conditions. The volatility of the yield curve and a βgood and bad newsβ variable, based on the changes in the yield spread through time, are used as inputs in the variance equation of the GARCH specification to help model the unexplained variance of the return series and improve the forecasts. The forecasts seem to improve in the case of office returns but slightly worsen for retail returns. Yield curve information was not relevant in the model of industrial returns. The findings of the present study suggest that situations of clusters of large and small errors arising from speculative behaviour should be tested in models and treated accordingly. They also show that the contributions of expectation variables such as the yield spread should be examined within forecasting methodologies more extensively. Omission of such series may come at the expense of producing more accurate forecasts. Copyright Β© 2002 Henry Stewart Publications