Are Sentiment Indicators Good Predictors of Yields?

The fall in property yields since 2001 and in particular in 2004 that resulted in strong total property returns both in the UK and the continent had three major implications for future research on property yields. First, it highlighted the need to assess models and methodologies of yield forecasting since the fall in yields and, especially the magnitude of the fall in a period of rather weak occupier markets, was not predicted by the industry. With the exception of papers such as McGough and Tsolacos (2001) and Krystaloyianni and Tsolacos (2004) the forecasting capacity of yield models is not assessed. Second, a belief is developing that fundamental models of property yields, those be models linking yields to the occupier market and the wider investment environment, were unable to predict the yield movements in 2003 and 2004, and that sentiment and behavioural factors played a role during that period. A key issue is therefore whether we have full knowledge of what these factors or indicators tell us about yield shifts in the short-run. Third, a clear need has emerged to model the direction in yields in the short-term (twelve months ahead). Asset allocators require information as to which sectors will see yield shifts before the stock selection process is initiated. Therefore early signals for yield movements offer significant information to property investors. This paper responds to these research requirements. It aims to identify how closely widely publicised sentiment indicators are correlated with yield movements and to assess their information content for near future yield movements. In particular it examines the usefulness of consumer and business surveys in forecasting the direction of retail yield shifts. To that effect, the empirical relationship between sentiment indicators and retail yields is explored through VAR models and causality tests. The aim of the empirical analysis is to construct VAR models to forecast the direction of yields two to four quarters out of sample. That is whether yields will be lower or higher in two and four quarters ahead than their current level, using advance information to replicate what a practitioner would do. The forecasts of the VAR models are then compared to those of an autoregressive model and other naÔve methodologies. The forecasts are estimated on a recursively basis but the performance of models in turning points receives specific attention.