Data Pooling for Early-Stage Price Forecasts

Many clients rely heavily on pretender price forecasts, provided by the Quantity Surveyor (QS), for their investment decisions. During the preliminary design stage, it is very common in practice for Q.S. to use historical building price data on which to base the forecast of the target project ? typically basing the forecast on the known price of one most similar project to the target. However, as statistician Beeston (1974) points out, a potentially more powerful method is to use the mean price of a suitable sample of similar projects on the rationale that there the price of the target project is one of a "family of prices" for that project, and for which the prices of those in the sample are a proxy. The major difficulty with this is what Flanagan (1980) has termed the "homogeneity" problem, in that the bigger the sample size the less similar are the projects in the sample. In statistical parlance this means that, although increased sample size reduces the standard error of the mean at the same time the sample becomes less representative of the population from which the target project is conjectured to belong. In other words, making the best use of available data involves a trade off between the need to restrict the sample of data used to that most relevant to the forecasted project while at the same time maximizing the sample size involved. In 2001, Skitmore offered an approach to solving this in the risk analysis contect by empirically examining the effects of all possible pooling (sampling) permutations on forecasting errors with a view to selecting the pooling arrangement that most minimises the spread of errors.. This paper develops this idea for early stage forecasting. Using out-of-sample mean square errors to measure the forecasting accuracy a method is presented for finding the best pooling arrangement of the available data according to the characteristics of the target project. This involves the use of crossvalidated multivariate regression analysis to pool data into subgroups ? typically within the building floor area, project type, number of bidders, etc.