Time series forecasts of the construction labour market in Hong Kong: the Box‐Jenkins approach

Abstract

Labour resources are invaluable assets in the construction industry. Nurturing a quality workforce and promoting stable employment for construction personnel have often been advocated as part and parcel of an industrial policy. Yet, the future labour market of the industry is always uncertain, and there is a need for estimating future labour market conditions as an aid to policy formulation and implementation. The Box‐Jenkins approach has been applied to develop Autoregressive Integrated Moving Average (ARIMA) models to analyse and forecast five key indicators in the construction labour market of Hong Kong: employment level, productivity, unemployment rate, underemployment rate and real wage. This approach can be adopted in more complex and diverse labour markets subject to the properties of the utilized data series. Quarterly time‐series statistics over the period 1983–2002 are used in this study. The predictive adequacy of the models derived is evaluated with out‐of‐sample forecasts in comparison with actual data, based on the mean absolute percentage error (MAPE) and the Theil's U statistics. The results indicate that except for construction employment, the proposed forecasting models have reasonably good predictive performance. Among the five case studies, the most accurate is the construction real wages model. In addition, we conclude that univariate projection is not an appropriate method for forecasting construction employment in Hong Kong. Multivariate structural forecasting analysis should be adopted in order to obtain more accurate estimates. The developed models can be used to provide benchmark estimates for further analysis of the construction labour market and the projections offer valuable information and early signals to training providers and employment policy makers.

Keywords:

• Box‐Jenkins approach
• time‐series forecast
• labour market
• construction industry
• Hong Kong

Acknowledgement

The authors gratefully acknowledge the contribution of Mr William W.L. Hui, MPhil, to this study.

Notes

1. If p is the total number of parameters estimated, ; , where n is the sample size.