Figure 8.2 Explanatory or Causal Relationship
8.5 Construction demand modelling and forecasting applications
Literature indicates that the methods used for demand prediction in the construction industry are
mainly extrapolative, that is, usually involving projection o f trends, with subjective adjustments
(Hillebrandt, 1974 and Fellows, 1987). Little use is made o f more sophisticated techniques such as
univariate or multivariate time-series analysis, econometrics and the leading indicators approach.
Sparkes and McHugh (1984) made a similar observation in the British manufacturing industry.
They noticed that although an increasing number o f companies appreciated the importance o f
forecasting, the methods used were predominantly naive. A lack o f knowledge and working
experience, particularly for methods involving the use of computers, was found to be the reason for
the lack o f use o f more advanced forecasting techniques. According to Oshobajo and Fellows
(1989), the use o f more sophisticated techniques would provide more reliable predictions for the
construction industry and consequently minimise the effects o f periodic shocks that have
characterised the industry. Several demand models have since been developed using more advanced
forecasting methods applied.
8.5.1 Investigation of leading indicators for the prediction of UK contractors' workloads (total new orders)
Oshobajo and Fellows (1989) explored the applicability o f the Leading Indicators technique to
model and predict contractors' new orders in the UK. The study identified variables that indicate
trends o f contractors' workload and used these identified leading indicators to build causal models
for forecasting new orders. At the same time, a time-series model was also developed, using the
univariate Box-Jenkins approach, to serve as a benchmark with which to compare the forecast
performance o f the causal model built using leading indicators.
The Leading Indicators technique was applied because the construction industry is susceptible to
economic fluctuations, which makes the fore-knowledge o f turning points and the magnitude o f the
fluctuations in economic trends provided by the leading indicators o f great significance. The study used both government and professional organisations' published data. New Orders was used as the
dependent variable, and the independent variables comprised Investment (GDFCF), RIBA New
Commissions, RIBA Work at Production Drawing Stage and Interest Rate. The first stage o f the
modelling process entailed an inspection of the time series o f these variables to determine if features
such as seasonal, cyclical and irregular components, and trend are present in them. Time plots for
each o f the variables were produced to facilitate the inspection o f each series. The second stage involved the testing o f the series for non-stationarity in their variances so as to determine the
appropriate transformations required. The method suggested by O'Donovan (1983) was adopted which involved dividing each o f the series into subsets relating to the size o f the seasonal periods.
The standard deviation and the mean for each series were plotted against each other to highlight the
underlying pattern in the respective series. When no evidence o f non-stationarity in the variance was
indicated, the appropriate differencing o f the respective variables was adequate to yield stationary
series. These stationary series were subsequently used to build the models. The procedure suggested
by Box and Jenkins (1976), which involves the stages o f identification, estimation and diagnostic checking, was used to develop the univariate time-series model. For the Leading Indicators model,
the classical decomposition technique and cross-correlation analysis were applied to determine the
relationship between the dependent variable and the independent variables, followed by the use of
multiple regression analysis to build regression equations o f the identified leading indicators to
The study identified Interest Rate as the only leading indicator o f New Orders. Both the classical
decomposition technique and the cross-correlation analysis revealed that Interest Rate consistently
led New Orders inversely by one to two quarters at a significant level. RIBA New Commissions was
found to be a coincident indicator o f New Orders. Alternative models were computed by regressing
the dependent variable New Orders on different time lags o f both Interest Rate and New Orders
itself. Forecasts were subsequently made using these models. The predictive accuracy o f the models
was compare with that o f the benchmark model. It was concluded that the performance of the
Leading Indicators models improved forecast accuracy by 5 to 12 per cent.
8.5.2 Thai construction industry: demand and projection
The study by Tang et al. (1990) was an attempt to gain further insight into the demand for construction activities in Thailand. The two objectives o f the study were: to estimate the demand
fimctions for construction activities in Thailand; and to project the growth of the Thai construction
industry for the next five years. Three types o f construction were considered: residential, non- residential and 'other' (mainly public projects). The demand function for each type o f construction
was estimated using regression analysis.
To develop regression models, the most important task is to determine suitable independent
variables. In the study, specific factors that affect the demand for residential construction were found to include National Income per Capita, Relative Price Index, Rate o f Household Formation, Size of
Population and Interest Rate. Those for demand for non-residential construction comprised
Corporate Savings, Industrial Production Index, Number of Tourists, Gross Domestic Product and
Exports. Factors found to affect the demand for other' construction were Government Revenue,
Value-added from Public Utilities and Government Expenditure. These various factors were
respectively used to estimate the demand functions for residential, non-residential and other' construction using the Stepwise regression technique. This technique selects the significant factors
and combines them to estimate the demand function for each type o f construction. Those factors
found to be insignificant at the 20 per cent confidence level were omitted. A linear relationship was
assum ed in the demand functions o f residential and others' construction, and a log-linear
relationship was specified for non-residential construction. The estimated demand functions were
subsequently used to project total demand for construction in three different scenarios. Each
scenario assumed different growth rates o f the social and economic factors used for estimating the
The results of the Stepwise regression analysis indicated that National Income per Capita, Relative
Price Index, Size o f Population and Interest Rate were the major determinants o f demand for
residential construction. Industrial Production Index and Corporate Savings were the determinants identified for non-residential construction, while Government Revenue and Value-added from Public
Utihties were those identified for other' construction. The demand models were found to have fairly
high R-square values, indicating their goodness o f fit. The projection o f total demand for
construction produced favourable results for the Thai construction industry over the next five years, especially for pubhc construction which will play a significant and dominant role in the growing Thai
economy, accounting for over half o f the total construction.
8.5.3 Preliminary investigation into formulating demand forecasting model for industrial construction
Leading indicators were employed by Killingsworth (1990) to develop a demand forecasting model
for industrial construction using multiple regression analysis. The study involved, firstly, identifying
the determinants that were highly correlated with industrial construction; secondly, identifying which
o f these determinants were leading indicators o f industrial construction; thirdly, determining the time
lag between the movements in these indicators and the corresponding movements in industrial
construction; and finally, applying a multiple regression procedure using industrial construction and
the lagged indicators to compute the demand model. United States government data were used for
the study. They were used to quantify industrial construction and its determinants namely, population, interest rates, economic shocks, demand for products, surplus manufacturing capacity,
business expectations and profit expectations.
The modelling process comprised the following procedures:
1) The data were first adjusted to remove the effects o f inflation and random fluctuations.
2) The data were then transformed into relative terms, that is, the annual rate o f change
was calculated for industrial construction and each o f its determinants.
3) Line graphs were plotted to compare industrial construction volume with each o f its
determinants.
with industrial construction, and to determine the time lags o f these leading indicators.
5) A multiple regression procedure was applied to compute the demand model using
industrial construction as the dependent variable and the lagged indicators as the
independent variables. A linear relationship was assumed in the model.
6) Test statistics were used to determine the model accuracy and reliability.
The results o f the graphical analysis indicated that correlation exists between industrial construction
and interest rates, demand for products and economic shocks at a consistent time lag o f two, three
and six quarters respectively. The other determinants were not found to be correlated with industrial
construction. The regression model developed based on the three identified leading indicators
produced a high adjusted R-square value, indicating that the model was significant in explaining the
dependent variable. Based on this high R-square value, it was assumed that the demand model
would produce forecasts o f acceptable accuracy.
8.5.4 Models of UK private sector quarterly construction demand
Models o f UK private sector quarterly construction demand were developed by Akintoye and
Skitmore (1994) using a priori selected leading indicators. Their study was prompted by a clear indication o f an increasing level of private investment in the UK over the period o f 1974 to 1988,
making it appropriate to consider the nature o f private sector demand. The models were produced
for three groupings o f private sector new orders: housing, commercial and industrial construction.
The indicators selected for the study were GNP, price level, real interest rate, unemployment and
manufacturing profitability. They were regarded as the general factors o f construction demand for
the UK and used as independent variables in the models.
The modelling process for each type of construction involved several stages. Firstly, time lags of
eight quarters for each independent variable were specified as it was anticipated that there would be
lead relationships between the dependent and independent variables. This was based on a priori
considerations that more than one time period was required to exert the influence o f all the past
changes in an independent variable. Hence, a maximum lag of eight quarters was used as this was
considered a long enough period for the influence of a change in a factor on the private-sector
lead) were created for each of the variables, giving a total o f 45 independent variables applied in each
initial model. Secondly, the strengths o f the relationships between construction demand and these
lagged variables were estimated by a standard ordinary least-squares (OLS) step-wise multiple
regression procedure. Thirdly, the variables that met the step-wise regression analysis criteria were
selected. Finally, the demand functions using only the selected variables were re-estimated. A log-
linear relationship was assumed for all three models based on two reasons: the need to determine
elasticities o f response o f the dependent variable to independent variables; and the original data o f the independent variables exhibited non-linear scatter when plotted against the dependent variable.
The step-wise regression procedure identified the significant indicators o f the three types of
construction. For housing construction demand, it was found that demand is significantly correlated
with price level, GNP and real interest rate. The relationships with unemployment and
manufacturing profitability were unsupported by the data. For commercial construction demand, its
trends could be explained by the trends in real interest rate, manufacturing profitability, GNP and
unemployment level. Demand for industrial construction was found to be correlated with only GNP
and manufacturing profitability. Several conclusions were drawn from these results. Firstly, only housing construction demand is responsive to changes in price level, unlike commercial and
industrial construction demand. This suggested that private sector housing demand may increase
with an elastic response to a given fall in the price level. Secondly, the positive inelastic relationship
between industrial construction demand and GNP supported the importance of national income or
economic conditions in private-sector industrial investment. Thirdly, the negatively and inelastically
relationship between unemployment and commercial construction demand implied that increasing unemployment has a declining effect on commercial construction investment generally; and
commercial building demand responds immediately to changes in real interest rates. Finally, as
expected, manufacturing profitability was only relevant to private sector commercial and industrial
construction investment. Although the log-linear regression models had high R-square values,
indicating good-fit models, their ex-post forecasting performance was poor. It was concluded that further variables need to be investigated and included to improve the accuracy o f these models.
8.6 A critique of the prevailing forecasting methods used for construction demand