The development and validation of models for assessing risk impacts on construction cash flow forecast

Various 'short cut' approaches have been adopted in cash flow forecasting. However, the majority of these approaches failed to consider the issues of risks inherent in construction. As such, a wide variation is observable between the predicted cash flow profile and the actual. This study attempts to model the variation between predicted and actual cash flow due to inherent risks in construction. The aim of the model is to provide a decision support tool for the construction contractor so that in case of risk occurrence impacting cash flow forecast during construction, the new cash flow profile can be determined without recourse to cumbersome calculations. Data were obtained through three sets of postal questionnaire surveys and empirical data collection from Construction Company’s archive. The first two questionnaire surveys tried to explore different dimensions of risk in order to determine the one to focus attention on for model development. The first questionnaire survey, which was a pilot study, focused on the probability/impact assessment of risk variables involved in cash flow forecasting. The 21 risk variables examined in this survey were those impacting the positive cash flow (cash in or value flows). The second questionnaire survey focused on the extent of occurrence/impact assessment of risk occurrence in cash flow forecasting. The 26 risk variables examined in this investigation were a modification of and addition to the risk variables in the first questionnaire. The risk variables in the second survey were those impacting the negative cash flow (cash out or cost flows). Responses to the surveys were analysed using various statistical techniques, including mean response analysis, ANOVA and factor analysis. The results of the analysis from the second questionnaire survey were found suitable for further use in model development. From the 26 risk variables investigated, 11 significant risk variables were determined which were focused on for model development. Moreover, using factor analysis, the 26 risk variables were reduced to six factors, which were also used for model development. The impacts of these significant and factorised risk variables on cost flow forecast were then investigated through two sets of data for the purpose of model development. One was through a third questionnaire survey done on a project-by-project basis. The second was by collection of archives data on predicted and actual cost flow from completed construction projects in order to determine their variation at 30%, 50%, 70% and 100% completion stages. Quantity Surveyors on each of the projects were requested to score on a Likert type scale, the extent of occurrence of the determined significant and factorised risk variables. The data sets were used to develop risk/impact assessment models using multi linear regression as well as Artificial Neural Networks (ANN). Models developed from the first and second data sets were validated using 20 and 15 virgin data sets respectively. Models developed from the archives data, using the 11 significant risk variables and employing the ANN back propagation algorithm was found to perform better than others. Curve fitting using the technique of logit transformation and curves' goodness of fit measure using the SDY (standard deviation about the estimate of Y) measure shows that the model performs creditably well in comparison with models developed in related studies.