Forecasting and Decision Making under Uncertainty

Organisations tend to make capital investment decisions based on fixed demand projections. Managers focus much attention and resource on the production of "accurate" demand forecasts, or hire consultants to do so, and then make a cost-minimising or value-maximising investment decision that meets these point forecasts. My practical experience with hospital planners in the UK and US has highlighted the challenge of making such investment decision. For example, a strategic level investment decision at the hospital typically asks for the required number of hospital beds. Hired consultants calculate bed demand for the lifespan of the infrastructure, which can be 30-50 years. They then design the hospital based on the "point"demand forecast (in terms of bed numbers) and operational efficiency targets set by their client. But, how reliable is this point forecast of bed demand 30-50 years ahead if we find it difficult to predict exactly how many beds will be needed, even in the next month? Will the decision to build a fixed number of hospital beds based on the point demand projection be clinically effective and economically efficient? If demand is higher or lower than expected and the design is rigid, then the infrastructure will not be able to cope with high or low demand, resulting in overcrowding or empty beds. A good value-for-money hospital infrastructure therefore needs to be flexible, to allow effective adaptation to unpredictably changing circumstances.

One of my PhD papers illustrates design flexibility as an important value driver typically in a large scale and long term capital investment project using a case study based on the development of the Cambridge University Hospitals NHS Foundation Trust campus. Examples of design features that make hospital buildings flexible include shell space, where areas are built but not yet medically equipped, or suitable structural foundations of building to allow additional floors at a later time. In my paper, I also lay out a stylised example of articulating the value of flexibility to enable the practitioners to make an economic case for flexible infrastructure.

The foremost important step towards designing flexibility and making an economic case for such flexibility is the recognition of uncertainty associated with point forecasts of value drivers or projects, such as demand, exchange rates and commodity prices. Once uncertainty is appreciated, it then becomes natural to introduce flexibility as an integral part of the design challenge. Traditional approaches to model uncertainty in forecasts assume that we have full knowledge about the true model which generated data. However, one rarely has sufficient knowledge of the underlying process to specify a correct model in practice. Two of my PhD papers focus on developing empirical and nonparametric methods to model future uncertainty which avoid the need to make assumptions about the data generating models for the validity of the methods. My PhD papers demonstrate the robustness of such methods, both analytically and numerically, against the cases of misspecifying the data generating models. I also investigate managerial implication of such methods in exchange rate forecasting and product/service demand forecasting for inventory planning. Based on the probabilistic forecasts output by the methods, one can make more informed investment decisions in an uncertain environment.