The modelling of risk and uncertainty should involve several stages. These are summarised in this note. It is important to note that each stage is important, and if not performed adequately, the quality of the results, conclusions and decisions could be severely affected to the point that the analysis is wrong, very inaccurate or misleading.
Identification. This involves identifying which factors of a real-life situation affect the outcome (whose risk we are trying to assess, such as the value of a contract, the cost of a project, and so on). It can be helpful to work backwards from the outcome asking: “What would make this item’s value different (to the base case, or what we have assumed)?” One can ask this question again for the first set of factors identified, until one has a set that is complete and realistic.
Classification. This involves assessing whether each factor is a choice (controllable) variable or a risk (uncertain) one. For example, to assess the value of a contract, the contract parameters (e.g. duration, price charges for services etc.) would typically be choice variables whereas other items (e.g. cost to deliver the promised services) could be uncertain. Choice factors are to be optimised, whilst risk factors are to be assessed using risk management techniques.
Nature and relationships. This involves describing the nature of each risk e.g. discrete, continuous, compound, single or multiple occurrence, the effect on other variables, and dependencies or correlations.
Data and calibration. This involves collecting or analysing the data available to calibrate (including using expert judgment) the data and assumptions used in the model. It could relate to the parameters of probability distributions used as inputs, to correlation, or to model outputs, as well as to the choice of distributions e.g. based on using data fitting algorithms to select the appropriate distribution.
Modelling and mapping. This involves building each risk (or its effect) into a model. If a non-risk model already exists, it may involve adding new line items, or restructuring the model to be able to capture the risks and include their effects. The amount of work involved depends on how well any existing model has anticipated risks or uncertainties. For example, a model in a which a sensitivity analysis can be run on some items is usually one where risk analysis on the same items is easy to include. One frequent challenge is where base case models do not allow for time-delay sensitivity on sub-components of the model, even as these may be critical for project economics (e.g. a delay to the launch of revenues). It can often be helpful to use add-in software to Excel (such as @RISK) for this step, given the large set of pre-defined distributions, correlation and graphics possibilities, for example.
Running and drawing inferences. Monte Carlo simulation is usually the easiest way to run a risk model, that is to assess the aggregate effect of many uncertain factors that may occur simultaneously. Once again, software such as @RISK can help to perform this step effectively, for example due to the graphical and statistical reporting features.
Decision. The steps above should inform a final decision. Insights gained from the analysis should include a much more complete understanding of the drivers of value in the situation, additional measures to optimise project design, and to mitigate risks, the ability to integrate risk tolerances into decision-making, and a more accurate quantitative assessment of the projects’ economics or likely success.