Systems dynamics, modelling and scenario planning
In recent posts (a,b) we’ve discussed the relationship between scenario planning, which is essentially qualitative in approach, and mathematical modelling. It seems that this is likely to become a lively area of discussion over the coming years. And now a paper from the US Transportation Research Board claims to be a “unique joint scenario/modelling approach”.
The paper is titled “The Effects of Socio-Demographics on Future Travel Demand”, clearly an interesting area where radically different perspectives of social behaviour need to be transformed into quantitative forecasts of demand patterns. The paper claims that in a unique approach, the researchers have married system dynamics (SD) models with scenarios to produce a range of forecasts for transport demand.
A basic SD model simulates the demographic evolution of a regional population to 2050, examining the impact on travel demand. The scenarios are exogenous to the SD model and influence the demographic evolution and travel demand by manipulating SD model parameters that are linked to scenario assumptions.
Four core scenarios are developed and modelled, but apparently these can be adjusted with other more refined assumptions. A “cross-impact matrix” is used to map these scenarios onto the key model parameters (eg population growth, household structure, income distribution). A large number of variables were considered and a wide range of metrics produced as output.
The four scenarios outline different possible paths to explain how socio-demographics may influence travel demand over the next 30–50 years. While similarities among the scenarios exist, what is important for anticipating and preparing for change are the critical uncertainties, or driving forces, that cause one path to emerge over another.
How successful an approach is this? Clearly any SD model can handle sensitivity analyses of its key variables. So if you can map your scenarios onto these variables then you have a useful predictive tool. But in my view the authors have not really addressed the fundamental philosophical problem – scenarios may identify issues outside the normal realm of forecasting, so almost by definition a model built on the structural assumptions of the status quo will miss critical factors. Does that mean that any attempt at long-range forecasting is doomed? Not necessarily, and to be give the authors credit, the identification of weak signals and emerging trends can, as they suggest, be built into quantitative approaches.
To my mind, this whole challenge may be a mirage. Neither scenarios nor model forecasts are “right”. The point of the exercise of thinking about the future, using whatever method, is to open up decision-makers minds to the possibilities and range of outcomes. And then they may be in a position to develop “robust decisions in uncertain times” – as we say at SAMI.
Written by Huw Williams.