How Accurate Are the Pre-recession Models?

How Accurate Are the Pre-recession Models?


S Subramanian, Imperial College London; S Kohli, Steer Davies Gleave, UK


Socioeconomic data is used to model traffic flows in the pre-recessionary period. These models are then used to ‘forecast’ traffic flows from 2008 to 2010. These are significantly different from actual flows with major implications for forecasts.


Key macro-economic indicators such as GDP, household consumption and industrial production have been used as the drivers of travel demand in various models, especially in toll road projects for several years. The trends of traffic demand observed during the recent recessionary period across Europe have indicated major differences in the forecasts obtained from these models and what happened in reality. In this paper we explore this divergence of reality from model forecasts and delve into the causes behind the change in travel behaviour during the recent recessionary period.

This paper uses traffic flows and socioeconomic indicators from the beginning of the decade until the 2007 to estimate traffic flow models for four European countries - UK, France, Germany and Italy. These models are then used to "forecast" annual traffic flows for years 2008 to 2010. In each of those years, there are significant differences between the flows "forecasted" by the models and actual traffic flows for those years. In particular, preliminary results indicate that following the estimation period (2000 to 2007), the models consistently under predict the volumes of light vehicle flows and over predict heavy vehicle flows. This finding is robust across all four countries surveyed. We conclude that models that are estimated based on data from "growth" years perform poorly when used to forecast for "recession" years. This finding has significant implications for the methods used to calculate average, optimistic and pessimistic forecasts.

This research uses panel data from four European countries over a period of a decade (2000 to 2010, where available). Traffic flows of light vehicles and heavy vehicles are analysed separately. Flows of light vehicles are hypothesized to be driven by per capita GDP, household consumption expenditures and employment, indicators commonly used in toll road forecasting. Heavy vehicle flows are hypothesized to be driven by gross GDP growth and industrial production. In both cases, domestic fuel prices are included in the models in order to account for their influence on travel demand. These hypothesized links between the chosen indicators and traffic flows are borne out by the good model fit (R-square > 0.85 for all countries) for the estimation period (2000 ? 2007).

The timeframe is conceptually divided into two parts: pre-recession i.e. pre-2008 and post-recession i.e. post-2008. Pre-2008 data is used to estimate a simple linear model of traffic flow for each country that is dependent on the socioeconomic indicators discussed above. The estimated model is then used to ?predict? flows for the two combined periods (2000 to 2010). The difference between the actual flows and the predicted flows is found to be significantly higher in the post-2008 period. Apart from the magnitude of differences, two trends are prominent. The models fitted to pre-2008 data consistently over-predict traffic volumes of heavy vehicles and under-predict traffic volumes of light vehicles.

These preliminary findings highlight the difficulty of forecasting travel demand for future macroeconomic scenarios. They also highlight the practical necessity of forecasting separately for growth years and recession years. In particular, estimating models to data from previous recession years could provide insights into how the relationships between socioeconomic indicators and traffic behaviour change in gloomier economic times. The research is on-going and we will report on our further findings in this paper including establishing what could be done to improve the accuracy of these models for recessionary periods and recommendations on the definition of upside and downside scenarios.


Association for European Transport