6. Forecasting and evaluation
6.1 Projection of demand growth
Many transport problems are a combination of deterministic and stochastic processes that may vary over time. Time series analysis is the statistical technique used to study such problems. From a stochastic point of view, traffic flow on a particular link of a network could be regards as consisting of four components: trend (Tr), seasonal (Se), cyclic (Cy), and random (Ra) components. The trend component may results from long-term growth in traffic. Seasonal variation may result from different flows at different times of the year. Cyclic components can results from long-term economic changes. The random component may results from short-term variations in traffic flow. When there is an additive relationship between the components, they can be combined as in equation EQ 6.1 and Figure 10 below.
[EQ 6.1]
The loads on a transport network reflect the time-dependent variations in social, economic, industrial, agricultural and recreational activities in the area it serves, as well as long-term trends in the levels of those activities. For instance, traffic data – such as hourly or daily traffic volumes – that are indicative of these loads, must be recorded as time dependent data. The distinguishing feature of time‑dependent data is that they come from processes that are undergoing continual change. If we are to understand the data, we must extract the components of change that are involved and separate the random effects from the trends and cycles that influence the data. Stationary processes (those whose parameters are stable over time) offer the possibility of repeating observations in order to uncover the degree of variability existing in the dat. Time series data do not permit repetition of observations as data collected at one point in time will, of necessity, differ from those collected at other times.
Figure 10: Components of a time series
Methods for the analysis of time series data, including the use of moving averages and autocorrelation coefficients, are outlined in Taylor, Bonsall and Young (2000, pp 416–22) and described more fully in textbooks such as Chatfield (1984).
Time series data are important information sources for transport analysis. They include data on system performance and impacts over time, such as passenger movements or road fatalities, as well as economic performance and activity data (such as quarterly GNP statistics and fuel sales) and socio-economic data (such as population and employment data sets). Gargett and Perry (1998), Amoako (2002) and Sutfcliffe (2002) provide recent examples of the use of time series data in analysis of freight movements in Australia.
6.2 Forecast horizon
In the past, the practice has been to develop a forecast for one year, usually the year of opening of the initiative. Given that demographic projections are usually available for a number of forecast years, it is recommended that at least two forecast years are used – one for the opening year and the other 10 years after opening.
Using the second forecast year enables issues such as fare level changes, value of time and vehicle operating cost increases, and land use changes to be explicitly input into the model and may result in a more robust forecast and appraisal outcome.
6.3 Network options
The ‘do-minimum’ network should be based on the validated Reference (Base) Year network and should include all the supply-side proposals (such as committed highway and public transport infrastructure) and operational proposals that are expected to be implemented by the forecast year.
The ‘do-something’ network should be based on the ‘do-minimum’ network with the difference between the two networks being the project being appraised and any other changes, such as a service scenario, that are different to that in the ‘do-minimum’ network.
6.4 Sensitivity tests
Sensitivity tests around a ‘do-minimum’ case should be undertaken in order to identify the robustness of the forecasts to changes in assumptions. Some examples of sensitivity tests that could be undertaken are:
- Different unit rates for travel time, vehicle operating costs, public transport wait times and transfer penalties
- Changes in public transport fare levels, parking charges, road pricing
- Changes in the demographic assumptions (i.e. population and employment levels)
- Ranges of growth in travel demand
- Changes in model parameter values (ie affecting routeing, distribution and mode choice responses)
- Different economic growth assumptions
- An assessment of complementary schemes.