1 INTRODUCTION
When forecasting from inside an organisation, it can be difficult at times to be aware of the wide world of factors outside an organisation that can influence demand. This particular case study is a good example of missing self-evident factors which clearly have a significant influence on future demand.
2 LONG TERM RAIL FORECAST
In 2011, Irish Railways (IE) published the 2030 Rail Network Strategy Review[1] which included a long term demand forecasting model.
The regression model used in that particular case was:
Passengers/Population = 4.532 + 0.143 * GNP/Population
In other words, the analysts who undertook the work determined that the only significant determinant of railway demand to be considered was GDP per capita.
Figure 1 shows the fit of the model against the data for the latest eighteen years of data available. 84% of the variation in passenger demand was explained by the one variable, namely Gross National Product per person, which, superficially, is a very creditable result.
Figure 1: Actual and Modelled Data for the Republic of Ireland
Inviting though this result looks, the technique involves much more than
just fitting a line. A closer examination of Figure
2above
seems to show that, as the number of trips per population increases, the gap
between actual and predicted per capita journeys appears to widen.
Figure 2: Author’s calculation of the Error Term in the AECOM Model
As well as Figure 2 providing grounds for concern as the error term increases in magnitude from 2004 onwards, the error appears to exhibit a downward drift from 1992 to 2004; this drift would not be apparent from Figure 1, which underlines the importance of examining the model’s errors. In other words, the randomness of the error term, which is an important part of a robust model, is questionable. If the error term had settled down in the latter years of the period observed, the earlier bias in the error term could perhaps have been discounted , but, clearly, this has not happened.
The purpose of developing a model such as this is to enable forecasts in passenger demand. As the model ended in 2009, it is instructive to see how the model performed for the three subsequent years, based on actual growth in population and GNP (as recorded by Ireland’s Statistical Office, CSO), in Figure 3.
Figure 3: Actual and Modelled Demand on Irish Railways 2010-2014
Source: CSO Ireland, Author’s calculations
So, we can see that the model significantly overestimated demand in the
five years immediately following the study period and the quality of the model
deteriorates over the five years, as the analysis of the errors suggested was likely. Ireland suffered a
significant drop in GNP during the recession which started in late 2008 and the
Irish economy did not reach 2007 levels of GNP again until the end of 2014.
There are clearly other factors at work which have contributed to the model
overestimating future demand.
3 THINKING OUTSIDE THE (RAILWAY) BOX
Although the forecast exclusively on Ireland’s overall economic wellbeing, as measured by GNP per capita, what else was happening in Ireland? One issue which arose during research for my previous book “Business Planning in Transport[2]” was a major increase in the motorway network, which grew from 423km in 2008 to 1,224km by 2014 or almost triple the size in six years.
The longest distance between major cities in Ireland is the 266km between Dublin and Cork or under three hours’ drive. The motorway linking the two cities was built in stages and was finally completed in 2010. The car journey time is comparable with rail, which has an average journey time of 2½ hours. However, to compensate for the superior accessibility of cars at the origin and destination, the rail journey needs to be faster than car to compensate for the extra journeys involved (origin to railway station, railway station to destination).
The next question is whether the number of journeys has increased as a result of the motorway network expansion.
It is notable that, compared to rail travel in Ireland, which declined by nearly 20% over 5 years, road travel has stayed reasonably constant over the same period, with a decline in car vehicle kilometres of just 1.3% from 2008 to 2013. Rail’s declining market share is likely to be influenced, at least in part, by and improved motorway network.
An alternative means of examining competition from the road sector is the traffic count method. Figure 4 shows that over the period 2008 to 2014, motorway traffic between Limerick, Cork and Dublin increased by around 8% in six years.
Figure 4: Motorway Traffic Count M7, Ireland
Source: National Roads Authority, Ireland
So, what we have are various bits of evidence which are reasonably consistent. What we may infer from what we have is that, on the key Dublin-Cork route at least – the railway’s flagship route – the impact of the opening of the motorway is likely to have had a negative effect on rail demand.
Moreover, at the time that the forecast was undertaken (presumably early 2011, as the forecast was published in October of that year), the expansion of Ireland’s motorway network was at the very least well planned, if not actually underway; this makes the omission of such an important influencing factor on railway demand somewhat perplexing, to say the least.
The lesson that can be drawn from this exercise is that it is important
to consider several aspects of the environment in which a business operates: a
key message in “The Art and Science of Forecasting”.
[1] http://www.irishrail.ie/media/irishrail_28febfinal_part11.pdf
[2] https://www.amazon.com/Business-Planning-Transport-Adam-Simmons/dp/1908135824