Alan Davies in Crikey has written an excellent post on Brisbane's Airport Link toll road that specifies some of the details of the demand modelling and posits a couple of conclusions as to why it went wrong.
By now, 136,000 vehicles per day were forecast to use Airport Link at the end of the initial three month toll-free period. But on October 18 – the first day tolls were introduced – traffic was only 53,000 vehicles per day.
That's how far out estimates have been. He notes that another set of modellers had a more sober view of demand:
Back in July, Brisbane transport modelling firm Veitch Lister Consulting released the results of (independent) modelling showing estimated traffic on Airport Link should rise to around 85,000 vehicles on a normal weekday (during school term) by April 2013. From that date the initial discounted toll will be replaced by a “medium toll”.
Veitch Lister estimates demand should drop to around 72,000 vehicles per day by October 2013 . However by November 2013 when the ramp-up to “full tolling” will be complete, the firm estimates the road might carry in the region of only 60,000 vehicles per day.
That compares with BrisConnection’s Product Disclosure Statement, which said 195,000 vehicles were expected to use the road 15 months after opening.
It's worth reminding readers that the toll in place, now, on the Airport Link is heavily discounted (49% the "full toll" level), and that for the road to be viable it needed to increase tolls to that level around a year after the road opened. So such an independent assessment is still out, but is at least closer to reality. Would investors have put their money into the road if those results had been in the prospectus?
That raises the first issue - which is the incentives around getting it right. Davies suggests they are badly wrong, and one reason is that those who undertake the forecasts get paid regardless of how accurate they are.
There seems to be a systemic issue with forecasting demand for major transport projects. Maybe the incentives for over-optimism on the part of the promoters are too strong.
The way projects are put together appears to underestimate demand risk. As this press report from back in 2010 notes, “despite the failure of projects elsewhere, BrisConnections is sticking resolutely to the optimistic predictions for the Airport Link.”
Perhaps part of the problem is many of those involved in putting together major projects get most of their fees irrespective of what ultimately happens with patronage. The total underwriting and associated fees mentioned in the Airport Link Product Disclosure Statement were $89 million.
Of course, if investors demanded an independent set of forecasts, it might change things, but imagine if the whole business model for demand forecasting became incentivised around accuracy, within a range. In other words, what if the modellers got a proportion of fees withheld for a set period, which would be paid if the forecasts were within a confidence range of say +/- 15%? This would inflate fees charged to undertake such forecasts, but wouldn't that be preferable to pouring billions of dollars into projects that are simply not viable?
The other conclusion Davies draws is that the value of time placed in transport models does not reflect what people are prepared to pay to save that time. In other words, $3 worth of delay does not mean a motorist is willing to pay $2.50 to save that. It just doesn't work like that.
This phenomenon isn’t new. There’s a long history of motorists, including commercial vehicle operators, going to extraordinary lengths to avoid using toll roads.
BrisConnections estimates Airport Link provides a 12 minute journey time from Bowen Hills to Toombul, compared to 25-29 minutes on alternative routes. The toll is much lower than conventional estimates of the value of the time drivers could save by using Airport Link.
On the face of it, drivers who don’t use the toll road are behaving “irrationally”. Perhaps many aren’t as good at estimating the value of their time as theory assumes. Or maybe there’s a quirk of human psychology at play – possibly many simply don’t see money and time as being readily interchangeable.
In a motoring environment where almost all trips are absent any direct pricing, the presence of a toll magnifies the cost of the toll to the driver, at least in the short term, but possibly in the medium term as well. This boils down to the tens of thousands of motorists every day making that time/value tradeoff, suggesting the value of time of congestion isn't as high as modellers suggest.
In conclusion, this all calls for some fundamental steps that may be taken in modelling future toll roads:
1. Get a completely independent demand/revenue modeller on top of the assigned one. Treat it as peer review if you must, but a second opinion is looking increasingly critical.
2. Get both demand/revenue modellers to give confidence assessments of their forecasts.
3. Seek demand/revenue modellers on the basis of payment for success, so that a portion of fees are retained if the forecasts prove to be accurate within a specific range. Be careful not to incentivise overly conservative forecasts (the obvious response of modellers will be to forecast very low figures), or otherwise no project will be worth pursuing.
4. Use stated preference surveys to verify the value of time estimates used for forecasts. Do this every time for every new project in different locations, because it is clear that value of time is far more fluid and individual than government agencies often think they are required to assume.