A Toll Road Is Financed on a Traffic Forecast That Is Usually Wrong

The Financing Structure

A toll road is typically financed as a stand-alone project on its own vehicle and the debt is repaid with toll proceeds. Lenders look to the project and not the sponsor

This means that all funding is based on a forecast of how many vehicles will use a road that does not yet exist over a concession period that typically lasts thirty years or more

The Forecasting Record

Studies of completed toll roads have found that actual traffic in the early years is well below expectations often by twenty to forty percent and that errors lean heavily in one direction

A random forecast error would produce overestimates and underestimates to approximately the same extent. Consistent overestimation indicates something systematic

A Worked Example: How a 30 Percent Traffic Miss Becomes a Default

It seems that a deficit of twenty to forty percent can be overcome. Pass it off as project financing and it is not because the structure amplifies it twice. Here is the arithmetic in simplified form with round numbers chosen for clarity

The base case. The forecast is 50,000 vehicles per day with an average toll of $4. Annual gross revenue is 50,000 times 4 times 365 which is 73.0 million. Operating costs i.e. maintenance toll systems and personnel amount to 18.25 million per year. Net revenue available for debt service is 54.75 million

The debt service of the project is 40.0 million per year. debt service coverage ratio which is net income divided by debt service is 54.75 divided by 40.0 or 1.37 times. Lenders are comfortable. There is 37 percent wiggle room before a dollar of interest goes unpaid

Now apply the documented deficit. Traffic is 30 percent lower than expected with 35,000 vehicles a day. This is exactly within the twenty to forty percent range that the studies describe so this is not a disaster scenario. It is the central expectation once you know the record

Gross revenue becomes 35,000 times 4 times 365 or 51.1 million. Here's the first amplifier: Operating costs barely change. A highway carrying 35,000 vehicles needs almost the same maintenance toll gates and staff as one carrying 50,000. Call operating costs remain unchanged at 18.25 million

Therefore net income is 51.1 minus 18.25 which is equal to 32.85 million. Coverage is 32.85 divided by 40.0 or 0.82 times

Read the two columns opposite each other. Traffic fell 30 percent. Net income fell 40 percent because fixed costs do not decrease with traffic. Coverage fell comfortably below one meaning the project cannot pay its lenders outside of its operations. Capital is worthless long before that point

Then the second amplifier which is what turns a bad year into a failed project. The obvious response from the administration is to increase the toll. To restore coverage to 1.0 times net income of 40.0 million is needed that is gross income of 58.25 million. Through 35,000 vehicles a day that means a toll of 58.25 million divided by 12,775 million annual crossings which is equivalent to $4.56. A price increase of 14 percent

But the reason traffic failed in the first place is that drivers were more willing to take the free alternative than the model assumed. Raising the price by 14 percent in a demonstrably price-sensitive population sends more people to the free route further reducing traffic requiring a higher toll again. That loop has a name in the industry and has ended several concessions

These figures are illustrative. The mechanism is not. A forecast error that studies describe as typical produces through fixed operating costs and fixed debt service a project that defaults on its senior debt while the highway itself functions perfectly

Why Willingness to Pay Is Underestimated

The most consistent modeling error has to do with how much drivers value time. Models assign a value to time saved and predict that drivers will pay tolls that cost less than that value

In practice drivers avoid tolls at far higher rates than the models predict choosing free routes even when the time cost exceeds the toll. Whether this is irrational or reflects something the models fail to capture it means that demand is more price-sensitive than assumed

My own suspicion is that the models measure the wrong thing. A toll is a visible detailed payment that is made repeatedly and people seem to hate it much more than they hate an equivalent amount of diffuse cost. Twenty minutes of your life are expensive and invisible. Four dollars is cheap and staring back at you every morning

The Ramp Up Problem

Traffic on a new road gradually increases as drivers learn the route and development develops around them. Forecasts model this increase and generally assume that it will happen faster than it does

This is hugely important for financing because debt service begins immediately while revenues arrive slowly. A highway that eventually reaches its expected traffic can still default in the first few years and many have

Put the worked example on a timeline and the point becomes sharper. If the first year's traffic is half of the final steady state coverage in the first year is well below the 0.82 times calculated above and no amount of being right about the eighth year helps a project that can't pay its lenders in the second year

Case Study: Brisbane's Clem7 and a Forecast That Missed by Eighty Percent

The clearest example on record is the Clem Jones tunnel known as Clem7 in Brisbane Australia

The tunnel opened in March 2010 funded through a listed vehicle called RiverCity Motorway. The traffic forecast used to raise the money anticipated about 100,000 vehicles a day. Actual traffic once the free introductory period ended and tolling began was about 21,000 per day

This is not a deficit of twenty to forty percent. It is an error of about eighty percent and the above arithmetic explains what happened next without requiring further details. RiverCity Motorway went into receivership in February 2011 less than a year after the tunnel opened. Investors who had provided approximately A$700 million in capital lost virtually everything

The part that makes Clem7 unusual is what followed. Investors filed a class-action lawsuit not against the operator but against the traffic forecaster arguing that the prospectus projections had no reasonable basis. Very few forecast errors have been litigated so directly and the case forced an uncomfortable public conversation about whether a traffic forecast in a fundraising document is an engineering estimate or a marketing document

Meanwhile the tunnel continued to operate. It was purchased bankrupt for a small fraction of its construction cost and Brisbane motorists continued to use it exactly as before. The road was successful. The financing fell through. That's the pattern

Clem7 is extreme but not isolated. The Indiana Toll Road was leased to a private consortium in 2006 for $3.8 billion due to long-term traffic projections filed for bankruptcy in 2014 after traffic and revenue declined during the recession and was sold again in 2015. Same structure same failure mode different continent

The Recurring Failure Pattern

The sequence is repeated in all countries. An optimistic forecast supports high leverage. Traffic disappoints. The project cannot pay its debt. Lenders and shareholders take losses and the concession is restructured or returned to the public authority

The road continues to function at all times because it is a working asset. Failure is more financial than physical and the eventual owner acquires a working road for a fraction of what it cost to build

It's worth saying clearly: the losses fall on investors who mispriced traffic risk and the public usually ends up with the infrastructure. Whether that is an acceptable way to finance roads is a legitimate question with arguments on both sides

Where This Critique Goes Too Far

I just devoted most of an article to arguing that toll road forecasts are systematically wrong. That argument is strong and routinely taken beyond what the evidence supports

The free road is not free. The idea that the public wipes out infrastructure at a discount is satisfying and incomplete. Private capital that disappears once does not return under the same conditions. It comes back demanding higher returns tighter covenants and public guarantees and those costs are paid by taxpayers on each subsequent project. The deal seems good on the path where the loss occurred and worse along the program

Availability payments move risk rather than eliminate it. Making the public authority pay for opening the road seems like the obvious solution and is often the right answer. It also means that the taxpayer is now financing a road that no one uses with no market signals to reveal that fact. The demand risk assumed by a lender produces a bankruptcy and revaluation. The demand risk assumed by a government produces an item that persists for thirty years

Not all categories work the same. The dismal track record belongs primarily to stand-alone brand-new toll roads with no history of operation. Managed lanes added to an existing congested corridor with dynamic pricing have a materially better track record because the demand they forecast is already observable rather than hypothetical. Treating all tolled infrastructure as an asset class is the same mistake forecasters are accused of making

The optimism bias literature has its own selection problem. Failed projects are studied because they are interesting. Roads that met expectations do not generate documents litigation or articles like this. I believe the direction of the bias is real and well evidenced. I am less confident that the average magnitude cited in the studies is the average magnitude in the world

How I Would Underwrite a Toll Road

If someone handed me the financing of a toll road to look at I would work in this order and almost none of it involves the traffic model itself

I would first reconstruct the coverage table from the worked example above at three levels of traffic: the sponsor's forecast that forecast less than thirty percent and that forecast less than fifty percent. If the structure does not survive the middle column the agreement is not supported by the historical record and everything the sponsor says about why this path is different is a story that would require extraordinary evidence to accept

Second I would look at the split between fixed and variable operating costs because that ratio is what turns a loss of traffic into a collapse in revenue. A highway with genuinely variable costs is a much safer asset than one where maintenance and tolling infrastructure are compromised regardless of its use

Third you would look for the free alternative on a map and drive it or at least look carefully at what it costs a traveler in minutes. Each willingness-to-pay assumption in the model is actually a statement about that specific alternative route and is testable in a way that the rest of the forecast is not

Fourth I would check who prepared the forecasts who paid them and whether their previous forecasts on comparable roads have been published based on the results. Clem7 made that issue litigious. It should have been standard by now

Fifth I would look for interest capitalized through the raise. Its absence tells me that the sponsor either fully believes in the raise assumption or needs the primary leverage figure to work both of which are reasons for me to be more careful

My general honest opinion is that these assets are frequently good stocks and frequently bad stocks and that the gap between those two judgments is where money is lost. That is an opinion formed by reading the record not professional advice

What Better Practice Looks Like

Improvements that have worked involve forecasting from the history of comparable projects rather than the details of the project making explicit adjustments for the well-known optimism bias capitalizing interest during the acceleration period so that early shortfalls do not cause default and using availability payments where the public authority pays for opening the road and retains the traffic risk itself

That last option is honest about who is best positioned to take on demand risk which is often the public sector and not lenders

The Bottom Line

Toll roads are funded based on long-term traffic forecasts that they overestimate with surprising consistency driven by optimism bias the selection of optimistic projects for funding and models that underestimate how much drivers avoid paying

Arithmetic explains why a familiar-sounding error is fatal. On a highway that provides for 50,000 vehicles per day a thirty percent traffic shortfall reduces net revenue by forty percent because operating costs are fixed and coverage drops from 1.37 times to 0.82 times. The road is full of cars and the project is in default and raising the toll to fix it scares away the same drivers who proved they were price sensitive

Clem7 forecast 100,000 vehicles a day and transported about 21,000 went bankrupt within a year of opening and ended up with investors suing the forecaster. The tunnel is still there and busy. Typically the road works and the financing doesn't and structures that recognize who should bear the risk of the lawsuit work better than those that pretend it can be transferred