Traffic Risk in PPPs, Part II: Bias in traffic forecasts—dealing with the darker side of PPPs


This page in:

Photo: Susanne Nilsson| Flickr Creative Commons

This is the second of a three-part series on traffic PPPs.

"It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so."
“The Big Short” 
Forecasting traffic accurately is a very difficult and thankless task, as I explained in my previous blog: Traffic Risk in Highway PPPs, Part I: Traffic Forecasting. As such, this gives rise to very real financial risks if these forecasts turn out to be wrong. This risk has crystallized many times as manifested in high-profile distressed projects, bankruptcies, renegotiations and bailouts.

So what’s driving the inaccuracy and resulting risk in traffic forecasts? In the Public-Private Infrastructure Advisory Facility (PPIAF) and Global Infrastructure Facility (GIF) publication, Toll Road PPPs: Identifying, Mitigating and Managing Traffic Riskwe postulate that forecasting inaccuracy comes from three sources:

  • Error: The inaccuracies that result from the errors of the forecasting method itself and that are internal to the forecasting process. These are effectively the result of (involuntary) human error that occurs during the development of the traffic study and specifically occur when the traffic forecaster is trying to establish the existing demand for travel in the study area (the so-called “in-scope” traffic). Examples include sampling errors in traffic surveys and traffic model errors.
  • Uncertainty: These are the inaccuracies that are typically out of the control of the traffic forecaster. They represent the changes in the external environment that occur over time and that were not foreseen or assumed at the time the traffic study forecasts were originally developed. Examples include large increases in fuel prices, economic recessions, or development of competing facilities.
  • Bias: Traffic forecasts can be artificially high to facilitate a specific goal of a project party, for example, a bidder trying to develop a winning bid for a project or a government official trying to ensure a project achieves government approval.
Error and uncertainty should, in theory at least, cancel each other out. Forecasters, on average, are just as likely to be over-forecasting as under-forecasting because of errors and uncertainties in their forecasts. For example, you might over-count some vehicles at one count site, but just as likely under-count somewhere else.

But empirical evidence has shown that it is all a bit more nefarious than that. Many studies have found that traffic forecasts skew toward overestimation, indicating that systematic biases are present in the forecasting process.

Bias is therefore the real enemy in the forecasting process.  Error and uncertainty are just a reflection of our imperfect knowledge of the existing and future state, and as we stated in the first blog, we need to go easy on the transport economist fraternity on this. But bias, in its many forms, is something much more deliberate and has even led recently to legal proceedings against forecasting firms who stand accused of deliberately inflating traffic forecasts.

This is why we have given it special attention to bias in the second of the PPIAF Issue Briefs on traffic risk, Delusion, Distortion, and Curses: Bias in Traffic Forecasting, where we outline what causes bias and how governments can act to reduce it through the project cycle.

We hope you'll find it useful reading.

Read these next:


Toll-Roads PPPs: Identifying, Mitigating and Managing Traffic Risk

Related Posts

Traffic Risk in Highway PPPs, Part I: Traffic Forecasting — It’s ok to be wrong, just try to be less wrong

Traffic Risk in PPPs, Part III - Allocating Traffic Risk: Prophet & Loss



Matt Bull

Senior Infrastructure Finance Specialist

Join the Conversation

Dr. Mohamed Taher Abdelrazik Hamada,Ph.D
June 20, 2017

Financial risks that make insurance companies pay a lot of money of thir terrible consequences. Managing traffic risk is essential to avoid such loss in lives and and money that are refelected negatively on traffic forecasts..
Some errors occur as a result of the forecasting process itself , maybe lack
experience is one of the main reasons.
These errors can be avoided by having trafficing models that can be put as examples to be avoided in later experiences.
Also uncertainity of the surrounding environment circumstances such as the unexpected huge raise in the prices of the vehicles , or the world fuel prices
and the unfair competitiveness among companies and firms that build highways or even sell vehicles.
Beside all that is the routine governmental issues that hinder the steps of the acceleration of the process of accuracy and forecasting of the sharing investment firms .
Accuracy in forecasting can make errors as less as possible .
Bias , of course is a dangerous situation that blocks any type of a forecasting process , it may damage the the forcasting process because it tries to avoid it's system of objectivity .
All researchers in the field of Traffic Risks in the PPPS should be aware of the causes of bias and try collectivily to eliminate them.
Yours Very Respectfully,
Dr. Mohamed Taher Abdelrazik Hamada, Ph.D
Retired Professor at Strayer University, USA

Jacques Cook
June 20, 2017

I am frankly surprised that the reliability of traffic projections has just now surfaced as an issue worthy of special attention. As a practitioner in the several PPP transactions, I was always uneasy with how much weight was accorded to the traffic forecasters, based mostly on their expertise. Since most of these toll roads were structured around limited recourse project financing with revenues generated by user fees, investors and lenders were almost always taking a big market risk unless government was offering a minimum revenue guarantee or other credit enhancements served to mitigate these risks. It was rarely considered whether the traffic forecasters should bear some of this risk. If they did, they might be more cautious and less likely to inject bias into their equations.