Congestion pricing: practical problems

I’ve been pushing congestion pricing on this blog for a couple of years now.   I’ve argued ad nauseum that variable congestion pricing would get traffic flowing at the least cost, and would enhance general welfare better than the continual and counterproductive expansion of highways. 

To be honest, though, I need to disclose the practical problems with congestion pricing.

Economists first began lecturing traffic engineers about congestion pricing 40 or so years ago.   The conversation’s gone something like this:

Economist:  “You guys need to stop building roads to alleviate congestion.  If you price the roads right, you can eliminate congestion and enhance overall welfare.” 

Traffice engineer:  “I hear what you’re saying, but we don’t know how to implement congestion pricing.”

Economist:  “It’s easy.  Just charge a high enough price to get traffic flowing.”

Traffic engineer:  “But we don’t know how to do that.”

Economist:  “What’s not to understand?  Just charge a high enough price to get traffic flowing.”

Traffic engineer:  “You don’t understand.  We don’t know how to figure out what that price is and we don’t know how to communicate that information to drivers in time for it to do any good and we’d have to price other roads in order for the system to work and we don’t know how to price networks.  Plus, everyone hates tolls and I like my job.”

Economist:  “I’m sorry.  I wasn’t listening.  I’m working on an interesting paper on the Herfindalh-Hirschman index and models of stochastic processes in hyperbolic discounting.”

Traffic engineer:  “Whatever.”

The traffic engineer goes back to work widening roads.

Here’s the information problem:  The optimal congestion toll should be set just high enough to achieve free-flow (45 mph) traffic.  But if the toll is set too high, it will induce too many drivers to shift to other times, routes or modes of transportation.  That’s bad, too (at least if you ignore other externalities like pollution.)

Traffic engineers can generally predict the high-demand days, but there’s a fair amount of randomness in traffic patterns.  Some days an unusually large number of drivers just happen to drive to work at the same time.

The optimal toll therefore should be variable — the greater the demand, the higher the toll.  But that’s very hard to implement as a practical matter.  How do we get would-be drivers the information they need to make timely decisions?  There’s no point in raising prices on drivers once they’ve entered the road; raising prices can no longer influence their behavior (except perhaps to launch them into a homicidal rampage). 

Price changes might affect the behavior of drivers who are about to enter the highway.  But they are just a fraction of the drivers targeted by congestion pricing.  Congestion prices are also intended to shift drivers’ time of travel and mode of transportation.  That requires getting them the price in advance, in real time (via the internet, for example).  But that, in turn,  creates a real risk of herd behavior.  If the posted price is high, most drivers will respond by taking alternate routes or leaving too late.  If the posted price is low, drivers will rush to their cars to take advantage of the low tolls, creating a sudden surge in demand and unnecessary spikes in prices.  There’s a sort of Heisenberg uncertainty principle at play.

There are other practical problems.  For example, charging tolls might hurt businesses that have based their investments on the assumption that roads would remain free.  Pricing’s impact varies industry by industry, and consumer by consumer.  The Oregon Transportation Commission has commissioned several white papers examining some of the practical problems.

Most experts therefore assume that the best pricing system will be a second-best system, like cordon charges or charges fixed by time of day.  These systems erode some of the advantages of congestion pricing but, in my opinion, are still vastly superior to the unpriced alternative.  London, for example, has had good success with its cordon-pricing system.  (See VTPI’s case study here.)

And congestion pricing is still very new.  We’ve had the technology for electronic toll collection for only a few years; congestion pricing isn’t feasible with toll booths.  Let’s experiment and get some data.  I’m confident the engineers will eventually develop effective models.  But there will be some glitches at first, and we’ll need some patience.