The View of Congestion Pricing from Texas

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

Trafficengineers can generally predict the high-demand days, but there's afair amount of randomness in traffic patterns. Some days an unusuallylarge number of drivers just happen to drive to work at the same time.

Theoptimal toll therefore should be variable -- the greater the demand,the higher the toll.  But that's very hard to implement as a practicalmatter. How do we get would-be drivers the information they need tomake timely decisions? There's no point in raising prices on driversonce they've entered the road; raising prices can no longer influencetheir behavior (except perhaps to launch them into a homicidalrampage). 

Pricechanges might affect the behavior of drivers who are about to enter thehighway.  But they are just a fraction of the drivers targeted bycongestion pricing. Congestion prices are also intended to shiftdrivers' time of travel and mode of transportation. That requiresgetting them the price in advance, in real time (via the Internet, forexample). But that, in turn, creates a real risk of herd behavior. Ifthe posted price is high, most drivers will respond by taking alternateroutes or leaving too late. If the posted price is low, drivers willrush to their cars to take advantage of the low tolls, creating asudden surge in demand and unnecessary spikes in prices.  There's asort of Heisenberg uncertainty principle at play.