Flo Doesn't Know the Lucas Critique

In the March issue of Automobile magazine, Ezra Dyer recounted his experience with Progressive insurance's "snapshot" system, the latest innovation in trading privacy for money (which we all seem pretty well inured to these days).  Customers who sign up for snapshot receive discounts for "safe driving," which is measured with a device that transmits data to Progressive from your car using the cell network.

Dyer explains:

According to the rules, Snapshot can generate a discount but not a surcharge -- unless you live in Rhode Island. The device logs your speed, but that's not a factor in the calculations because Progressive doesn't know where you are -- you might be doing 65 mph in a 70 zone or 45 mph through a car wash (although one wonders if a few trips into the triple digits would disqualify you from a safe-driver discount). The deciding factors are what time of day you drive, how far you drive, and how forcefully you brake.

The reason for the braking criteria is that "gentle braking apparently correlates to low insurance claims".  Anyone familiar with the Lucas critique will spot the problem with this, as Dyer does: 

If you're approaching a yellow light, Snapshot is an incentive to risk running the red rather than hitting the brakes. If a deer jumps out in front of you, Snapshot would prefer that you swerve into the oncoming lane rather than mash that brake pedal.

That is, the past relationship between braking behavior and driving safety reflects the behavior of agents under one set of incentives - if you change their incentives, their behavior will change, and the previous relationship between braking and safe driving will no longer be valid.

What Progressive really needs is a "structural" model that embodies the underlying preferences of their customers, which will be invariant to the policy change.  Of course that's a much harder thing to do (as macroecomists have discovered over the past several decades...).