At a recent talk, the speaker made the interesting observation that one of the problems in Bihar (one of India’s traditionally under-performing states) has been that it’s not corrupt enough. Following the famous “Fodder Scam” (buying lots of fodder for non-existent cattle), a signal was sent out that there should not be a whiff of corruption from projects in the state. The bureaucracy clamped down and for 15 years the best way to avoid accusations of corruption was to do absolutely nothing.

Finding empirical evidence for the hypothesis that “greater vigilance leads to less action” is difficult, but this is precisely what Abhijit Banerjee, Shawn Cole and Esther Duflo manage to do in their paper on vigilance and bank lending. The figure below says it all:

The blue line is the change in credit (logs) plotted against time relative to the time when an officer in a bank-branch faces an anti-corruption charge—so -1 is one quarter before the event, 2 is 2 quarters after and so on. Credit falls the quarter of the enquiry—not only for the officer who is investigated but the entire bank branch. As the authors put it, “relative to an unaffected branch, a branch in which an officer is accused of corruption experiences a 20 percent decline in credit over a period of two years.”

While one way to view this is that the costs of corruption include the externalities imposed on others in the bank, the other approach follows the classic statement by Tirole on “optimal collusion”—the equilibrium amount of corruption may not be “zero” if it’s too costly to design contracts that can enforce this. Or, as in the case of Bihar, anti-corruption drives stifle innovation and action and as the speaker put it, it may be better to have a road with a 20 percent (illegal) markup than none at all.