Friday, November 25, 2011

When random taxation is better

As many governments are scrambled for new revenue, tax evaders are falling under more intense scrutiny than usual. Southern European countries are particularly well known to be a paradise for tax evasion, as authorities are rather week and corruptible. As enforcement is rather difficult to improve, could an institutional change work better?

Stéphane Gauthier and Guy Laroque think they have found a solution, which is to randomize taxes. This builds on the fact the risk tolerance varies across tax payers and that the latter can face stiff penalties (or inconvenience) when caught evading. Suppose the skilled worker is more risk averse than the unskilled one, and you want to redistribute from skilled to unskilled, but cannot observe skills (or risk aversion). The skilled will try to pretend to be unskilled to avoid taxation. But if taxes are random, then he has fewer incentives to do so. If risk aversion goes the other way, then of course, tax evasion becomes worse. The authors offers no evidence on the correlation of risk aversion and skills or income, thus it is hard to tell how useful this scheme would be, and how welfare would be improved.

NB: The authors could have just spent a couple of seconds at the paper before submitting it. The forgot to BibTeX it, and none of the references show up. This is unfortunately a carelessness that I have encountered too often with French economists.


Kansan said...

This paper highlights one of the big problems with theory papers in public economics: no shred of an attempt to prove the results are relevant in any sensible way, for example by drawing on some empirical evidence. What is the use of theorems if all they say is that anything could happen under some circumstances, and we have no idea whether those circumstances have a positive probability of existing?

Anonymous said...

Good call on the French carelessness. I was wondering whether I was the only one noticing it. The most stunning I saw was a tenure file with obviously wrong journal names.