Randomly Assigned Policies

Polities choosing their own policies is never optimal

The FDA is a very stringent regulatory body. There are many drugs which have been tested elsewhere, and regarded as safe, which are not allowed here. We cannot use the same sunscreens that Europe does, as the FDA does not share the opinion of the EU. Metamizole, a non-steroidal anti inflammatory drug, has been banned in the United States since 1979, is available by prescription in the EU, and is available over the counter in Brazil. Metamizole probably isn’t dangerous, but Thalidomide was. It was banned in the US, but allowed in Europe, to disastrous results. In short, whatever the optimal policy is, it isn’t being applied everywhere. But since we don’t know what the optimal policy is until it’s carried out, what is the optimal variance in regulations?

If polities are free to choose their own policies, we have suboptimally little variance in policies. What’s more, the adoption of different policies is hardly ever exogenous, which makes us less able to determine the true effects. The optimal policy is to randomly assign policies, and then converge toward optimal as time approaches infinity.

This is because policy experiments are a classic information externality. Imagine every policy has a continuous range of possible implementations. There could be an optimal tax rate, for instance, and we want to know the right number. We can think of the consensus as being on average accurate, like an efficient market. Every county in America would prefer to have more information about the effects of policies, but would lose, on average, from implementing it themselves. The way out is an agreement to experiment, and to bear the costs among themselves equally.

There are some possible problems with this proposal. First, the size of the unit might matter, such that estimates derived from a smaller polity might lack external validity. A policy could be welfare enhancing if a nation enacts it, but welfare decreasing if a smaller entity enacted it. For example, a tax to fund a public good might be welfare enhancing for a nation, but welfare decreasing for a city, if people can easily migrate to avoid the effects; or a gun control bill could be beneficial if enacted at the national level, but not at the county level, as people could easily evade it. The national government could get around this by writing in many possible versions of a bill, and choosing which version enters into law through odds specified at the bill’s passage.

Another problem is risk-aversion. Perhaps people prefer a certain bad policy, to an uncertain alternation between good and bad policies. Imagine if we switched back and forth between a tariff and no tariff, and because of fixed costs, firms neither moved overseas nor invested at home. Alternatively, perhaps polities fear the negative effects of bad policies more than they would gain from good ones. This would simply narrow the variance in policy.

I do fear that, for many questions, we won’t have a sufficiently large sample to converge to something close to a point. Still, even if we are just providing bounds, it’s still good to do. Even when we cannot undertake the maximalist program proposed here, we should still test more programs. What is good enough for foreign aid to other countries should be good enough for here.