The `lemons model' in milk procurement
One of the classic stories of India of old is that of Amul, which brought new technology into milk procurement. When the farmer brought his shipment of milk to the Amul front-end, a centrifuge was used to measure the characteristics of this shipment, and based on this payments were made. See this blog post by Alok Parekh, Naman Pugalia and Mihir Sheth. This eliminated the incentives for aduleration of milk by the farmer, which used to be done by adding in water or by skimming the cream.
We can think about this differently. Suppose the centrifuge was not there at the front end. Then the buyer of the milk faced asymmetric information about the characteristics of the milk that were being offered to him. Generally we expect that faced with this `lemons' problem, the buyer would bid low prices for the milk.
So to some extent, the ability of Amul to pay higher prices for milk is not about the greatness of cooperatives when compared with profit-oriented firms: it was about the injection of new technology which removed this asymmetric information.
The interesting puzzle is: in that age, why was Amul the pioneer in buying centrifuges? Why did no private firm buy centrifuges and create a winning business model around milk?
Penalty structure under incomplete detection
Another nice idea that we have understood is the relationship between the probability of getting caught and the penalty. Suppose the fee required for parking is Rs.10 and suppose the probability of getting caught when illegally parked (i.e. without paying the fee) is 10%. Then it's sensible to set the penalty for getting caught at Rs.100 so that even a risk-neutral person will prefer to play by the rules.
This can be applied in the problem of milk procurement. Suppose we say to the farmer: We'll trust you and accept your milk, but on a sampling basis, one in ten farmers will be tested.
Suppose a person added 2 litres of water to his shipment of milk and suppose the price of milk was Rs.10 a litre. In that case, he was trying to steal Rs.20 by palming off low quality milk. But there was only a 10% probability of getting caught, because only one in ten farmers is tested. So the penalty he should face should be Rs.200. If this is done, the risk-neutral farmer is agnostic between playing fair and cheating, even if only one in ten farmers is tested.
The advantage of this strategy is that for 90% of the farmers, the deadweight cost of putting a sample into the centrifuge is eliminated.
This idea is, of course, general:
- Sometimes, we are in situations (as in market manipulation in finance) where we know that even the best regulator in the world will only catch some of the crooks. So we should estimate what fraction of the crooks are getting caught, and then multiply up the size of theft that was attempted. That is, the right way to think of disgorgement is not that the bad guys should fork up the money that was stolen, but that the penalty imposed by the government should be equal to the size of theft divided by the probability of detection.
- Sometimes, while comprehensive checking is feasible, it's quite expensive, and it's efficient to deliberately only do checking on a sampling basis. A fairly modest scale of randomised checking (e.g. 5%) can do the trick, coupled with a 20x multiplication factor against the size of the theft that was attempted. This would yield a 95% reduction in the amount of checking that is required. This is the idea in the milk example above.