The Danger of Optimization
RISKVIEWS was recently asked “How do insurers Optimize Risk and Reward?”
The response was “That is dangerous. Why do you want to know that?” You see, a guru must always answer a question with a question. And in this case, RISKVIEWS was being treated as a guru.
Optimizing risk and reward is dangerous because it is done with a model. Not all things that use a model are dangerous. But Optimizing is definitely dangerous.
One definition of optimizing is
“to make as perfect as possible.”
Most often, optimization means taking maximum possible advantage of the diversification effect. You will often hear someone talking about the ability to add risk without adding capital. Getting a free ride on risk.
There are two reasons that optimizing ends up being dangerous…
- The idea of adding risk without adding capital is a misunderstanding. Adding risk always adds risk. It may well not add to a specific measure of risk because of either size or correlation or both, but the risk is there. The idea that adding a risk that is low correlation with the firm’s predominant risk is a free ride will sooner or later seep into the minds of the people who ultimately set the prices. They will start to think that it is just fine to give away some or all of the risk premium and eventually to give up most of the risk margin because there is thought to be no added risk. This free risk idea will also lead to possibly taking on too much of that uncorrelated risk. More than one insurer has looked at an acquisition of a large amount of the uncorrelated risk where the price for the acquisition only makes sense with a diminished risk charge. But with the acquisition, the risk becomes a major concentration of loss potential and suddenly, the risk charge is substantial.
- In almost all cases, the best looking opportunities, based on the information that you are getting out of the model are the places where the model is in error, where the model is missing one or more of the real risks. Those opportunities will look to have unusually fat risk premiums. To the insurer with the incorrect model, those look like extra margin. This is exactly what happened with the super senior tranches of sub prime mortgage securities. If you believed the standard assumption that house prices would never go down, there was no risk in the super senior, but they paid 5 – 10 bps more than a risk free bond.
The reliance on a model for optimization is dangerous.
That does not mean that the model is always dangerous. The model only becomes dangerous when there is undue reliance is placed upon the exact accuracy of the model, without regard for model error and/or parameter uncertainty.
The proper use of the model is Risk Steering. The model helps to determine the risks that should be held steady, which risks would be good to grow (as long as the environment stays the same as what the model assumes) and which risk to reduce.