« Prediction and Forecasting »

Limits of Models in Decision

In today’s Financial Times columnist John Kay has a very insightful piece on the limits of models in decision making. He discusses the downfall of Amaranth, a hedge fund, which lost billions of dollars, in part, because its investors did not fully understand the full scope of uncertainties associated with their investment strategies. Kay highlights an important distinction between what he calls “in model” risk and “off model” risk. In model risk refers to the uncertainties that are associated with the design of the model, in data inputs, randomness, and so on. Modelers use techniques such as Monte Carlo analysis to get a quantitative sense of model uncertainties. Off model risk refers to the degree of conformance between a model and the real world. Models by their nature are always simplifications of the real world. As in the case of Amaranth, often hard lessons of experience remind us that as powerful as models are, they can also reinforce bad decisions. As Kay writes,

When someone does attach a probability to a forecast, they have – implicitly or explicitly – used a model of the problems. The model they have used accounts for in-model risk but ignores off-model risk. Their forecasts are therefore too confident and neither you nor they have much idea how over-confident they are. That is why mathematical modeling of risk can be an aid to sound judgment, but never a complete substitute.