Anon

291

CHAPTER

14

Model Selection

A

fter reading this chapter you will understand:

■ (^) The notion of machine learning.
■ (^) The difference between an approach based on theory and an approach
based on learning.
■ (^) The relationship between the size of samples and the complexity of
models that can be learned.
■ (^) The concept of overfitting.
■ (^) The use of penalty functions in learning.
■ (^) The concept of data snooping.
■ (^) The concept of survivorship bias.
■ (^) The concept of model risk.
■ (^) Methods for mitigating model risk.
■ (^) Model averaging.
In the previous chapters in this book, we described the most important
financial econometric tools. We have not addressed how a financial modeler
deals with the critical problem of selecting or perhaps building the optimal
financial econometric model to represent the phenomena they seek to study.
The task calls for a combination of personal creativity, theory, and machine
learning. In this chapter and the one to follow we discuss methods for model
selection and analyze the many pitfalls of the model selection process.

Physics and Economics: Two Ways of Making Science

In his book, Complexity, Mitchell Waldrop describes the 1987 Global Econ-
omy Workshop held at The Santa Fe Institute, a research center dedicated
to the study of complex phenomena and related issues.^1 Attended by

(^1) M. Mitchell Waldrop, Complexity (New York: Simon & Schuster, 1992).