Setting Up the Problem

Joseph Breeden

Every modelling problem begins with the same steps:

• • specify the problem;

• • choose the variables; and

• • validate the data.

This is the boring stuff compared with designing fancy models or the thrill of discovering something new in the results, but the best model can easily be defeated by poor design or poor data. This chapter is about beginning with a sound foundation.

MODELLING OPPORTUNITIES

Although portfolio forecasting has been around a very long time, it is often only used in limited ways. For almost any activity that occurs within a portfolio, we can ask: “Where will it happen?”, “When will it happen?”, “How much will it happen?”

Retail portfolios of all types have scoring incorporated; many have made significant use of customer scoring tools to refine marketing, target promotions and manage risk. However, few businesses outside banking have incorporated sound methods for portfolio modelling.

Portfolio analysis is much more than just a loss forecast for a bank or a revenue forecast for a telecommunications company. How many accounts will that telecom’s mobile phone network need to support? Which regions will grow faster? How many users must