Colloquium Details
Modeling a Company - Forecasting Behavior
| Author: | Juan Flores University of Michoacan, Mexico |
|---|---|
| Date: | January 26, 2006 |
| Time: | 15:30 |
| Location: | 220 Deschutes |
| Host: | Dejing Dou |
Abstract
The state of a company can be portrayed by means of its financial statements: the balance sheet and the earnings statement. From those financial statements, we can extract and derive measures and indicators that will form the company's portrait (v.g. assets, liabilities, leverage, etc.). Those statements are the standard byproduct of an accountant's work, and they are available in the company files, typically from the beginnings of the company. Let us say we can portray a company using m indicators; we will call those indicators State Variables. Of course the amount of information contained in a portrait depends on the number of variables to be considered; which variables to include in the State Variables may depend on the type of company we are modeling.
Using Genetic Programming, we are modeling a company through those State Variables. Each State Variable is represented as a monthly time series, with n observations per variable. Having a model of those variables, we can forecast the company's behavior in the short range. We have performed a first experiment, modeling each variable independently. A second experiment tries to include a subset of all m variables the model of each variable.
This experiment may disclose undiscovered relations among the models' variables, while forecasting allows us to draw conclusions like "the company will go to bankruptcy in a short time", or "the company will recover from this critical state in three months", and so on. It is convenient to note that our model is a forecasting model, and does not include the concept of causality, so it cannot be used for explanatory purposes, nor for diagnosis.