An important role of the management accountant is to evaluate the data that managers use when making decisions. For example, how objective are the data? Are the data an estimate or carefully measured? What data and costs are relevant for the decision at hand?
The term exploratory data analysis refers to any activity that reveals deeper insight into a dataset. This activity may include numeric analysis, such as finding the mean, median, minimum, and maximum values for a feature like annual income.
Before the data science team builds a sophisticated model, Paige explores the PD dataset to understand its size and contents. There are 500,000 funded loans in the dataset from the years 2015 and 2016, of which 100,000 are defaults and 400,000 are repays . The overall default rate is 20% .
Each record in the PD dataset has a number of features that describe the characteristics of the loan and the borrower. The target feature is loan status. This is the feature the Sierra Investments team wants to predict (default or repay). The independent features of each record are the loan and borrower characteristics Paige plans to use to predict whether a loan will default or be repaid. Some independent features are from the past, such as credit score, number of mortgages, and number of bankruptcies. Other features are from the present, such as a borrower’s employment, annual income, amount of loan, and the stated purpose of the loan. (See columns under “Phase 1” in Exhibit 11-3.) Examples of purpose are refinancing credit card debt or home renovation. PD has done a careful job of verifying information such as employment and annual income because its business model is based on being transparent and having high-quality data that lenders can trust when making their decisions.
| Phase 1 (Before) | Decision to Invest | Phase 2 (After) | |
|---|---|---|---|
| Past Data | Present Data | “Future” Data | |
|
| Yes or No |
|
Paige wonders whether the purpose of the loan is a feature she should include in her analysis. Should she have an a priori reason about why the purpose of the loan might help distinguish between default and repay loans? Or should she simply include this feature and let the analysis tell her whether it matters? Management accountants could include many independent features in the model because it is easy and inexpensive to handle large quantities of data. At the same time, it is important to understand why a particular feature might affect the target feature. Otherwise, as we discussed in Chapter 10, we might mistake correlation for causation. This type of critical thinking about what data should go into a model is a key role of the management accountant. However, as we shall see later in the chapter, there are data analytic methods that explicitly select those feature variables that best explain the target feature.