How to Conduct Discriminant Analysis to Predict a Company's Power?

Finance ManagementBanking & FinanceGrowth & Empowerment

What is Discriminant Analysis?

Discriminant analysis is a widely used tool in Machine Learning, Statistics, and Finance. It is a manner to classify the targets based on some assumptions, and then use them to predict the future of a process. As is obvious, since investors and analysts want to predict a company’s power in the future, the tool is of prime importance to them.

The discriminant analysis relies upon continuous independent variables to form a pattern that shows relationships between two parameters or predictive equations that may satisfy the relationship for the data sets. These equations are then used to classify the dependent variables.

When the data are clubbed into two groups, it is called DFA or Discriminant Function Analysis. When data is clubbed into more than two groups, then the process is known as Multiple Discriminant Analysis or MDA. Multiple Discriminant Analysis is also referred to as canonical varieties analysis or CVA.

Assumptions of Discriminant Analysis

When the applicable assumptions are satisfied, discriminant analysis can provide excellent details about the predictive classification.

Following are the major assumptions in the Discriminant Analysis function −

  • The independent variables are in a normal distribution. Most of the variables from real-life applications either have a normal distribution or tend to lend themselves to a normal distribution curve.

  • The variances across the given categories are assumed to be similar or the same across the level of predictors. Quadratic DA is more suited to these assumptions.

    However, linear DA also follows the assumption. The presence of outliers can be monitored here. So, making variances stable is a major need that can be satiated by logarithmic transformations.

  • The predictor variables are taken as independent ones. Having a correlation between them can diminish the power of analysis. To get rid of this issue one can replace or remove variables to ensure independence.

  • The samples are also considered to be independent in the analysis. It is a fair assumption when the population is large.

How to Conduct Discriminant Analysis to Predict a Company’s Power?

Discriminant analysis is done when the categories of output are known beforehand or when one wants to successfully categorize the datasets. If the output categories are not known beforehand, clustering has to be done.

In the following mentioned below discriminant analysis is conducted to predict a company’s health −

Formulating the Problem

The first idea is to find the problem at hand. In the case of determining a company’s power, the problem is whether the company would be successful or it will fail and go bankrupt.

After finding the problem the independent variables and categories of results have to be identified. Some of the variables may include net working capital to total assets, retained earnings to total assets, and EBIT/total assets. The number of independent variables can be as many as possible.

Samples are divided into two groups–analysis is used for estimation while the validation group is used to check the results.

Finding the Discriminant Function

The discriminant function is written as −

$$\mathrm{D\:=\:b0\:+\:b1X1\:+\:b2X2\:+\:\ldots+bkXk}$$

Here, ‘D’ is the discriminant score, ‘b’ represents the coefficients for the predictor variables ‘X’. when ‘X’ is known, one needs to estimate the values of ‘b’.

Finding the Importance of the Discriminant Function

The function we get from the above-mentioned process must be significant statistically. For this purpose, the eigenvalues of the function are checked. The better the eigenvalue the better the chances of the function being correct.

Interpreting the Results

It is important to check the influence of the predictors by their coefficients. A better absolute standard co-efficient value points toward better discrimination.

Assessing the Validity

You must now categorize the data according to the discriminant score and one decision rule. After the classification of the validation sample, the percentage of correct classifications should be measured. This is a crosschecking for the validations.

raja
Updated on 23-May-2022 12:21:45

Advertisements