Models to Predict Bankruptcy

¶ … Discriminate Analysis and Other Models to Predict Bankruptcy

Type of Data Used

Selection of Keywords

Authenticity and Validity

Methods Used for the Prediction of Bankruptcy

Discriminant analysis

Logit analysis

Neural networks

Distance to Default

There has been a constant increase in the attempts that are made to predict bankruptcy because of the deteriorating consequences that are associated with this phenomenon. These consequences include the following: The negative social and economic consequences for the investors and creditors who are associated with the bankrupted organization. The social and economic consequences that the competitors and government, who are associated with the affected organization, would confront. This research will explore various methods that are used for the prediction of bankruptcy. It will highlight the functioning of these models. In addition to that, this paper will also list down the advantages and disadvantages of all of the discussed models as well.

Using Discriminate Analysis and Other Models to Predict Bankruptcy

Introduction

There has been a constant increase in the attempts that are made to predict bankruptcy because of the deteriorating consequences that are associated with this phenomenon. (Santos & Cortez et al., 2006)These consequences include the following:

The negative social and economic consequences for the investors and creditors who are associated with the bankrupted organization. (Santos & Cortez et al., 2006)

The social and economic consequences that the competitors and government, who are associated with the affected organization, would confront. (Santos & Cortez et al., 2006)

The two basic kinds of models that are generally adopted for the prediction of bankruptcy are discussed in the following section:

Accounting ratios-based models: These models consist of the statistical techniques, which include discriminant analysis and logistic regression models. (Santos & Cortez et al., 2006)

Market-based models. In this category the KMV model of Moody was adapted. (Santos & Cortez et al., 2006)

This research will explore various methods that are used for the prediction of bankruptcy. It will highlight the functioning of these models. In addition to that, this paper will also list down the advantages and disadvantages of all of the discussed models as well.

Research Approach

This methods that have been used for the collection, analysis and interpretation of the data, in the course of this research, are highlighted by the following section. In addition, this section also discusses the strategy that has been deployed for the presentation of this data under this research.

Research Method

For the purpose of this research, the research has utilized a mixed research methodology. Such a methodology consists of a combination of the characteristics of both the qualitative as well as the quantitative research methodology. (Saunders and Lewis et al., 2003)

Mixed Research Methodology

The assumptions used by this type of research methodology are Post positivist and constructivist assumptions in nature. In addition to that, this type of methodology also uses pragmatic assumptions. (Saunders and Lewis et al., 2003)

This type of research methodology generally employs a mixed strategy of inquiry. The experimental, quasi-experimental, narrative and ethnographic designs are included in the strategy of inquiry of this research. (Saunders and Lewis et al., 2003)

The mixed research methodology deploys a balanced combination of both emerging as well as predetermined methods of research. A combination of statistical data, close ended questions as well as open ended questions is deployed by such research methodology for the purpose of collection and evaluation of data. (Saunders and Lewis et al., 2003)

In addition to that, a number of other sources, including performance, attitude, observation and census data, image analysis, audio visual data, text analysis and field observation, are also deployed by the above mentioned research methodology for the collection of data and information for the purpose of the research under consideration. (Saunders and Lewis et al., 2003)

Type of Data Used

The two primary types of data can be named as secondary data and primary data. Primary data can be defined as the new and fresh data that is first handedly collected by the researcher himself. This data is generally collected by the researcher for the purpose of the research that he or she aims to conduct. A number of varying tools are used by the researcher to collect this type of data. These tools include questionnaires, interviews, field observations and experiments etcetera. (Saunders and Lewis et al., 2003)

Secondary data, on the other hand, can be defined as the type of data which exists previously in relation to the topic under consideration. Such data is collected by various individual researchers, academic institutions and business organizations etcetera due to a wide range of reasons. This type of data can generally be accessed by the common public in an easy manner. The basic sources of this data include, published books, e-books, published journals, online journal articles, peer reviewed researches and other internet-based sources. (Zikmund, 2003)

The data used for the purpose of this research was secondary in nature. The researcher gathered data for the purpose of this research from various sources that existed in relation to the issue under consideration. These sources included published books, e-books, published journals, online journal articles, peer reviewed researches and other internet-based sources. (Saunders and Lewis et al., 2003)

Selection of Keywords

In order to collect appropriate and effective data about the topic under consideration, it is necessary for the researcher to develop a strong and valid relationship between the topic of the research and the keywords used to search for data. This is because appropriate selection of keywords enables the researcher to get an access to the effective sources of data in a less time consuming manner. Effective selection of keywords makes the data collection process efficient and enables the researcher to collect relevant data in an easy manner. (Cooper and Schindler, 1998)

The keywords that were deployed for the purpose of this research included, 'using discriminate analysis and other models to predict bankruptcy', 'efficiency of the deployment of discriminate analysis and other models to predict bankruptcy', 'advantages and drawbacks of using discriminate analysis and other models to predict bankruptcy.' And 'efficiency of discriminate analysis and other models to predict bankruptcy' (Cooper and Schindler, 1998)

Authenticity and Validity

Appropriate importance was given to the validity and authenticity of the resources that were deployed for the purpose of data collection under the course of this research. Only the sources that were published by authentic publishers were deployed for data collection in this research. In addition to that, if the sources did not give detailed and appropriate information about the publisher then these sources were not include in this research. Only the sources which gave complete, appropriate and authentic information about the publisher of the data content were used for the purpose of data collection under the course of this research. (Cooper and Schindler, 1998)

Apart from that, only the sources, which consisted of the data content that was relevant to the topic under consideration were included in the research. If it was evaluated that the data content of a source did not share a strong relationship with the topic under consideration, using discriminate analysis and other models to predict bankruptcy, then such a source was not used for data collection under the course of this research. (Cooper and Schindler, 1998)

Data Presentation

The data presentation strategy, which was deployed under the research, is discussed in the following section:

This research of a number of critical arguments because of the nature of topic that was explored. The data presentation strategy, therefore, put appropriate emphasis on keeping the data content of this research simple. The complex arguments, which were included in this research, were converted into simpler arguments in order to facilitate the readers so that they may be able to understand data content easily and extract relevant information from it. (Cooper and Schindler, 1998)

Even though the data content was made simple and easy to understand, it was ensured that the original data did not lose its underlying essence. All the major fact in relation to the data, therefore, were included in the research, even if in a simpler and easy to understand manner. (Cooper and Schindler, 1998)

The form in which it was extracted from the primary source is the form in which it was presented in the research. In other words, the data was presented in its original form. The component of data that was extracted first from the primary source was presented first in the research, the element that was extracted after it was included after it and so on. (Cooper and Schindler, 1998)

In addition to that, it was also emphasized by the data presentation strategy that a strong interrelation between various components of data shall be developed. The basic reason behind the development of this interrelation was to establish a pattern in the research so that the readers might be able to grab the information included in the research in an easy and effective manner. (Cooper and Schindler, 1998)

Methods Used for the Prediction of Bankruptcy

A number of models are used for the prediction of bankruptcy. This section of the paper discusses the various models that are used for bankruptcy prediction.

1. Discriminant analysis

Discriminant analysis operates by trying to develop a linear combination that consists of two or more variables and has the ability to discriminate between two priori defined grouped in an effective and efficient manner. This discrimination is achieved through the deployment of a statistical rule that maximizes the variance that exists between the groups relative to the variance that exists within the group. (Karamzadeh, 2013)

The above mentioned relationship is expressed in the form of ratio of between the group and within the group variance. (Back & Laitinen et al., 1996) The linear combination is derived by the discriminate analysis on the basis of an equation that takes the following form:

Z = W1X1+ W2X2+...+WnXn

(Back & Laitinen et al., 1996)

Where,

Z represents the discriminant score

Wi (i=1, 2, ..., n) represents the discriminant weights

Xi (i=1, 2, ..., n ) represents independent variables or the financial ratios (Back & Laitinen et al., 1996)

On the basis of the above mentioned equation each firm gets an individual discriminate scores. This score is then compared to a cut off value. This comparison determines to which group a particular firm belongs. (Back & Laitinen et al., 1996)

If the variables that are included in various groups follow a multivariate normal distribution and all of the groups have equal covariance matrices, then in such conditions a discriminate analysis performs in a very effective and accurate manner. (Back & Laitinen et al., 1996)

It has, however, been determined through empirical testing that generally the failing firms violate the condition of normal distribution. In addition to that, the condition of equal covariance matrices of the group is also often violated. (Back & Laitinen et al., 1996)

In addition to that, if the stepwise procedure is employed then the multicollinearity among the independent variables also acts as one of the major and serious problems. It has, however, been identified through empirical testing that the assumptions in relation to normal distribution of variables were not weakening the classification capability of the model, instead they were weakening the prediction ability of the model. (Back & Laitinen et al., 1996)

The two basic methods that have been deployed for the determination of the discriminate models are:

Simultaneous (Direct) Method: This method is based on construction of models on theoretical grounds. It ensures that the model is ex-ante defined and then deployed in discriminant analysis. (Back & Laitinen et al., 1996)

The Stepwise Method: The stepwise method goes for the construction of the models through the selection of a subset of variables, which ca be used to produce a good discrimination model. These variables are selected through forward selection, backward elimination, or stepwise selection. (Back & Laitinen et al., 1996)

A number of other modelling strategies are also suggested by various professionals, an example of which could be the one suggested by Hosmer and Lemeshow (1989). It has been identified by a number of studies that the major weakness of the stepwise method is that it overlooks the economic importance of the variables and put increased emphasis of the statistical grounds of the variables. (Back & Laitinen et al., 1996)

The modelling strategy that is suggested by Hosmer and Lemeshow, however, is less mechanical in nature. This is because it allows the opinions of the analysts, in relation to the variables, to be included in the model. (Back & Laitinen et al., 1996)

2. Logit analysis

Logistic regression analysis has also been used to determine the relationship that exists between binary or ordinal response probability variables and explanatory variables. (Klobucnik & Sievers, 2013)

By the method of maximum likelihood, the Logit analysis fits the method for logistic regression between binary or ordinal response probability variables and explanatory variables. Ohlson (1980) was among one of the first users of Logit analysis in financial situations. (Klobucnik & Sievers, 2013)

Similar to the discriminate analysis, this method also weights various independent variables and assign a Z score, which represents the probability of failure, to each of the organization that is included in a sample. (Klobucnik & Sievers, 2013)

The basic advantage of this method is that, unlike discriminate analysis, this method does not assume that all the normality of distribution and equality of covariance matrices among the groups. (Klobucnik & Sievers, 2013) In order to predict predicting a bankruptcy nonlinear effects are incorporates and cumulative logistic functions are used by this method, i.e.,

(Klobucnik & Sievers, 2013)

Logistic regression is restricted to the prediction of discrete sets, unlike linear regression. Therefore, the dependent variables that are included in logistic regression are restricted. These variables are restricted to a discrete number set. (Klobucnik & Sievers, 2013)

Another major disadvantage of this method is the underlying assumption of linearity between the variables that is deployed by the method. Furthermore, this method restricted to only between subject designs and does not address the within subject designs. (Klobucnik & Sievers, 2013)

3. Neural networks

A large number of processing elements, which are known as neurons, and connections between them are two important components of an artificial neural network. A function is implemented by this method. Through this function a set of input values, represented as x, are mapped to a set of output values, represented as y. The function, therefore, is represented as y = f (x). (Makeeva & Neretina et al., 2012)

The neural network aims at find out the best approximation of the function. Through the deployment of the weights, which are associated with each of the neurons, the above mention approximation is coded into the neurons of the network. (Makeeva & Neretina et al., 2012)

A genetic algorithm is another method to select the variables that would be a part of the network. The Darwinian evolution is simulated by a genetic algorithm. A population of chromosomes is maintained by a genetic algorithm. Here, chromosome can be defined as the solution of the problem that we wish to solve. (Makeeva & Neretina et al., 2012)