CLASSIFICATION BASED CREDIT RISK ANALYSIS THE CASE OF LENDING CLUB AADIGUPTA PRIYA GULATISIDDHARTHA P. C HAKRABARTY Abstract

2025-04-27 0 0 211KB 9 页 10玖币

侵权投诉

CLASSIFICATION BASED CREDIT RISK ANALYSIS: THE CASE OF LENDING CLUB

AADI GUPTA *PRIYA GULATI †SIDDHARTHA P. CHAKRABARTY ‡

Abstract

In this paper, we performs a credit risk analysis, on the data of past loan applicants of a company named

Lending Club. The calculation required the use of exploratory data analysis and machine learning classiﬁca-

tion algorithms, namely, Logistic Regression and Random Forest Algorithm. We further used the calculated

probability of default to design a credit derivative based on the idea of a Credit Default Swap, to hedge against

an event of default. The results on the test set are presented using various performance measures.

Keywords: Credit risk; Classiﬁcation algorithm; Exploratory data analysis

1 INTRODUCTION

Lending Club, headquartered in San Francisco, was the ﬁrst peer-to-peer lending institution to offer its secu-

rities through the Securities and Exchange Commission (SEC) and enter the secondary market [1]. This article is

essentially a case study, on how ﬁnancial engineering problems can be addressed using Machine Learning (ML)

and Exploratory Data Analysis (EDA) approaches. Lending Club specializes in extending different types of loans

to urban customers, which is decided on the basis of the applicant’s proﬁle. Accordingly, the data considered in

this work, contains information about past loan applications and whether they were “defaulted” or “not”. One of

the main objectives of this work is the calculation of the Probability of Default (PD) and (in case of a default)

the determination of Loss Given Default (LGD), Exposure at Default (EAD), and ﬁnally, the Expected Loss (EL),

making use of the historical data. Also, using the “Recovery Rate” (estimated while calculating “LGD” from

“EAD”) and the PD, a simple credit derivative is implemented, based on the concept of Credit Default Swaps

(CDS) which can be used to hedge against such defaults. It may be noted that the total value a lender is exposed

to when a loan defaults, is the EAD and the consequent unrecoverable amount for the lender is the LGD [2].

Accordingly, EL is deﬁned as,

EL =EAD ×LGD ×PD.

The driver of this work is the notion of “Classiﬁcation Algorithm”, which weighs the input data (of the applicant,

in this case) to classify the input features into positive and negative classes [3] (default on the loan or not, in

this case). Accordingly, we consider two “Classiﬁcation Algorithms”, namely, the Logistic Regression and the

Random Forest.

The main idea behind Logistic Regression is to determine the probability of a particular data point belonging to

the positive class (in the case of binary classiﬁcation) [4]. The model does so by establishing a “linear relationship”

between the independent and the dependent variables. The weights of the linear relations is determined through

*Department of Mathematics, Indian Institute of Technology Guwahati, Guwahati-781039, India, e-mail: aadi18@alumni.iitg.ac.in

†Department of Mathematics, Indian Institute of Technology Guwahati, Guwahati-781039, India, e-mail: gulati18@alumni.iitg.ac.in

‡Department of Mathematics, Indian Institute of Technology Guwahati, Guwahati-781039, India, e-mail: pratim@iitg.ac.in

arXiv:2210.05136v1 [q-fin.RM] 11 Oct 2022

the minimization of a Loss Function, which is achieved by using the “Gradient Descent Optimization Algorithm”.

Since Logistic Regression ﬁrst predicts the probability of belonging to the positive class, therefore it creates a

linear decision boundary (based on a threshold, set by the user), separating the two classes from one another. This

decision boundary can now be represented as a conditional probability [5]. Implementation of the Random Forest

algorithm involves the training stage construction of several decision trees [6], and predictions emanating from

these trees are averaged to arrive at a ﬁnal prediction. Since the algorithm uses an average of results to make

the ﬁnal prediction, the Random Forest algorithm is referred to as an ensemble technique. Decision Trees are

designed to optimally split the considered dataset into smaller and smaller subsets, in order to predict the value

being targeted [7, 8]. Some of the criterion used to calculate the purity or impurity of a node, include Entropy

and Gini Impurity. In summary, a decision tree splits the nodes on all the attributes present in the data, and then

chooses the split with the most Information Gain, with the Decision Tree model of Classiﬁcation and Regression

Trees (CART), being used in this paper.

The ﬁrm based model uses the value of a ﬁrm to represent the event of default, with the default event being

represented by the boundary conditions of the process and the dynamics of the ﬁrm value [9]. In particular, we

refer to two well-established models. Firstly, we mention the Merton model based on the the seminal paper of

Black and Scholes [10], which is used to calculate the default probability of a reference entity. In the context,

the joint density function for the ﬁrst hitting times is determined [11]. Secondly, we have the Black-Cox model,

which addresses some of the disadvantages of the Merton model [10]. In order to hedge against credit risk, the

usage of credit derivatives is a customary approach [12, 13, 14], with Credit Default Swaps (CDS) being the most

common choice of credit derivatives. This type of contracts entail the buyer of the CDS to transfer the credit risk

of a reference entity (“Loans” in this case constitute the reference entity) to the seller of the protection, until the

credit has been settled. In return, the protection buyer pays premiums (predetermined payments) to the protection

seller, which continues until the maturity of the CDS or a default, whichever is earlier [10]. The interested reader

may refer to [10] for the formula of CDS spread per annum, to be used later in this paper. Another widely used

credit derivative, albeit more sophisticated than the CDS are the Collateralized Debt Obligation (CDO), which is

a structured product, based on tranches [14]. CDOs can further classiﬁed into cash, synthetic and hybrid. The

interested reader may refer to [14] for a detailed presentation on pricing of synthetic CDOs.

2 METHODOLOGY

The goal of this exercise is the approximation of a classiﬁcation model, on the data considered (from the

peer-to-peer lending company Lending Club), in order to predict as to whether an applicant (whose details are

contained in the considered database) is likely to default on the loan or not. Accordingly, to this end, it is necessary

to identify and understand the essential variables, and take into account the summary statistics, in conjunction

with data visualization. The dataset used for the estimation of PD, EAD, LGD and EL was obtained from Kaggle

[15]. The data contains details of all the applicants who had applied for a loan at Lending Club. There were

separate ﬁles obtained for the accepted and the rejected loans. The ﬁle “accepted loans” was only used, since the

observations were made on the applicants who ultimately paid the loan, and those who defaulted on the loan. The

data involved the details of the applicants for the loan, such as FICO score, loan amount, interest rate, purpose of

loan etc. Machine Learning (ML) algorithms were applied on this data to predict the PD after data exploration,

data pre-processing and feature cleaning.

The feature selection was performed on the considered data, bearing in mind the goals of predictive modelling.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

CLASSIFICATIONBASEDCREDITRISKANALYSIS:THECASEOFLENDINGCLUBAADIGUPTA*PRIYAGULATISIDDHARTHAP.CHAKRABARTYAbstractInthispaper,weperformsacreditriskanalysis,onthedataofpastloanapplicantsofacompanynamedLendingClub.Thecalculationrequiredtheuseofexploratorydataanalysisandmachinelearningclassica-tionalgor...

展开>> 收起<<

CLASSIFICATION BASED CREDIT RISK ANALYSIS THE CASE OF LENDING CLUB AADIGUPTA PRIYA GULATISIDDHARTHA P. C HAKRABARTY Abstract.pdf

共9页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

CLASSIFICATION BASED CREDIT RISK ANALYSIS THE CASE OF LENDING CLUB AADIGUPTA PRIYA GULATISIDDHARTHA P. C HAKRABARTY Abstract

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: