“Examining determinants of loan default: An empirical analysis on credit factors

in Thai savings and credit cooperatives”

Klangjai Sangwichitr
AUTHORS
Panern Intara

Klangjai Sangwichitr and Panern Intara (2024). Examining determinants of loan

default: An empirical analysis on credit factors in Thai savings and credit
ARTICLE INFO
cooperatives. Investment Management and Financial Innovations, 21(4), 323-
332. doi:10.21511/imfi.21(4).2024.26

DOI http://dx.doi.org/10.21511/imfi.21(4).2024.26

RELEASED ON Monday, 25 November 2024

RECEIVED ON Tuesday, 17 September 2024

ACCEPTED ON Thursday, 14 November 2024

LICENSE This work is licensed under a Creative Commons Attribution 4.0 International
License

JOURNAL "Investment Management and Financial Innovations"

ISSN PRINT 1810-4967

ISSN ONLINE 1812-9358

PUBLISHER LLC “Consulting Publishing Company “Business Perspectives”

FOUNDER LLC “Consulting Publishing Company “Business Perspectives”

NUMBER OF REFERENCES NUMBER OF FIGURES NUMBER OF TABLES

38 0 4

© The author(s) 2024. This publication is an open access article.

businessperspectives.org
 Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

Klangjai Sangwichitr (Thailand), Panern Intara (Thailand)

Examining determinants
BUSINESS PERSPECTIVES of loan default: An empirical
LLC “СPС “Business Perspectives”
Hryhorii Skovoroda lane, 10,
Sumy, 40022, Ukraine
analysis on credit factors
www.businessperspectives.org in Thai savings and credit
cooperatives
Abstract
Savings and credit cooperatives (SACCOs) are crucial institutions in promoting finan-
cial accessibility. SACCOs provide financial loans to individuals who may not have
access to traditional banking. SACCOs take their own risk to get loan defaults from the
offerings because member loans are approved without checking the members’ credit
background by SACCO committees. This study aims to investigate factors influenc-
ing loan defaults of savings and credit cooperatives in Thailand. Based on the savings
and credits cooperative database in November 2023, the cooperative has emergency
Received on: 17th of September, 2024 loans, regular loans, and special loans totaling 11,441 contracts. In this study, all loan
Accepted on: 14th of November, 2024 contracts of this cooperative were used to analyze. The data were divided into two
Published on: 25th of November, 2024 categories of debt classification, including (1) non-default status and (2) default status.
The data were analyzed using logistics regression to select the highest accuracy model.
© Klangjai Sangwichitr, Panern Intara,
Furthermore, the finding reveals that the highest accuracy model, at 99.78%, contains
2024 five variables, including interest rate, collateral value, remaining contract duration,
outstanding debt, and installment amount. The savings and credit cooperatives institu-
tion should adjust the loan interest rates according to economic conditions. Moreover,
Klangjai Sangwichitr, D.B.A., closely monitoring members with high remaining debt would help the institution pre-
Assistant Professor of Business vent loan defaults, and the institution should also create a conservative loan approval
Administration, Department of policy to reduce its loan default.
Business Administration, Faculty
of Management Sciences, Prince of
Songkla University, Thailand. Keywords loan default, savings and credit cooperatives, financial
institution, logistic regression, Thailand
Panern Intara, M.B.A., Department
of Business Administration, Faculty JEL Classification G17, G23, G51
of Management Sciences, Prince
of Songkla University, Thailand.
(Corresponding author)
INTRODUCTION
Savings and credit cooperatives (SACCOs) provide financial ser-
vices to individuals who may not have access to traditional bank-
ing (Assawawongsathien et al., 2017), support economic expan-
sion (Namwong & Janyasuprab, 2018; Sfar & Ben Ouda, 2016), and
enhance the financial welfare of their members and communities
(Keawmanee, 2016). Changes in interest rate policy announced by the
Bank of Thailand (BOT) directly impact the investment opportunities
of savings and credit cooperatives in several ways, such as higher bor-
rowing costs and pressure on loan demand. Higher interest rates may
reduce the demand for loans from cooperative members, potentially
impacting the cooperatives’ revenues and growth prospects. Although
This is an Open Access article, large SACCOs have various investment opportunities that can gener-
distributed under the terms of the ate reasonable returns, it cannot be denied that lending to its members
Creative Commons Attribution 4.0
International license, which permits for various purposes remains a significant source of revenue and profit
unrestricted re-use, distribution, and for the cooperative. Due to the significant roles of SACCOs for their
reproduction in any medium, provided
the original work is properly cited. members, communities, and the economy, any financial instability ex-
Conflict of interest statement: perienced by SACCOs would not only disrupt their functioning but
Author(s) reported no conflict of interest also have consequences for the community, economy, and all mem-

http://dx.doi.org/10.21511/imfi.21(4).2024.26 323
Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

bers’ financial stability. The default risk of SACCOs refers to the probability that borrowers within the
cooperative will fail to repay their loans as agreed. The financial health of SACCOs is heavily dependent
on the repayment behavior of their members (McKillop et al., 2020). High default rates may gradually
weaken a cooperative’s financial stability, resulting in financial losses that impact all members.

1. LITERATURE REVIEW Department, 2021). The SACCOs being discussed

AND HYPOTHESES have a large internal cash surplus (The Savings
and Credits Cooperatives, 2023). This cash flow
Understanding the factors influencing member presents both opportunities and challenges for the
debt repayment behaviors is crucial for develop- cooperative. If the cooperative cannot find suffi-
ing effective credit analysis strategies, thereby re- cient investment opportunities to cover its cost of
ducing NPL risks. Implementing credit scoring as capital – including the cost of debt (interest pay-
a statistical tool for credit analysis can enhance ments) and the cost of equity (dividends) – or if it
loan consideration efficiency and help mitigate takes an overly conservative approach to invest-
long-term NPLs for the cooperative. ment or credit analysis, its revenue generation and
long-term profitability might be adversely affected.
In the 19th century, the first saving and credit coop- Conversely, if the cooperative adopts an aggressive
eratives were established in Germany by Herman investment strategy, it might increase revenue but
Schultze-Delitsche and Freidrich Reifeisen. Then also face higher risks amid rising interest rates, po-
Luigi adopted the principles of the German saving tentially leading to an increase in non-performing
and credit cooperatives and established the first loans (NPLs). The Civil and Commercial Code
saving and credit cooperatives in Italy. After that B.E. 2535 defined loan defaults as any contract
saving and credit cooperatives principles were owner who fails to make their payment by the due
well known and were established by labor unions date. On the other hand, Cooperative Auditing
and organizations around the world (Galor, 1999). Department (2016) defines loan default as any con-
In Thailand, SACCOs were established to promote tact owner who fails to pay the principal or interest
savings and loans to a group of people who are in within 1-60 days from the date. It means that if the
the same organization (Cooperative Promotion debtor fails to make a payment for more than 60
Department, 2021). In SACCOs, funds are raised days, the contract would be defined as loan default.
from owners’ capital, and the owners get divi- The prediction model can be used as a tool promot-
dends in return every year based on the cooper- ing a warning system and ensuring an organiza-
atives’ performance. The SACCOs’ uses of funds tion’s financial stability (Hammond et al., 2023).
come from their investment and loan interest
payments, which would create their performance Credit Scoring is a statistical analysis of a bor-
(Galor, 1999). The SACCOs with good structure rower’s historical debt repayment information
and financial capital may have better perfor- and behavior. Financial institutions use credit
mance (Laliwan & Potipiroon, 2022). However, scoring to perform a predictive model assessing
board committees would be the key people who the likelihood of repaying a loan for a new loan
make short- and long-term regulations (Kaplan & application. This analysis evaluates the current
Mccay, 2004). The SACCO provides credit to its situation of the loan applicant, including per-
members to create income by receiving full repay- sonal data, financial data, and historical repay-
ment and interest payments. ment records (Onay & Öztürk, 2018; Petrides et
al., 2022; Van Gool et al., 2012). Credit scoring
SACCOs were established to help their members plays a critical role in finance and banking as it
promote savings and to provide loans to members allows lenders to have a better understanding of
when needed (Galor, 1999; Omeke et al., 2019; borrowers by predicting the likelihood of future
Laliwan & Potipiroon, 2022). In Thailand, there loan repayments. Moreover, this score reflects
are 3 loan categories offered by savings and credit the economic credibility of an individual and is
cooperatives, including emergency loans, general used as part of the loan approval process. Thus,
loans, and special loans (Cooperative Promotion this scoring is a risk management tool used to

324 http://dx.doi.org/10.21511/imfi.21(4).2024.26
 Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

evaluate the credibility of borrowers (Boughaci H1: The loan interest rate impacts the repayment
& Alkhawaldeh, 2019). It is important to find fac- status.
tors for model predictions used as credit scoring.
H2: The collateral value impacts the repayment
There is much research focusing on seeking credit status.
factors influencing loan defaults (Steenackers &
Goovaerts, 1989; Nguyen, 2015; Bravo et al., 2015; H3: The approved loan amount impacts the re-
Barua et al., 2021; Saha et al., 2023; Thomas et al., payment status.
2023; Abdou et al., 2017; Lokesha & Hawaldar,
2019). The results reveal that borrower characteris- H4: The remaining contract duration impacts the
tics, the down payment rate, credit term, remain- repayment status.
ing contract term, interest rate, and approved loan
amount impact the probability of loan default. In H5: The outstanding debt impacts the repayment
Thailand, Pattarapornpairot (2020) studied the status.
variables influencing loan default prediction, in-
cluding the approved loan amount, remaining H6: The monthly installment impacts the repay-
contract term, interest rate, outstanding debt, and ment status.
collateral. Similarly, Notanonda (2000) investigat-
ed factors leading to default on residential loans
at a commercial bank in Chiang Mai, finding that 2. RESEARCH METHODOLOGY
income and the borrower’s age were key predictors,
aligning with Chushim’s (2013) study on factors af- Based on the literature review, this study’s inde-
fecting non-performing loans (NPLs) for housing pendent variables are loan interest rate, collateral
loans, which highlighted age, income, debt-to-in- value, the approved loan amount, remaining con-
come ratio, loan-to-value ratio, and monthly in- tract duration, outstanding debt, and monthly
stallments as significant factors. installment. The dependent variable is the repay-
ment status, which represents the loan defaults.
In addition, there are many methods in order to Logistic regression analysis is used to analyze data.
create prediction models (Zhu, 2019; Zhu et al.,
2023; Nguyen, 2015; Sariannidis et al., 2023) such This study collected data from the database of a
as Decision tree, Random forest, Neural networks, large savings and credit cooperative whose assets
K nearest Neighbor, Logistic regression, XGBoost, exceeded 30,000 million Thai baht in November
and LightGBM. Ali Albastaki (2022) proved that 2023. The analysis utilized all contracts from this
Logistic Regression is the best model for a loan cooperative, totaling 11,441 contracts. The data
default prediction system. Research constructing were divided into two categories of debt classifica-
loan default prediction models typically utilizes tion: (1) non-default status and (2) default status.
Logistic Regression statistics as in commercial
bank, student loan, and public mortgage loans; the This study divided the data into two parts: part
results reveal that factors affecting loan repayment (1) 80% proportion, totaling 9,163 contracts, was
are approved loan amount, remaining contract selected using accidental sampling. This includes
term, collateral, interest rate, and outstanding 8,973 contracts with non-default status and 190
debt as influential (Thiansomboon, 2017; Imsuk, contracts with default status. Part (2) 20% propor-
2018; Kamphod, 2019; Maichandrang, 2021) In tion, totaling 2,287 contracts, was also selected us-
other countries, mortgage loan default can be pre- ing accidental sampling. This includes 2,222 con-
dicted from interest rate and monthly payments tracts with non-default status and 56 contracts
(ANTAR, 2024). with default status. Data were randomly selected
from debtors across the two categories of debt
The purpose of the study is to seek factors influ- classification. Part (1) was used for model cre-
encing loan defaults among the members of the ation by logistic regression analysis, while part (2)
savings and credit cooperatives in Thailand. The was used to validate the model’s accuracy. Table
research hypotheses are formulated as follows: 1 shows the proportion of 80% and 20% samples

http://dx.doi.org/10.21511/imfi.21(4).2024.26 325
Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

categorized by research variables, including the Model 2 uses five variables, excluding the ap-
approved loan interest rate, collateral value, ap- proved loan amount (Vol):
proved loan amount, remaining contract duration,
outstanding debt, and monthly installment. Y β 0 + β1 Int + β 2Col + β3Term
=
(2)
+ β 4 Loan + β5 Pay.
The first part of the study analyzes the factors af-
fecting loan default among the savings and credit Model 3 uses four variables, excluding the ap-
cooperatives members, using data from November proved loan amount (Vol) and collateral value
2023, representing 80% of the sample. The analy- (Col):
sis is conducted using logistic regression models
=
with six variables: loan interest rate (Int), collat- Y β 0 + β1 Int + β 2Term + β 3 Loan + β 4 Pay. (3)
eral value (Col), approved loan amount (Vol), re-
maining contract duration (Term), outstanding Model 4 uses three variables, excluding the ap-
debt (Loan), and monthly installment (Pay), with proved loan amount (Vol), collateral value (Col),
loan status (Y) as the dependent variable. Six dif- and outstanding debt (Loan):
ferent models are analyzed:
Y = β 0 + β1 Int + β 2Term + β3 Pay. (4)
Model 1 uses all six variables:
Model 5 uses three variables, excluding the ap-
Y = β 0 + β1 Int + β 2Col + β3Vol + β 4Term (1) proved loan amount (Vol), collateral value (Col),
+ β5 Loan + β 6 Pay. and monthly installment (Pay):

Table 1. Sample loan data from the savings and credit cooperatives, divided by 80% and 20%
proportions, by variable
Proportion of 80% Proportion of 20%
Non-default Default Non-default Default
Variables Frequency Frequency Frequency
Frequency
(total Proportion (total Proportion (total of Proportion Proportion
(total of 56)
of 8,973) of 190) 2,222)
Loan interest rate (Int)
3.95%-10.47% 8,971 0.9998 170 0.8947 2,222 1.0000 47 0.8393
10.48%-17% 2 0.0002 20 0.1053 0 0.0000 9 0.1607
Collateral (Col)
Less than 1,000,000 7,431 0.8282 134 0.7053 1,848 0.8317 36 0.6429
1,000,000 – 5,000,000 1,482 0.1652 51 0.2684 361 0.1625 18 0.3214
More than 5,000,000 60 0.0067 5 0.0263 13 0.0059 2 0.0357
Approved loan amount (Vol)
Less than 1,000,000 6,790 0.7567 125 0.6579 1,698 0.7642 34 0.6071
1,000,000 – 5,000,000 2,122 0.2365 60 0.3158 510 0.2295 20 0.3571
More than 5,000,000 61 0.0068 5 0.0263 14 0.0063 2 0.0357
Remaining contract duration (Term)
Less than 1,000,000 3,881 0.4325 73 0.3842 949 0.4271 16 0.2857
1,000,000 – 5,000,000 1,852 0.2064 68 0.3579 456 0.2052 22 0.3929
More than 5,000,000 3,240 0.3611 49 0.2579 817 0.3677 18 0.3214
Outstanding debt (Loan)
Less than 1,000,000 7,761 0.8649 151 0.7947 1,928 0.8677 42 0.7500
1,000,000 – 5,000,000 1,185 0.1321 35 0.1842 290 0.1305 12 0.2143
More than 5,000,000 27 0.0030 4 0.0211 4 0.0018 2 0.0357
Monthly installment (Pay)
Less than 10,000 7,451 0.8304 154 0.8105 1,831 0.2041 44 0.7857
10,001 – 25,000 1,294 0.1442 28 0.1474 346 0.0386 9 0.1607
25,001 – 40,000 172 0.0192 4 0.0211 35 0.0039 1 0.0179
More than 40,000 56 0.0062 4 0.0211 10 0.0011 2 0.0357

326 http://dx.doi.org/10.21511/imfi.21(4).2024.26
 Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

=Y β 0 + β1 Int + β 2Term + β3 Loan. (5) lation coefficients of the variables are presented
in Table 2.
Model 6 uses five variables, excluding the monthly
installment (Pay): Table 2, displayed the correlation coefficients be-
tween the variables, shows the following signifi-
Y = β 0 + β1 Int + β 2Col + β3Vol (6) cant relationships: collateral value (Col) is posi-
+ β 4Term + β5 Loan. tively correlated with the approved loan amount
(Vol) (r = .798**), outstanding debt (Loan) (r =
.632**), and installment amount (Pay) (r = .629**).
3. RESULTS Moreover, the approved loan amount (Vol) is posi-
tively correlated with the outstanding debt (Loan)
The correlation coefficients between the six vari- (r = .713**) and installment amount (Pay) (r =
ables are used to observe the direction of the re- .677**). Finally, the remaining contract duration
lationship between the two variables. The corre- (Term) is positively correlated with the outstand-
Table 2. Correlation coefficients of predictive variables
Variable Correlation and P-value Int Col Vol Term Loan Pay
Pearson Correlation 1 .006 –.003 –.014 .006 .012
Int
Sig. (2-tailed) – .594 .792 .193 .581 .266
Pearson Correlation .006 1 .798** .005 .632** .629**
Col
Sig. (2-tailed) .594 – 0.000 .647 0.000 0.000
Pearson Correlation –.003 .798** 1 –.007 .713** .677**
Vol
Sig. (2-tailed) .792 0.000 – .512 0.000 0.000
Pearson Correlation –.014 .005 –.007 1 .022* –.101**
Term
Sig. (2-tailed) .193 .647 .512 – .037 .000
Pearson Correlation .006 .632** .713** .022* 1 .659**
Loan
Sig. (2-tailed) .581 0.000 0.000 .037 – 0.000
Pearson Correlation .012 .629** .677** –.101** .659** 1
Pay
Sig. (2-tailed) .266 0.000 0.000 .000 0.000 –

Notes: ** Correlation is significant at the 0.01 level (2-tailed), and * Correlation is significant at the 0.05 level (2-tailed).

Table 3. Results of logistic regression analysis of variables leading to loan default

Coefficient
Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
and S.E.
Coefficient –11.074 –11.065 –10.777 –10.347 –10.845 –11.109
Constant
S.E. 0.823 0.822 0.816 0.804 .816 0.824
Coefficient 6.371** 6.365** 6.306** 6.252** 6.283** 6.313**
Int S.E. 0.75 0.75 0.748 0.746 .748 0.749
Exp(B) 584.778 581.422 547.952 518.904 535.913 551.733
Coefficient 0.885** 0.934** – – – 0.743*
Col S.E. 0.316 0.222 – – – 0.314
Exp(B) 2.422 2.545 – – – 2.103
Coefficient 0.073 – – – – –0.085
Vol S.E. 0.33 – – – – 0.329
Exp(B) 1.075 – – – – 0.918
Coefficient –0.097 –0.097 –0.073 –0.032 –.050 –0.048
Term S.E. 0.093 0.093 0.091 0.088 .090 0.091
Exp(B) 0.907 0.907 0.93 0.969 .951 0.953
Coefficient 0.382 0.401 0.865** – – 0.089
Loan S.E. 0.278 0.267 0.254 – – 0.259
Exp(B) 1.465 1.493 2.376 – – 1.093
Coefficient –0.596** –0.59** –0.31 0.156 0.584** –
Pay S.E. 0.215 0.213 0.211 0.142 .176 –
Exp(B) 0.551 0.554 0.734 1.169 1.795 –
Nagelkerke R Square – 0.102 0.102 0.092 0.086 .091 0.097

http://dx.doi.org/10.21511/imfi.21(4).2024.26 327
Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

ing debt burden (Loan) (r = .022**) but negatively Table 3 reveals that at a significance level of 0.01,
correlated with the installment amount (Pay) (r = the interest rate (Int) significantly affects loan
-.101**), with a significance level of 0.01. default probability across all models, with coeffi-
cients of 6.371, 6.365, 6.306, 6.252, 6.283, and 6.313,
Base on Table 3, the models are formulated as respectively, and odds ratios of 584.778, 581.422,
follows: 547.952, 518.904, 535.913, and 551.733, respective-
ly. The coefficient and odds ratio of Model 1 can
Model 1 uses all six variables: be explained as a coefficient of 6.371; this means
that a 1% increase in the interest rate increas-
Y= −11.074 + 6.371Int + 0.885Col
(7) es the probability of loan default by 6.371 times.
+0.073Vol − 0.097Term + 0.382 Loan Moreover, collateral value (Col) is significant at
−0.596 Pay. the level of 0.01 in Models 1, 2, and 6 with coef-
ficients and odds ratios: (1) Model 1: coefficient =
Model 2 uses five variables, excluding the ap- 0.885, odds ratio = 2.422; (2) Model 2: coefficient =
proved loan amount (Vol): 0.934, odds ratio = 2.545; and (3) Model 6: coeffi-
cient = 0.743, odds ratio = 2.103. For Model 1, a co-
Y= −11.065 + 6.365 Int + 0.934Col (8) efficient of 0.885 means that a 1,000 THB increase
−0.097Term + 0.401Loan − 0.59 Pay. in collateral value increases the probability of loan
default by 0.885 times. Outstanding debt (Loan)
Model 3 uses four variables, excluding the ap- is significant at the level of 0.01 in Model 3 with
proved loan amount (Vol) and collateral value a coefficient of 0.865 and an odds ratio of 2.375.
(Col): Installment amount (Pay) is significant at the lev-
el of 0.01 in models 1, 2, and 5 with coefficients
Y= −10.777 + 6.306 Int − 0.073Term (9) and odds ratios: (1) Model 1: coefficient = –0.596,
+0.865 Loan − 0.31Pay. odds ratio = 0.551; (2) Model 2: coefficient = –0.59,
odds ratio = 0.554; and (3) Model 5: coefficient =
Model 4 uses three variables, excluding the ap- 0.584, odds ratio = 1.795. For Model 1, a coefficient
proved loan amount (Vol), collateral value (Col), of –0.596 means that a 1,000 THB increase in the
and outstanding debt (Loan): installment amount decreases the probability of
loan default by 0.596 times.
Y= −10.347 + 6.252 Int − 0.032Term (10)
+0.156 Pay. To validate the model’s accuracy, the second part
of the study predicts loan default using 20% of the
Model 5 uses three variables, excluding the ap- sample data, with 2,278 contracts. The prediction
proved loan amount (Vol), collateral value (Col), uses the logistic regression models to calculate Y
and monthly installment (Pay): for each contract and determine the probability of
default. Based on model 1, the model’s prediction
Y= −10.845 + 6.283Int − 0.50Term (11) accuracy is 2.38% when substituting the value into
+0.584 Loan. the probability equation (Boateng & Abaye, 2019;
Thiansomboon, 2017) as follows.
Model 6 uses five variables, excluding the monthly
installment (Pay): The result reveals that there were 2,222 non-de-
fault contracts; the equation correctly predicted
Y= −11.109 + 6.313Int + 0.743Col (12) 53 contracts. For the loan defaults of 56 contracts,
−0.085Vol − 0.048Term + 0.089 Loan. the equation correctly predicted 1 contract. This
means that a total of 2,278 contracts were used to

e −11.074 + 6.371Int + 0.885Col + 0.073Vol − 0.097Term + 0.382 Loan −0.596 Pay

P( loan defaults ) = . (13)
1 + e −11.074 + 6.371Int + 0.885Col + 0.073Vol − 0.097Term + 0.382 Loan −0.596 Pay

328 http://dx.doi.org/10.21511/imfi.21(4).2024.26
 Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

Table 4. Result validation

Non-default Default
Model Correct Incorrect Correct Incorrect
Frequency Proportion Frequency Proportion Frequency Proportion Frequency Proportion
Model 1 53 0.0238 2,169 0.9762 1 0.0238 55 0.9762
Model 2 2,217 0.9978 5 0.0022 56 0.9978 0 0.0022
Model 3 1,841 0.8285 381 0.1715 46 0.8285 10 0.1715
Model 4 1,684 0.7581 538 0.2419 42 0.7581 14 0.2419
Model 5 967 0.4350 1,255 0.5650 24 0.4350 32 0.5650
Model 6 1,933 0.8700 289 0.1300 49 0.8700 7 0.1300

validate the model; Model 1 correctly predicted 54 4. DISCUSSION

contracts, as shown in Table 4.
This study’s logistics regression result reveals that
According to the validation method used in Model loan default can be predicted by interest rate, col-
1, the equation correctly predicted 2,217 non-de- lateral value, remaining contract duration, out-
fault accounts and 56 default accounts. Out of the standing debt, and monthly installment. The re-
2,278 accounts used for validation, the model cor- sult aligns with the study by Tiensomboon (2017)
rectly predicted 2,273 accounts, resulting in an ac- on factors affecting housing loan defaults, indi-
curacy of 99.78% in Model 2. The result confirms cating that the significant factors are consistent
that H3 is not a predictor for the loan default. across different types of loan analyses. Moreover,
the variables used as loan default predictors align
For Model 3, the equation correctly predicted with many previous studies, such as Barua et al.
1,841 non-default accounts and 46 default ac- (2021), Omek et al. (2023), Saha et al. (2023), and
counts. Out of the 2,278 accounts used for valida- Thomas et al. (2023), which all concluded that
tion, the model correctly predicted 1,887 accounts, credit term, interest rate, outstanding debt, im-
resulting in an accuracy of 82.85%. The equation pact significantly to the probability of loan de-
correctly predicted 1,684 non-default accounts fault. One possible explanation for the findings’
and 42 default accounts. Out of the 2,278 accounts consistency with many previous studies is that,
used for validation, the model correctly predicted despite the fact that savings and credit coopera-
1,727 accounts, resulting in an accuracy of 75.81% tives differ from commercial banks in that they
in Model 4. exclusively offer their loan to members, the credit
analysis procedures of these financial institutions
For Model 5, the equation correctly predicted 967 should be identical to those of commercial banks.
non-default accounts and 24 default accounts. Out Crucial factors such as interest rate, outstanding
of the 2,278 accounts used for validation, the mod- debt, credit term all need to be carefully consid-
el correctly predicted 991 accounts, resulting in an ered. The result on interest rate can be a predic-
accuracy of 43.50%. tor for loan default. A loan contract with a higher
interest rate would have a higher monthly loan
Finally, the equation correctly predicted 1,933 payment than a contract with a lower interest rate,
non-default accounts and 49 default accounts. Out reflecting the higher opportunity of loan default.
of the 2,278 accounts used for validation, the mod- If a borrower has liquidity issues, he or she will
el correctly predicted 1,982 accounts, resulting in not be able to repay the loan. Moreover, high col-
an accuracy of 87.00% in Model 6. lateral value and high outstanding debt balance
will cause loan default since it would be difficult
Thus, Model 2 provides the highest accuracy at for a borrower to make a repayment in case of in-
99.78%, suggesting it is the most accurate for pre- sufficient liquidity. On the other hand, a shorter
dicting loan default with the five variables: interest contract duration makes a lower risk of loan de-
rate (H1), collateral value (H2), remaining contract fault because the outstanding debt is not as high
duration (H4), remaining debt (H5), and monthly as when the contract begins. Moreover, the higher
installment (H6). monthly installments reflect the borrower’s good

http://dx.doi.org/10.21511/imfi.21(4).2024.26 329
Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

financial stability, then the likelihood of loan de- database. Second, this study implies only the
fault would be avoided. Logistics Regression method to create a predic-
tion model. Future studies should consider time
The limitations of this study are that, first, it series data analysis and compare several models
is difficult to collect time series data from the to confirm predictions.

CONCLUSION
The study aims to investigate factors influencing loan defaults among the members of the savings and cred-
it cooperatives in Thailand. The finding reveals that the highest accuracy factors are interest rate, collateral
value, remaining contract duration, outstanding debt, and installment amount. The savings and credits
cooperative institution should adjust the loan interest rates according to economic conditions. Moreover,
closely monitoring members with high remaining debt would help the institution prevent loan defaults,
and the institution should create a conservative loan approval policy to reduce its loan default. Moreover,
the prediction results benefit stakeholders such as the savings and credit cooperatives institution, micro-
finance institutions, and regulators. The savings and credit cooperatives institution should adjust the loan
interest rates according to economic conditions. Moreover, closely monitoring members with high out-
standing debt would help the institution prevent loan defaults. On the other hand, the institution should
create a conservative loan approval policy, such as considering the appropriateness of the loan amount
relative to collateral value, increasing monthly installment for loans without collateral to reduce default
probability, and extending repayment periods for loans with collateral to reduce default probability.

AUTHOR CONTRIBUTIONS
Conceptualization: Klangjai Sangwichitr, Panern Intara.
Data curation: Klangjai Sangwichitr.
Formal analysis: Klangjai Sangwichitr, Panern Intara.
Investigation: Klangjai Sangwichitr, Panern Intara.
Methodology: Klangjai Sangwichitr, Panern Intara.
Project administration: Panern Intara.
Supervision: Klangjai Sangwichitr, Panern Intara.
Validation: Panern Intara.
Visualization: Klangjai Sangwichitr.
Writing – original draft: Klangjai Sangwichitr, Panern Intara.
Writing – review & editing: Klangjai Sangwichitr, Panern Intara.

ACKNOWLEDGMENTS
The research for the work featured in this article is funded by the Prince of Songkla Savings and Credit
Cooperatives, Limited.

REFERENCES
1. Abdou, H., Abdallah, W., Mul- 2. Ali Albastaki, A. (2022). Loan tics, Loan Specific Parameters, and
keen, J., Ntim, C., & Wang, Y. Default Prediction System. Thesis. Property Conditions on Mortgage
(2017). Prediction of financial Rochester Institute of Technology. Default Risk. Theoretical and
strength ratings using machine Retrieved from https://repository. Practical Research in Economic
learning and conventional tech- rit.edu/cgi/viewcontent.cgi?article Fields, 15(2), 481-501. https://doi.
niques. Investment Management =12544&context=theses org/10.14505/tpref.v15.2(30).24
and Financial Innovations, 14(4),
3. Antar, M. (2024). Evaluating the 4. Assawawongsathien, S., Srichart,
194-211. https://doi.org/10.21511/
Impact of Borrower Characteris- K., & Nudam, R. (2017). The role
imfi.14(4).2017.16

330 http://dx.doi.org/10.21511/imfi.21(4).2024.26
 Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

and risks of the savings cooperative erative Banks to the Regional 20. Lokesha, & Hawaldar, I. (2019).
system. Puey Ungphakorn Institute Economic Growth: Empirical Evi- Impact of factors on the utiliza-
for Economic Research. Thailand: dence from France. International tion of agricultural credit of banks:
Bangkok. Journal of Economics and Financial an analysis from the borrowers’
Issues, 6(2), 508-514. Retrieved perspective. Banks and Bank Sys-
5. Barua, S., Gavandi, D., Sangle, P.,
from https://www.econjournals. tems, 14(1), 181-192. https://doi.
Shinde, L., & Ramteke, J. (2021). com/index.php/ijefi/article/
Swindle: Predicting the Probability org/10.21511/bbs.14(1).2019.16
view/1750
of Loan Defaults using CatBoost 21. Maichandrang, S. (2021). An
Algorithm. 2021 5th Interna- 13. Galor, Z. (1999). Credit and Saving Analysis of debt default probability
tional Conference on Computing Cooperatives: A new conceptual of bank client: A Case study of
Methodologies and Communication approach. Retrieved from https:// teacher and local government of
(ICCMC) (pp. 1710-1715). Erode, www.findevgateway.org/sites/ Krung Thai Bank, Hod Branch.
India. https://doi.org/10.1109/IC- default/files/publications/files/ Chaing Mai University. Retrieved
CMC51019.2021.9418277 mfg-en-paper-saving-and-credit- from http://cmuir.cmu.ac.th/jspui/
cooperatives-a-new-conceptual- handle/6653943832/73697
6. Boateng, E., & Abaye, D. (2019). A approach-2003.pdf
Review of the Logistic Regression 22. McKillop, D., French, D., Quinn,
Model with Emphasis on Medical 14. Hammond, P., Opoku, M.,
B., Sobiech, A. L., & Wilson, J.
Research. Journal of Data Analysis Kwakwa, P., & Amissah, E. (2023)
S. (2020). Cooperative financial
and Information Processing, 11(4), Predictive models of going con-
institutions: A review of the litera-
190-207. https://doi.org/10.4236/ cerns and business failure. Cogent
ture. International Review of Fi-
jdaip.2019.74012 Business & Management, 10(2),
nancial Analysis, 71, 1-11. https://
1-24. https://doi.org/10.1080/2331
7. Boughaci, D., & Alkhawaldeh, A. 1975.2023.2226481 doi.org/10.1016/j.irfa.2020.101520
A. (2019). A cooperative classifi-
15. Imsuk, S. (2018). Factors affecting 23. Namwong, S., & Janyasuprab, P.
cation system for credit scoring.
the credit default for the loan of the (2018). The Roles of Thai Coop-
Smart Technologies and Innova-
land bank administration institute eratives in New Economic Age.
tion for a Sustainable Future.
(public organization). Thamasat Walailak Journal of Social Science,
Proceedings of the 1st American
University. Retrieved from https:// 11(1), 143-183. Retrieved from
University in the Emirates Inter-
digital.library.tu.ac.th/tu_dc/fron- https://so06.tci-thaijo.org/index.
national Research Conference (pp.
tend/Info/item/dc:142217 php/asi/article/view/110702/86746
11-20). Dubai, UAE. https://doi.
org/10.1007/978-3-030-01659-3_2 16. Kamphod, K. (2019). Factors Af- 24. Nguyen, H. T. (2015). How is credit
fecting the Default on Student Loan scoring used to predict default
8. Bravo, C., Thomas, L. C., & Weber,
R. (2015). Improving credit scor- Fund (SLF) in Thailand. Chaing in China? (EconomiX Working
ing by differentiating defaulter Mai University. Retrieved from Papers 2015-1). University of Paris
behavior. Journal of the Opera- https://cmudc.library.cmu.ac.th/ Nanterre, EconomiX. Retrieved
tional Research Society, 66(5), frontend/Info/item/dc:143095 from https://ideas.repec.org/p/
771-781. https://doi.org/10.1057/ drm/wpaper/2015-1.html
17. Kaplan, I., & Mccay, B. (2004).
jors.2014.50 Cooperative research, co-man- 25. Notananda, V. (2000). Factors
9. Chushim, K. (2013). Factors caus- agement and the social dimen- causing defaults on housing-loan
ing the non-performing loans on sion of fisheries science and repayment of a commercial bank
housing-loan of a commercial bank. management. Marine Policy, 28, in Muang district, Chiang Mai
Thammasat University. Retrieved 257-258. https://doi.org/10.1016/j. Province. Chiang Mai Uni-
from https://digital.library.tu.ac. marpol.2003.08.003 versity. Retrieved from http://
th/tu_dc/frontend/Info/item/ 18. Keawmanee, S. (2016). The Role repository.cmu.ac.th/han-
dc:97119 of Cooperatives in Economic and dle/6653943832/21665

10. Cooperative Auditing Depart- Social Development by Members 26. Omeke, M., Ngoboka, P., Nkote, I.,
ment. (2016). The Registrar of of Thanto Agricultural Coop- & Kayongo, I. (2019) The relation-
eratives Limited, Yala Province. ship between complexity behavior
Cooperatives Regulations on the
Bangkok: Sukhothai Thammathi- and enterprise growth: A case of
Classification of Loan Quality and
rat Open University. Retrieved savings and credit cooperatives in
the Provision for Doubtful Debts.
from https://ir.stou.ac.th/han- Uganda. Cogent Business & Man-
Retrieved from https://www.cad.
dle/123456789/5470 agement, 6(1), 1-15. https://doi.org
go.th/ewt_dl_link.php?nid=30451
&filename=regul2_05_1 19. Laliwan, S., & Potipiroon, W. /10.1080/23311975.2018.1564421
(2022). Board Capital, Organiza-
11. Cooperative Promotion De- 27. Onay, C., & Öztürk, E. (2018). A
tional Capital and Organizational
partment. (2021). Background. review of credit scoring research
Performance of Agricultural and
Retrieved from https://cpd.go.th/ in the age of Big Data. Journal of
Non-agricultural Co-operatives
knowledge/general-coop.html Financial Regulation and Compli-
in Thailand. ABAC Journal, 42(2),
195-215. https://doi.org/10.14456/ ance, 26(3), 382-405. https://doi.
12. El Hancha Sfar, F., & Ben Ouda, O.
abacj.2022.11 org/10.1108/JFRC-06-2017-0054
(2016). Contribution of Coop-

http://dx.doi.org/10.21511/imfi.21(4).2024.26 331
Investment Management and Financial Innovations, Volume 21, Issue 4, 2024

28. Pattarapornpairot, N. (2020). making based on machine learn- Young adults’ default intention:
Examining the relationship between ing (ML) techniques. Annals of influence of behavioral factors
macroeconomic variables on non- Operations Research, 294, 715-739. in determining housing and real
performing loans from commercial https://doi.org/10.1007/s10479- estate loan repayment in India.
banks in Thailand. Thamasat Uni- 019-03188-0 International Journal of Hous-
versity. https://doi.org/10.14457/ ing Markets and Analysis, 16(2),
32. Steenackers, A., & Goovaerts, M. J.
TU.the.2020.826 426-444. https://doi.org/10.1108/
(1989). A credit scoring model for
29. Petrides, G., Moldovan, D., personal loans. Insurance: Math- IJHMA-01-2022-0012
Coenen, L., Guns, T., & Verbeke, ematics and Economics, 8(1), 31- 36. Van Gool, J., Verbeke, W., Sercu,
W. (2022). Cost-sensitive learning 34. https://doi.org/10.1016/0167- P., & Baesens, B. (2012), Credit
for profit-driven credit scoring. 6687(89)90044-9 scoring for microfinance: is it
Journal of the Operational Research worth it? International Journal
33. The Savings and Credit Coopera-
Society, 73(2), 338-350. https:// of Finance & Economics, 17,
tives. (2023). Annual report 2023.
doi.org/10.1080/01605682.2020.1 103-123. https://doi.org/10.1002/
Songkhla: The Prince of Songkla
843975 ijfe.444
University Savings and Credits
30. Saha, A., Hock Eam, L., & Goh Cooperative, Limited. Retrieved 37. Zhu, L. (2019). Predictive Model-
Yeok, S. (2023). Housing loan from https://psucoop.psu.ac.th/ ling for Loan Defaults. UCLA.
default in Malaysia: an analytical home/ebook_report/detail/17/ Retrieved from https://escholar-
insight and policy implications. data.html ship.org/uc/item/3rs9b3d6
International Journal of Hous-
34. Thiansomboon, S. (2017). Factors 38. Zhu, X., Chu, Q., Song, X., Hu,
ing Markets and Analysis, 16(2),
causing default on housing loan: P., & Peng, L. (2023). Explainable
273-291. https://doi.org/10.1108/
case study of commercial bank. prediction of loan default based
IJHMA-01-2022-0002
Thammasat University. Retrieved on machine learning models.
31. Sariannidis, N., Papadakis, S., from https://digital.library.tu.ac.
Data Science and Manage-
Garefalakis, A., Lemonakis, C., & th/tu_dc/frontend/Info/item/
ment, 6(3), 123-133. https://doi.
Kyriaki-Argyro, T. (2020). Default dc:126741
org/10.1016/j.dsm.2023.04.003
avoidance on credit card portfolios
35. Thomas, S. S., George, J. P.,
using accounting, demographical
Godwin, B. J., & Siby, A. (2023).
and exploratory factors: decision

332 http://dx.doi.org/10.21511/imfi.21(4).2024.26