Hard or Soft Information on Determining
P2P NPLs:
Evidence From Leading P2P Platforms
Research Proposal
Words:1999-excluding-references
Abstract
The internet-based economy has challenged financial intermediation through rise of Peer-to-
Peer lending. Unlike commercial banks, lenders are not experts in evaluating credit risks,
suffering information asymmetry, and disadvantaged when facing the borrower. This research
proposal in P2P lending is within the stream of literature which focuses on determining
factors on credit risk. The research aims to determine whether hard or soft information can
serve as a predictor of non-performing loans and examines this across markets. Data is
sourced from leading platforms, across 5-years, and use relevant explanatory variables.
2
Introduction
Digitalisation in finance and the rise of ‘Fintech’ has caused ‘disruption’ in banks’ role as
intermediary between borrowers and savers (Thakor 2020). New business models have
emerged, such as peer-to-peer (P2P) lending model where borrowers are directly matched to
savers (Gomber, Koch & Siering 2017). Through a bidding process, potential lenders offer a
loan amount, to which the platform combines appropriate bids into a single borrower’s loan
(Thakor 2020). P2P is unsecured, providing small loans to individuals and SMEs who
encounter difficulties from commercial credit alternatives (Bachmann et al. 2011). P2P has
developed significantly in recent years, with estimates that P2P will grow to almost $300
billion by 2022 (Thakor 2020).
A critical issue, however, is information asymmetry (IA) between borrowers and lenders
(Emekter et al. 2015). Whereas banks can accurately assess credit quality and precisely
monitor borrowers to secure loan repayment, P2P firms lack these capabilities (Iyer et al
2016). This causes lender disadvantage due to less information about creditworthiness of
borrowers, causing adverse selection (Akerlof 1970) and moral hazard (Stiglitz and Weiss
1981). Difficulties of regular monitoring are exacerbated by the online environment where
parties do not physically meet (Emekter et al. 2015). Instead, creditworthiness and loan
success is screened based upon both hard and soft information (H&SI). Hard information is
information that can be quantified, stored easily, and communicated impersonally; including
indicators used by banks such as debt-to-income ratio, number of outstanding loans, credit
inquires, etc. (Lin 2009). Soft information also can contain valuable factors for mitigating
against the inability of predicting creditworthiness (Cornee 2017). These are qualitative (Lin
2009) and are subjective based on borrower’s social circumstance (Dorfleitner et al. 2015),
and includes friendship networks, and textual descriptions.
This reliance on H&SI to screen for borrowers meant risk and risk management becomes a
greater issue than traditional lending models (Pope and Sydnor 2011). Indeed, IA meant Iyer
et al. (2016) has identified credit risk to be most important, and mechanisms to control it, ex
ante, is paramount (Yan, Yu, & Zhao 2015). The emphasised risk is of default and untimely
repayment, stemming from insufficient credit checking, and transparency. Typically,
borrowers were unable to access credit from traditional facilities, hence risk for P2P is much
greater.
3
Motivation
I am interested in P2P lending as a new field of banking and motivated for my research to
have impact on existing knowledge. I am also interested in P2P as a new investment
opportunity.
Research Question: To what extent do factors of hard and soft information in P2P lending
serve as predictors of non-performing loans in P2P?
Non-performing loans (NPLs) in P2P are loans that are not current for 30+ days (Lending
Club 2020); and this is constant for most lending platforms. This is contrasted to most
literature of P2P loans being in ‘default’ which means cannot be repaid.
Contribution/Significance
The question is significant, as it provides novel research in the stream of literature which
investigates H&SI determinants on credit/default risks. As demonstrated by the literature-
review, most focus upon only the influence of hard information on probability of default (PD)
and risk of default. The few later studies examining the impact of soft information are
arguably inadequate. Hence, the proposed research addresses two gaps. Firstly, my research
will be first to examine IA determinants (H&SI) and its impact on arising of NPLs. This
enriches research into ex ante credit risk of P2P, specifically of non-repayment. NPLs assume
loan hasn’t been defaulted yet, but in arrears. Secondly, previous research has taken an
either/or approach to H&SI whereas my research will examine both H&S. As
Malekipirbazari & Aksakalli (2015) affirms, even borrowers with high credit rating (hard
information) have a high chance of default, so rich user data of soft information should also
be examined. Additionally, this investigation will examine both developed markets through
US platform and emerging markets through Chinese platform, hence making a cross-
geographic contribution.
Hence, my results obtained will add value as lenders can recognise which drivers are most
significant to predicting NPLs. Non-repayment is an important risk for lenders, as it
encapsulates the cash flow concerns for investors, and hence his/her earnings volatility. When
a loan becomes NPL status, there is significant chance that it will be defaulted. Hence, by
understanding the determinants of NPL, IA can be reduced, and helping investors mitigate
against non-repayment - investment risks managed. My research has practical implications,
as P2P will continue to grow exponentially, especially with arising of hedge funds and
SID:480221342
4
institutional investors (Thakor 2020), and recent explosion in emerging markets such as
China.
Literature Review
Online P2P lending is still a relatively novel field of research. As a response to growing
popularity of these online platforms, three main streams of literature have emerged (Emekter
et al. 2015). The first stream reviews the history and advent of P2P lending. Second stream
examines the performance of P2P loans for a certain risk characteristic. But as demonstrated
by my research problem, the focus of this literature review will be on the third stream -
investigating determining factors that contribute to funding success and credit risk of
borrowers.
Firstly, we must consider the literature concerning information asymmetry, specifically for
P2P. IA occurs when borrowers are more knowledgeable about their ability to meet
repayments than lenders (Serrano-Cinca, Gutierrez-Nieto & Lopez-Palacios 2015). P2P
lenders lack screening and monitoring capabilities of banks; hence this information gap leads
to adverse selection (Akerlof 1970). IA in finance also leads to credit rationing (Stiglitz and
Weiss 1981), where intermediated financial institutions are cannot distinguish between risky
and less risky borrows, and hence disintermediated P2P allow these borrowers to capital.
Since IA is the main obstacle for lenders to minimise credit risk (Yan et al. 2015), the third
stream of literature examines how H&SI provided by the borrower, can be a determinant of
such risk.
Earlier research in this third stream looked at selective factors that determine the credit risk of
borrowers. For example, Gomez and Santor (2008) used Canadian micro-credit data to find
that default rates were lower for group lending than individual lending.
Research on impact of hard information beyond financial metrics arose in late 2000s, with the
rise of the Prosper lending platform. Datasets employed have come from publicly available
transaction data from Lending Club and Prosper (Zhou, Zhang, & Luo 2018), and studies in
from PPDai in Asia (Wang et al. 2019). The development of literature thus focused and
continues to attract interest, on probability of default (PD), from hard information (Jiang et al.
2017) - developed through statistical analysis and machine learning. Statistically, Seranno-
Cinca et al. (2015) employed time-series data and found hard factors explaining default
SID:480221342
5
include annual income, housing situation, credit history, level of debt, and loan purpose.
Seranno-Cinca et al. employs binary logistic regression in evaluating PD and survival
analysis on when a borrower will default. Machine learning to evaluate hard information on
PD include neural network (Hajek 2011) and random forest (Malekipirbazari & Aksakalli
2015). Emekter et al (2015), demonstrated that hard information such as debt-to-income ratio,
credit grade, FICO score, and revolving line utilisation also had a significant impact in loan
defaults, and these results reflected ‘Cox Proportional Hazard’ test which shows a positive
relationship between default and credit risk.
However, research concerning the contribution of both H&SI for loan evaluation was
initiated by Iyer et al (2009), who showed that a third of credit risk can be inferred from both
hard and soft data. Indeed, soft information can be a valid aid in mitigating asymmetric
information (Dorfleitner et al. 2016). Lin, Prabhala, & Viswanathan (2013) examines soft
information social connection and found that strength of a borrower’s social networking
relationship was an important determinant for lowering default risk and increasing funding
success. Lin et al., further posits that the extent of friendships was linked to reduction of ex
post default rates on Prosper (2013). Additionally, studies made by Duarte, Siegel, & Young
(2012), suggests that borrowers who looked more trustworthy had greater creditworthiness
and had a lower probability of default and increased funding success, and Larrimore et al.
(2011) suggests that extensive quantitative descriptions also contributed to reducing risk of
default.
Recent explosion of P2P lending in China has also triggered contemporaneous research, such
as where IA exists, market-based interest rate generated did not fully reflect default risk, and
H&SI of the borrower contributed to predicting such risk (Zhou et al. 2018). Interestingly, in
China, most soft factors could predict PD, and used to assess the creditworthiness of
borrowers (Wang et al. 2019).
As such, whilst there has been reasonable development of hard information on PD, soft
information is still largely insufficient. Indeed, research which relates these determinants to
P2P NPLs has yet to be explored, despite being extensively researched in traditional banking
(Klein, 2014). Hence, this proposed research will address this, by researching NPLs in the
context of H&S determinants, and hence guiding investors and P2P platforms to more
accurately evaluate risks.

SID:480221342
6

Hypothesis
From aforementioned question, two tentative hypotheses obtained:
H1: Hard information in P2P lending will serve as a better predictor of NPLs than soft
information.
This is because a significant amount of riskiness can be captured by these hard information
and the selected explanatory variables are determinants of PD. However, soft information is
still a sound predictor.
H2: Soft information can be a more accurate predictor of NPLs in emerging markets, than in
developed counterparts.
This is because in Asian markets, social networks through online platforms are well-
entrenched and these networks provide rich source of soft data for predicting NPL. Loose
privacy laws meant soft information are readily available (Wang et al. 2019).

Methodology
The research will regress NPLs against a selection of H&SI, to analyse the extent of
contribution of each determinant, over a 5-year time frame.
Data Inputs/Sources
Loan data from Lending Club, the biggest US P2P firm (Serrano-cinca, et al. 2015), will be
used. Lending Club is a good representative of P2P firms in advanced economies. For Asian
markets, data from PPDai, leading Chinese platform, will be used (Wang et al. 2019). H&SI
data on both platforms is publicly available for each borrower. A sample of 50,000 loans
from each will be analysed from 2015-2019. The chosen 5-year time frame keeps
macroeconomic factors relatively stable.
Variables
Dependent: Will use the descriptor of ‘NPL’ and ‘not NPL’.
Independent: This investigation will consider selected H&SI that have influence on NPLs. A
further analysis of literature will be performed to examine more variables. Currently, hard
information will include loan purpose, loan amount, loan period, borrower’s interest rate,
platform credit grade, delinquency 2 years, inquiries made in last 6 months and debt-to- 
7
income ratio and revolving line utilisation (Serrano-Cinca et al. 2016). Soft include
employment length, friendship networks, group affiliation, description length, description
complexity, friend endorsement, employment lengths (Iyer et al. 2016). These are currently
tentative.
Regression Model
To analyse the cross-sectional, time series data, model will be binary logistics regression,
where has been benchmarked from previous literature (Serrano-Cinca et al. 2016). The model
is useful, as linear regression cannot work with categorically dependent variables
(Mollenkamp 2017):
NPL = β0 + β1 xi1 + β2 xi2 + … + βk xin+ ε,
Where dependent variable is the binary NPL status, xi1-n represents each of the
aforementioned 9 hard and 7 soft information, and β1-k are coefficients, and ε is error term.
The regression will run for both Lending Club and PPDai datasets.
8
References
Akerlof, G, A 1970, ‘The Market for "Lemons": Quality Uncertainty and the Market
Mechanism’, The Quarterly Journal of Economics, vol. 84, pp. 488-500.
Bachmann, A, Becker, A, Buerckner, D, Hilker, M, Lehmann, M & Tiburtius, P 2011,
‘Online peer-to-peer lending – a literature review’, Journal of Internet Banking and
Commerce, vol. 16, pp. 1-18.
Cornee, S, ‘The Relevance of Soft Information for Predicting Small Business Credit Default:
Evidence from a Social Bank’, Journal of Small Business Management, vol. 57, no.3.
Dorfleitner, G, Priberny, C, Schuster, S, Stoiber, J, Weber, M, de Castro, I & Kammler, J
2016, ‘Description-text related soft information in peer-to-peer lending – Evidence
from two leading European platforms’, Journal of Banking & Finance, vol. 64, pp.
169-187.
Duarte, J, Siegel, S, & Young, L 2012, ‘Trust and credit: the role of appearance in peer-to-
peer lending, Review of Financial Studies, vol. 25, pp. 2455-2484.
Emekter, R, Tu, T, Jirasakuldech, B, & Lu, M 2015, ‘Evaluating credit risk and loan
performance in online Peer-to-Peer (P2P) lending’, Applied Economics, vol. 47, no.1,
pp. 54-70.
Gomber, P, Koch, J, & Siering, M 2017, ‘Digital Finance and FinTech: current research and
future research directions’, Journal of Business Economics, vol. 87, pp. 537-580.
Gomez, R, & Santor, E 2003, ‘Do peer group members outperform individual borrowers? A
test of peer group lending using Canadian micro-credit data’, Working Paper 2003-
22, Bank of Canada, Ottawa, ON.
Hajek, P 2011, ‘Municipal credit rating modelling by neural networks’, Decision Support
Systems, vol. 51, no. 1, pp. 108-118.
Iyer, R, Khwaja, A, I, Luttmer, E, F, P & Shue, K 2009, ‘Screening in New Credit Markets:
Can Individual Lender Infer Borrower Creditworthiness in Peer-to-Peer Lending?,’
Working Paper Series, Harvard University, John F. Kennedy School of Government.
Iyer, R, Khwaja, A, I, Luttmer, E, F, P & Shue, K 2016, ‘Screening Peers Softly: Inferring
the Quality of Small Borrowers’, Management Science, vol. 62, pp. 1554-1577.
9
Jiang, C, Wang, Z, Wang, R, & Ding, Y 2017, ‘Loan default prediction by combining soft
information extracted from descriptive text in online peer-to-peer lending’, Annals of
Operations Research, vol. 266, pp. 511-529.
Klein, N 2014, ‘Non-performing loans in CESEE: determinants and impact on
macroeconomic performance’, Social Science Electronic Publishing, vol. 13, no. 72,
pp. 17-37.
Larrimore, L 2011, Jiang, L, Larrimore, J, Markowitz, D, & Gorski, S, ‘Peer to Peer Lending:
The Relationship Between Language Features, Trustworthiness, and Persuasion
Success’, Journal of Applied Communication Research, vol. 39, no. 1.
Lending Club 2020, What do the different Note statuses mean?, viewed 17th May 2020,
https://help.lendingclub.com/hc/en-us/articles/216109367-What-do-the-different-
Note-statuses-mean-.
Lin, M 2009, ‘Peer-to-Peer Lending: An Empirical Study’, AMCIS 2009 Doctoral
Consortium, 17.
Lin, M, Prabhala, N, & Viswanathan, S 2013, ‘Judging borrowers by the company they keep:
friendship networks and information asymmetry in online peer-to peer lending’,
Management Science, vol. 59, pp. 17-35.
Malekipirbazari, M, & Aksakalli, V 2015, ‘Risk assessment in social lending via random
forests.’ Expert Systems with Applications, vol. 42, no. 10, pp. 4621-4631.
Mollenkamp, N, 2017, ‘Determinants of Loan Performance in P2P Lending’, 9th IBA
Bachelor Thesis Conference, pp. 1-15.
Pope, D, G, & Sydnor, J, R 2011, ‘What’s in a picture? Evidence of discrimination from
Prosper.com’, Journal of Human Resources, vol. 46, no. 1, pp. 53-92.
Serrano-Cinca, C, Gutierrez-Nieto, B, & Lopez-Palacios 2015, ‘Determinants of Default in
P2P Lending’, PLOS One, vol. 10, pp. 1-22.
Stiglitz, J, E, Weiss, A 1981, ‘Credit rationing in markets with imperfect information’.
American Economic Review, vol. 71, pp. 393-410.
Thakor, A, V 2020, ‘Fintech and banking: What do we know?’ Journal of Financial
Intermediation, vol. 41, pp. 1-13.
10
Wang, C, Zhang, W, Zhao X, & Wang J 2019, ‘Soft information in online peer-to-peer
lending: Evidence from a leading platform in China’ Electronic Commerce Research
and Applications, vol. 36.
Yan, J, Yu, W, & Zhao, J, L 2015, ‘How signaling and search costs affect information
asymmetry in P2P lending: the economics of big data’ Financial Innovation, vol. 1,
no. 19.
Zhou, G, Zhang, Y, Luo, S 2018, ‘P2P Network Lending, Loss Given Default and Credit
Risks’ MDPI, vol. 10, pp. 1-15.