Who Benefits from Alternative Data for Credit Scoring? Evidence from Peru

Bibliography

Credit scores determine a person’s life chances. The credit scores we’re all used to, calculated by Equifax, Experian, or TransUnion, take as inputs a person’s payment history, loans, current debt, and similar financial information. But that world is changing. Modern alternative data models for credit scoring can go so far as to include an individual’s educational record, criminal history, shopping behavior, or telephone patterns. Activists, regulators, and scholars have expressed serious concerns about these new credit systems. Do they classify applicants on unfair or arbitrary grounds? Do they perpetuate, or even amplify, bias and pre-existing inequality? Participants in this conversation tend to assume that the new credit scoring models are a departure from a stable historical norm in which lenders made credit decisions solely based on individuals’ loan repayment history and similar financial inputs. But that’s not right. The new models recapitulate a story from the mid-twentieth century, when a new credit scoring industry, relying on newly developed statistical modeling techniques, looked to a broad range of information: How many years had the person been at the same address? Did he have a telephone? What zip code did he live in? For the new method’s proponents, all data – including the applicant’s race and religion -- was fair game. The new credit scoring crystallized a growing sense that computers, and the new computer age, had no room for fully fleshed human beings. Opponents charged that the new technology enabled and replicated bias, seized on spurious correlations, and generated arbitrary results. They saw it as stripping away agency from credit applicants, based on apparently arbitrary criteria, and as reinforcing social and economic hierarchy. More fundamentally, they argued that it was inconsistent with basic human dignity. The technology was short-lived and has largely been forgotten. By the early 1990s, lenders––for economic rather than public-policy reasons––had moved to the model we’re familiar with today, in which credit scores are based solely on applicants’ credit history and related financial information. However, the story of 1970s-era credit scoring is still relevant today, and it provides lessons as we confront today’s use of machine-learning algorithms to categorize people and predict their future behavior.

An important development in personal lines of insurance in the United States is the use of credit history data for insurance risk classification to predict losses. This research presents the results of collaboration with industry conducted by a university at the request of its state legislature. The purpose was to see the viability and validity of the use of credit scoring to predict insurance losses given its controversial nature and criticism as redundant of other predictive variables currently used. Working with industry and government, this study analyzed more than 175,000 policyholders’ information for the relationship between credit score and claims. Credit scores were significantly related to incurred losses, evidencing both statistical and practical significance. We investigate whether the revealed relationship between credit score and incurred losses was explainable by overlap with existing underwriting variables or whether the credit score adds new information about losses not contained in existing underwriting variables. The results show that credit scores contain significant information not already incorporated into other traditional rating variables (e.g., age, sex, driving history). We discuss how sensation seeking and self-control theory provide a partial explanation of why credit scoring works (the psycho-social perspective). This article also presents an overview of biological and chemical correlates of risk taking that helps explain why knowing risk-taking behavior in one realm (e.g., risky financial behavior and poor credit history) transits to predicting risk-taking behavior in other realms (e.g., automobile insurance incurred losses). Additional research is needed to advance new nontraditional loss prediction variables from social media consumer information to using information provided by technological advances. The evolving and dynamic nature of the insurance marketplace makes it imperative that professionals continue to evolve predictive variables and for academics to assist with understanding the whys of the relationships through theory development.

Working Paper 2009-9 March 2009 Abstract: The literature has documented a positive relationship between the use of credit scoring for small business loans and small business credit availability, broadly defined. However, this literature is hampered by the fact that all of the studies are based on a single 1998 survey of the very largest U.S. banking organizations. This paper addresses a number of deficiencies in the extant literature by employing data from a new survey on the use of credit scoring in small business lending, primarily by community banks. The survey evidence suggests that the use of credit scores in small business lending by community banks is surprisingly widespread. Moreover, the scores employed tend to be the consumer credit scores of the small business owners rather than the more encompassing small business credit scores that include data on the firms as well as on the owners. Our empirical analysis suggests that credit scoring is associated with increased small business lending after a learning period, with no material change in the quality of the loan portfolio. However, these quantity and quality results appear to vary depending on the way in which credit scores are implemented in the underwriting process. JEL classification: G21, G28, L23 Key words: banks, small business, credit scoring I. Introduction Commercial bank lending to small businesses has received a great deal of research attention over the past two decades. The overriding issue in this literature is one of credit availability, given that small firms have historically faced significant difficulties in accessing funding for creditworthy (i.e., positive net present value) projects due to a lack of credible information. Small businesses are typically much more informationally opaque than large corporations because small firms often do not have certified audited financial statements to yield credible financial information on a regular basis. As well, these firms typically do not have publicly traded equity or debt, yielding no market prices or public ratings that might suggest their quality. To address the informational opacity problem, financial institutions use a number of different lending technologies (e.g., Berger and Udell 2006). One lending technology that has recently received considerable research attention is small business credit scoring (SBCS). This technology confronts the opacity problem by combining personal financial data about the owner of the business with the relatively limited information about the firm using statistical methods to predict future credit performance. Consumer credit scoring (CCS) has been widely used for many years in retail credit markets (e.g., mortgages, credit cards, and automobile credits), but SBCS is a more recent phenomenon. Most large U.S. banks did not adopt SBCS until the mid-1990s due to concerns regarding firm heterogeneity and nonstandardized loan documentation (e.g., Mester 1997). As discussed below, some banks instead use the consumer credit scores of small business owners to evaluate small business loan applications. The application of CCS to small business lending has not been previously studied. The empirical literature studying the effects of SBCS has documented significant favorable effects of this lending technology on small business credit availability, broadly defined. Specifically, the adoption of SBCS is empirically associated with 1) increases in the quantity of lending (Frame, Srinivasan, and Woosley 2001, Frame, Padhi, and Woosley 2004, Berger, Frame, and Miller 2005); 2) more lending to relatively opaque, risky borrowers (Berger, Frame, and Miller 2005); 3) lending within low-income as well as high-income areas (Frame, Padhi, and Woosley 2004); and 4) lending over greater distances (DeYoung, Glennon, and Nigro 2008). (1,2) See Berger and Frame (2007) for a more comprehensive review of these studies. While the extant research provides some important information about SBCS, this literature is hampered by the fact that all of the empirical studies are based on a single survey of the largest U. …

An 'authorized user' is a person who is permitted by a revolving account holder to use an account without being legally liable for any charges incurred. The Federal Reserve's Regulation B, which implements the 1974 Equal Credit Opportunity Act, requires that information on spousal authorized user accounts be reported to the credit bureaus and considered when lenders evaluate credit history. Since creditors generally furnish to the credit bureaus information on all authorized user accounts, without indicating which are spouses and which are not, credit scoring modelers cannot distinguish spousal from non-spousal authorized user accounts. This effectively requires that all authorized user accounts receive similar treatment. Consequently, becoming an authorized user on an old account with a good payment history may improve an individual's credit score, potentially increasing access to credit or reducing borrowing costs. As a result, the practice of 'piggybacking credit' has developed. In a piggybacking arrangement, an individual pays a fee to be added as an authorized user on an account to 'rent' the account's credit history. This paper provides the first comprehensive look at authorized user accounts in individual credit records and how their importance differs across demographic groups. Our analysis suggests that piggybacking credit can materially improve credit scores, particularly for individuals with thin or short credit histories. We also evaluate the effect that eliminating authorized user accounts from credit scoring models would have on individual credit scores. Our results suggest that removing this information has relatively little effect on credit scores, but may reduce model predictiveness.

The most important new development in the past two decades in the personal lines of insurance may well be the use of an individual's credit history as a classification and rating variable to predict losses. However, in spite of its obvious success as an underwriting tool, and the clear actuarial substantiation of a strong association between credit score and insured losses over multiple methods and multiple studies, the use of credit scoring is under attack because there is not an understanding of why there is an association. Through a detailed literature review concerning the biological, psychological, and behavioral attributes of risky automobile drivers and insured losses, and a similar review of the biological, psychological, and behavioral attributes of financial risk takers, we delineate that basic chemical and psychobehavioral characteristics (e.g., a sensation‐seeking personality type) are common to individuals exhibiting both higher insured automobile loss costs and poorer credit scores, and thus provide a connection which can be used to understand why credit scoring works. Credit scoring can give information distinct from standard actuarial variables concerning an individual's biopsychological makeup, which then yields useful underwriting information about how they will react in creating risk of insured automobile losses.

The basic concept of credit scoring is to assess an individual`s payment ability as well as the specific individual`s credit default risk, hence determining an individual`s creditworthiness. Based on the credit score, financial institutions, insurance companies, telecommunication companies and other businesses decide whether consumers are eligible for a mortgage, credit card, auto loan, and other credit products. However, in many countries, potential tenants and insurance applicants also use credit scores extensively for screening. Accordingly, Credit Bureaus (CB) or Consumer Reporting Agencies (CRA) exert an essential gatekeeper function for important economic areas of consumers’ everyday life. However, when examining CBs globally, there are considerable differences in the use of data to calculate credit scores. Interestingly, the influence of CBs on credit rating receives little to no attention in academic research. This is particularly evident in the absence of a framework for classifying Credit Bureaus. Therefore, 24 traditional and non-traditional Credit Bureaus operating in 17 different countries are analyzed. First, the study identifies the different data types underlying credit reports and credit scores. Second, CBs are classified and clustered according to the type of information used for credit scoring. Furthermore, promising areas of research, in particular the ethical conflict between data protection and economic participation are highlighted.

A granular computing-based approach to credit scoring modeling

PurposeThe purpose of this paper is to investigate the mechanism how the platform obtains and uses undisclosed information to determine individual borrowers’ credit score and to examine the effectiveness of credit scoring in predicting default. The motivation stems from the fact that there is little evidence about the role of P2P platform, which has been positioned as a kind of information intermediary.Design/methodology/approachUsing a sample of 5,176 unsecured P2P loans having expired before December 31, 2015 on Renrendai.com and an approach of two-stage regression, the paper first estimates the undisclosed information embedded in credit score by regressing credit score on four types of public information about a borrower’s creditworthiness. Then, the authors use a Logit regression to examine the role of the excess information in predicting the default probability.FindingsThe certification information provided by the platform is the most important determinant for a borrower’s credit score and the undisclosed information embedded in credit score can predict the loan performance better than the public information of posted listings. Moreover, the predictive ability of the undisclosed information is better for high-risk borrowers than for low-risk ones.Research limitations/implicationsProviding a credit score for each individual is a way for P2P platforms to play an information intermediary role. More evidence about whether or how a platform plays its role are worthy to be exploited by investigating a platform’s operating policies in detail and doing cross-platform comparative studies.Practical implicationsThe results about the effect of various types of information on loan performance can provide an insightful guidance for P2P platforms to optimize their mechanism on information disclosure and credit scoring.Originality/valueThe existing literature mainly focuses on the effects of information voluntarily disclosed by borrowers and the behaviors of investors on P2P lending outcomes. The paper highlights the information intermediary role played by the platform and presents empirical evidence that credit scoring for individual borrowers is a way for P2P platforms to promote the direct lending for individual.

Credit scoring — the quantitative and statistical techniques which assess the credit risks when lending to consumers — has been one of the most successful if unsung applications of mathematics in business for the last fifty years. Now though, credit scoring is beginning to be used in relation to other decisions rather than the traditional one of assessing the default risk of a potential borrower. Lenders are changing their objectives from minimizing defaults to maximizing profits; using the internet and the telephone as application channels means lenders can price or customize their loans for individual consumers. The introduction of the Basel Capital Accord banking regulations and the credit crunch following the problems with securitizing sub prime mortgage mean one needs to be able to extend the default risk models from individual consumer loans to portfolios of such loans. Addressing these challenges requires new models that use credit scores as inputs. These in turn require extensions of what is meant by a credit score. This book reviews the current methodology for building scorecards, clarifies what a credit score really is, and the way that scoring systems are measured. It then looks at the models that can be used to address a number of these new challenges: how to obtain profitability based scoring systems; pricing new loans in a way that reflects their risk and also customise them to attract consumers; how the Basel Accord impacts on way credit scoring; and how credit scoring can be extended to assess the credit risk of portfolios of loans.

Credit risk management has become a must in this era due to the increase in the number of businesses defaulting. Building upon the legacy of Kealhofer, McQuown, and Vasicek (KMV), a mathematical model is introduced based on Merton model called KMV-Merton model to predict the credit risk of firms. The KMV-Merton model is commonly used in previous default studies but is said to be lacking in necessary detail. Hence, this study aims to combine the KMV-Merton model with the financial ratios to determine the firms’ credit scores and ratings. Based on the sample data of four firms, the KMV-Merton model is used to estimate the default probabilities. The data is also used to estimate the firms’ liquidity, solvency, indebtedness, return on asset (ROA), and interest coverage. According to the weightages established in this analysis, scores were assigned based on those estimates to calculate the total credit score. The firms were then given a rating based on their respective credit score. The credit ratings are compared to the real credit ratings rated by Malaysian Rating Corporation Berhad (MARC). According to the comparison, three of the four companies have credit scores that are comparable to MARC’s. Two A-rated firms and one D-rated firm have the same ratings. The other receives a C instead of a B. This shows that the credit scoring technique used can grade the low and the high credit risk firms, but not strictly for a firm with a medium level of credit risk. Although research on credit scoring have been done previously, the combination of KMV-Merton model and financial ratios in one credit scoring model based on the calculated weightages gives new branch to the current studies. In practice, this study aids risk managers, bankers, and investors in making wise decisions through a smooth and persuasive process of monitoring firms’ credit risk.

This article examines the ways institutions involved in mortgage lending assess credit risk and how credit risk relates to loan performance. An increasingly prominent tool used to facilitate the assessment of credit risk in mortgage lending is credit scoring based on credit history and other pertinent data, and the article presents new information about the distribution of credit scores across population groups and how credit scores relate to the performance of loans. In addition, this article takes a special look at the performance of loans made through nontraditional underwriting practices and "affordable" home lending programs.

The explosive growth of data in banking sector is common phenomena. It is due to early adaptation of information system by Banks. This vast volume of historical data related to financial position of individuals and organizations compel banks to evaluate credit worthiness of clients to offers new services. Credit scoring can be defined as a technique that facilitates lenders in deciding to grant or reject credit to consumers. A credit score is a product of advanced analytical models that catch a snapshot of the consumer credit history and translate it into a numeric number that signify the amount of risks that will be generated in a specific deal by the consumer. Automated Credit scoring mechanism has replaced onerous, error-prone labour-intensive manual reviews that were less transparent and lacks statistical-soundness in almost all financial organizations. The credit scoring functionality is a type of classification problem for the new customer. There are numerous data classification algorithm proposed and each one has its pros and cons. This independent study focuses on comparing three data classification algorithms namely: Naïve Bayes, Bayesian Network and Bagging, for credit scoring task. An extensive series of experiments are performed on three standard credit scoring datasets: (i) German credit dataset, (ii) Australian credit dataset and (iii) Pakistan credit dataset. One of the main contributions of this study is to introduced Pakistan credit dataset; it is collected from local credit repository, and transformed accordingly to be used in the study. The studies compare the experimental results of different selected algorithms for classification, their standard evaluation measures, performance on the three datasets, and conclude the major findings.

Assessing credit risk of commercial customers using hybrid machine learning algorithms

The purpose of this article is twofold: first, we show how algorithms have become increasingly central to financial credit scoring; second, we draw on this to further develop the anthropological study of algorithmic governance. As such, we describe the literature on credit scoring and then discuss ethnographic examples from two regulatory and commercial contexts: the US and Denmark. From these empirical cases, we carve out main developments of algorithmic governance in credit scoring and elucidate social and cultural logics behind algorithmic governance tools. Our analytical framework builds on critical algorithm studies and anthropological studies where money and payment infrastructures are viewed as embedded in their specific cultural contexts (Bloch and Parry 1989; Maurer 2015). The comparative analysis shows how algorithmic credit scoring takes different forms hence raising different issues in the two cases. Danish banks seem to have developed a system of intensive, yet hidden credit scoring based on surveillance and harvesting of behavioural data, which, however, due to GDPR takes place in restricted silos. Credit scores are hidden to customers, and therefore there has been virtually no public debate regarding the algorithmic models behind scores. In the US, fewer legal restrictions on data trading combined with both widespread and visible credit scoring has led to the development of a credit data market and widespread use of credit scoring by ‘affiliation’ on the one hand, but also to increasing public and political critique on scoring models on the other.