AI helps you reading Science

AI generates interpretation videos

AI extracts and analyses the key points of the paper to generate videos automatically


pub
Go Generating

AI Traceability

AI parses the academic lineage of this thesis


Master Reading Tree
Generate MRT

AI Insight

AI extracts a summary of this paper


Weibo:
Because of the lifealtering consequences that can flow from a faulty or unfair credit score, regulators must ensure that innovators proceed responsibly and have strong legal incentives to ensure that their scoring decisions are transparent, accurate, unbiased, and fair

Credit Scoring in the Era of Big Data

Yale Journal of Law and Technology, no. 1 (2017): 5

Cited by: 4|Views112
Full Text
Bibtex
Weibo

Abstract

TABLE OF CONTENTS TABLE OF CONTENTS I. INTRODUCTION II. TRADITIONAL CREDIT-ASSESSMENT TOOLS III. ALGORITHMS, MACHINE LEARNING, AND THE ALTERNATIVE CREDIT-SCORING MARKET A. Introduction to basic terminology and concepts B. How traditional credit-modeling tools compare to alternative, tools C. Using machine learning to build a big-data cre...More

Code:

Data:

0
Introduction
  • One day in late 2008, Atlanta businessman Kevin Johnson returned home from his vacation to find an unpleasant surprise waiting in his mailbox.
  • Kevin, who is black, was running a successful Atlanta public relations firm, was a homeowner, and had always paid his bills on time, rarely carrying a balance on his card.[2] Kevin's father, who had worked in the credit industry, had taught him the importance of responsible spending and, "because of his father's lessons, Ron Lieber, American Express Kept a (Very) Watchful Eye on Charges, N.Y. TIMES (Jan. 30, 2009), http://www.nytimes.com/2009/01/31/your-money/creditand-debit-cards/3 lmoney.html?pagewanted-all [https://perma.cc/QQ7P-
Highlights
  • One day in late 2008, Atlanta businessman Kevin Johnson returned home from his vacation to find an unpleasant surprise waiting in his mailbox
  • It was a letter from his credit card company, American Express, informing him that his credit limit had been lowered from $10,800 to a mere $3,800.1 While Kevin was shocked that American Express would make such a drastic change to his limit, he was even more surprised by the company's reasoning
  • Kevin's father, who had worked in the credit industry, had taught him the importance of responsible spending and, "because of his father's lessons, Ron Lieber, American Express Kept a (Very) Watchful Eye on Charges, N.Y
  • The company had deemed him a risk because, as the letter put it, "[o]ther customers who ha[d] used their card at establishments where [Kevin] recently shopped have a poor repayment history with American Express."[4] When Kevin sought an explanation, the company was unwilling to share any information on which of businesses - many of them major retailers - contributed to American Express's decision to slash Kevin's limit by more than 65 percent. 5
  • Because of the lifealtering consequences that can flow from a faulty or unfair credit score, regulators must ensure that innovators proceed responsibly and have strong legal incentives to ensure that their scoring decisions are transparent, accurate, unbiased, and fair
Results
  • The company had deemed him a risk because, as the letter put it, "[o]ther customers who ha[d] used their card at establishments where [Kevin] recently shopped have a poor repayment history with American Express."[4] When Kevin sought an explanation, the company was unwilling to share any information on which of businesses - many of them major retailers - contributed to American Express's decision to slash Kevin's limit by more than 65 percent. 5.
  • (3) If a consumer plaintiff initiates an action pursuant to this Section and the Attorney General elects to proceed with the action, the consumer plaintiff shall receive at least 15 percent but not more than 33 percent of the proceeds of the action or settlement of the claim, depending upon the extent to which the consumer plaintiff substantially contributed to the prosecution of the action
Conclusion
  • Big-data credit-scoring tools may, as their proponents claim, emerge as a way to ensure greater efficiency in underwriting while expanding access to the underbanked and to historically neglected groups.
  • This zeal to "build a better mousetrap" must be tempered against its possible perils.
  • Because of the lifealtering consequences that can flow from a faulty or unfair credit score, regulators must ensure that innovators proceed responsibly and have strong legal incentives to ensure that their scoring decisions are transparent, accurate, unbiased, and fair.
Funding
  • The company had deemed him a risk simply because, as the letter put it, "[o]ther customers who ha[d] used their card at establishments where [Kevin] recently shopped have a poor repayment history with American Express."[4] When Kevin sought an explanation, the company was unwilling to share any information on which of businesses - many of them major retailers - contributed to American Express's decision to slash Kevin's limit by more than 65 percent. 5
  • (2) If a consumer plaintiff initiates an action pursuant to this Section and the Attorney General does not elect to proceed with the action, the consumer plaintiff shall receive an amount not less than 33 percent and not more than 50 percent of the proceeds of the action or settlement
  • (3) If a consumer plaintiff initiates an action pursuant to this Section and the Attorney General elects to proceed with the action, the consumer plaintiff shall receive at least 15 percent but not more than 33 percent of the proceeds of the action or settlement of the claim, depending upon the extent to which the consumer plaintiff substantially contributed to the prosecution of the action
Study subjects and analysis
consumers: 1001
http://www.myfico.com/CreditEducation/WhatsNotInYourScore. aspx. [https://perma.cc/3LUN-NRCD]. 38 Out of a survey population of 1,001 consumers. See FED

consumers: 211
For instance, by its terms ECOA does not clearly prohibit discrimination on the basis of a consumer's sexual orientation. While some courts have interpreted ECOA's prohibition on sex discrimination as encompassing claims where an individual was denied access to credit because he or she did not comply with the lender's expectations regarding gender norms, 211 consumers may find it difficult to challenge lender discrimination based on sexual orientation. Proving a violation of ECOA is burdensome, and the use of highly complex big-data credit-scoring tools may only exacerbate that difficulty

full-time equivalent employees: 5
DEFINITIONS. As used in this Act: (1) "Consumer" means any individual or group of individuals, including households, family groups, and small businesses having 5 full-time equivalent employees or fewer. (2) "Credit score" means any numerical or descriptive assessment of a consumer's creditworthiness

full-time employees: 5
Section 1. Definitions " Consumer: Refers to any individual person or group of persons including households, family groups, and small businesses with fewer than five full-time employees. This definition ensures that the Act applies regardless of whether the covered entity is dealing with an individual, a group of persons, or a small family-owned business. * Credit Score: A numerical or descriptive assessment of a consumer's creditworthiness. * Credit Assessment Tool: a system, model, technique, factor, set of factors, report, or any other mechanism used to score assess consumer creditworthiness

Reference
  • 7 Tracy Alloway, Big data: Credit where credit's due, FINANCIAL TIMES (Feb. 4, 2015), http://www.ft.com/cms/s/0/7933792e-a2e6-11e4-9c06-
    Locate open access versionFindings
  • 19 See Yu, supra note 10, at 27 (FICO first introduced its flagship score in 1981).
    Google ScholarFindings
  • 20 Ashoka, Banking the Unbanked: A How-To, FORBES (Jun. 14, 2013), http://www.forbes.com/sites/ashoka/2013/06/14/banking-the-unbanked-a-howto [https://perma.cc/PD4J-VFDT].21 See Amy Traub, Discredited:How Employment Credit Checks Keep Qualified Workers out of A Job, DEMOS 1 (Feb.2013), http://www.demos.org/discredited-how-employment-credit-checks-keepqualified-workers-out-job [https://perma.cc/82CK-ZGNG].22 See FED. TRADE COMM'N, USING CONSUMER REPORTS: WHAT LANDLORDS NEED
    Locate open access versionFindings
  • 30 Danielle Keats Citron & Frank Pasquale, The Scored Society: Due Process for Automated Predictions,89 WASH. L. REV. 1, 8-9 (2014).
    Google ScholarLocate open access versionFindings
  • 32 See id. 33 Blake Ellis, Millions Without Credit Scores not so Risky After All, CNN MONEY (Aug. 14, 2013), http://money.cnn.com/2013/08/14/pf/credit-scores [https://perma.cc/AA8Y-ZU3K].
    Locate open access versionFindings
  • 40 For example, in 2014, the Huffington Post reported on 69 year-old veteran who was forced out of his home as a result of an erroneously-reported debt on a credit card that he never held. The debt, which he disputed, remains on his credit score to this day. See Hunter Stuart, It's DisturbingLikely that Your Credit Report is Wrong, HUFFINGTON POST (Aug. 11, 2014), http://www.huffingtonpost.com/2014/08/1 l/credit -report-bureau-mistakesn 5661956.html [https://perma.cc/Q83N-3JBV].41 Rome Neal, Mistaken Identity in Credit Report, CBS NEWS (July 31, 2002), http://www.cbsnews.com/news/mistaken-identity-in-credit-report extensive attempts to rectify the error, Ms. Thomas finally sued TransUnion, ultimately winning a multi-million dollar verdict.
    Locate open access versionFindings
  • 43 In 2013, a similar fate befell yet another, different Judy Thomas. According to a report by 60 Minutes, Ohio resident Judy Thomas discovered that her credit reports inaccurately contained information on the debts of a Utah woman, Judy Kendall. As a result of the false information on her reports, Ms.
    Google ScholarFindings
  • 45 See Marcus Tewksbury, The 2013 Big Data Planning Guide for Marketers, EXPERIAN
    Google ScholarLocate open access versionFindings
  • 52 There may also be inherent randomness or noise in the system being studied. A straightforward application of a mathematical formula would ignore the inherent noise in the system. See HAROLD J. KUSHNER & G. GEORGE YIN, STOCHASTIC APPROXIMATION ALGORITHMS AND APPLICATIONS 2 (1997).
    Google ScholarLocate open access versionFindings
  • 53 Suresh Venkatasubramanian, When an Algorithm Isn't, MEDIUM (Oct. 1, 2015), https://medium.com/@geomblog/when-an-algorithm-isn-t-
    Findings
  • 56 Kevin P. Murphy, MACHINE LEARNING: A PROBABILISTIC PERSPECTIVE (2012).
    Google ScholarLocate open access versionFindings
  • 62 See Rob Berger, A Rare Glimpse Inside the FICO Credit Score Formula, DOUGHROLLER (Apr. 30, 2012), http://www.doughroller.net/credit/a-rareglimpse-inside-the-fico-credit-score-formula [https://perma.cc/8VD7-6JSX].63 Id.64 Id.
    Findings
  • 70 Founded in 2009, ZestFinance offers big-data credit-scoring tools to providers of payday loans (short-term, high-interest loans), while also offering such loans through its affiliate, ZestCash.
    Google ScholarFindings
  • 75 U.S. Patent App. No. 14/276,632 (filed May 13, 2014), http://www.google.com /patents/US20150019405
    Findings
  • 82 See, e.g., Brief for Center for Dig. Democracy as Amicus Curiae Supporting Respondents, Spokeo, Inc. v. Robins, 135 S.Ct. 1892, No. 13-1339, 2015 WL 5302538, at "12-13; see also supra note 16.
    Google ScholarLocate open access versionFindings
  • 83 U.S. Patent No. 9,100,400 (filed Aug. 2, 2012).
    Google ScholarFindings
  • 98 As Kate Crawford points out, when target variables are under-inclusive due to gaps in the data set, there may be "signal problems" where "some citizens and communities are overlooked or underrepresented." Kate Crawford, Think Again: Big Data, FOREIGN POLICY (May 9, 2013), http://foreignpolicy.com/2013/05/10/think-again-big-data [https://perma.cc/EF5B-XK8E].99 See Toon Calders and Indre Zliobaite, supra at note 96, 46-47 ("Computational models typically [assume]...that (1)the characteristics of https://digitalcommons.law.yale.edu/yjolt/vol18/iss1/5
    Findings
  • 101 See, e.g., Blake Ellis, Millions Without Credit Scores not so Risky After All, CNN MONEY (Aug. 14, 2013), http://money.cnn.com/2013/08/14/pf/creditscores [https://perma.cc/4GD4-PPN5].
    Locate open access versionFindings
  • 119 See Singer, supra note 8. According to Singer, as of 2012, Acxiom's "servers process[ed] more than 50 trillion data 'transactions,"' and that the company's
    Google ScholarFindings
  • 132 David Bollier, The Promise and Peril of Big Data, ASPEN INSTITUTE, at 5, 14 (Jan. 1, 2010), https://www.aspeninstitute.org/publications/promise-peril-bigdata [https://perma.cc/RB5Q-6TJ5].
    Findings
  • 141 Credit Scoring, ELECTRONIC PRIVACY INFORMATION CENTER (2016), https://epic.org/privacy/creditscoring [https://perma.cc/W94Z-HGWP].
    Locate open access versionFindings
  • 147 Christopher P. Guzelian et al, Credit Scores, Lending, and Psychosocial Disability, 95 B.U. L. Rev. 1807, 1815 (2015).
    Google ScholarLocate open access versionFindings
  • 149 Feature selection is an important component of any machine-learning application. Feature selection increases the signal to noise ratio by eliminating irrelevant input variables. See Isabelle Guyon & Andre Elisseeff, An introductionto variableandfeature selection, 3 J. MACHINE LEARNING RES. 1157 (2003).
    Google ScholarLocate open access versionFindings
  • 164 See Robinson + Yu, supranote 11, at 28; see also generally S. Rep. No. 91-169 (1969). 165 15 U.S.C. § 1681(b) (2012).
    Google ScholarFindings
  • 170 McCready v. EBay, Inc., 453 F.3d 882 (7th Cir. 2006) (information pertaining to an anonymous computer username does not qualify under definition of "consumer report"). 171 See Robinson + Yu, supra note 11, at 17.
    Google ScholarFindings
  • 172 Fiscella v. Intelius, Inc., 2010 WL 2405650 (E.D. Va., June 10, 2010).
    Google ScholarFindings
  • 173 Fuges v. Southwest Fin. Serv., Ltd., 707 F.3d 241, 253 (3d Cir. 2012) (finding that it was not unreasonable for the defendant to interpret the FCRAs definition of a consumer report as excluding information about encumbrances on a property, even if the property was owned by an identifiable consumer).
    Google ScholarFindings
  • 182 It should be noted that the FCRA also specifically excepts actors that only acquire data "first-hand" from consumers, see 15 U.S.C. § 1681a(d)(2)(A)(i) (2012), a flexibility that may have particular importance for online lenders that use detailed applications. See also Robinson + Yu, supra note 11, at 28.
    Google ScholarLocate open access versionFindings
  • 185 See, e.g., Consent Decree, United States v. Spokeo, Inc., No. CV12-05001 MMM (C.D. Cal., June 7, 2012).
    Google ScholarFindings
  • 188 See 15 U.S.C. §1681b(a) (2012).
    Google ScholarLocate open access versionFindings
  • 189 See Trans Union Corp. v. FTC, 245 F.3d 809, 812-16 (D.C. Cir. 2001) (confirming the FTC's finding that lists containing the names and address of individuals who have auto loans, department store credit cards, or mortgages, qualified as consumer reports under the FCRA, and that the sale of such lists for target marketing purposes was a violation of the Act).
    Google ScholarLocate open access versionFindings
  • 190 FairCredit Reporting, supra note 168, at § 7.1.2.1 (citing 15 U.S.C. § 1681e(a) (2012)); Levine v. World Fin. Network Nat'l Bank, 437 F.3d 1118 (11th Cir. 2006). 191 15 U.S.C. § 1681e(b) (2012).
    Google ScholarLocate open access versionFindings
  • 192 See FairCreditReporting,supranote 168, at § 4.2.3. 193 15 U.S.C. § 1681m(a) (2012); see also Fair Credit Reporting, supra note 168, at § 3.3.6. 194 15 U.S.C. § 1681g (2012). 195 15 U.S.C. § 1681i(a)(2012).
    Google ScholarFindings
  • 196 For example, the FCRA does not apply to companies that collect and maintain their own data on consumers, and use it internally rather than selling it. See 15 U.S.C. § 1681a(d)(2)(A)(i) (2012). As a practical consequence, online lenders that acquire their information first-hand from consumers or through automated web-crawling will not be subject to the FCRA. See Robinson + Yu, supra note 11, at 28.
    Google ScholarLocate open access versionFindings
  • 197 Yu ET AL., supra note 10, at 24; see also Persis S. Yu & Sharon M. Dietrich, Broken Records: How Errors by Criminal Background Checking Companies Harm Workers and Businesses, NCLC (Apr. 2012), www.nclc.org/issues/broken-records.html [https://perma.cc/SVY9-X24R].
    Locate open access versionFindings
  • 199 See 15 U.S.C. § 1681c (2012).
    Google ScholarLocate open access versionFindings
  • 208 Edith Ramirez, Chairwoman, FTC, ProtectingPrivacy in the Era ofBig Data, Remarks Before the International Conference on Big Data from a Privacy Perspective, at *5 (June 10, 2015), availableas 2015 WL 3824154. 209 15 U.S.C. § 1691(a)(1) (2012). 210 15 U.S.C. § 1691(a)(2) (2012).
    Google ScholarLocate open access versionFindings
  • 211 See, e.g., Rosa v. Park W. Bank & Trust Co., 214 F.3d 213, 215 (1d Cir. 2000)
    Google ScholarLocate open access versionFindings
  • 219 See, e.g., Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 657-58 (1989)
    Google ScholarLocate open access versionFindings
  • 226 See, e.g., Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 998 (1988).
    Google ScholarFindings
  • 227 Beaulialice v. Fed. Home Loan Mortgage Corp., No. 8:04-CV-2316-T-24-EA, 2007 WL 744646 (M.D. Fla. Mar. 6 2007), offers a rare example of a challenge to a credit-scoring algorithm under the disparate impact theory. Unfortunately, Beaulialiceprovides little insight into how a court might view a disparate impact claim in the credit-scoring setting as the case was dismissed on the ground that the plaintiffs claims were barred by the doctrine of "unclean hands." Id.
    Google ScholarLocate open access versionFindings
  • 228 See infra p. 202, Julius Adebayo, Mikella Hurley & Taesung Lee, Model Fairness and Transparency in Credit Scoring Act (FaTCSA). The Model FaTSCA is reproduced with permission of its authors. As currently drafted, the Model FaTSCA has been optimized for enactment at the state level.
    Google ScholarLocate open access versionFindings
  • 229 See id. at § 3(a), p. 204.
    Google ScholarFindings
  • 230 See id. at § 3(b), p. 204.
    Google ScholarFindings
  • 231 See Danielle Keats Citron & Frank Pasquale, The Scored Society: Due Process for Automated Predictions,89 WASH. L. REV. 1, 25 (2014).
    Google ScholarLocate open access versionFindings
  • 232 See 15 U.S.C. § 1681i(a) (2012).
    Google ScholarLocate open access versionFindings
  • 233 See infra pp. 206-207, Model FaTSCA, at §§ 4(d), 4(g), 5.
    Google ScholarFindings
  • 234 See id. at § 4(d), p. 206.
    Google ScholarFindings
  • 235 See id. at § 6, pp. 207-208.
    Google ScholarFindings
  • 236 See, e.g., 15 U.S.C. § 1681c (2012).
    Google ScholarLocate open access versionFindings
  • 239 See infra pp. 206-207, Model FaTSCA, at § 4(b)-(c). Credit scorers would be permitted, for example to consider a borrower's age pursuant to the limitations already imposed by ECOA. See McCoy, supra note 237, at § 8.02[1] [a] [ii].
    Google ScholarFindings
  • 240 See infra, pp. 206-207, Model FaTSCA, at § 4(b)-(c). 241 See id. at 407, at § 4(e). 242 See id. at § 4(g). 243 See id. 244 Data scientists and lawyers have already proposed technical solutions to such problems as discrimination by proxy. One such group of experts, for example, proposes a method that could be used to "repair" training datasets at the outset to eliminate implicit bias, thereby avoiding the risk that factors like race or gender will be weighted in a final scoring model. See Michael
    Google ScholarFindings
  • 246 See infra p. 205, Model FaTSCA, at § 4(a).
    Google ScholarFindings
Author
Mikella Hurley
Mikella Hurley
Your rating :
0

 

Tags
Comments
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn
小科