最有看点的互联网金融门户

最有看点的互联网金融门户
全新的互联网金融模式国际资讯监管与政策

CFPB向网贷平台Upstart发出无异议函

近日,美国消费者金融保护局(后文简称CFPB)向在线借贷公司Upstart发出无异议函。无异议函代表一家由某一政府机构(比如CFPB)监管的公司可以在该公司提议的某些特定情况下免于执法或监管行动。CFPB的目标是针对新兴产品和服务探索消费者友善的创新做法。这份声明也是该机构首次颁布此类文件,重要性可见一斑。CFPB今年早些时候就利用非传统数据扩展贷款发放的益处与风险进行了公开咨询。

说到用替代数据担保贷款,人们第一反应想到的基本都是美国之外的公司。因为许多国家没有像美国这样的保护措施,这为使用替代数据提供了许多可能,比如进行贷款决策。替代数据包括社交数据和收集数据(比如Tala和Lenddo,这两家公司也在Lend Academy博客节目中被提及过)。

Upstart由谷歌前员工创建,现在许多员工也来自谷歌。他们脱颖而出是因为依靠在谷歌的学习利用替代数据、机器学习和人工智能来做出更好的贷款决定。他们一直使用的一些数据点包括借贷者的教育背景。

CFPB认为:

替代数据包括手机账单、租金、电子交易信息以及其他的一些与个人金融行为较远的信息。这份调查同样研究了利用新兴科技进行担保活动,比如机器学习愈加被运用到提供在贷款过程中提供新视角,改善贷款决定。

未来,Upstart将会向CFPB提供借贷和合规信息来降低消费者风险,并帮助CFPB理解替代数据在真实世界中对于借贷决策的影响。CFPB感兴趣的是如何使用替代数据向更多人以可负担的利率发放贷款。这封无异议函与《平等信贷机会法案和条例B》有关。

Upstart的无异议函申请书很有趣。以下这段是其中我觉得最有趣的,Upstart介绍了他们的担保如何为借贷者改善了利率。

为了检验将Upstart的担保模型中的变量仅限于信用报告和申请者自报的收入的影响,我们建立了一个模型("限定版模型"),这一模型中没有非传统变量。2016年9月,我们将这个限定版模型和我们的生产模型比较后发现,限定版模型推进的利率比生产模型高。具体来说,在评估同一组借贷者(2014-2016年间通过Upstart借贷的随机个人样本)时,限定版模型给出的平均年利率为23.5%,而我们的生产模型推荐年利率为16.7%。差别高达680基点。评估所有Upstart申请者时(而不只是成功通过Upstart借贷的人),限定版模型推荐的年利率比生产模型多了560个基点。限定版模型考虑的借贷者间违约率的不同较少,所以导致平均利率偏向高风险型利率。所以平均来看,我们的生产模型比起只考虑传统变量的模型确实为借贷者提供了更好的借贷条款。

我们联系了Upstart的联合创始人兼CEO Dave Girouard,以下是他的意见:

这封无异议函证明了另类担保如AI和机器学习对于加强可负担借贷提供十分重要,而这一点可以在借贷法律规定内做到。对于Upstart和整个行业来说这是一个大事件。

结论

正如Dave所说,这不仅是对Upstart也是对整个市场借贷平台行业来说是个大事件。许多公司如今都在使用机器学习来做出贷款决定。虽然CFPB的这封无异议函"仅针对该公司的特定情况和行为,并不代表支持使用任何特定变量或建模技巧",它确实标明对于探索新的担保方法来扩大借贷发放的开放态度。

The no-action letter is an indication that the CFPB is open to alternative data being used to expand access to credit.

Today, the Consumer Finance Protection Bureau (CFPB) announced a no-action letter to Upstart, an online consumer lender. A no-action letter essentially allows a company that is regulated by a government entity, in this case the CFPB, to operate under certain circumstances as proposed by the company without enforcement or supervisory action. For the CFPB the goal is to explore consumer-friendly innovations for emerging products or services. The announcement today by the CFPB is significant as it is the first no-action letter of its kind. The CFPB previously sought public feedback on the benefits and risks of using unconventional data to extend credit earlier this year.

When we talk about underwriting with alternative data we often talk about companies that operate outside of the US. Since many countries lack the protections similar to what we have in the US, it opens up the possibilities of using alternative data for many different things, including making credit decisions. Some examples include social data and mobile phone data. Two great examples of this include Tala and Lenddo which have both been featured on the Lend Academy podcast.

Upstart was founded by ex-Googlers and many of the employees hail from Google. They have differentiated themselves, leaning on their learnings at Google to use alternative data, machine learning and artificial intelligence to make better credit decisions. Some of the data points they have used historically has been using information around a borrower's education. According to the CFPB's news release:

Alternative data could include things such as bill payments for mobile phones and rent, electronic transactions such as deposits and withdrawals, and other information that may be less closely tied to a person's financial conduct. This inquiry also looked at the use of emerging technologies for underwriting, such as the expanded use of machine learning to potentially identify new insights and improve decisions in the credit process.

According to the news release, Upstart will report lending and compliance information to the CFPB to mitigate risk to consumers and aid in the Bureau's understanding of the real-world impact of alternative data on lending decision-making. The CFPB is interested in learning how using alternative data can help extend credit to more borrowers at affordable rates. The no-action letter relates to enforcement of the Equal Credit Opportunity Act and Regulation B.

The request for a no-action letter (linked at the end of this blog post) is an interesting read. Below is a passage I found particularly interesting where Upstart shares just how much their underwriting has improved rates for borrowers:

To examine the effect of limiting Upstart's underwriting model to variables derived solely from the credit report and applicant-reported income, Upstart built a model ("the limited model"), which does not rely on non-traditional variables. A September 2016 comparison between the limited model and Upstart's production model found that the limited model recommended higher interest rates than the Upstart production model. Specifically, in evaluating the same pool of borrowers (a random sampling of individuals who actually received loans through Upstart between 2014-2016), the limited model set an average APR of 23.5% while Upstart's production model recommended an average APR of 16.7%. That is a difference of 680 basis points. In comparing all Upstart applicants (as opposed to just individuals who received loans from Upstart), the limited model recommended an APR that was 560 basis points greater than the Upstart production model. The limited model explains much less of the variance in default rate across the Upstart borrower pool, resulting in a convergence toward the higher-risk mean. Thus, on average, the Upstart model provides better terms to borrowers than a model relying solely on traditional variables.

We reached out to Dave Girouard, CEO and Co-Founder of Upstart who provided this statement:

The no-action letter validates that alternative underwriting, including AI/ML, is critical to improving access to affordable credit, and that it can be done within the bounds of fair lending laws. It's a big day for Upstart and for the industry overall.

Conclusion

As Dave pointed out this is big news not just for Upstart but for all of marketplace lending. Many companies, not just Upstart. are using machine learning today to make lending decisions. While the CFPB no-action letter "is specific to the facts and circumstances of the particular company and does not serve as an endorsement of the use of any particular variables or modeling techniques", it does signify an openness in looking at new underwriting approaches that can expand access to credit.

There have been whispers in the industry over the years that Upstart may eventually fall foul of the CFPB. This no-action letter can put to rest those whispers for now. Upstart is a leader in alternative data and they are one of the larger online consumer lenders in the US. It's going to be interesting to see what this news could mean for the entire industry longer term and it's fascinating to see how much alternative data can help in underwriting borrowers.


用微信扫描可以分享至好友和朋友圈

扫描二维码或搜索微信号“iweiyangx”
关注未央网官方微信公众号,获取互联网金融领域前沿资讯。

发表评论

发表评论

您的评论提交后会进行审核,审核通过的留言会展示在下方留言区域,请耐心等待。

评论

您的个人信息不会被公开,请放心填写! 标记为的是必填项

取消

潘, 妍媛未央编辑团队

101
总文章数

TA还没写个人介绍。。。

CFPB发布现金贷上限新规,市场反映较为平稳

潘, 妍媛 11-02

谷歌、高盛跻身全球顶级区块链机构投资者行列

Arjun Khar... 10-20

美国监管机构CFPB重拳整治发薪贷、车贷和“气球贷”

高旭 10-13

拓展支付能力,谷歌有意收购HTC?

Steven And... 09-14

发薪日贷款新规出台,CFPB监管新动作引发关注

常笑 09-01

版权所有 © 清华大学五道口金融学院互联网金融实验室 | 京ICP备17044750号-1