金融科技新纪元：数字足迹驱动的信用评估

需积分: 3 39 浏览量更新于2024-07-09 收藏 1.12MB PDF 举报

"这篇研究论文探讨了金融科技（FinTech）领域的一个新兴趋势——利用数字足迹进行信用评分。通过对超过250,000个观察值的分析，研究发现，消费者的数字足迹，如网络浏览历史和注册信息，可以提供与传统征信机构相当甚至更丰富的信用评估信息。此外，这种基于数字足迹的信用评分方法在区分潜在违约者方面的效果与传统信用评分卡相近。论文指出，这些发现可能对金融中介机构的运营模式、未开户人群的信贷接入，以及数字空间内消费者、企业和监管机构的行为产生深远影响。" 在金融科技领域，信用评估是至关重要的环节，传统方式依赖于消费者的信用历史、收入证明等信息。然而，随着互联网和移动设备的普及，人们的数字足迹变得日益丰富，这为信用评估提供了新的途径。这篇由Tobias Berg、Valentin Burg、Ana Gombovi和Manju Puri合作完成的NBER工作论文（Working Paper 24551）揭示了这个现象。研究显示，即使是最简单的数字足迹变量，如网页访问记录和注册信息，也能有效地预测消费者的信用风险。这暗示着金融机构可以利用这些数据来更精确地评估客户的信用状况，特别是对于那些缺乏传统信用记录的人群，如年轻消费者或未开户群体。这种方法可能降低信贷准入门槛，增加金融包容性，帮助更多人获得金融服务。同时，这种信用评分方法的效率和准确性与传统的信用评分卡不相上下，意味着金融机构可以考虑将其纳入信用决策过程，优化风险管理和贷款审批流程。这对于金融中介机构的业务模型创新具有重大意义，可能推动他们更加重视数据驱动的决策策略。此外，论文的结论也对消费者行为产生了影响，因为个人在数字世界的行为将直接影响其信用评估。企业，尤其是金融科技公司，可能会调整其产品和服务，以更好地利用这些数据。而监管机构则需要面对新的挑战，如何在保护消费者隐私和促进金融创新之间找到平衡，制定相应的法规和政策。总结起来，这篇研究论文揭示了数字足迹在信用评分中的潜力，预示着金融科技的未来趋势，即利用大数据和非传统信息来源进行信用评估，这对于整个金融行业来说是一个革命性的变化。这一发现不仅可能改变金融服务的提供方式，也可能影响社会经济的多个层面，包括消费者的信贷机会、企业的市场策略以及监管框架的发展。

The rest of the paper is structured as follows. Section 2 provides an overview about the institutional

setup and data. Section 3 provides empirical results. Section 4 discusses further implications of our

findings. Section 5 concludes.

2. Institutional setup, descriptive statistics, and the digital footprint

2.1 Institutional setup

We access data about 270,399 purchases from an E-commerce company selling furniture in

Germany (similar to “Wayfair” in the U.S.) between October 2015 and December 2016. Before purchasing

an item, a customer needs to register using his or her name, address and email. Judging the creditworthiness

of its customers is important because goods are shipped first and paid later.

The claims in our data set are

therefore akin to a short-term consumer loan.

The company uses information from two private credit bureaus to decide whether customers have a

sufficient creditworthiness. The first credit bureau provides basic information such as whether the customer

exists and whether the customer is currently or has been recently in bankruptcy. This score is used to

screen out customers with fraudulent data as well as customers with clearly negative information.

The

second credit bureau score draws upon credit history data from various banks (credit card debt and loans

outstanding, past payment behavior, number of bank accounts and credit cards), sociodemographic data, as

well as payment behavior data sourced from retail sales firms, telecommunication companies, and utilities.

This credit bureau score is requested for purchases exceeding € 100 and we consequently restrict our data

set to purchases for which the company requested a credit bureau score.

We label those customers for

whom a credit bureau exists “scorable customers”.

Customers can choose to pay upfront instead of paying after shipment of the products. Customers paying upfront

are not included in our data set. Paying after shipment, so called “deferred payment”, is by far the dominant payment

type: more than 80% of customers choose to pay after shipment if this method is offered to a customer.

The firm switched the credit bureau that provides this basic information in July 2016. Results are very similar pre-

July-2016 and post-July-2016.

The company requests the credit bureau score if the customer’s shopping cart amount exceeds € 100, even when the

customer ultimately purchases a smaller amount.

The E-commerce company uses the credit bureau score together with the digital footprint

(discussed further below) to screen out borrowers with a predicted default rate exceeding 10 percent.

Restricting our data set to orders exceeding € 100 and excluding customers with a very low

creditworthiness has the benefit of making our data set more comparable to a typical credit card, bank loan

or peer-to-peer lending data set.

After the purchase, the items are sent to the customer together with an invoice. The customer has

14 days to pay the invoice. If the customer does not pay on time, three reminders (one per email, two per

email and letter) are sent out. A customer who does not pay after three reminders is in default and the claim

is transferred to a debt collection agency, on average 3.5 months from the order date.

2.2 Descriptive statistics

Our data set comprises 270,399 purchases between October 2015 and December 2016. The credit

bureau score is available for 254,808 observations (94% of the sample) and unavailable for 15,591

observations (6% of the sample). Non-existence is due to customers being unscorable, i.e., not having a

sufficient credit history that would allow the credit bureau to calculate a credit score. In the following and

throughout the entire paper, we distinguish between scorable and unscorable borrowers, i.e. those with and

without a credit bureau score. As shown in Figure 1a, the purchases are distributed roughly even over time

with slight increases in orders during October and November, as typical for the dark season. Table 2

provides descriptive statistics for both subsamples, variable descriptions are in Table 1.

[Table 1 and Table 2, Figure 1a, Figure 1b, Figure 2]

In the sample with credit bureau scores, the average purchase volume is EUR 318 (approximately

USD 350) and the mean customer age is 45.06 years. On average, 0.9% of customers default on their

payment. Our default definition comprises claims that have been transferred to a debt collection agency.

The credit bureau score ranges from 0 (worst) to 100 (best). It is highly skewed with 99% of the

The average time between the order date and the date a claim is transferred to the debt collection agency is 103

days in our sample, i.e., approximately 3.5 months.

observations ranging between 90 and 100. The average credit bureau score is 98.11, the median is 98.86.

Figure 2 provides the distribution of credit bureau scores together with (smoothed) default rates. The

average credit bureau score of 98.11 corresponds to a default rate of approximately 1% and default rates

grow exponentially when credit bureau scores decrease, with a credit bureau of 95 corresponding to a 2%

default rate and a credit bureau of 90 corresponding to a 5% default rate. Note that default rates are not

annualized but constitute default rates over a shorter window of approximately 3.5 months.

Descriptive statistics for the sample without credit bureau score are similar with respect to order

amount and gender, with age being somewhat lower (consistent with the idea that it takes time to build up a

credit history) and default rates being significantly higher (2.5%).

2.3 Representativeness of data set

Our data set is largely representative of the geographic distribution of the German population

overall. As can be seen from Figure 1b, the share of observations in our sample closely follows the

population share for all the 16 German states. Furthermore, the mean customer age is 45.06 years,

comparable both to the mean age of 43.77 in the German population as well as to the mean age of 45.24

reported by Berg, Puri, and Rocholl (2017) in a sample of more than 200,000 consumer loans at a large

German private bank. Our sample is restricted to customers of legal age (18 years and older) and less than

5% of the customers are older than 70. The age distribution in our sample therefore resembles the age

distribution of the German population aged 18-70: the interquartile range of the German population aged

18-70 ranges from 31-56, compared to an interquartile range of 34-54 in our sample.

The average default rate in our sample is 1.0% (0.9% for scorable customers, 2.5% for unscorable

customers). As discussed above, these default rates constitute default rates over a window of approximately

4 months, implying a scaled-up annualized default rate of 3.0%. We compare our default rate to other

studies in Appendix Table A.1. Berg, Puri, and Rocholl (2017) report an average default rate of 2.5% in a

sample of more than 200,000 consumer loans at a large German private bank; the major German credit

bureau reports an average default rate of 2.4% (2015) and 2.2% (2016) in a sample of more than 17 million

剩余50页未读，继续阅读

weixin_38610277

粉丝: 8
资源: 906

金融科技新纪元：数字足迹驱动的信用评估

20210527_叶鑫_论文展示1

碳足迹管理平台E-R图

城市水足迹研究对作物水足迹有什么借鉴意义

解释“区域尺度水足迹空间异质性解析”研究的必要性

全球遥感水足迹的研究方向有哪些创新点

用科技词汇润色下句：区域耗水-产量耦合的粮食生产水足迹遥感模型构建及调控效用评价

遥感水足迹模型研究应该怎么做

扩写此项目特色：构建区域水足迹与实体水耦合调控机制，为分析农业生产中水资源利用过程提供全局整体视角

uni-app编写个人历史浏览足迹如何实现，一般缓存多少条历史记录

最新资源