Experimental Design Details
Experimental setting – the online P2P lending platform and borrowers
We conduct the experiment on a P2P lending platform, which is one of the largest nationwide P2P lending platforms in China. The platform collects money from individual lenders via loan programs of various interest rates, minimum loan amounts, and program lengths. The platform’s borrowers can be broadly categorized into three types – individuals, business owners and firms. Programs of different interest rates, maximum loan amounts, and loan periods are designed specifically suitable for each type of borrowers. Borrowing can also happen both online and in its offline stores.
In this experiment, we focus only on individual borrowers. To apply for a loan, potential borrowers need to fill in an application form and provide required supporting documents including a copy of national identification card, household register, income certificate and marital status certificate, etc. The annual interest rate ranges between 3.96%-23.99% at the time of the experiment depending on the loan size, payment conditions and supporting documents provided.
After receiving the loan, borrowers need to repay the designated principal, interest and service charge in full amount monthly on the due date. To facilitate repayment, the platform sends reminder text messages or non-interactive voice phone calls from 2 days before the due date to the due date on a daily basis, specifying the due date and the due amount to be repaid as well as the official ways of repayment. If a borrower fails to repay the due amount on the due date, the platform keeps sending messages and also randomly assign the case to an operator to make phone calls. The operators follow the platform’s general guidelines and procedures to communicate with the delinquents and collect the overdue loan. They have some discretion on exactly how to communicate with the delinquents, but insulting, slandering, harassing the delinquent and their contact persons, and threatening to report the delinquency to the credit rating system or deducting the loan amount are strictly prohibited and punished. The delinquent also needs to pay the default penalty as the overdue amount times the penalty interest rate (usually daily 0.1%) times the number of days overdue, if she is overdue for over four days.
In this experiment, our eligible loan delinquents are selected based on the following criteria: (1) overdue loan amount is over 1000 CNY for online programs and 5000 CNY for offline programs; (2) number of days overdue is over 180 days for online programs and 450 days for offline programs. The population of eligible loan delinquents is drawn a week prior to the commencement of the experiment, together with loan information (online/offline program, program title, overdue loan amount, number of days overdue, etc.) as of when the population is drawn. We further obtain additional loan information (loan size, interest rate, loan period, loan purpose, repayment frequency, etc.) and individual information (including gender, date of birth, educational level, place of residence, marital status, annual income, occupation, etc.) after the experiment intervention starts.
Treatments and experimental procedure
We use a 1 × 3 factorial design in which we vary the financial incentive and reporting loan delinquency information. We implement the experimental manipulation in two stages, one stage of intervention lasting for one month. In the first stage, in the Control treatment, the delinquents are reminded to repay the overdue loan by above-described standard procedure. In the Deduct treatment, in addition to the standard procedure as in the Control treatment, the subjects are further informed that a proportion of the overdue amount would be deducted if they repaid the deducted overdue amount in full within one month. In the Deduct&Report treatment, in addition to informing the subjects of the deduction as in the Deduct treatment, they are further informed that, if they could not repay the deducted overdue amount in full within one month, their delinquency, including individual identification information and the overdue loan amount, would be reported to the credit score system. All the treatments are implemented by operators based on standard procedures that are carefully designed and recorded and the frequency of contact is on a daily basis.
In the second stage, for subjects who have not repaid the deducted overdue amount in full in the Deduct&Report treatment, the platform actually reports their delinquency to the social credit rating system and informs these delinquents right after the report. Subjects are also informed that the platform would update their record as “having fully repaid the loan” in the social credit rating system once the platform received their deducted repayment in full. In the Control and the Deduct&Report treatments, the implemented treatment conditions are identical to those in the first stage.
Sampling
Given most of the delinquents are difficult to reach, we choose 4500 delinquents with the smallest number of days overdue, who, according to the platform’s experience, have relatively high probability to be contactable among all eligible loan delinquents. We stratify the sample by online and offline programs, and then randomly assign all the sampled delinquents with matched overdue loan amount into these three treatments.
Operators are randomly assigned a treatment, within which they are randomly assigned to manage several cases. To keep the manipulation comparable, all operators are instructed by the same supervisor. To avoid contamination, operators assigned to different treatments come from different working divisions, who normally do not know the operators in other divisions, and neither know that they are participating in a scientific study nor are informed about the existence of the other treatments.
The key data constructed from the experiment include:
1. Individual level data on who repays the overdue loan
2. Individual level data on how much each repayment of overdue loan is repaid
3. Individual level data on when each repayment of the overdue loan is repaid