Herding in P2P Lending Market: Rational Inference or Irrational Trust? Pei Ping, Zhang Ke Department of Finance and Insurance Business School Nanjing University 2016/5/22 1
Content Introduction Literature review Aim of study Methodology Empirical results Conclusion 2016/5/22 2
Introduction Feature of P2P lending Massive lenders Social aspect Unprofessional lenders Advantages of analyzing herding in P2P Controlled variable Pre-fixed price Discern rational herding from irrational 2016/5/22 3
Literature review Definition and classification of herding Herding is everyone doing what everyone else is doing, even when their private information suggests doing something quite different. Rational and irrational Herding in P2P lending market What we have known What is unknown: no credit score system, first 24 hours, auto-bid 2016/5/22 4
Aim of Study To examine lenders’ behavior in the circumstance of no widely accepted credit score system the first few hours of bidding process the condition of both auto and manual bidding Research question The existence of herding The type of herding 2016/5/22 5
Methodology Identify herding bidi,t=β1bidi, t−1+β2amounti, t−1+γXi,t+δZi+μi+ei,t bidi,t denote the number of biddings that loan i receives during its t th hour amounti, t−1 denote the amount of prior biddings that loan i has received in its ( t-1) th hour Time varying Xi,t captures the time effects. It includes: ������� �,� � � , ℎ��� �,� � � , Day-of-Week, Start-day , Month Time unvarying Zi captures the loan fixed effects. It includes: grade , term of loan , rate , overdue , no_paid , success 2016/5/22 6
Methodology Distinguish rational herding from irrational bid_mi,t=β1bid_mi,t−1+β2bid_ai,t−1+β3amount_mi,t−1+ β4amount_ai,t−1+γXi,t+δZi+μi+ei,t bid_mi,t−1 denote the number of manual biddings that loan i receives during its ( t-1) th hour bid_ai,t−1 denote the number of auto biddings that loan i receives during its ( t-1) th hour amount_mi,t−1 denote the amount of prior manual biddings that loan i has received in its ( t-1) th hour amount_ai,t−1 denote the amount of prior auto biddings that loan i has received in its ( t-1) th hour 2016/5/22 7
Empirical results Data and summary statistics We collect all the loan requests posted on Renrendai platform from October 2010 to January 2015. The initial dataset contains 454,584 loan requests. Then we excluded all loans without any bids, which are 334,377 loan requests. Our final dataset therefore includes a total of 120,207 loan requests which have received 4,856,413 biddings. It is notable that 113,718 out of 120,207 loan requests are fully funded in 24 hours after they are first posted on the platform. 2016/5/22 8
Empirical results 2016/5/22 9
Empirical results Existence of herding We find that the lag bid has a significant positive effect on bid when the impact of percent funded on bidding is controlled. The herding effect is much higher after we control the time limit effect. 2016/5/22 10
Empirical results 2016/5/22 11
Empirical results Classification of herding When control the effect of percent funded and time limit, we find both lag bid_m and lag bid_a have significant effect on bid_m. It suggests that herding in P2P market consists of both rational and irrational herding. 2016/5/22 12
Empirical results 2016/5/22 13
Empirical results Robustness check Alternative definition of bidding: bidding amount instead of bidders number GMM method to estimate dynamic formulation VIF test for multicollinearity 2016/5/22 14
Conclusion We find that lenders appear to imitate others' behavior and herding exists in the P2P market when we control for the percent funded and time limit effect. Besides rational herding, there are significant evidence that lenders would follow others' behavior blindly and ignore the information they obtain. 2016/5/22 15
Future research Different level of disclosure on multiple P2P lending platforms and the type of investor’s herding behavior. Social media, Natural Language Processing (NLP) method and investor behavior in P2P market. 2016/5/22 16
Thanks ！ 2016/5/22 17
Recommend
More recommend
