SWUFE IFS Thinking Fast Not Slow Evidence from

SWUFE, IFS Thinking Fast, Not Slow: Evidence from Peer-to-Peer Lending Li Liao, Tsinghua University Zhengwei Wang, Tsinghua University Jia Xiang, Tsinghua University Jun Yang, Indiana University 1

Research questions � Do investors who make decisions in a very short time (under fast-thinking) tend to make mistakes? ◦ We answer such a question in a setting of P 2 P lending in China; ◦ The bidding process of a credit loan typically completes in a few minutes; ◦ All investors observe the characteristics of the loan and borrower, as well as the bidding process in real time; � Does paying much attention to interest rates lead to better or worse outcomes: higher or lower return (abnormal IRR) and default? � Do past experiences help mitigate investors’ tendency in making such mistakes under fast-thinking? 2

Figure 1. Sample of loans listed on the Renrendai. com (1)

Figure 1. Sample of loans listed on the Renrendai. com (2)

Figure 2. Procedure for borrowing and repaying on Renrendai

Related literature (1) � Information contents in P 2 P lending (Prosper. com) ◦ Rational herding (Zhang and Liu 2012); ◦ Usage of soft and non-standard information for screening (Iyer et al. 2009); ◦ Certification of friendship network (Lin et al. 2013); ◦ Information in voluntary disclosure (Michels 2012); � Behavioral ◦ ◦ biases in P 2 P lending Home bias (Lin and Viswanathan 2015); Racial bias (Pope and Sydnor 2011); Beauty premium (Revina 2008); Trust in P 2 P lending decisions (Duarte et al. 2012).

Related literature (2) � Cognitive biases in psychology and economics ◦ Kahneman and Tversky (1974), Kahneman (2011); �“Dual-system” – “System 1” and “System 2” thinking: �System 1 thinking is fast, automatic, involuntary, and often unconscious; �System 2 thinking is slow and effortful, and tends to be more rational. ◦ Behavioral intervention and education programs help young people slow down and reflect on their automatic thoughts reduces the rates of arrests and readmission to jail, and improves school engagement and graduation rates (Heller et al. 2017); ◦ In experimental settings, faster thinking is associated with greater risktaking (Cella et al. 2007; De. Donno and Demaree 2008; and Candler and Pronin 2012). 7

Related literature (3) � Limited attention in financial markets ◦ Investors pay attention to more salient details (Benartzi and Lehrer 2015); ◦ Individual investors are net buyers of attention grabbing stocks (Barber and Odean 2008); ◦ Google search frequency is associated with the attention of retail investors and predicts stock prices in the next 2 weeks (Da et al. 2011); ◦ Investor reactions to a firm’s earnings surprise are much weaker and post -announcement drift is much stronger when more firms announce earnings in the same day (Hirshleifer et al. 2009); � Individual investors in the stock market learn about their ability through trading ◦ High-ability investors tend to trade more, and low-ability ones stop trading (Seru et al. 2010). 8

Data (1) Renrendai. com: 2012. 09— 2014. 12 (data extracted in March, 2016); ◦ Renrendai（人人贷） was founded in May 2010, and the cumulative trading volume reached 23. 0 billion CNY by November 2016; Borrowers provide: ◦ The average rate of return is 11. 58% in the year 2015; ◦ Loan amounts range from 3, 000 to 500, 000 CNY; ◦ There are eight repayment terms for credit loans: 3 months, 6 months, 9 months, 12 months, 15 months, 18 months, 24 months, and 36 months. Lenders submit: ◦ The bidding amount is multiples of 50 CNY, and the minimum bidding amount is 50 CNY; ◦ Once the requested amount on a listing is fully funded (in 7 days), the loan is complete and the funding process stops. During the bidding process, the time and amount of bids are posted online and investors have access to the bidding information in real-time. 9

Data (2) � From September 1, 2012 to December 31, 2014, there are in total 270, 929 credit loan applications at Renrendai; � Out of which 95. 50% fail in the audit process and get withdrawn, not seen by investors; � Nearly all loan applications that pass the audit get fullyfunded; � Out of the 11, 897 fully funded loans, 10, 385 have been paid off or defaulted (as of March, 2016). 10

Data (3) Renrendai guarantees the repayment of the principal. � 90% of Renrendai loans are fully funded in under eight minutes, with the 25 th percentile at 42 seconds and the 75 th percentile at 180 seconds; � Fee charged to borrowers by Renrendai: ◦ 0 -5% upfront, depending on credit rating; ◦ 0. 1 -0. 35% monthly on balance, depending on credit rating; ◦ 1% penalty for pre-payment. 11

Figure 3. Probability density distribution of Ln(Duration) from 2012 to 2014

Loan characteristics and market conditions Variable Obs. Mean S. D. p 1 p 25 p 50 p 75 p 99 Duration (Seconds) 10, 385 290. 930 1, 581. 505 4 42 80 180 2, 972 Default 10, 385 0. 175 0. 380 0 0 1 Interest (%) 10, 385 12. 704 2. 204 10 11 12 13 20 Amount (¥) 10, 385 3, 000 8, 000 Term (Months) 10, 385 10. 301 7. 076 3 6 9 12 36 Rm 10, 385 0. 037 0. 068 -0. 117 -0. 004 0. 025 0. 063 0. 247 Rf (%) 10, 385 2. 924 0. 355 2. 75 2. 8 3 4. 25 25, 371. 750 39, 667. 600 14, 000 27, 000 200, 000

Borrower characteristics Variable Obs. Mean S. D. p 1 p 25 p 50 p 75 p 99 HR 10, 385 0. 712 0. 453 0 0 1 1 1 Male 10, 385 0. 873 0. 333 0 1 1 Age 10, 385 32. 889 7. 024 23 28 32 37 52 Bachelor 10, 385 0. 298 0. 457 0 0 0 1 1 Master. Or. Above 10, 385 0. 023 0. 151 0 0 1 Employ(3– 5 yrs) 10, 385 0. 220 0. 414 0 0 1 Employ(5 yrs+) 10, 385 0. 347 0. 476 0 0 0 1 1 Income(¥ 5, 000– 10, 000) 10, 385 0. 267 0. 442 0 0 0 1 1 Income(¥ 10, 000– 20, 000) 10, 385 0. 140 0. 348 0 0 1 Income(¥ 20, 000– 50, 000) 10, 385 0. 143 0. 350 0 0 1 Income(¥ 50, 000+) 10, 385 0. 130 0. 336 0 0 1 Ln(Income/debt) 10, 385 1. 233 1. 038 0 0 1. 339 1. 929 3. 773 House 10, 385 0. 555 0. 497 0 0 1 1 1 Mortgage 10, 385 0. 217 0. 412 0 0 1 Car 10, 385 0. 408 0. 492 0 0 0 1 1 Car. Loan 10, 385 0. 080 0. 272 0 0 1

Characteristics of fast loans vs. other loans Difference t-statistics p-value 28. 236 0. 000 2. 253 0. 024 -23. 572 0. 000 Fast Loans 13. 711 10. 568 9. 276 Other Loans 12. 360 10. 209 9. 752 Rm Rf (%) 0. 019 2. 949 0. 043 2. 916 -15. 379 4. 074 0. 000 HR Male Age Bachelor Master. Or. Higher Employ(3– 5 yrs) Employ(5 yrs+) Income(¥ 5, 000– 10, 000) Income(¥ 10, 000– 20, 000) Income(¥ 20, 000– 50, 000) Income(¥ 50, 000+) Ln(Income/Debt) 0. 787 0. 882 31. 313 0. 291 0. 022 0. 208 0. 304 0. 310 0. 127 0. 095 0. 061 1. 163 0. 686 0. 869 33. 427 0. 301 0. 024 0. 224 0. 361 0. 253 0. 145 0. 160 0. 153 1. 257 10. 030 1. 745 -13. 479 -0. 960 -0. 689 -1. 686 -5. 275 5. 749 -2. 234 -8. 268 -12. 327 -4. 026 0. 000 0. 081 0. 000 0. 337 0. 491 0. 092 0. 000 0. 026 0. 000 Interest Rate (%) Term (months) Ln(Amount) (¥)

Empirical results Greater sensitivity of the bidding duration to interest rate in the fastthinking group; � Loans in the fast-thinking group have a lower return and higher default rate; � Abnormal IRR and interest rate are negatively correlated in the fastthinking group, and positively correlated in the slow-thinking group; � Investors learn to avoid participating in the fast-thinking group; � Introduction of mobile app in July, 2014: � ◦ More noticeable interest rate on a smaller screen; ◦ Biding duration becomes even more sensitive to interest rate – chasing for high yield; � Two experiments: attention and learning. 16

IRR and Abnormal IRR

Table 2. Marginal effect of Fast on the performance of loans (1) Fast Interest Ln(Amount) Term Rm Rf HR Male Age Bachelor Master. Or. Above Employ(3– 5 yrs) Employ(5 yrs+) Pre-Match Abnormal. IRR -0. 231*** (-5. 264) 0. 033*** (3. 463) 0. 397*** (10. 617) -0. 123*** (-10. 919) -0. 084*** (-32. 163) 0. 092 (0. 426) -0. 591*** (-18. 282) -0. 070 (-1. 616) -0. 006** (-2. 517) 0. 082** (2. 425) 0. 284*** (2. 844) -0. 034 (-0. 844) -0. 093** (-2. 389) Default 0. 031*** (7. 830) 0. 005*** (5. 283) 0. 036*** (10. 469) 0. 025*** (24. 751) 0. 003*** (11. 295) -0. 181*** (-9. 183) 0. 294*** (99. 884) 0. 018*** (4. 645) 0. 002*** (8. 443) -0. 031*** (-10. 174) -0. 063*** (-6. 914) -0. 014*** (-3. 851) -0. 010*** (-2. 679) Post-Match Abnormal. IRR Default -0. 176*** 0. 024*** (-2. 810) (4. 444) 0. 010 0. 009*** (0. 696) (7. 204) -0. 030 0. 072*** (-0. 496) (13. 737) -0. 210*** 0. 028*** (-10. 227) (16. 089) -0. 088*** 0. 004*** (-13. 645) (6. 381) 2. 583*** -0. 284*** (6. 507) (-8. 352) -0. 968*** 0. 276*** (-12. 768) (42. 500) -0. 334*** 0. 039*** (-3. 687) (5. 028) -0. 045*** 0. 006*** (-8. 156) (12. 860) 0. 317*** -0. 088*** (4. 637) (-15. 007) 0. 722*** -0. 069*** (3. 519) (-3. 951) 0. 236*** -0. 037*** (3. 196) (-5. 828) -0. 171** -0. 017*** (-2. 291) (-2. 658)

Table 2. Marginal effect of Fast on the performance of loans (2) Income(¥ 5, 000– 10, 000) Income(¥ 10, 000– 20, 000) Income(¥ 20, 000– 50, 000) Income(¥ 50, 000+) Ln(Income/debt) House Mortgage Car. Loan Constant Verification Fixed Effects Observations R-squared Pre-Match Abnormal. IRR 1. 073*** (11. 228) 0. 173*** (2. 995) 0. 016 (0. 222) -0. 220** (-2. 232) 0. 446*** (8. 335) -0. 065* (-1. 862) 0. 324*** (8. 013) 0. 005 (0. 146) -0. 046 (-0. 816) -4. 182*** (-5. 854) YES 62, 054 0. 054 Default 0. 005 (0. 612) -0. 026*** (-4. 900) -0. 010 (-1. 527) -0. 033*** (-3. 725) 0. 023*** (4. 684) -0. 003 (-1. 024) -0. 039*** (-10. 707) 0. 009*** (2. 654) 0. 027*** (5. 373) -0. 200*** (-3. 077) YES 62, 054 0. 253 Post-Match Abnormal. IRR Default -0. 804*** 0. 103*** (-5. 359) (8. 044) 0. 751*** -0. 056*** (7. 375) (-6. 432) 0. 814*** -0. 043*** (6. 100) (-3. 766) 1. 167*** -0. 115*** (6. 738) (-7. 753) -0. 521*** 0. 065*** (-6. 245) (9. 077) -0. 282*** 0. 015** (-3. 695) (2. 300) 0. 457*** -0. 046*** (5. 480) (-6. 435) 0. 481*** 0. 011 (5. 033) (1. 306) -0. 691*** 0. 033*** (-5. 459) (3. 003) -3. 605*** -0. 645*** (-2. 882) (-6. 011) YES 24, 290 0. 275 24, 290 0. 422

Introduction of quantile regression � Loss Function OLS Regression Quantile Regression Squared Loss Absolute Loss

Table 3. Marginal effect of Interest on Ln(Duration) at different quantile of Ln(Duration) Interest Ln(Amount) Term Rm Rf HR Male Age Bachelor Master. Or. Above Employ(3– 5 yrs) Employ(5 yrs+) OLS Regression -0. 202*** (-40. 133) 0. 502*** (20. 876) 0. 073*** (12. 720) 2. 572*** (14. 237) -1. 109*** (-10. 439) -0. 117*** (-4. 260) -0. 023 (-0. 783) 0. 004** (2. 193) -0. 051** (-2. 263) -0. 208*** (-3. 195) -0. 007 (-0. 263) 0. 023 (0. 936) 0. 05 Quantile -0. 245*** (-21. 447) 0. 511*** (9. 303) 0. 029** (2. 422) 1. 449*** (3. 548) -0. 274 (-1. 297) -0. 020 (-0. 370) -0. 058 (-1. 009) 0. 002 (0. 572) -0. 030 (-0. 656) -0. 187 (-1. 528) -0. 023 (-0. 454) -0. 012 (-0. 232) Ln(Duration) Quantile Regression 0. 25 quantile 0. 75 Quantile -0. 180*** -0. 185*** (-39. 821) (-26. 446) 0. 383*** 0. 528*** (17. 576) (16. 584) 0. 034*** 0. 093*** (6. 494) (12. 977) 3. 180*** 2. 930*** (19. 307) (12. 214) -0. 417*** -1. 589*** (-4. 293) (-11. 811) -0. 036 -0. 174*** (-1. 463) (-4. 915) -0. 023 -0. 018 (-0. 865) (-0. 483) 0. 005*** 0. 003 (3. 515) (1. 171) -0. 015 -0. 077*** (-0. 739) (-2. 699) -0. 058 -0. 280*** (-1. 000) (-3. 423) -0. 015 0. 006 (-0. 654) (0. 196) -0. 002 0. 028 (-0. 106) (0. 865) 0. 95 Quantile -0. 159*** (-10. 340) 0. 597*** (9. 876) 0. 134*** (10. 064) 2. 241*** (4. 976) -2. 214*** (-9. 002) -0. 299*** (-4. 306) -0. 036 (-0. 519) 0. 001 (0. 323) -0. 072 (-1. 313) -0. 365** (-2. 403) -0. 095 (-1. 505) 0. 031 (0. 515)

Table 3. Marginal effect of Interest on Ln(Duration) at different quantile of Ln(Duration) (2) Income(¥ 5, 000– 10, 000) Income(¥ 10, 000– 20, 000) Income(¥ 20, 000– 50, 000) Income(¥ 50, 000+) Ln(Income/debt) House Mortgage Car. Loan Constant Verification Fixed Effects Day-of-Week Fixed Effects Hour-of-Day Fixed Effects Number of Observations Adjusted/Pseudo- R 2 OLS Regression 0. 105* (1. 699) 0. 174** (2. 201) 0. 262*** (2. 749) 0. 329*** (2. 790) -0. 092*** (-2. 779) -0. 032 (-1. 255) -0. 085*** (-3. 119) -0. 020 (-0. 685) -0. 012 (-0. 314) 7. 254*** (16. 715) YES YES 0. 05 Quantile -0. 084 (-0. 578) -0. 161 (-0. 866) -0. 144 (-0. 648) -0. 282 (-1. 022) 0. 051 (0. 665) 0. 010 (0. 186) -0. 100* (-1. 865) -0. 040 (-0. 693) -0. 006 (-0. 086) 4. 051*** (4. 585) YES YES 10, 385 0. 399 10, 385 0. 207 Ln(Duration) Quantile Regression 0. 25 quantile 0. 75 Quantile 0. 120** 0. 022 (2. 116) (0. 273) 0. 162** 0. 096 (2. 224) (0. 945) 0. 219** 0. 169 (2. 495) (1. 372) 0. 315*** 0. 234 (2. 889) (1. 524) -0. 090*** -0. 030 (-2. 968) (-0. 699) -0. 048** -0. 030 (-2. 075) (-0. 912) -0. 034 -0. 074** (-1. 368) (-2. 152) -0. 005 -0. 001 (-0. 185) (-0. 021) 0. 018 -0. 011 (0. 514) (-0. 226) 5. 791*** 8. 858*** (14. 615) (15. 995) YES YES YES 10, 385 0. 220 10, 385 0. 269 0. 95 Quantile 0. 319** (2. 157) 0. 447** (2. 338) 0. 694*** (3. 028) 0. 857*** (2. 962) -0. 243*** (-3. 001) -0. 081 (-1. 306) -0. 142** (-2. 167) -0. 011 (-0. 151) -0. 034 (-0. 378) 10. 220*** (9. 609) YES YES 10, 385 0. 295

Figure 3. Marginal effect of Interest on Ln(Duration) at different quantiles of Ln(Duration)

Additional analyses � Investor learning (Table 4) � Introduction of the mobile app (Table 5) � Two experiments (Tables 6 and 7) 24

Table 4. Summary statistics Panel A. Entire sample Variable Obs. Bidt Accu. Bidst Months. From. Firstbidt (Months) Monthly. Bid. Amountt-1 (¥) Ln(1+Monthly. Bid. Amount)t-1 (¥) Weighted. Average. Returnt-1 (%) Proportion. Defaultt-1 (%) Activeness (%) 288, 944 288, 944 Mean S. D. p 1 p 25 p 50 p 75 p 99 0. 176 0. 381 1 0 0 0 1 25. 939 66. 368 1 2 5 19 439 12. 348 9. 327 0 4 10 18 34 944. 249 9, 461. 214 0 0 56, 950 1. 590 2. 962 0 0 9. 793 8. 964 6. 327 0 0 12. 524 13. 803 16. 396 5. 074 17. 371 0 0 100 67. 191 30. 479 13. 693 38. 372 77. 778 100

Table 4. Summary statistics Sub-sample (including only those months in which investors lend) Variable Obs. Mean S. D. ΔLn(Duration)t (Seconds) 50, 420 0. 327 1. 003 Proportion. Fast. Bidst (%) 50, 420 16. 000 30. 988 0 Accu. Bidst 50, 420 46. 893 100. 941 8. 329 Months. From. Firstbidt (Months) 50, 420 9. 341 p 25 p 50 p 75 p 99 0. 211 0. 882 3. 357 0 0 16. 667 100 2 5 13 39 562 1 3 7 13 35 -1. 709 -0. 343 Monthly. Bid. Amountt-1(¥) 50, 420 3, 764. 13 21, 374. 89 0 0 300 1, 850 57, 000 Monthly. Bid. Amountt(¥) 50, 420 3, 999. 51 19350. 54 50 200 700 2, 450 53, 800 Ln(1+Monthly. Bid. Amount)t-1 (¥) 50, 420 4. 549 3. 642 0 0 5. 707 7. 523 10. 951 Weighted. Average. Returnt-1 (%) 50, 420 11. 109 4. 861 0 11 13. 000 13. 981 16. 532 Proportion. Defaultt-1 (%) 50, 420 5. 189 14. 696 0 0 0 3. 922 100 50, 420 77. 459 26. 746 100 Activeness (%) 12. 828 60. 606 87. 901

Introduction of learning models �

Table 4. Learning models with individual heterogeneity and survival controls Panel B. Experience measured by the total number of loans on Renrendai(1) Simple learning model Dependent variable Accu. Bidst Accu. Bids. Squaredt Ln(Monthly. Bid. Amount)t-1 Weighted. Average. Returnt-1 Proportion. Defaultt-1 Ln(Duration)t Learning model with heterogeneity Proportion Ln(Duration)t Fast. Bidst 2 nd Stage Bidt= 1 ΔLn(Duration)t ΔProportion Fast. Bidst 3. 01 e-4* -0. 004 4. 917 e-4** -0. 001 0. 013*** 0. 006*** -0. 188*** (1. 655) (-0. 689) (1. 971) (-0. 106) (65. 337) (3. 635) (-3. 157) -1. 55 e-6** 1. 94 e-5 -1. 66 e-06* 3. 65 e-6 -3. 59 e-5*** -1. 43 e-5** 4. 38 e-4* (-2. 376) (0. 864) (-1. 762) (0. 132) (-50. 567) (-2. 150) (1. 948) -0. 028*** 0. 450*** -0. 022*** 0. 391*** 0. 163*** 0. 005*** -0. 019 (-26. 405) (12. 576) (-20. 118) (10. 685) (161. 846) (2. 986) (-0. 372) -0. 006*** 0. 159*** -0. 005*** 0. 153*** 0. 031*** 0. 003 -0. 081 (-6. 343) (5. 175) (-4. 814) (4. 826) (46. 103) (1. 267) (-1. 085) -4. 51 e-4* -0. 001*** 0. 001 -0. 002*** 0. 049** (-1. 811) (-0. 062) (-2. 582) (0. 149) (-4. 157) (2. 471) 0. 467*** (18. 967) 1. 412* (1. 669) 0. 393*** (15. 805) 2. 029** (2. 382) (-9. 884) 0. 008*** (57. 336) -2. 716*** (-96. 582) Activeness Constant Selection models (1 st Stage)

Table 4. Learning models with individual heterogeneity and survival controls Panel B. Experience measured by the total number of loans on Renrendai (2) Simple learning model Dependent variable Observations R-squared Individual fixed effects Time fixed effects Log likelihood Joint test of ρt =0 (t=3, …, 28) F(25, 42569) Pr > F Ln(Duration)t 50, 420 0. 337 No Yes Proportion Fast. Bidst 50, 420 0. 216 No Yes Learning model with heterogeneity Ln(Duration)t 50, 420 0. 337 Yes Proportion Fast. Bidst 50, 420 0. 216 Yes Selection models (1 st Stage) 2 nd Stage Bidt= 1 ΔLn(Duration )t ΔProportion Fast. Bidst Control 42, 600 0. 138 42, 600 0. 079 268. 37 0. 000 144. 40 0. 000 288, 944 0. 251 Yes -100800. 73

Table 4. Learning models with individual heterogeneity and survival controls Panel C. Experience measured by the # of months lapsed from first investment at Renrendai (1) Simple learning model Dependent variable Months. From. Firstbidt Months. From. First. Bid. Squaredt Ln(Monthly. Bid. Amount)t-1 Weighted. Average. Return t-1 Proportion. Defaultt-1 Ln(Duration)t Learning model with heterogeneity Proportion Ln(Duration)t Fast. Bidst Selection models (1 st Stage) 2 nd Stage Bidt= 1 ΔLn(Duration)t ΔProportion Fast. Bidst 0. 008*** -0. 043 0. 011*** -0. 035 0. 009*** 0. 302*** -6. 011*** (5. 712) (-0. 883) (7. 547) (-0. 709) (4. 708) (18. 510) (-10. 842) -1. 27 e-4*** -0. 002 -1. 537 e-4*** -0. 003** -1. 818 e-4*** -0. 002*** 0. 050*** (-2. 801) (-1. 563) (-3. 251) (-2. 024) (-4. 581) (-6. 893) (4. 110) -0. 026*** 0. 419*** -0. 018*** 0. 351*** 0. 182*** -0. 006*** 0. 175*** (-24. 867) (11. 739) (-16. 757) (9. 460) (180. 470) (-3. 486) (3. 023) -0. 008*** 0. 193*** -0. 007*** 0. 186*** 0. 040*** -0. 004 0. 035 (-8. 669) (6. 210) (-7. 223) (5. 848) (60. 611) (-1. 593) (0. 466) -1. 817 e-4 -0. 005 -3. 743 e-4 -0. 003 -0. 002*** 0. 035* (-0. 728) (-0. 586) (-1. 452) (-0. 353) (-12. 439) (-2. 982) (1. 764) 0. 003*** Activeness (9. 443) Constant 0. 436*** 1. 601* 0. 246*** 2. 037** -2. 136*** (17. 329) (1. 850) (2. 619) (2. 341) (-46. 891)

Introduction of the mobile app � Renrendai launched its mobile app on July 30, 2014; � The screen of a mobile phone is much smaller; � Interest rate is much more salient: ◦ Located near the top ◦ In the middle of the screen, and ◦ Shown in orange font (only information not in black) � Credit rating not shown on the screen; � The updated status such as “ 99% funded” also attracts investor attention. 32

Figure 4. A sample loan listed on Renrendai mobile app

Table 5. Sensitivity of duration to interest rate under fast thinking around mobile app introduction Interest Rate× 61 -90 days. Before. Mobile. APP Interest Rate× 31 -60 days. Before. Mobile. APP Interest Rate× 1 -30 days. After. Mobile. APP Interest Rate× 31 -60 days. After. Mobile. APP Interest Rate× 61 -90 days. After. Mobile. APP Interest Rate×Over 90 days. After. Mobile. APP 0. 1 Quantile -0. 196*** (-23. 586) 0. 032 (1. 219) 0. 015 (0. 579) -0. 035 (-1. 300) -0. 084*** (-3. 311) -0. 107*** (-3. 430) 0. 028 (0. 823) -0. 004 (-0. 234) Ln(Funding Duration) 0. 15 quantile 0. 20 Quantile -0. 187*** -0. 180*** (-25. 282) (-23. 312) 0. 023 0. 017 (1. 009) (0. 721) 0. 011 0. 006 (0. 481) (0. 261) -0. 027 -0. 021 (-1. 109) (-0. 814) -0. 079*** -0. 072*** (-3. 532) (-3. 078) -0. 092*** -0. 120*** (-3. 310) (-4. 141) 0. 026 0. 015 (0. 848) (0. 456) -0. 012 -0. 036** (-0. 861) (-2. 385) 0. 25 Quantile -0. 184*** (-25. 570) -0. 001 (-0. 036) 0. 014 (0. 660) -0. 021 (-0. 880) -0. 056** (-2. 576) -0. 115*** (-4. 287) 0. 028 (0. 950) -0. 049*** (-3. 503)

Experiment on attention � Experiment conducted in Jan. 2017 at PBCSF, Tsinghua University; � 60 participants are divided into two groups of 30 (by even or odd student ID); � Trained together for 30 minutes using 50 loans randomly drawn from loans listed during 10/4/2014 and 11/3/2013, 30 days before a random date near the middle of our sample period, 11/4/2013; ◦ Given loan and borrower characteristics, and loan performance; ◦ No communications allowed among participants. � Fast-thinking group was required to choose one out of five loans within 42 seconds; � Slow-thinking group had more time to select one loan from the same five loans; � Investment was conducted separately for two groups. 35

Table 6. Experiment on attention: Loan characteristics under fast vs. slow thinking Interest Rate (%) Term (Months) Amount (¥) HR Fast thinking 16. 267 14. 9 9, 000 0. 267 Slow thinking 15. 067 14. 1 7, 700 0. 067 Difference t-statistics p-value 3. 016 0. 004 0. 316 0. 753 1. 573 0. 121 2. 121 0. 038

Experiment on learning � Experiment conducted in June 2017 at PBCSF, Tsinghua University; � 60 different participants divided randomly into two groups of 30 (by even or odd student ID); � Trained together for 30 minutes using 50 loans randomly drawn from loans listed during 10/4/2014 and 11/3/2013; � In Round 1, all participants select one out of five given loans from 11/4/2013 to invest; � Group 1 participants were surveyed before knowing selected loan performance, and before round 2 investment; � Group 2 participants were surveyed after round 2 investment. 37

Figure 5. A screen shot of survey questions from Survey Monkey

Table 7. Experiment on learning (1) Panel A. The difference in the percentage of factor selected between Groups 1 and 2 Interest Rate Credit Rating Term Amount Others Intuition Score Group 1 50. 00% 23. 30% 16. 70% 3. 30% 5. 47 Group 2 13. 30% 66. 70% 10. 00% 4. 07 t-statistics 3. 266 -3. 685 0. 75 -0. 46 1. 000 3. 55 p-value 0. 002 0. 001 0. 456 0. 647 0. 332 0. 001 Panel B. The difference in decision time between the first and second rounds Groups 1 and 2 Group 1 Group 2 First Round 112. 833 113. 3 112. 367 Second Round 97. 033 102 92. 067 t-value 2. 474 1. 161 2. 430 p-value 0. 015 0. 250 0. 018

Table 7. Experiment on learning (2) Panel C. The impact of default on the investing style of investors Default Constant Observations R-squared Ln(Decision Time) Groups 1 and 2 0. 564*** 0. 488*** (4. 948) (3. 006) 4. 164*** 4. 100*** (50. 795) (30. 957) 60 30 0. 297 0. 244 Interest Rate Credit Rating Group 2 -0. 400*** 0. 550*** (-3. 528) (3. 485) 0. 400*** 0. 300** (4. 320) (2. 328) 30 30 0. 308 0. 302 Intuition Score -1. 950*** (-3. 503) 5. 700*** (12. 540) 30 0. 305

Conclusion � We examine behavioral biases in decision making under fast -thinking using a unique setting of P 2 P lending in China – Renrendai (人人贷); � Investors tend to over-weigh interest rates and under-weigh risk under fast-thinking, which results in lower abnormal IRR and higher default rate; � Mobile app exacerbates the tendency of chasing for yield; � More experienced investors are less prone to making such mistakes. Your comments and suggestions are greatly appreciated. 41