Understanding Peer Effects in Financial Decisions: Evidence from a Field Experiment

Transcription

1 Understanding Peer Effects in Financial Decisions: Evidence from a Field Experiment Leonardo Bursztyn UCLA Anderson joint with Florian Ederer, Bruno Ferman, and Noam Yuchtman CEGA Research Retreat November 2, 2012

2 Motivating Questions Do peers decisions have a causal impact on investment or consumption choices?

3 Motivating Questions Do peers decisions have a causal impact on investment or consumption choices? If so, through which channel? Social learning: a change in beliefs that results from observing another person s revealed preference Social utility: a direct effect on one s utility of another person s possession of an asset or product

4 Motivating Questions Do peers decisions have a causal impact on investment or consumption choices? If so, through which channel? Social learning: a change in beliefs that results from observing another person s revealed preference Social utility: a direct effect on one s utility of another person s possession of an asset or product Understanding channels through which a peer s choice affects one s own is especially important in finance Social learning: herding and informational cascades Banerjee (1992), Bikhchandani et al. (1992), Chari and Kehoe (2004) Social utility: e.g., keeping up with the Joneses Relative consumption exogenously (Abel, 1990; Gali, 1994; Campbell and Cochrane, 1999) or endogenously (DeMarzo et al., 2004, 2008)

5 Identification Challenges Identifying peer effects notoriously difficult (e.g., Manski, 1993) One needs: Exogenous variation in a peer s purchase of a good or asset to identify the peer effect

6 Identification Challenges Identifying peer effects notoriously difficult (e.g., Manski, 1993) One needs: Exogenous variation in a peer s purchase of a good or asset to identify the peer effect Identifying the channel through which peer effects work is even more difficult One needs to de-couple: 1. The revealed preference decision of the peer to purchase the good or asset 2. The peer s possession of the good or asset Tricky, as revealed preference and possession typically observed together

7 This Paper With a financial brokerage in Brazil, we implemented an experimental design that allows us to identify overall peer effects in the purchase of a new financial asset, and to disentangle the channels Lottery to determine whether individuals who choose to purchase an asset are actually allowed to make the investment (de-couple decision to buy from possession)

8 Outline 1. Experimental design 2. Results 3. Discussion

9 Outline 1. Experimental design 2. Results 3. Discussion

10 Experimental Design

11 Experimental Design

12 Experimental Design

13 Experimental Design

14 Experimental Design

15 Experimental Design

16 Experimental Design

17 Experimental Design

18 Experimental Design

19 Designing and Selling the Asset - Requirements (1) 1. Possibility of learning from one s peers decisions New, risky asset created by the brokerage specifically for the study: combination of an actively-managed, long/short mutual fund, and a real estate note 2. Sufficient demand for the asset that individuals wished to purchase it Attractive rate on the real estate note; pilot sales with purchase rate 50% 3. No secondary market: goal is to identify impact of learning from peers decisions to purchase the asset, rather than learning after purchase. We also do not want peer pairs to jointly make decisions about selling the asset The brokerage structured the asset as having a fixed term with no resale

20 Designing and Selling the Asset - Requirements (2) 4. Limited entry to the fund to justify a lottery to implement investors decisions to purchase the asset The brokerage limited entry into the fund, and implemented the lottery to confirm or reject purchase decisions 5. No other opportunity to purchase the asset New asset designed for the study and only chance to purchase during phone call by broker. Asset more attractive than the ones usually available to clients High stakes: investors could only invest new resources on the asset; minimum investment of R$2,000 (US$1,100); median investor salary is R$4,300 per month

21 Implementation I Between January 26 and April 3, 2012, brokers called 150 pairs of clients whom brokers had previously identified as having a social connection

22

23 Outline 1. Experimental design 2. Results 3. Discussion

24 Treatment Effects (1) Asset purchase rates of investor 2 s (conditional on investor 1 wanting to purchase the asset): Control (no peer effect) (Condition 1): 42.3%

25 Treatment Effects (1) Asset purchase rates of investor 2 s (conditional on investor 1 wanting to purchase the asset): Control (no peer effect) (Condition 1): 42.3% Learning only (Condition 2): 70.8% p-value of Condition 2 = Condition 1: 0.040

26 Treatment Effects (1) Asset purchase rates of investor 2 s (conditional on investor 1 wanting to purchase the asset): Control (no peer effect) (Condition 1): 42.3% Learning only (Condition 2): 70.8% p-value of Condition 2 = Condition 1: Learning plus possession condition (Condition 3): 92.8% p-value of Condition 3 = Condition 1: p-value of Condition 3 = Condition 2: 0.047

27 Treatment Effects (1) Asset purchase rates of investor 2 s (conditional on investor 1 wanting to purchase the asset): Control (no peer effect) (Condition 1): 42.3% Learning only (Condition 2): 70.8% p-value of Condition 2 = Condition 1: Learning plus possession condition (Condition 3): 92.8% p-value of Condition 3 = Condition 1: p-value of Condition 3 = Condition 2: Negative Selection condition: 39% p-value of Negative Selection = Condition 1: Covariates balance

28 Take-up by Investor 2 Figure 2: Investor 2 s take-up rates % 71% 93% 39% Condition 1 Condition 2 Condition 3 Negative selection Group resents the mean (and 95% confidence intervals) of take-up rates for each group of investor 2 s.

29 Treatment Effects (2) Results unchanged if we include individual covariates, broker FE, and rest of the sample (investor 1 s and remaining investor 2 s) Regressions Similar results if we consider other outcome variables: amount invested; invested more than minimum required Other outcomes

30 Heterogeneity of Social Learning (1) If investors exhibit heterogeneous levels of financial sophistication, one would expect: An investor with a higher degree of financial sophistication is less likely to follow the revealed preference decisions of other investors An investor with a higher degree of financial sophistication is more likely to influence the decisions of other investors through his revealed preference decisions We use independently-coded occupational categories to indicate financial sophistication and test these predictions These tests can help rule out some alternative stories for our findings

31 Heterogeneity of Social Learning (2) Strong and significant learning effects by financially unsophisticated investor 2 s; no learning by sophisticated investor 2 s Weaker evidence for learning from financially sophisticated/unsophisticated investor 1 s, though right direction Regressions Alternative hypotheses External validity

32 Outline 1. Experimental design 2. Results 3. Discussion

33 Summary High-stakes field experiment implemented to identify causal effect of a peer s choice to purchase an asset Also able to separately identify effect of social learning, and additional effect of asset possession Both important components of the overall peer effects that we observe

34 Our Contributions Our results point to both social learning and social utility channels in generating peer effects in an important setting: investment choices Neither the labor/applied micro literature on peer effects (e.g., Cai et al., 2009, Moretti, 2011) nor the finance literature (e.g., Duflo and Saez, 2003, Beshears et al., 2011) has attempted to use experimental variation to separately identify causal roles played by multiple channels of peer influence

35 Implications Asset prices driven by social learning may be destabilizing, if herds build on very little information Greater provision of information might break herds, and limit financial instability Asset prices driven by social utility can also be destabilizing But, greater availability of information won t solve the problem

36 Applications Our methodology can be applied in a range of other settings: Marketing using social networks Facebook likes ; Groupon; product give-aways Adoption of new technologies: New methods of production may be adopted after peers adopt them because of learning or a desire to conform, coordinate, or due to network externalities Health-promoting behaviors

37 Thank you

38 Covariates Balance Results Investor 2 conditional on investor 1 wanted to purchase the asset Control Learning Learning p-value of test: N only + possession (1)=(2)=(3) N=26 N=24 N=28 (1) (2) (3) (4) (5) Age (2.16) (2.55) (2.98) Gender (=1 If male) (0.095) (0.098) (0.094) Married (0.097) (0.090) (0.092) Single (0.100) (0.095) (0.092) Earnings 4,000 4,000 4, (782) (534) (1,941) Financially sophisticated occupation (0.10) (0.13) (0.12) Relationship with investor 1 (=1 if family) (0.10) (0.10) (0.10) Notes: for earnings, we display medians.

39 Treatment effects regressions Treatment effects Dependent variable Wanted to purchase the asset (1) (2) (3) (4) Mean under no information (Condition 1) (0.099) Learning alone 0.285** 0.298** 0.328** 0.278** (Condition (2) - Condition (1)) (0.136) (0.140) (0.134) (0.127) Learning and possession 0.505*** 0.540*** 0.552*** 0.500*** (Condition (3) - Condition (1)) (0.110) (0.122) (0.123) (0.111) Negative selection (0.114) (0.124) (0.118) (0.117) Investor (0.106) Possession alone 0.220** 0.242** 0.224* 0.222** (Condition (3) - Condition (2)) (0.106) (0.109) (0.124) (0.108) Broker fixed effects No Yes Yes Yes Individual controls No No Yes Yes Sample includes investor 1 s No No No Yes N R

40 Additional Outcome (1) - Amount invested Average amount invested by investor 2 s (conditional on investor 1 wanting to purchase the asset): Control (Condition 1): R$884.6 Learning only (Condition 2): R$1,833.3 p-value of Condition 2 = Condition 1: Learning plus possession (Condition 3): R$3,517.9 p-value of Condition 3 = Condition 2: Again, results unchanged if we include individual covariates, broker FE, and rest of the sample

41 Additional Outcome (2) - Invested more than minimum required Percentage of investor 2 s investing above R$2,000 minimum (conditional on investor 1 wanting to purchase the asset): Control (Condition 1): 3.8% Learning only (Condition 2): 25% p-value of Condition 2 = Condition 1: Learning plus possession condition (Condition 3): 53.6% p-value of Condition 3 = Condition 2: Again, results unchanged if we include individual covariates, broker FE, and rest of the sample Back to slides

42 Heterogeneity of Social Learning Learning by a financially sophisticated investor? Yes No p-value of (1) (2) test (1)=(2) Panel A: no controls Learning alone ** (Condition (2) - Condition (1)) (0.231) (0.219) Panel B: full specification Learning alone ** (Condition (2) - Condition (1)) (0.214) (0.215) Mean (no information; peer chose the asset) (Condition (1)) (0.135) (0.179)

43 Heterogeneity of Social Learning Learning from a financially sophisticated investor? Yes No p-value of (1) (2) test (1)=(2) Panel A: no controls Learning alone (Condition (2) - Condition (1)) (0.191) (0.423) Panel B: full specification Learning alone (Condition (2) - Condition (1)) (0.184) (0.324) Mean (no information; peer chose the asset) (Condition (1)) (0.146) (0.197) Heterogeneity

44 Evaluating Alternative Hypotheses (1) Guilt or getting ahead of the Joneses: different implications for social learning, but heterogeneity suggest they are not main forces Side payments: investor 1 did not know about investor 2 s call, calls on the same day, investor 2 could not talk to investor 1, no larger learning effect among family members Revealed preference is indication of peer s portfolio: so social learning has some anticipated possession effect. Sophisticated inference. Also, usually $10,000 limit No mention of investor 1 in condition (1): investor 2 might worry about decision being shared. But vast majority only connected to one other client; and inference of greater utility if another peer received the offer and purchased the asset is sophisticated

45 Evaluating Alternative Hypotheses (2) No update about demand for the asset and future asset prices since asset only sold then and no resale Trust: likely not an issue, ongoing relationship with brokers Lottery: no effect of lottery in pilot, also used in oversubscribed IPO s Supply side Broker manipulation to increase sales Brokers incentivized to sell assets, not to confirm hypotheses Results do not change with broker experience; or according to presence of an RA at the brokerage Heterogeneity

46 External Validity Common form of sales (over 70% in the brokerage) Not unusual sample of clients within the brokerage Not unusual subsample that chose to purchase the asset Financial advisors and brokers very common in the US as well Heterogeneity