Understanding Mechanisms Underlying Peer Effects: Evidence from a Field Experiment on Financial Decisions

Pairs of investors, with social linkages among one another, from a client pool of a Brazilian brokerage firm were identified. One investor from the pair was randomly chosen to receive an offer of a new investment product. If the investor expressed interest to buy the product, he or she was entered into a lottery to determine whether they actually can buy it. If the lottery is positive, the investor was allowed to purchase, if not the investor did not get a chance to purchase the asset despite willingness to do so. The second investor was offered the same product but either (i) received no information about investor 1’s decision (ii) received information the investor 1 wanted to buy the same product but did not do so (iii) received information that investor 1 wanted to buy the product and was able to do so. The difference between (i) and (ii) is how much social learning contributes to purchase decision and the difference between (ii) and (iii) gives the contribution of social utility.

2012-01-26

2012-04-03

Purchase of financial asset by second investor.

Purchase decision by second investor is measured directly.

First randomization: The list of participants in the study was chosen from the clientele list of a Brazilian brokerage firm for whom one other peer could be identified with social linkages. One hundred and fifty pairs of investors, thus a total of 300 investors, were chosen in this manner. Within each pair one was randomly chosen to be investor 1 – the investor who first receives the offer to purchase the asset.

Second randomization: Once investor 1 expressed willingness to buy the asset, he or she was entered into a lottery that generated a random number between 0 and 100. If the number was greater than 50, the investor was allowed to purchase the asset, otherwise the investor could not do so. If Investor 1 refused to buy the asset in the beginning he/she was not entered into the lottery.

Third randomization: Investor 2 received the same offer as investor 1 and was randomized to assign what kind of information he/she will receive about investor 1. Investor 2 either received no information, or complete information about the investment decision and the purchase outcome. Because of the first lottery, which allowed only half of the willing investor 1s to purchase the asset, some investor 2s will be told that their peer wanted to purchase this asset but did not do so, while some will be told that their peer wanted to buy the asset and did so.

A post-intervention survey was conducted to test the financial knowledge of all the participants to see if the actual or perceived financial sophistication of one’s peer had an influence on the how investor 2s chose to act on information about the purchase of the asset by their friend.

Randomization done in office by a computer

Three types of randomization was carried out:

First randomization: within investor pairs to decide which investor was to receive the offer first.

Second randomization: to decide if investor 1, who wanted to purchase the asset, will be allowed to do so.

Third randomization: to decide if investor 2 will receive full or no information about investor 1’s purchase decision and purchase outcome.

No

150 investor pairs

150 investor 2 decisions

Condition A (Associated Investor 1 wanted/no information) = 26
Condition B (Associated Investor 1 wanted but did not) = 24
Condition C (Associated Investor 1 wanted and got it)= 28
Condition A^neg (associated investor 1 did not want)= 72

The main comparisons in this experiment are based on investor 2’s whose associated investor 1’s accept the offer. Therefore, investor 1’s take-up rate could not be too low because in this case there would be few investor 2’s in this condition. Also, this take-up rate could not be too high because in this case it would not be possible to detect effects (since average take-up rate cannot be greater than 1). The brokerage firm allowed us to have 150 pairs of investors in our sample. We piloted the offer to expect an average take-up rate of 50% for investor 1’s, so we expected to have 25 investor 2’s in each treatment arm. With this sample size, we expected to have approximately 70% power for tests with significance level of 10% to find effects of at least 30 percentage points in comparisons between either treatment conditions (B or C) and the control group (condition A).

Understanding Mechanisms Underlying Peer Effects: Evidence from a Field Experiment on Financial Decisions

Using a high-stakes field experiment conducted with a financial brokerage, we implement a novel design to separately identify two channels of social influence in financial decisions, both widely studied theoretically. When someone purchases an asset, his peers may also want to purchase it, both because they learn from his choice (“social learning”) and because his possession of the asset directly affects others’ utility of owning the same asset (“social utility”). We randomize whether one member of a peer pair who chose to purchase an asset has that choice implemented, thus randomizing his ability to possess the asset. Then, we randomize whether the second member of the pair: (i) receives no information about the first member, or (ii) is informed of the first member’s desire to purchase the asset and the result of the randomization that determined possession. This allows us to estimate the effects of learning plus possession, and learning alone, relative to a (no information) control group. We find that both social learning and social utility channels have statistically and economically significant effects on investment decisions. Evidence from a follow-up survey reveals that social learning effects are greatest when the first (second) investor is financially sophisticated (financially unsophisticated); investors report updating their beliefs about asset quality after learning about their peer’s revealed preference; and, they report motivations consistent with “keeping up with the Joneses” when learning about their peer’s possession of the asset. These results can help shed light on the mechanisms underlying herding behavior in financial markets and peer effects in consumption and investment decisions.

Burnsztyn, Leonardo, Florian Ederer, Bruno Ferman, and Noam Yuchtman. 2014. "Understanding Mechanisms Underlying Peer Effects: Evidence from a Field Experiment on Financial Decisions." Econometrica 82(4):1273-1301.

https://www.povertyactionlab.org/sites/default/files/publications/893%20Peer%20Effects%20July2014.pdf