The influence of upward social comparison on retail trading behaviour

Online investing is often facilitated by digital platforms, where the information of peer top performers can be widely accessible and distributed. However, the influence of such information on retail investors’ psychology, their trading behaviour and potential risks they may be prone to is poorly understood. We investigate the impact of upward social comparison on risk-taking, trading activity and investor satisfaction using a tailored experiment with 807 experienced retail investors trading on a dynamically evolving simulated stock market, designed to systematically measure various facets of trading activity. We find that investors presented with an upward social comparison take more risk and trade more actively, and they report significantly lower satisfaction with their own performance. Our findings demonstrate the pitfalls of modern investment platforms with peer information and social trading. The broad implications of this study also provide guidelines for improving retail investor satisfaction and protection.

trading activity in an environment that mimics a recent financial technology trend.Second, we demonstrate that experienced retail investors exposed to outcomes of top traders have lower satisfaction from their performance and increase the fraction of risky assets in their portfolios.Third, we show that they increase different facets of trading activity in heterogeneous ways, suggesting that online trading activity could help explain the drivers of the trading behaviour resulting from social influence.
'Social influence is defined as change in an individual's thoughts, feelings, attitudes, or behaviours' 29 .Social comparison is a form of social influence in which an individual changes their behaviour, judgment or attitudes as a result of observing others' behaviour.Social comparison refers to people's tendency to compare themselves with others 30 , to reduce uncertainty about themselves by evaluating their abilities 31 .In goal-directed behaviour, people compare their own performance to a reference group that is better than they are, in upward social comparison, to increase their performance, or to a worse reference group, termed downward social comparison, to preserve or boost their self-perception 32 .Both mechanisms constitute systems that regulate motivation to act and degree of present and future effort.They can also result in an emotional response.Feeling worse can trigger negative affect, whereas feeling better than others can trigger positive affect 33 .
Information about others' performance sets a social reference point (SRP), and this can influence what individuals aspire to have even if this aspiration deviates from what they can rationally hope to obtain 16 .This aspiration, motivated by the desire to get ahead of others, influences risk-taking in investment tasks independently of whether the goal is achievable or not 15,16 .SRP influences investment risk-taking in tournament experiments, which list investors in a performance-based rank (see 34 for a systematic review).A tournament is characterized by two motivator components: a monetary reward that depends on rank position and a nonmonetary incentive to outperform peers.Nonmonetary tournament components increase risk-taking only in underperforming professionals who report valuing socially relative performance more than monetary performance, whereas underperformers generally take more risk than overperformers 35 .
Presenting social comparison information leads to higher risk-taking, even when it is irrelevant to the payoff and not attached to tournament incentives 12 .Kirchler et al. 36 experimentally showed that inducing upward social comparison through presenting financial outcomes of top investors increased risk-taking in an asset allocation task.Individuals also adjust their risk-taking levels to that of a better player in two-player investment games where one outperforms the other 37 .Such adjustments also occur beyond tournaments: investors adjust the risk of their portfolios when they receive bonuses dependent on a market index that indicates the performance of all market players 13 .A possible explanation of these findings is that relative performance implicitly creates a tournament reward structure 38 .
An experimental asset market is an experimental simulation of a financial market in which several humans trade artificial assets with each other using computer software.In such experiments, upward performance comparison 38 and larger wealth inequality among market players 39 leads to higher market prices and more inflated market bubbles.Also, novice investors underperform when their peers report very high returns 20 .A potential driver of this observation could be that sharing of the most successful performances encourages risky stock trading behaviour, especially in less experienced investors.Likewise, selectively communicating only positive information to investors causes higher stock market participation rates and consequently higher risk-taking 40 .
Copy trading is a form of social trading in which top social traders, who usually exhibit high trading performance, take the role of signal providers and are followed by other users.Research indicates that followers of the eToro trading platform are prone to take more risk, and they tend to overreact when the signal providers take more risk 5 , while followers on a European social trading platform increase their activity after signal followers had posted comments 10 .Experiments about copy trading found that information about the success of other investors leads to increased investment risk-taking, where participants choosing the riskiest asset are much more likely to copy this decision from another player than when choosing any of the less risky assets 41 .Investors also report that social influence through peer recommendations generally impacts their trading frequency and volume 22 .On social trading platforms, social influence can override rational decisions, raising questions about the use of these platforms 5 .This is especially important because followers prefer more-risky high-volatility stocks and trade more frequently, but obtain more negative returns 14 .A possible explanation for this observation is that that the followers perform less well because trades made after the signal is provided result in a lower performance than before-signal trades 14 .
Upward social comparison not only impacts how people behave but also how they feel.Comparing oneself to richer people decreases people's life satisfaction 42 .Upward social comparison in financial investment results in lower satisfaction with one's performance 38 , while engaging in online upward social comparison decreases positive affect 43 .Thus, affective and cognitive drivers of behaviour interplay [44][45][46] , such that affect moderates cognitive processes preceding actions 47 .Consequently, peer effect in risk-taking behaviour, such as investing, is a combination of conscious informed choices influenced by peers motivating individuals to execute particular strategies and affect arising from social comparison 48 .
However, data beyond risk-taking is needed to capture the psychological drivers of financial decisions because the timing of these decisions plays a crucial role 49 .Various aspects of trading activity, including trading frequency, average transaction volume and cumulative volume of all trades, can provide a much rounder picture of the potential psychological drivers of trading behaviour 25 .In dynamic decision tasks such as trading and investment, reaching the goal of generating high positive returns requires a series of decisions which need to be taken at the right times.The consecutive decisions are dependent on each other, such that earlier decisions determine the context and options available in subsequent decisions 50 .This causes the state of each decision to change continuously 51,52 .Experimental designs that incorporate these features and enable measurement of trading activity in a realistic and controlled way are scarce 53 .
Overall, the literature reviewed here indicates that social comparison can result in increased risk-taking in trading and investment 5,[11][12][13][14][15][16]25,36,40,41 and increased trading activity 5,9,10,18,25 measured in various ways which www.nature.com/scientificreports/ are rarel linked to interpretations of cognitive and affective interpretations of these goal directed behaviours.Some studies 19,22 investigated trading activity, but not in the context of social influence. Also, previos literature indicates that upward social comparison may be linked to a lower satisfaction from one's own financial and economic outcomes 38,42 .To gain a better understanding of the effects of upward social comparison on trading behaviour,we programmed a dynamic artificial stock market substantially extending the features of 17 to test two hypotheses preregistered on the Open Science Framework (https:// osf.io/ we48d/).Our hypotheses derived from the literature discussed here relate to risk-taking and three facets of trading activity: Hypothesis 1 Upward social comparison will lead to increased risk-taking, where risk-taking is defined as the mean fraction of the risky asset in a portfolio throughout one experimental round.
Hypothesis 2 Upward social comparison will lead to increased trading activity: a.Leading to a larger number of transactions, where the number of transactions refers to the number of separate executed trades in one experimental round, b.Leading to higher volume of each individual trade, where the volume corresponds to the mean quick-trade button size, and c.Leading to higher overall trading volume, where the overall trading volume refers to the cumulative volume of all shares traded in one round.
Hypothesis 2c is not a direct mathematical composite of hypotheses 2a and 2b, but it embeds them.

Results
We developed a task that presented participants with historical day closing stock prices every 0.8 s (see Fig. 1) in two trading rounds.At the start of each round, their endowed capital was split equally between stocks and cash.In each round, our online experimental task displayed tick-by-tick 252 consecutive prices of one stock and enabled a participant to continuously buy and sell their shares in various volumes without transaction fees using six quick-trade buttons.On online platforms, buy buttons are a call-to-action feature 54,55 that moderates the execution of a transaction.In our task, the buttons were designed to measure trading activity in three ways: www.nature.com/scientificreports/ (1) number of transactions, (2) average trading volume, and (3) overall volume of all traded shares.Participants could sell as many shares as they owned, and they could purchase as many shares as they wanted if they had enough cash.After the first round, participants in the experimental condition received information about the performance of three other participants in addition to feedback about their own performance (see Fig. 1).The three other participants were first batch study participants who obtained very high earnings.Participants in the control condition only received feedback about their own performance but not about their peers.Following the trading task, we measured participants' satisfaction about their performance in the task.This experimental design extends the asset allocation design by 36 enabling measuring the influence of upward social comparison on trading activity.
In this difference-in-difference design with prescreened, financially literate participants, we first tried to replicate previous findings that upward social comparison increases risk-taking (c.f., Risk).Then, we aimed to substantially extend previous findings by investigating the impact of upward social comparison on trading activity and trading satisfaction in experienced retail investors.We systematically measured trading activity that may indicate affective drivers of trading behaviour by the number of transactions (c.f., Trans); we also measured cognitive drivers by average trading volume (c.f., Vol), and the mediation of cognitive drivers by affect by the overall volume of all traded shares (c.f., Vol_Sum).

Hypothesis testing
In accordance with our preregistered data analysis plan, we tested the differences between the control and the experimental conditions in round 2 using Wilcoxon rank sum tests with continuity correction.We repeated the analysis with less conservative Welch two sample t tests.Participants in the experimental condition had significantly riskier portfolios (W = 68,128, P < 0.001, d = 0.2; see Fig. 2A), and they used the decline in price to buy the risky asset.Therefore, our data clearly supports the first hypothesis.
We found weak support for Hypothesis 2a, which states that participants in the experimental condition will make more trades.The Wilcoxon test was not statistically significant (W = 75,591, P = 0.11) and had a weak effect size (d = 0.07), whereas the Welch test was significant (t = − 1.97, df = 561.33,P = 0.05).Further, participants in the experimental control condition showed a trend towards initiating larger transactions on average, partially supporting Hypothesis 2b.This difference was significant according to the Welch test (t = − 2.09, df = 797.51,P = 0.05) but not according to the Wilcoxon test (W = 75,512, P = 0.06), and the effect size was small (d = 0.1).The combined value expressed in experimental currency units of all shares bought and sold by a participant was significantly higher in the experimental condition than in the control condition (W = 72,374, P = 0.01, d = 0.13; t = − 2.55, df = 692.98,P = 0.01, see Fig. 2B), supporting Hypothesis 2c.
Descriptive statistics for all hypotheses are presented in Table 1.To verify that the effects that we report in this section are nonrandom patterns that became significant, we conducted a randomization test in which we shuffled the data 50,000 times by re-allocating the experimental condition to observed trading activity metrics.For each dependent variable tested in the hypotheses, we plotted the distribution of the W test statistics obtained from each reshuffling iteration in Fig. 3.We repeated this procedure with mean group differences Between conditions and compared this distribution to the difference observed between group medians.As Fig. 3 shows, the effects driving hypotheses 1 and 2c were due to differences between conditions starting from about experimental trading day 120 (one and a half minutes).Although an effect size of 0.13 is considered small 56 , a comparable real-world increase in US retail volume would correspond to about 600 million additional shares traded each day 57 .

Robustness checks
To ensure that the differences documented in round 2 were not attributable to a randomization failure such that pre-existing group differences might have been responsible for our observed effects, we tested the differences in risk-taking and trading activity in round 1, before the upward social comparison was induced.We used Wilcoxon tests on the differences between the control and experimental conditions, but found no differences in Risk (W = 78,335, P = 0. Further, we found that participants' earnings in neither round 1, nor round 2 differed between the conditions (Welch two-sample t test for ROI 1 : t = − 1.40, df = 776.89,P = 0.16, and for ROI 2 : t = − 0.99, df = 758.6,P = 0.32).This is to some extent to be expected given the absence of return predictability and low expected returns.Table 1.Descriptive statistics for the variables for testing differences between the control (C) and experimental (E) condition corresponding to each hypothesis.Variables are explained in the main text.Variable Vol is related to the values of the quick-trade buttons (see Fig. 1A).www.nature.com/scientificreports/ In addition to this, we tested for potential differences between participants in the control and in the experimental condition that could contribute to the differences in trading behaviour in round 2. According to Wilcoxon-sum rank tests, we found no differences in the self-reported propensity to take risk (W = 74,221, P = 0.07) and in three financial literacy questions (Question 1: W = 82,650, P = 0.28; Question 2: W = 78,634, P = 0.50; Question 3: W = 82,650, P = 0.28).The sample demographics were also similar with 39% of women and mean age of 40.01 years in the experimental condition and 36% of women and mean age of 41.10 years in the control condition indicating no statistical differences in age between the two conditions, t = 1.1235, df = 790.67,P = 0.26.
Finally, an alternative explanation for the observed differences in trading activity between the two conditions (Hypothesis 2c) is that this effect is driven entirely by differences in risk-taking (Hypothesis 1).This would happen if the additional trading activity in the experimental condition consisted exclusively of buying shares of the risky asset for the purpose of increasing portfolio risk.In such a case, an observed increase in trading volume would be mechanistically driven by an increase in risk appetite.To test this, we regressed Vol_Sum in round 2 on Risk in round 1 and a dummy variable for the experimental condition.Due to the distribution of Vol_Sum, we implemented a rank-based estimation method for linear models from the rfit R package 58 .The effect of condition on trading activity was substantially unchanged after accounting for concurrent differences in risk-taking (β = 0.058, P = 0.02).This implies that the higher trading activity observed in the experimental condition cannot be interpreted as an artefact or by-product of increased risk-taking.
Overall, these findings indicate that the social comparison induced in the experimental condition significantly increased risk-taking and trading activity.

Manipulation check
After the trading task, we asked participants to estimate the mean performance of all participants to verify that participants in the experimental condition perceived typical earnings in the task as higher than those in the control condition.According to a Wilcoxon test (W = 51,600, P < 0.001), participants in the experimental condition perceived the average earnings of all participants, expressed in experimental currency, to be significantly higher than the participants in the control group (Md Control = 500, Md Experimental = 1000, d = 0.46).Therefore, we conclude that our manipulation was effective.

Post-task satisfaction
In a preregistered analysis, we posed a question, 'How do you feel about your performance in the trading task?', adapted to our experimental design from 38 .In line with their results, we found that participants in the experimental condition were substantially less satisfied with their own performance than were participants in the control condition (Md Control = 5, Md Experimental = 4, d = 0.36, W = 1,022,889, P < 0.001).

Discussion
Social trading platforms make financial asset trading easily accessible to anyone with internet access and some disposable capital.They also provide access to a large amount of information about the financial strategies and behaviour of traders.The COVID-19 pandemic has accelerated the growing popularity of online trading, with many people staying at home and channelling their time and effort into learning retail trading 3,59 .In 2020, the market size of online trading amounted to 8.28 billion USD, with growth to 12.16 billion USD forecast by 2028 60 raising the question of how this growing trend impacts individuals.Relative performance information is omnipresent in the real-world markets, but laboratory studies systematically investigating its impact on retail investors are very scarce 38 .
To address the need to study human interaction with increasingly popular trading platforms, we investigated the influence of upward social comparison on risk-taking, trading activity and trading performance satisfaction with a dynamic trading task.Our results confirmed previous findings indicating that upward social comparison results in more risky trading behaviour 12,13,36,38,41 .Our findings add to previous research by experimentally demonstrating that upward social comparison also increases trading activity such as total traded volume.In contrast to the previous research demonstrating that upward social comparison increases trading activity 18,22,25 , we indicated three measures of trading activity which may allow us to identify whether the impact of social influence triggers affective or cognitive drivers of goal-directed behaviour.The trading behaviour we observed confirms the suggestion that even in financially literate investors, upward social comparison causes strategic trading behaviour to be moderated by affect; this corresponds to the psychological literature 44,46,[61][62][63][64] .
Overall, participants were ultimately less satisfied with their performance, resonating with previous findings that more experienced investors derive more satisfaction from their own performance being ahead of others' 35,36,38,65,66 .This effect could be because the reward centre of the brain responds more strongly to a performance generated by skills and effort than one resulting from luck 66 .Given that previous experience with asset trading is linked to the expectation of performing well on social trading platforms 65 , financially literate participants of this study would perceive their performance as less rewarding when facing upward social comparison.Neuroscientific findings indicate that compared to no social comparison 67 and downward social comparison 64 , upward social comparison is associated with increased brain activity in anterior insula, ventromedial prefrontal cortex (vmPFC), and dorsolateral prefrontal cortex (DLPFC), which are responsible for emotional processing (insula and vmPFC) and planning (DLPFC) 68 .The two conditions in our experiment did not differ in their final earnings, and, in contrast to tournament experiments, participants' compensation was independent of that of other participants.We surmise that when participants trade extensively yet fail to perform as well as better traders, such participants may view their effort as wasted, lowering their trading satisfaction.
A limitation of the current study is that we deliberately did not investigate the effect of downward social comparisons as previous studies had 11,12,38 and found that exposing participants to the worst investor's performance www.nature.com/scientificreports/led to less risky portfolio allocations.This decision was motivated by the fact that hardly any real-life social trading platforms show the worst performers.Moreover, success in finance and investment is much more salient than failure, and success stories are more viral than stories of loss 20,40 .Investment flows to the best-performing portfolio managers on social trading platforms 69 .Furthermore, early research on social comparison suggests an overall tendency of people to upward social comparison in performance situations 31,70 .Also, comparing oneself to richer people decreases people's life satisfaction more than comparing themselves to poor people improves their life satisfaction 42 indicating a smaller effect of downward social comparison in a financial setting 71 .Further research could examine the impact of social influence on signal providers rather than on their followers.Signal providers should be considered as investment managers without direct access to the follower's wealth 6 .Signal providers are more susceptible to the disposition effect-bias where an investor sells the well-performing assets, while keeping the badly-performing assets-than their followers, because they experience a fear of losing followers when admitting a poor decision 72 .Furthermore, the signal providers who receive more attention from their peers experience more excitement about trading, intensify their trading activity and increase risk-taking.Although the emotions and resultant increased trading activity wear off with time, the elevated risk-taking persists 73 , once again confirming the relation between social influence and trading behaviour.As a follow-up to this online experiment, future brain imaging and cognitive modelling studies could provide more in-depth insights to better understand the cognitive processes underlying the link between social comparison and trading.
Although previous studies showed no difference between using ECU and "real" currency 71 , another possible extension of this study would include upward social comparison including real rather than experimental currency.This would allow to capture the effect of social influence on investment behaviour in even more realistic setting.Related to this, future research could investigate the minimum difference between a person's earnings and the top traders' earnings required to induce change in trading behaviour.In addition to this, the current experimental design could be modified to offer copying trading strategies of the top performers, for the purpose of testing the impact of upward social comparison on copy-trading behaviour.Further extensions could involve modeling of trading behaviours and strategies and their change because of stimuli such as social influence.Potential models would account for differences in traders' portfolio risk and trading activity and would allow for model fitting with experimental data for testing the validity of these models.
Nevertheless, findings from the current study are particularly relevant for the nascent yet rapidly expanding research devoted to social trading mechanisms and platforms.Recent European regulation MiFID II, Directive 2014/65 74 aims to provide guidelines for disseminating only 'fair, clear and not misleading' materials to prospective investors, but it provides no explicit guidelines for presenting peer information either as part of marketing messages or integrated into platforms' user interfaces.This study shows that presenting the returns of top performers can increase other traders' risk-taking and effort.Excessive trading, which usually occurs on online trading platforms, leads to mental health problems in some users 75 .Presenting only the best-performing portfolios on social media platforms promotes herding in trading strategy selection, in which past winners are bought more frequently and past losers are more frequently sold 76 .However, followers often obtain worse performance than the leaders who initially implement strategies 14 .In contrast, open information exchange among users of social trading platforms could promote self-consciousness and reduce biases such as the disposition effect 8 .
In sum, this study contributes to the current literature on social influence and trading behaviour by demonstrating that social comparison impacts trading activity on an online trading platform.In a naturalistic experiment with financially literate participants, it investigates three, linked with each other, aspects at a time: risk-taking, three facets of trading activity and traders' satisfaction.It demonstrates that investigating various aspects of user activity on online trading platforms can help better understand the psychological triggers of user behaviour and potential biases.Further experimental research on how people respond to online and offline social influences in the domain of finance, and how new technologies facilitate these effects, should inform future policy efforts to maximize consumer welfare.It is necessary for behavioural research to facilitate safe market participation of the retail investors-regular members of our society investing their savings.We believe that this study provides a steppingstone towards a better understanding of the interplay between social and technological factors influencing retail investing.

Method
The study was preregistered on the Open Science Framework (https:// osf.io/ we48d/) and approved by the ETH Ethics Committee (2020-N-120).Anonymized data, analysis code and experimental materials are available on the Open Science Framework.All methods were performed in accordance with the relevant guidelines and regulation.This study uses a difference-in-difference, mixed experimental design with the upward social comparison as an independent between-subject variable and trading behaviour (risk and trading activity) measured as both within and between dependent variables.

Participants
In the period 17.11.2020-07.12.2020, we recruited 1622 CloudResearch Mechanical Turk workers, aiming to collect data from 800 participants, based on an a priori power analysis.1262 individuals entered the experiment and accepted the informed consent form, and after the prescreening, 807 people (300 female, M age = 40.62,Age range: 18-83) completed the full study.This resulted in a 64% pre-screening rate for consenting participants.Each prospective participant received 0.30 USD base payment regardless of whether they passed the initial screening, and everyone who completed the experiment received a variable performance-dependent bonus ranging from 1.38 to 1.66 USD (M Bonus = 1.51).The participants had to be verified by Amazon as individual investors directly holding investments in the US stock market.Their investor status had been previously checked by Cloud Research 77 , they were identified as 'high-quality' based on their earlier survey participation, and they

Figure 1 .
Figure 1.Social comparison information (A) and user interface of the trading task (B).

Figure 2 .
Figure 2. The distribution of portfolio risk between conditions (A) and trading activity expressed as Vol_Sum (B) in round 2. Note Horizontal bars indicate the group median, while larger dots indicate group means.In panel B, due to the presence of outliers the y-axis is bounded at 150,000 experimental currency units.