How Wise Is the Crowd? Timing and Selection Skill of Retail Social-Trading Portfolios

Social-trading platforms publish every trade of thousands of private investors. Aggregating such trades into a "crowd signal" is an old idea; whether the crowd actually knows something is an empirical question. We study 313,302 trades of 82 top-ranked portfolios on wikifolio.com (June 2012 to 23 February 2026; data cutoff 1 March 2026) and separate market timing from stock selection. Four results stand out. (i) Net crowd buying is a reaction variable: it follows past index weakness and rising volatility, and none of 54 predictive regressions on future DAX or S&P 500 returns has a Newey–West |t| above 1.7. (ii) Stock selection looks positive for buys (60-day alpha +0.79%, cluster t=2.3, return measured from the close of the day after the trade) and the buy-minus-sell difference is significant (+1.04 pp, t=3.3). A panel baseline of all days of the same stocks shows that the spread of buy returns is mostly proximity to the 52-week high, a known anomaly; after controlling for it, buys keep a modest increment of +0.92 pp over 60 days (t=2.9), sells none (+0.15 pp, t=0.4), gross of costs. (iii) Since 2020 the edge sits on the sell side: stocks sold in 2020–26 underperformed by 1.12% over 60 days (t=-2.3) while buy alpha at 60 days is indistinguishable from zero (+0.05%, t=0.1); sells after strong run-ups are too early. (iv) Very short holdings earn nothing (1–5 days: 36.8% win rate). We then report what happened when these facts were turned into a signal generator ("Maximale Crowd"): a contribution-weighted aggregation of traders (Budescu and Chen), a binomial consensus gate and a stop-loss hybrid. The system suffered from sparse trader overlap, a concentrated effective crowd (about 10 of 71 qualified traders carry the weight), tuning on 165 trading days, and a look-ahead bug that invalidated every backtest. We therefore claim no tradable edge. The paper documents data, methods, discarded ideas and pitfalls so that the work can be audited or continued.Read next: The look-ahead bug described here motivated the leak-resistant point-in-time backtesting. For event signals beyond the crowd, see the negative results on event-driven signals. Crowd-based candidates were later judged in a pre-registered out-of-sample test.This preprint is part of an eight-paper series on retail equity research from public data by Christopher Helm, covering German and Nordic small caps, backtesting methodology, data sources and AI-assisted research. All papers in the series:A Leak-Resistant Point-in-Time Backtesting Protocol and Engine for Retail Equity StrategiesThree Pre-Registered Bets: A One-Shot Out-of-Sample Test of Retail Strategy CandidatesWhat Did Not Work: Negative Results on Event-Driven Signals in German and Nordic Small CapsFree, Point-in-Time Public Data for German and Nordic Small-Cap Research: Sources, Coverage and PitfallsRequirements for an Autonomous Retail Order Engine Running Unattended for a DecadeRunning an Autonomous AI Agent on Quantitative Trading Research: Protocol, Supervision and Failure ModesFrom Pre-Announcement to Verified Facts: Design of a Primary-Source Report-Analysis Funnel for Small-Cap EquitiesChristopher Helm is an entrepreneur, consultant, and researcher. More research and articles on his blog.