Retail and day-trader outcomes
Large studies of individual traders, most of them using complete exchange or broker records rather than surveys.
97% of Brazilian day traders who persisted for at least 300 days lost money.
The study follows every individual who began day trading in the Brazilian equity futures market between 2013 and 2015. Of those who kept going for 300 days or more, only 0.4% earned more than a bank teller's starting pay (US$54 a day).
SourceFernando Chague, Rodrigo De-Losso and Bruno Giovannetti, Day Trading for a Living?, FGV EESP working paper (Textos para discussão) No. 525, . Figures confirmed in the paper's abstract. It is a working paper, not a journal article.
Less than 1% of day traders in Taiwan could predictably and reliably earn positive abnormal returns after fees.
Barber, Lee, Liu and Odean studied the trading records of day traders in Taiwan from 1992 to 2006.
SourceBrad M. Barber, Yi-Tsung Lee, Yu-Jane Liu and Terrance Odean, The cross-section of speculator skill: Evidence from day trading, Journal of Financial Markets 18, pp. 1–24, .
An estimated 97% of day traders in Taiwan are likely to lose money in their future day trading.
The same research group also finds that about three-quarters of day-trading volume comes from traders with a history of losses, and that unprofitable day traders are more likely to quit than profitable ones.
SourceBrad Barber, Yi-Tsung Lee, Yu-Jane Liu, Terrance Odean and Ke Zhang, Learning, Fast or Slow, The Review of Asset Pricing Studies 10(1), pp. 61–93, . Figures confirmed in the paper's abstract.
US households that traded the most earned 11.4% a year, while the market returned 17.9% (1991–1996).
Barber and Odean analysed 66,465 households with accounts at a large US discount broker. The average household earned 16.4% a year.
SourceBrad M. Barber and Terrance Odean, Trading Is Hazardous to Your Wealth: The Common Stock Investment Performance of Individual Investors, The Journal of Finance 55(2), pp. 773–806, .
Trading cost individual investors in Taiwan 3.8 percentage points of return a year, losses equal to 2.2% of Taiwan's GDP (1995–1999).
Measured on the complete trade data of the Taiwan Stock Exchange for all individual investors over five years.
SourceBrad M. Barber, Yi-Tsung Lee, Yu-Jane Liu and Terrance Odean, Just How Much Do Individual Investors Lose by Trading?, The Review of Financial Studies 22(2), .
CFDs, forex and leveraged retail products
Figures published by market regulators from client account data, most of them gathered before leverage limits and standard risk warnings were imposed on retail CFDs.
For the figure each UK and EU broker publishes today, see our tracker of retail CFD loss rates by broker, checked quarterly on the brokers' own sites.
74–89% of retail CFD accounts typically lose money, according to national regulators' analyses across the EU.
ESMA cited these analyses, with average losses per client of €1,600 to €29,000, when it banned binary options and restricted CFDs for retail clients in 2018. The range became the standard risk warning for providers without their own recent retail figures.
SourceEuropean Securities and Markets Authority (ESMA), ESMA agrees to prohibit binary options and restrict CFDs to protect retail investors, Press release, .
82% of UK CFD clients lost money, in the FCA's analysis of a representative sample of client accounts.
The sample covered client accounts at eight CFD firms; across them, the average result per client account was a loss of £2,200.
SourceFinancial Conduct Authority (FCA), FCA proposes stricter rules for contract for difference products (CP16/40), Press release and consultation paper, .
More than 89% of French retail forex and CFD clients lost money over 2009–2012.
The AMF studied 14,799 active clients of regulated providers over four years; the average result was a loss of nearly €10,900 per client.
SourceAutorité des marchés financiers (AMF), Forex : les particuliers perdants (La Lettre de l'Observatoire de l'épargne n°10), AMF savings observatory letter, in French, .
Among French forex and CFD clients who traded in all four years from 2009 to 2012, nearly 88% lost money, with an average loss of €26,745.
About 2,000 clients traded actively in each of the four years. They did worse than the whole population, and the AMF found no sign that experience improved results.
SourceAutorité des marchés financiers (AMF), Forex : les particuliers perdants (La Lettre de l'Observatoire de l'épargne n°10), AMF savings observatory letter, in French, .
74% of retail CFD clients of a sample of large Irish providers lost money in the two years to 31 December 2016, with an average loss of €2,700.
Figures gathered by the Central Bank of Ireland when it consulted on restricting the distribution of CFDs to retail investors.
SourceCentral Bank of Ireland, Consultation on the Protection of Retail Investors in relation to the Distribution of CFDs (CP107), Consultation paper, .
72% of Australian clients lost money trading CFDs other than margin FX over a 12-month period, and 63% lost money on margin FX.
Figures from ASIC's 2017 review of CFD issuers, restated when ASIC made its CFD product intervention order in October 2020.
SourceAustralian Securities and Investments Commission (ASIC), Public notice: Product intervention order in relation to contracts for difference, Public notice, .
The share of loss-making retail CFD accounts in Australia fell from a quarterly average of 64% to 50% after ASIC limited leverage in March 2021.
Data from 59 CFD issuers: the four quarters before the order took effect on 29 March 2021, against its first three months.
SourceAustralian Securities and Investments Commission (ASIC), Extension of the CFD product intervention order (Consultation Paper 348), Consultation paper, .
68% of Dutch retail clients made a loss trading turbos between June 2017 and June 2018.
About 35,000 retail investors at four large distributors; the average result was −€2,680 per client, and 88% of clients with more than 500 transactions made a loss, against 64% of those with fewer than 10.
SourceDutch Authority for the Financial Markets (AFM), Quantitative analysis of turbos distributed to retail clients in the Netherlands, Report, .
After the US capped retail forex leverage in 2010, high-leverage traders' losses fell by about 40% and their trading volume by about 23%.
Heimer and Simsek compare traders affected by the 2010 rules with unaffected ones (a difference-in-differences design) across three datasets.
SourceRawley Heimer and Alp Simsek, Should retail investors' leverage be limited?, Journal of Financial Economics 132(3), pp. 1–21, . Figures confirmed in the paper's abstract.
Algorithmic, electronic and high-frequency trading
Market size and the share of trading done by machines, from central banks, regulators and their economists.
Global foreign exchange turnover averaged $9.5 trillion a day in April 2025, up 27% from April 2022.
From the BIS Triennial Central Bank Survey of over-the-counter FX markets. The first release (30 September 2025) gave $9.6 trillion; the revised figure published in December 2025 is quoted here. April 2022: $7.5 trillion.
SourceBank for International Settlements (BIS), BIS Quarterly Review, December 2025: foreword on the 2025 Triennial Survey, BIS Quarterly Review, .
59% of FX trading was executed electronically in April 2025, a share virtually unchanged since the 2022 survey.
BIS analysis of how FX trades are executed, based on the 2025 Triennial Survey.
SourceBank for International Settlements (BIS), The FX trade execution landscape through the prism of the 2025 BIS Triennial Survey, BIS Quarterly Review, December 2025, .
Execution algorithms account for an estimated 10–20% of global FX spot trading, about $200–400 billion a day.
Estimate by a BIS Markets Committee study group from a survey of 70 market participants. The same report puts electronic execution of spot FX at about 70% in 2019, up from about 55% in 2010.
SourceBIS Markets Committee, FX execution algorithms and market functioning (Markets Committee Papers No 13), Report, .
Automated trading participation reached 80% in FX futures at CME Group, 67% in equity futures and 62% in interest-rate futures (November 2012 to October 2014).
Measured by CFTC economists from CME's flag for manual and automated orders, a wider category than high-frequency trading.
SourceRichard Haynes and John S. Roberts, Automated Trading in Futures Markets, US Commodity Futures Trading Commission (CFTC), Office of the Chief Economist, .
Latency-arbitrage races between high-frequency firms account for about 20% of trading volume in FTSE 100 stocks.
FCA economists measured the races in London Stock Exchange message data. They estimate that eliminating them would reduce the cost of liquidity by 17%, and extrapolate the races' value to about $5 billion a year across global equity markets.
SourceMatteo Aquilina, Eric Budish and Peter O'Neill, Quantifying the High-Frequency Trading “Arms Race”, Financial Conduct Authority (FCA) Occasional Paper 50, .
Crypto trading and crypto derivatives
Retail outcomes and market structure in crypto, from the BIS and the UK regulator, with one clearly labelled vendor estimate.
An estimated 73–81% of retail crypto investors had likely lost money on their initial bitcoin investment.
BIS economists combined crypto app downloads and usage in 95 countries (2015–2022) with bitcoin prices; the range depends on when and how much users are assumed to have bought.
SourceRaphael Auer, Giulio Cornelli, Sebastian Doerr, Jon Frost and Leonardo Gambacorta, Crypto trading and Bitcoin prices: evidence from a new database of retail adoption (BIS Working Paper 1049), BIS Working Papers, .
The median crypto app user who bought $100 of bitcoin a month would have lost $431 of the $900 invested by December 2022.
A BIS simulation in which each new app user buys $100 of bitcoin every month after downloading the app; on that basis, over four-fifths of users would have lost money.
SourceGiulio Cornelli, Sebastian Doerr, Jon Frost and Leonardo Gambacorta, Crypto shocks and retail losses (BIS Bulletin No 69), BIS Bulletin, .
8% of UK adults held crypto assets in 2025, down from 12% in 2024.
Online survey of 2,353 UK adults run by YouGov for the FCA, with fieldwork in August and September 2025.
SourceFinancial Conduct Authority (FCA), Cryptoassets consumer research 2025 (Wave 6), Research note, .
57% of UK retail accounts trading crypto exchange-traded notes made a loss between April 2018 and December 2019.
Client outcome data the FCA used to justify banning the sale of crypto derivatives and exchange-traded notes to retail consumers.
SourceFinancial Conduct Authority (FCA), Prohibiting the sale to retail clients of investment products that reference cryptoassets (PS20/10), Policy statement, .
The FCA estimated that banning the sale of crypto derivatives to retail consumers would reduce their losses by £19m to £101m a year.
The ban on CFDs, futures, options and exchange-traded notes referencing crypto assets took effect on 6 January 2021.
SourceFinancial Conduct Authority (FCA), Prohibiting the sale to retail clients of investment products that reference cryptoassets (PS20/10), Policy statement, .
Backtesting, overfitting and research reliability
How often backtested and published trading edges hold up on new data, from peer-reviewed finance research.
Backtest Sharpe ratios explained less than 2.5% of the variation in out-of-sample performance (R² < 0.025) across 888 trading algorithms.
Researchers at Quantopian compared the backtests of 888 algorithms with at least six months of out-of-sample results. A machine-learning model using many backtest features did better, with an R² of 0.17.
SourceThomas Wiecki, Andrew Campbell, Justin Lent and Jessica Stauth, All That Glitters Is Not Gold: Comparing Backtest and Out-of-Sample Performance on a Large Cohort of Trading Algorithms, The Journal of Investing 25(3), pp. 69–80, . Figures confirmed in the paper's SSRN abstract.
With five years of data, no more than 45 independent strategy variants should be tried, or the best of them is expected to reach an in-sample Sharpe ratio of 1 even if none has any skill.
Bailey, Borwein, López de Prado and Zhu compute the minimum backtest length for a given number of trials. With one year of data, trying 10 variants of a strategy without skill already gives an expected best in-sample Sharpe ratio of 1.57.
SourceDavid H. Bailey, Jonathan M. Borwein, Marcos López de Prado and Qiji Jim Zhu, Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance, Notices of the American Mathematical Society 61(5), pp. 458–471, .
At least 316 factors have been tested to explain stock returns, so a new factor should clear a t-statistic above 3.0 rather than the usual 2.0, according to Harvey, Liu and Zhu.
The authors catalogued factors from published papers and working papers and built a multiple-testing framework to set the significance hurdle.
SourceCampbell R. Harvey, Yan Liu and Heqing Zhu, … and the Cross-Section of Expected Returns, The Review of Financial Studies 29(1), pp. 5–68, .
Returns of 97 published stock-return predictors were 26% lower out of sample and 58% lower after publication.
McLean and Pontiff compare each predictor's returns in the original study's sample, after that sample ends, and after the paper is published; they read the 26% as an upper bound on data mining.
SourceR. David McLean and Jeffrey Pontiff, Does Academic Research Destroy Stock Return Predictability?, The Journal of Finance 71(1), pp. 5–32, .
65% of 452 published stock market anomalies failed the usual significance hurdle (|t| ≥ 1.96) when replicated with microcaps controlled; 82% failed a multiple-testing hurdle of 2.78.
Hou, Xue and Zhang rebuilt the anomalies with NYSE breakpoints and value-weighted returns, which limits the weight of the smallest stocks.
SourceKewei Hou, Chen Xue and Lu Zhang, Replicating Anomalies, The Review of Financial Studies 33(5), pp. 2019–2133, .
About 45% of apparent discoveries would be false if researchers ignored multiple testing, according to a study of more than 2 million generated trading strategies.
Chordia, Goyal and Saretto built the strategies from real data and derived the t-statistic hurdles that correct for multiple testing: 3.8 for time-series and 3.4 for cross-sectional tests.
SourceTarun Chordia, Amit Goyal and Alessio Saretto, Anomalies and False Rejections, The Review of Financial Studies 33(5), pp. 2134–2179, .
A counterpoint: 82.4% of US factors replicate when multiple testing is handled with a Bayesian model, in a study of 153 factors across 93 countries.
Jensen, Kelly and Pedersen count a factor as replicated when its risk-adjusted (CAPM alpha) return is significant, a different definition from the raw-return tests above, so the figures aren't directly comparable.
SourceTheis Ingerslev Jensen, Bryan Kelly and Lasse Heje Pedersen, Is There a Replication Crisis in Finance?, The Journal of Finance 78(5), pp. 2465–2518, .
Edgoro's own data
These figures come from Edgoro's own backtests on public exchange data, not from the sources above. Each study publishes its rules, costs and limits; they are simulations of the past, not track records, and they describe one market and one period.
Crypto scalping: 2,160 strategies against retail fees
- About 2,160 scalping configurations on Binance 1-minute data for BTC, ETH and SOL (January 2024 to November 2026): all six finalists that use market orders failed the sealed 495-day test.
- A hypothetical bot with perfect foresight of the next 1-minute candle averaged 0.066% per trade, against a retail round-trip cost of 0.22%.
- The one idea left, a resting buy order 2% below the price, made +7.0% in 43 trades, but the 95% confidence interval of its average trade (−0.26% to +1.14%) includes zero.
Leverage: 240 setups chasing +10% a day
- 240 leveraged setups on BTC, ETH and SOL perpetual futures (x5 to x100) aiming for +10% a day, with fees, funding and liquidations: 6 made money on average, the best +0.66% a day, and that rule lost money on 2020 to 2023 data.
- None of the 240 accounts grew when gains were compounded; the best ended with €8.73 of its starting €100.
- At x20 with no stop, a coin flip hit the +10% target on 86% of days, as often as the trend rule, and still lost €1,535 over the period through 70 to 77 liquidations.
The 200-day moving average filter on BTC, ETH and SOL
- Holding BTC, ETH or SOL only above the 200-day average (since 2018 for BTC and ETH, 2021 for SOL, at 0.30% per round trip) raised the Sharpe ratio on all three coins, from 0.67 to 0.74 on BTC.
- It cut the worst drawdown by only 16% on BTC, 18% on ETH and 31% on SOL, short of the one-third cut set in advance as the pass mark.
- In 2022 the filter returned 0.0% on BTC against −64.2% for buying and holding, yet it lost money on 25 of its 33 round trips on BTC.
A Bitcoin grid bot, restarted every month for 58 months
- A fresh ±10% spot grid on Bitcoin every month from January 2022 to October 2026, replayed on 1-minute data: positive in 64% of months, against 55% for holding.
- Its average month was −0.39%, against +2.03% for holding, and its worst month −34.04%.
- The price stayed inside the ±10% range for a whole month in only 12 of the 58 months.
Walk-forward test of moving-average settings on Bitcoin
- 25 moving-average settings on Bitcoin daily data (2018 to October 2026), re-chosen every six months: the chosen settings averaged an in-sample Sharpe ratio of 1.44 and 0.22 on the six months that followed.
- Walk-forward efficiency was 0.25: three quarters of the in-sample performance came from the selection.
- The setting that looked best in hindsight had a deflated Sharpe ratio of 61% after 25 tries, which can't be told apart from luck.
Updated as new data comes out. Add Edgoro as a preferred source on Google to see these first.
How these were checked
- Primary sources only. Each number was read in the regulator's or central bank's own page or PDF, or in the paper itself, not in an article quoting it. News coverage and statistics sites were used only to find the original.
- Abstracts are flagged. Where only a paper's abstract could be read, the source line says so.
- Context travels with the number. Who was measured, where and over which period matters: the share of losing CFD accounts, for example, was measured before European and Australian leverage limits, and fell after them.
- Dropped when unconfirmed. Popular figures we couldn't trace to a primary source, or whose primary text gave a different number, are not on this page. Vendor figures are labelled as such.
- Corrections. Spotted an error or a newer edition of a source? Write to us through the contact page; the page is re-verified and its date changes.
The terms used here, such as CFD, overfitting and the deflated Sharpe ratio, are defined in our trading glossary.
How to cite this page
Cite the primary source of each figure first. To cite this collection:
Edgoro, “Algorithmic trading statistics”, edgoro.com/research/algorithmic-trading-statistics, last verified November 12, 2026.
Trading involves substantial risk of loss and is not suitable for every investor. Edgoro is a software tool, not a broker or financial adviser: nothing it produces is investment advice, and past or backtested performance does not guarantee future results.