-
Results provide early evidence that BayesShield can identify potentially profitable trading signals by finding repeatable skill in trader activity
-
375-day backtest using Perpetuals’s production code generated a hypothetical 380% non-compounded return, a period in which Bitcoin fell 40.2%
-
The strategy achieved a return-to-drawdown ratio of 4.4, a measure comparing its overall return with the largest decline it experienced during the test
Perpetuals.com Ltd announced backtest results for version two of BayesShield, its patent-pending AI engine developed for UpsideOnly.
Read More on Fintech : Global FinTech Interview: AI and the future of fintech with Hugh Cumming, CTO, Vena
UpsideOnly is a risk-free paper trading platform where users simulate trades using virtual money across global equity, commodity, foreign exchange and crypto markets. The platform is powered by BayesShield, an AI model that uses machine learning to identify consistently successful traders. It generates automated signals when enough qualified traders independently take the same position at approximately the same time. These signals inform UpsideOnly’s trading strategies, with profits shared among eligible top-performing users.
Using the same production code that powers the live platform, Perpetuals replayed approximately 1.07 billion historical order fills. Over the 375-day test period, the strategy generated a hypothetical, non-compounded return of 380%, with a maximum drawdown, meaning its largest decline from a previous peak, of 27%. It achieved a return-to-drawdown ratio of 4.4, indicating that its overall gains were substantially greater than its largest setback during the test.
Starting with $400,000 in simulated deployed capital, the system generated $1.522 million in hypothetical profit, ending the test with $1.922 million. If replicated in live markets, the results point to a different role for AI in trading: identifying repeatable human skill across large volumes of market activity and turning it into automated trading signals, rather than trying to predict prices directly.
Encouragingly, the strategy remained resilient under market stress. It recorded a positive return in every calendar quarter and zero liquidations across the 375-days, a period when Bitcoin fell 40.2%. During the sharp market sell-off on October 10, 2025, the system was essentially flat, declining by just 0.7%, while Bitcoin fell 16% within hours.
“This massive backtesting effort has provided promising signals that our BayesShield AI algorithm is working as intended,” said Patrick Gruhn, CEO of Perpetuals. “Our early live, real-money results have been encouraging and generally consistent with this testing”
This testing protocol was designed to reduce bias and prevent the use of future information, drawing on data from more than 429,000 trading accounts, 70.8 million positions sourced from PM MTF Ltd and more than 22 billion historical retail trades.
UpsideOnly launched in May 2026 and now has more than 225,000 users and recorded $70 billion in simulated notional trading volume across 25 assets. The live trading period remains too limited to draw conclusions about future performance.
Catch more Fintech Insights : Global FinTech Innovations Are Transforming Banking into Continuous Financial Guidance
[To share your insights with us, please write to psen@itechseries.com ]
