- Introduction
- The Dataset Behind the Numbers
- Most Losing Trades Leave a Trail Before They Close
- Your Trading Habits Are Part of the Signal
- The Crowd Still Has an Edge Worth Measuring
- What Traders Can Take From This
- Put Your Market Calls to Work
Our models trained on more than 11.7 billion individual retail trades from over 1 million trader histories, which came from several years of live market activity. That’s a lot of late-night decisions. Some of what we found confirmed what experienced traders have said for decades. The rest has plenty to say about how you place your next order.
The Dataset Behind the Numbers
Eleven point seven billion is hard to picture. Divide it across our million-plus trader histories and you land under 12,000 trades per person on average, which is still years of clicking buy and sell for most of us.
Every trade came from a real account in a live, regulated market. Real money moved each time, so every record reflects a person with something to lose. Our models score each trade on more than 200 separate features & produce a probability of a win or a loss before the order goes in.
Most Losing Trades Leave a Trail Before They Close
Our first large test focused on one market: Bitcoin perpetual futures. We took 12 months of real retail trades, had our models score each one with only the information available at entry, and then compared those scores against what happened. The models flagged 92% of the trades that went on to lose money. A flag means the model rated the trade a likely loser before it ever opened.
That figure comes from a backtest, and we want to be upfront about it. A backtest replays history. Live markets add slippage & thin order books, plus a talent for humbling anyone who gets confident. A full year of data still told us something we had suspected for a long time: the majority of retail losses in leveraged markets follow patterns our models can spot ahead of time.
Your Trading Habits Are Part of the Signal
Price action explains part of a losing trade, the person who places it explains the rest. Our models read each trader’s own profit and loss history next to the mood of the wider market. A million personal records gave them plenty of repeat behavior to study.
Your trading history tells our models a lot about your next trade. Someone who revenge-trades after three bad days has a habit, and habits repeat – the fourth bad day tends to look like the first three. The good news runs the other way too. A trader with a steady, disciplined record keeps producing one, and our models learn to trust that record over time.
The Crowd Still Has an Edge Worth Measuring
A pile of 11.7 billion trades holds a lot of losers. It also holds traders who get it right week after week, and those people shaped how we built our models.
Our models read the mood of the whole pool & then weigh each trader by how reliable their record has been. Consistency scores higher than flash. A trader who grinds out 3% a week across 40 trades gives us a cleaner read than someone who doubles an account on Monday and hands it all back by Thursday (the Monday trader is more fun at parties, to be fair).
Short holding periods help as well. A call that opens and closes within a day or so tells us whether the trader read the move right. A position left open for six months picks up so much outside noise from earnings & rate decisions that the original idea becomes hard to judge at all, the trade outlives the reason for it.
What Traders Can Take From This
Four habits from this research work fine without a machine learning model behind them.
Measure every trade in percentage return. A $400 gain means one thing on a $2,000 position and something else on a $60,000 one, and percentage puts both on the same footing. Give any new setup 30 or 40 trades before you decide it works. Four trades tell you about luck and not much else.
Size your positions as though rent money were on the line, even in a practice account… the habits you build there are the habits you’ll keep. Write down why you entered each trade, too. A one-line reason that you write before the order goes in makes it much harder to fool yourself afterward.
Revenge trading deserves its own warning. It excels at one thing, and that thing is turning a bad Tuesday into a bad month.
Put Your Market Calls to Work
The research above powers the models behind UpsideOnly. Our paper trading platform gives every user a $100,000 virtual balance across 25 global markets, with stocks, indices, commodities, forex & crypto all available. Your simulated trades feed the same models that studied those 11.7 billion trades. When the models trade our own capital at a profit, we share 50% of those profits with eligible users. Only eligible users receive payouts, and you never put your own money into a position.
Some users add a small Participation Deposit to qualify for larger payouts. The deposit is refundable on demand and never at risk.
Leave a Reply