Confirmed! Is the "smart money" following the order still losing? Over 60% of trades are noise; you're only copying the market makers' "test orders"!

You did everything seemingly correctly. Found those high-performing wallets, studied their history, and identified traders in prediction markets who make money based on skill rather than luck. You set up copy trading and waited. But the results were confusing: some trades made money, others quietly lost. Eventually, you even wondered if flipping a coin might be better.

Here’s a rarely discussed fact in market analysis: finding smart money addresses only solves half the problem. The mainstream view is that identifying top traders and copying their actions can lead to profits, but this intuitive logic has a fatal blind spot.

A backtest on the past 90 days of trading activity from the top 2000 wallets in a major prediction market reveals the truth. The test included only market maker buys, each fixed at $100, totaling 1,147 trades. The method was straightforward, data clean. The results should alert all copy traders: among these top wallets, up to 62% of “confidence scores” are below 0.2.

This means that even elite traders mostly operate with low signal-to-noise ratios. These are exploratory positions, hedges, or casual attempts in markets they’re not fully confident in. They stem from curiosity, not certainty. When you copy an address, you’re not just copying its best decisions but all its actions, including moments of casualness.

A concept worth borrowing from quantitative trading is the Kelly criterion. Its core is that the optimal bet size depends not only on the win rate but also on confidence in that win rate. A trader with a 60% chance of success, backed by thorough research, has a vastly different betting value than someone who stumbles upon the same probability by chance.

Professional traders instinctively follow this principle. Their portfolios typically mix three types of trades: exploratory trades to test the market and liquidity; hedging trades for risk management; and well-researched, heavily invested “high-confidence” trades. On-chain, these look identical—a buy record. But only the last type is worth following.

“Confidence scores” are designed to reverse-engineer these trade types. They analyze signals such as the proportion of position size relative to total assets, repeated purchases in the same market, and historical accuracy in similar situations, compressing this information into a score from 0 to 1. The goal is to answer a core question: how confident is the trader themselves in this trade?

Backtest data shows that as filtering criteria become stricter, average profit/loss and win rate improve, but trading frequency decreases. This trade-off is objective—there’s no free lunch. The key is how you balance it.

To quantify this relationship, the analysis defines a “marginal efficiency” metric: how much average profit/loss increases for every 1% sacrifice in trading coverage. Data shows that confidence scores between 0.4 and 0.6 have the highest marginal efficiency, effectively filtering out the strongest signals in the dataset, even though such trades are relatively rare.

Therefore, there is no universal “optimal” threshold. A more rational approach is to offer three differentiated strategies based on trader type: for large capital traders needing volume to deploy capital, use the 0–0.2 range, accepting more noise for higher trading frequency; for average traders seeking balance, 0.2–0.4 is the default, maintaining 30%–60% opportunity while significantly improving returns and win rate; for small capital or highly selective traders, the 0.4–0.6 range yields fewer signals but each represents a high-confidence decision with the highest marginal efficiency.

The essence of copy trading is not imitation but translation. Most people understand it as “they buy, I buy.” A more effective understanding is: what does this trade convey about the trader’s belief, and how strong is that belief? A small exploratory position in a new market by one trader is different from another trader adding to a position held for two weeks; on record, they look the same, but the messages they send are entirely different.

That’s what confidence scores are for—a translation layer. Its purpose isn’t to predict whether a trade will be profitable but to answer a more fundamental question: does the trader truly believe it will be profitable? Because a trader with conviction will invest heavily, base decisions on historical experience, and keep adding to positions rather than testing the waters. Such traders are fundamentally different from casual experimenters. When you follow, you’re not just copying an address but its judgment. Now, you have a way to distinguish the difference.

Follow me: for more real-time analysis and insights into the crypto market! $BTC $ETH $SOL

#Gate February Transparency Report