How to Choose a Good Copy Trader: Four Checks, Scored on the Day You Would Have Made Them

Every guide to choosing a copy trader gives the same list: look at the track record, the drawdown, the return, the number of followers, the fee. None of them says what any of those things bought the people who used them. We can, because we hold the data to score the advice: every ninety-day copy that could have started on any day of any followed forex strategy, classed by what the card showed on that day, and the published follower pools of every followed crypto leader.

That is the test a "how to choose a good copy trader" article should pass. A criterion is only worth anything if a follower could have read it on the day they joined and if the next ninety days were then different from what they would have been otherwise. So on the forex side, RoboForex's copy-trading service, we asked exactly that: for each of 748,826 entry days across 1,789 strategies with a follower and a full ninety days of history, what did the curve show at that moment - how old was the account, how deep had it ever fallen, what number was on the card, what had the previous quarter done - and what did a ninety-day copy started that day return after the trader's fee. On the crypto side, Bybit, the board publishes for each leader what their followers collectively made over the trailing ninety days, and we cut that by the card's own fields. The forex data is as of 14 September 2026, the crypto data as of 12 September 2026.

The one-line answer: no check on the card raised a follower's odds of a winning quarter by much - passing all four of the ones that matter left the share of profitable copies at 59.7% against 58.5% without them. What the checks changed was the size of the losing quarter: the share of copies that lost half or more fell from 3.1% to 0.1%, and the tenth-worst copy in a hundred went from -20.3% to -5.4%. Choosing well is not how you find the winners. It is how you avoid the wipe-outs, and the price is a smaller median and more time when nothing happens.

What "good" has to mean for the follower

A good copy trader, for the person copying, is one whose next ninety days are worth having: the copy ends in profit after the fee, and when it does not, it does not end in a hole. That is the whole definition used here, and it is deliberately the follower's, not the leader's. Whether copy trading works at all is a separate question we answered earlier; this piece assumes you are going to copy someone and asks how to pick.

Two things about the method matter for reading the numbers. First, on the forex side every criterion is measured on the entry day, from the curve as it stood then. The account's age is its age that day; the deepest fall is the deepest fall it had shown by then; the number on the card is the cumulative return the card displayed that morning. Nothing from after the entry leaks into the class. This sounds obvious and is not: grouping strategies by their drawdown as it stands today, and then scoring windows from their whole life, lets a fall that happened inside the window decide which group the window sits in. The earlier article did that for a descriptive purpose; for advice, it would be circular, so here the classing is dated.

Second, on the crypto side that dating is not possible. Bybit publishes a card, not a curve, and the card's drawdown, ROI and win rate are measured over the same trailing ninety days as the followers' profit. So a Bybit row like "drawdown under 15" describes leaders whose quarter was calm and whose followers did well in that same calm quarter - a fact, and a useful one, but not a rule you could have applied before the quarter began. Where the two platforms sit side by side below, the forex figure is the one that answers "what if I had checked this first," and the crypto figure is the one that says whether the same pattern exists on a board that reports its followers directly. Read the direction across the two; do not read the size.

Unless stated otherwise, every forex figure gives equal weight to every entry day, so that a strategy with three years of curve contributes more windows than one with four months. Weighting each strategy equally moves the numbers a little and never changes the direction, and where it would matter the text says so.

Check one: the deepest fall the curve has shown so far

Start with the check that separated outcomes most cleanly, because it is the one most guides put third or fourth.

For every entry day we took the deepest peak-to-trough fall the curve had shown up to that morning - not the card's field, the curve itself, as a follower scrolling the chart could have read it. Where that fall was under 15%, 63.1% of ninety-day copies ended in profit after the fee, 2.3% lost half or more, and the tenth-worst copy in a hundred returned -12.8%. Where the curve had already fallen by half at some point, 52.4% ended in profit, 4.4% lost half or more, and the tenth-worst copy returned -27.9%. Between them, 58.9%, 2.6% and -19.4%. Every column moves the same way, and the tail moves fastest: the odds of a good quarter fall by ten points across the range, the size of a bad one roughly doubles.

The crypto board says the same thing more loudly, for the reason given above. Followers' pools were in profit for 66.8% of leaders whose ninety-day drawdown stayed under 15, for 47.9% at 15 to 50, and for 15.7% at 50 or worse; the last group alone summed to -$2.32 million, more than the whole board's loss. That is the quarter describing itself, not a forecast - but the direction matches the forex one, which was dated.

One detail decides how you apply this. On RoboForex the card's "Max drawdown" column is the platform's worst single step, not the deepest peak-to-trough fall, and it understates what a follower actually watched; we have documented that measure before. So we also classed the windows by the worst single-day step the curve had shown, the nearest thing to the card's field. It separates less: 59.4% profitable under 15%, 58.2% at 15 to 50%, 55.0% at 50% or worse. The number the card offers is the weaker version of the number that matters. Open the "All" tab of the strategy's yield chart and read the fall off the line, peak to trough, yourself. On Bybit the card's field is the right shape but the wrong memory: it lives inside the trailing window and a bad stretch ageing out takes the drawdown with it, which is why the leader card needs reading against a longer history.

Where it is: on the RoboForex card, the "Max drawdown" column on the list and the "Maximum drawdown, %" chart under Stats - both single-step; the "All" period of the yield chart is where the real fall lives. On the Bybit card, "Max drawdown" - measured over the window shown, so switch to the ninety-day view and remember it resets.

Check two: the number on the card

The biggest figure on any card is the return, and it is the one most followers sort by. What did it predict?

On the forex curves, the size of a positive card bought almost nothing in odds. Copies started when the card showed between 0 and 50% ended in profit 65.2% of the time; between 50 and 200%, 65.8%; above 200%, 67.7%. Two and a half points across a range that runs from "modest" to "quadrupled." What the bigger number bought was a fatter tail on both sides: the median copy rose from +3.3% to +6.7%, and the tenth-worst copy fell from -14.6% to -30.9%, with the share that lost half or more going from 2.2% to 5.2%. A card above 200% is a card whose curve moves a lot, and a copy inherits the movement in both directions.

The card that did predict something was the one in the red. Copies started when the cumulative return was zero or negative ended in profit 40.7% of the time - but 28.2% of them never moved at all, because a curve that is flat at zero is usually an account that has stopped trading, and a copy of an idle account earns nothing and pays nothing. Count only the copies that moved and a negative card still trailed: 56.7% of them ended in profit, against about 70% for any positive card counted the same way. So the headline return works as a filter at the bottom, not as a ranking at the top: below zero is a warning, and above zero the size of the number is mostly telling you how wide the swings will be.

The crypto board agrees, with its same-window caveat. Followers' pools were in profit for 69.3% of leaders whose ninety-day ROI sat between 0 and 20, for 67.0% between 20 and 100, and for 60.9% above 100. The 89 leaders above 100% carried 2,679 seats and summed to +$404,329, but that sum sits mostly on one account, which is what a fat tail looks like from the profitable side. Below zero, 13.9% - the quarter describing itself.

Where it is: the "Yield" column on the RoboForex list, which is the all-time cumulative return, and the "Yield, %" chart on the card with its period tabs; the "ROI" on the Bybit card, which is the trailing window's and not the account's life.

Check three: what the last quarter did

Followers do not arrive on random days. They arrive after the chart has looked good, which is exactly the moment the card is least informative about the next ninety days.

We classed every entry day by what the same curve had returned over the previous ninety days. After a quarter of +20% or more, 69.0% of the next ninety-day copies ended in profit - and 4.2% lost half or more, with the tenth-worst copy at -28.9%. After a quarter between 0 and +20%, the share in profit was slightly higher, 70.1%, while the share that lost half was 1.9% and the tenth-worst copy was -13.2%. The same odds of a good quarter; a bad one twice as deep. After a losing quarter, 49.3% - roughly a coin toss again - and after a quarter in which the curve did not move at all, 10.7%, because 82.8% of those copies stayed idle too.

The previous year tells the same story on a longer scale. After a year of +50% or more, 70.7% of copies ended in profit and 4.4% lost half; after a year between 0 and +50%, 62.3% and 1.4%. The stronger history bought eight points of odds and three times the tail.

Whether the entry day was a new high of the curve mattered as well, and in the direction the intuition does not expect: copies started on a day the curve had just set a new high ended in profit 68.4% of the time, against 54.6% for days below the high. That is not a reason to chase highs; the "below the high" group includes every day inside a drawdown and every idle stretch, which is where the poor copies live. It is a reason not to treat a fresh high as a sell signal for a copy, which some guides do.

The crypto board cannot be cut this way - it has no curve to look back along - and the nearest thing it offers is the ROI band above, where the leaders whose quarter was hottest had followers in profit least often among the positive bands.

Where it is: the period tabs on the RoboForex yield chart - "Three months" against "All" tells you whether you are arriving after a hot quarter; on Bybit, the 7-day, 30-day and 90-day toggles on the card do the same job in a shorter frame.

Check four: how long the record is

Length of track record is the first item on every list, and it is the one where the data has to be read most carefully.

Copies started in an account's first ninety days ended in profit 67.5% of the time; between 90 and 365 days, 60.0%; in the second year, 54.8%; after two years, 53.1%. Read plainly, that says younger is better, and it should not be read plainly. The early windows we can see belong to strategies that survived long enough to be on the list today. A strategy that blew up in its second month two years ago has no curve to score and no seat to count; our record of departed accounts reaches back only to 25 August, so adding them barely moves the figures (66.7%, 58.6%, 53.3%, 52.4%). The honest reading is narrower: even among the survivors, age did not buy a higher share of profitable quarters.

What age did buy was, again, the tail and the idle time. The share of copies that lost half or more fell from 4.5% in the first ninety days to 2.2% after two years; the tenth-worst copy went from -25.4% to -15.8%. And the share of copies that never moved rose from 5.0% to 16.8%, because old accounts on this list are often accounts that have gone quiet. Cross age with the fall and the pattern sharpens: among strategies two years old whose curve had never fallen 15%, only 41.2% of copies ended in profit and 30.2% never moved, while the tenth-worst copy was a mild -9.6%. An old, calm strategy is frequently a parked one. The group that combined a reasonable share in profit with a thin tail was the three-to-twelve-month strategies whose curve had not yet fallen 15%: 63.0% in profit, 1.9% lost half, tenth-worst copy -11.6%.

On the crypto board the age groups sit within six points of each other - 45.5% of followers' pools in profit under ninety days, 45.4% at 90 to 365, 44.5% in the second year - and the oldest leaders, past two years, were lowest at 39.8%. Length of record on its own was not the tell there either.

Where it is: the start of the "All" yield chart on the RoboForex card is the account's first trade; the Bybit card shows the leader's days on the board.

What the card shows that did not help

The fee. On neither platform did the trader's cut of profit sort good copies from bad in any direction you could use. On the forex list, strategies charging 5 to 15% had most of their windows in profit 66.2% of the time, those at 20 to 25% 67.1%, those at 30% or more 61.6%, and those charging nothing 58.6% - the free ones came last. On the crypto board, followers' pools were in profit for 44.1% of leaders at the default 10% and 41.0% above it, 42.2% at zero. The fee is a price, and what the price does to a copy is worth knowing, but it is not a criterion for choosing whom to copy. These forex fee groups are by today's card rather than the entry day, because our panel has only recorded fees since late August; the direction is the same on the board that does date them.

The win rate. Bybit prints one; RoboForex does not. The leaders with a win rate of 90% or more had the highest share of followers' pools in profit of any band, 64.4%, and the most negative sum of any band, -$994,334. A strategy that closes almost every trade at a gain is a strategy that holds its losers, and the two facts belong together - the smooth chart that ends badly is the same account seen from two sides. A high win rate is a description of the losing style, not a safety feature.

The crowd. The number of followers did predict, on both platforms and in the same direction: on the forex list, strategies with a hundred or more seats had most of their windows in profit 88.9% of the time against 56.9% for those with one to four, and on the crypto board the 31 leaders with a hundred or more followers were the only group whose pools summed positive. But the seat counts here are today's - the only ones our panel can date are three weeks old - and a follower joining early cannot see where the crowd will be. It is a fair tiebreak between two otherwise equal cards, and no more.

All four checks at once

Put the four dated checks together into one rule a follower could apply on the morning they join: the account is at least a year old, its curve has never fallen more than 15% from a high, the card is above zero, and the previous ninety days returned less than +20%.

On the forex curves, 143 strategies passed all four on at least one day, giving 31,624 entry days. Copies started on those days ended in profit 59.7% of the time. Copies started on the other 717,202 days - every day that failed at least one check, on every strategy - ended in profit 58.5% of the time. One point. That is the whole effect of the checklist on the odds of a winning quarter, and it is the finding this article exists to report, because it is not what the guides promise.

What the checklist did was remove the disaster. Of the copies started on passing days, 0.1% lost half or more, against 3.1% on the others; the tenth-worst copy in a hundred returned -5.4% against -20.3%. The price was paid in the middle and at the top: the median copy on a passing day returned +1.0% against +2.5%, and 15.1% of passing-day copies never moved at all, against 12.6%. Weighting strategies equally instead of days gives 62.1% against 61.6% in profit and 0.7% against 4.8% lost half - the same shape.

On the crypto board the same rule, read off today's card, splits the leaders far more sharply: followers' pools were in profit for 79.0% of the 204 leaders who passed and 39.5% of the 1,566 who failed, with sums of +$61,483 and -$1,942,760. The width of that gap is the width of the same-window effect - a leader whose drawdown stayed under 15 over the quarter is a leader whose quarter was calm, and calm quarters are where followers' pools are in profit. It confirms that the pattern is real on a board that reports followers directly. It does not tell you the rule would have picked those leaders in advance. The forex figure, which was dated, is the one that says what advance picking is worth: a thinner tail, and about the same odds.

Where each check lives on the card

Neither card shows the one thing the follower most needs, which is the follower's own outcome: the forex platform does not publish it, and the crypto platform publishes it as a pooled total that mixes everyone who ever joined. The rating we keep on the forex side reads every listed strategy daily and scores it on its whole curve, and the crypto rating reads the full board rather than the top of it, so that the fall a card has forgotten is still on record. The scoring rules are on the methodology page.

So, how to choose a trader to copy

Read the fall off the chart yourself, peak to trough, and prefer a curve that has never given back more than 15%. Treat the return on the card as a filter - above zero - rather than a ranking, and treat a very large one as a promise of large swings. Do not arrive right after a +20% quarter; the same odds are available a quarter later with half the tail. Do not pay for age as such; pay for age without a deep fall and without a long idle stretch. Ignore the fee when choosing and the win rate always. And then expect what the data says to expect: about six copies in ten will end the quarter in profit whatever you do, and the checks decide how the other four end.

The practical consequence is that choosing is only half the job. If a good choice cannot lift the odds much above the field's, then how much you put into one strategy and how many you hold decide the rest, because the quarter that goes wrong is the one thing the checks reliably shrink but cannot remove.

Every figure above is our own computation from the platforms' per-account figures: RoboForex curves and offers read on 14 September 2026 through the same public API the platform's own pages use, with each ninety-day copy classed by what the curve showed on its entry day and the fee charged as the platform's help pages describe; Bybit's per-leader ninety-day fields read on 12 September 2026 for every leader with at least one follower. Forex follower outcomes are simulated on the trader's curve and assume proportional copying without spread or swap; crypto follower outcomes are the platform's published pools. Windows overlap and are not independent trials; the early history of strategies that left the list years ago cannot be scored, so the age figures describe survivors. Seats are open subscriptions, not people or money. We will re-run both computations as the panels grow.