Position Sizing Beyond the 1% Rule: Fixed Fractional, Kelly, and Volatility-Adjusted Sizing

You already know the rule this article starts from: risk 1% to 2% of account equity per trade, and derive position size by dividing that dollar risk by the distance from entry to stop. That calculation, R-multiples, and the correlation evidence all live in managing risk in crypto, and stop placement itself lives in where to place a crypto stop-loss. This article takes both as inputs and asks the next question: given a risk budget and a stop distance, which sizing model turns them into a number, and what does each model do to your equity curve.

Why fixed fractional and fixed dollar diverge, and which one you actually want

Both models take the same risk budget. Fixed fractional recomputes the dollar risk from current equity before every trade. Fixed dollar fixes it once, usually from starting equity, and leaves it there. The difference is not cosmetic and it does not average out.

A worked comparison over eight trades

Start with $10,000, risk 2% per trade, and assume every trade resolves at either −1R or +2R. The sequence is five losses followed by three wins. The fixed dollar trader stakes $200 every time; the fixed fractional trader stakes 2% of whatever equity is left.

Run the same eight results in the opposite order and both models finish on exactly the same numbers: $10,167.88 and $10,200. Fixed fractional's final equity is a product of per-trade multipliers and fixed dollar's is a sum of per-trade amounts, and neither operation cares about order.

What happens on a streak that does not end

The models only genuinely part company when the losing run is long enough to matter.

What these examples actually tell you

Fixed fractional cannot reach zero, and that is a property of multiplication rather than a property of discipline. Multiplying by 0.98 an unlimited number of times produces a number that decays toward zero without arriving. Fixed dollar subtracts a constant, and subtraction does arrive: at 2% of a $10,000 start, the fiftieth consecutive loss lands exactly on zero. Order independence holds for both models right up to that point, and then it stops holding for fixed dollar, because a sequence that hits zero truncates and there is nothing left to trade the remaining results with. That truncation is the whole of the "cannot be ruined" claim. It is not a claim that fixed fractional protects you: 36% of your starting capital after fifty losses is not a survivable outcome in any practical sense, it is simply a non-zero one.

The price of that floor is a slower recovery, and the eight-trade table shows exactly where you pay it. Fixed fractional is ahead by $39.21 at the bottom of the drawdown and behind by $32.12 at the end of the recovery, because it staked $180.78 on the first winning trade while the flat trader staked $200. It de-risks into weakness by construction, which means it also re-risks into strength late. The cleanest version of that: ten losses then ten wins, all at exactly ±1R and 2% risk, leaves the fixed dollar trader at precisely $10,000 and the fixed fractional trader at $9,960.07, because 0.98 × 1.02 = 0.9996 and that shortfall compounds ten times. A symmetric round trip costs the fractional sizer 0.4% and costs the flat sizer nothing.

So the choice is a statement about what you expect next, not about which model is better. Use fixed fractional when the account is the thing you are compounding and you intend to trade the same strategy for years: it scales you up automatically as equity grows and it removes the decision of when to increase size. Use fixed dollar when the account is a fixed allocation you top up or draw down from an outside source, when you are inside a defined evaluation period, or when the strategy's edge is measured in R and you want the trade count rather than the equity path to be what the sample is testing.

The Kelly criterion: what it optimizes, and why full Kelly is not a target

Kelly's 1956 paper answers a narrower question than most people assume. It does not maximize expected wealth. It maximizes the exponential rate of growth of capital, defined as the long-run average of the log of the wealth multiple, and Kelly is explicit that maximizing expected wealth instead means betting everything, which leaves you "broke with probability one" if you keep playing.

For a bet where you risk a fraction f of equity, win with probability p, and a win pays b times what you risked, the optimizing fraction is:

f* = (p × b − q) ÷ b, where q = 1 − p.

Three assumptions carry the result, and each is shaky in crypto. Kelly treats p and b as known parameters, not estimates with standard errors. It requires independent, identically distributed outcomes, which trades taken in the same regime are not. And it assumes the distribution you specified is the real one, whereas a two-outcome or log-normal model has no room for the days an asset moves five or ten times its recent range, which are precisely the days the sizing decision is decided on.

A worked example on a realistic crypto win rate

Take a trend strategy with a 40% win rate and winners that average 2R, a distribution the risk management guide shows is comfortably profitable, with expectancy of 0.40 × 2 − 0.60 = +0.20R per trade.

Full Kelly: f* = (0.40 × 2 − 0.60) ÷ 2 = 0.20 ÷ 2 = 0.10. Risk 10% of equity on every trade.

That is five to ten times the 1–2% rule, and it is the correct answer to the question Kelly asked. Now change one number.

What a five-point error in the win rate does

Suppose you backtested the same strategy and measured a 45% win rate instead of 40%, an error well inside the sampling noise of a few hundred trades.

Full Kelly: f* = (0.45 × 2 − 0.55) ÷ 2 = 0.35 ÷ 2 = 0.175. Risk 17.5% of equity.

A five-point overestimate of the win rate raised the recommended size by 75%, and put you at 1.75 times the true Kelly fraction. Here is what that costs, computed on the true 40% distribution.

The trader who overestimated the win rate by five points ends up with less than the trader who deliberately halved, on a strategy that genuinely works. Push a little further and it gets worse than "less": at 20% of equity, twice true Kelly, the growth rate is 7% of optimal, and past roughly 20.4% it turns negative. A strictly profitable strategy, sized wrongly, loses money.

Why the error is asymmetric, in one line

Figure 1 is the whole argument. Thorp's continuous treatment of Kelly gives the growth rate at a fraction c of full Kelly as c × (2 − c) of the maximum, and that is the curve plotted.

Read it in three steps, using the parameters as they are plotted.

1. Find c = 1 on the horizontal axis. The curve peaks there at 100%. This is full Kelly, and it is a maximum, which means the curve is flat around it: at c = 0.75 and c = 1.25 the growth rate is still 93.8%. Being slightly wrong about the size costs almost nothing.
2. Compare c = 0.5 and c = 1.5. Both read 75%. Underbetting by half and overbetting by half cost exactly the same growth. On its own that looks symmetric.
3. Now follow the curve past c = 2 and look for a mirror image on the left. At c = 2 the growth rate is zero; at c = 2.5 it is −125% and still falling. There is no value of c on the left-hand side that produces a negative growth rate; the curve simply approaches zero as c approaches zero. Underbetting is bounded by doing nothing. Overbetting is not bounded by anything. That asymmetry is why an edge estimate you are unsure of should be shaded down rather than up, and it is the mathematical form of Thorp's observation that forecast excess returns are more likely to be too high than too low.

The curve is the log-normal result. On the discrete two-outcome example above, half Kelly retains 76% of the growth rate rather than exactly 75%, and the zero-growth point sits at 2.04 times Kelly rather than exactly 2. Same shape, marginally different constants.

Fractional Kelly, and the drawdown number that matters

Thorp gives three results for betting a fraction c of the Kelly amount, and the third is the one that decides the argument:

• Growth rate scales as c(2 − c) of the maximum. Half Kelly keeps three quarters of it.
• The standard deviation of the growth rate scales as c, so half Kelly is exactly half as volatile in the continuous model and 0.51 times as volatile on the discrete example above.
• The probability of your bankroll ever being reduced to a fraction x of its starting value is x raised to the power 2m/s². At full Kelly that exponent is 1, so the probability of ever halving your account is exactly one half.

Substituting the fractional-Kelly expressions for growth and standard deviation into that third result makes the exponent 2/c − 1. So the chance of ever halving the account is 1 in 2 at full Kelly, 1 in 8 at half Kelly, and 1 in 128 at quarter Kelly, against growth rates of 100%, 75% and 44% of the maximum. That is the trade "use fractional Kelly" is actually describing: giving up a quarter of the growth rate buys a four-fold reduction in the probability of ever halving.

Figure 2 plots that probability across the same horizontal axis as Figure 1, so the two curves can be read against each other.

Read the two figures together, because neither is the argument on its own.

1. Take c = 0.5 on both. Figure 1 gives 75% of the maximum growth rate. Figure 2 gives a 12.5% chance of ever halving the account.
2. Move both to c = 1. Growth picks up its last 25%, and the halving probability goes to 50% — four times higher for that final quarter.
3. Now go to c = 1.5 and compare it with c = 0.5. Figure 1 reads 75% at both points, exactly as noted above. Figure 2 reads 12.5% at the first and 79.4% at the second. Identical growth rate, more than six times the probability of ruin. Whenever you are betting above full Kelly there is a smaller bet with the same expected growth and a fraction of the drawdown, which is why the region to the right of c = 1 is never the rational place to stand.

Both figures are the continuous model, which is what makes them directly comparable to each other. Read against the discrete two-outcome example instead and the constants shift the way they shifted for the growth rate: the shape holds, the exact numbers do not. Use the curves for the shape of the trade-off and the worked table above for the arithmetic of any specific bet.

What this example actually tells you

Kelly is a diagnostic, not a size. Its real output is the number 10%, which says two things at once: this edge is large enough that a 1% risk leaves growth on the table, and your estimate of that edge would have to be accurate to a fraction of a percentage point for 10% to be defensible. Use it to check the risk you have already chosen. If your rule sits above full Kelly on your own measured statistics, you are overbetting a strategy that works.

Then treat half Kelly as a ceiling rather than a target. Thorp recommends choosing f between half the estimated Kelly fraction and the estimate itself, capping the growth penalty at 25% while protecting against the estimate being wrong. On the 40% / 2R example that is a 5% risk, still well above the conventional rule, and the conventional 1–2% is quarter Kelly or tighter, defensible precisely because nobody knows their true win rate to five points.

Volatility-adjusted sizing: the variable that breaks a flat percentage rule

This is the part most crypto sizing advice omits, and it is the part that decides whether the risk you wrote down is the risk you took.

The same 1% is not the same risk

Two positions on a $10,000 account, both sized to risk 1%, which is $100. The trader uses a flat 5% stop on both, because 5% is what the rule says.

• BTC at $100,000, with a 14-day average true range of $2,000, or 2% of price. A 5% stop sits 2.5 ATR from entry.
• A mid-cap alt at $2.00, with a 14-day ATR of $0.16, or 8% of price. A 5% stop sits 0.625 ATR from entry.

Both positions risk $100. Neither risks it with the same probability. A stop placed closer than one average daily range is hit by an ordinary session with no news in it; a stop 2.5 ranges away is not. The expected loss per trade is the risk amount multiplied by the probability of realizing it, and a flat percentage rule controls only the first term.

The consequence is worse than a bad stop. Because the alt is stopped out far more often, the strategy's measured win rate on that asset is depressed by an artifact of the sizing rule rather than by anything about the asset, which then feeds a wrong number into everything above.

Converting a stop distance into a size, in volatility units

Fix the stop in volatility units instead. Where to place it in the first place is stop-loss territory; take 1.5 × ATR(14) as the input and let the size fall out.

One note on units first. ATR as a percentage of price is the volatility proxy used in this subsection, and it is not the same statistic as the standard deviation of daily returns: it is a mean of daily ranges rather than a dispersion of closes, and it usually reads higher. It is used here because it is stable, comparable across assets, and available on any exchange chart. Every result below is a ratio between two assets, so the choice of proxy moves the absolute notionals and not the relative sizing.

The notionals differ by a factor of 4.0, which is exactly the inverse of the volatility ratio: 8% ÷ 2% = 4. That is not a coincidence and it is the point of the section. Once the stop is measured in volatility units, ATR sizing is inverse-volatility sizing with a stop attached.

The same idea without a stop

Positions you hold without a hard stop (a spot allocation, a basket, a slow trend follow) use the same formula written directly:

Position notional = (target daily equity volatility in dollars) ÷ (asset's daily volatility)

Here the volatility term is the standard deviation of daily returns rather than ATR, because the square-root-of-time scaling below is a property of a standard deviation and not of an average range. Targeting 0.5% of a $10,000 account, or $50 of daily movement per position:

• BTC at a 2.0% daily return standard deviation: $50 ÷ 0.02 = $2,500 of notional, which annualizes to roughly 38% by 2.0% × √365.
• The alt at 8.0%: $50 ÷ 0.08 = $625 of notional, roughly 153% annualized.

Same 4:1 ratio. The dollar amounts differ from the ATR version because the risk being held constant is different (daily fluctuation rather than loss at a stop), but the relative sizing is identical, because both are one over volatility.

The same rule applied to a book, not a trade

The effect is easier to see across three positions than one. Hold BTC at 2% daily, SOL at 5%, and a mid-cap alt at 8%, on the same $10,000 account at 1% risk each. Compare a flat 5% stop on all three against a 1.5 ATR stop on all three, and measure what each position contributes to the book's daily movement.

The flat-stop book states three equal 1% risks and delivers a book where one position accounts for over half the daily movement. The volatility-scaled book splits it three ways exactly, and the equality is not approximate: each position's notional is the dollar risk divided by 1.5 times its volatility, so notional multiplied by volatility comes to the same $66.67 for every asset by construction. The flat-stop book also carries more notional in total, $6,000 against $5,500, while producing 50% more daily movement, $300 against $200. More exposure for a worse-balanced book is the specific combination the flat rule keeps producing.

Choosing the lookback, and why recent volatility is predictive at all

The reason any of this works is a property of volatility rather than a property of the model. Harvey and co-authors open their study of volatility targeting with it directly: volatility is persistent, or clusters, and high volatility over the recent past tends to be followed by high volatility in the near future. That observation is what makes a backward-looking estimate usable as a forward-looking input, and it is the empirical finding underneath the whole approach.

Their own estimator is exponentially weighted squared returns at a 20-day and a 90-day half-life. In practice the trade-off runs like this:

• Short lookback (10 days, or a 20-day half-life). Reacts fast, and turns the position over often. Every resize is a round trip through the fee schedule, which the next section prices.
• Long lookback (60 days, or a 90-day half-life). Stable sizes and low turnover, at the cost of being wrong for weeks after a regime changes.
• Exponential weighting rather than a flat window. Removes the artifact where a single old shock drops out of a rolling window and the position jumps for no reason that happened today.

The failure mode: the model sizes you largest right before the break

Volatility targeting is slow by construction, because the estimate can only include moves that have already happened. Take a stretch of 2.00% daily moves, then a single 10% day, and watch what each estimator does.

Every one of those cuts happens after the 10% day, which was taken at the pre-shock size. And because volatility clusters, the elevated regime the position now sits in is exactly where the fat tail lives. The model's protection arrives one move late, permanently, and no lookback fixes it: a shorter window arrives sooner and cuts more, but it also cuts on noise and pays fees for the privilege.

What this example actually tells you

Stop expressing risk as a percentage of price and start expressing it as a multiple of the asset's own range. A 5% stop is a different instrument on BTC than on a mid-cap alt. A 1.5 ATR stop is the same instrument on both, which is the only condition under which comparing your results across assets means anything.

Expect the notional to look wrong, because it is supposed to. A 33% of equity BTC position next to an 8% of equity alt position looks backwards to anyone reading notional as risk. It is correct: the two positions carry the same risk, and the alt is smaller precisely because it moves more.

Do not treat the vol-target size as a floor when volatility is genuinely spiking. The model will keep handing you a size after a regime break because it has not seen enough of the new regime to know. Pair it with a hard rule that is not derived from the estimate: a maximum notional, a maximum number of concurrent positions, or simply not trading the first two days after a market-wide gap.

Correlation-aware sizing: ten 1% positions can be one 10% position

The per-trade layer is now solved. The portfolio layer undoes it if you ignore it.

Ten alt positions each risking 1% are ten independent 1% risks only if the alts are independent. They are not. The risk management guide computes this from real daily closes: average pairwise correlation across BTC, ETH, SOL, AVAX and XRP was 0.22 in a calm two-week window and 0.95 in a panic window. At 0.95 the ten positions are, for sizing purposes, one position with ten times the risk budget, and it will be one position on the day it matters.

The usable fix is not a covariance matrix. It is a budget at the level above the trade:

1. Group your universe into correlation clusters, defined by what they trade on rather than by sector labels. Large-cap beta, L1 ecosystem tokens, DeFi governance tokens, and memecoins are typically four clusters; in a drawdown they collapse toward one, which is why the next step exists.
2. Assign each cluster a risk budget, and treat the cluster as the position. If your per-position rule is 1%, cap the cluster at something like 2%, not 10%.
3. Divide the cluster budget among its members. Four alt positions inside a 2% cluster budget get 0.5% each, not 1% each.
4. Cap total risk across clusters at a number you can name before the fact, because in a genuine deleveraging event the clusters converge too.

Volatility targeting interacts with this in a way worth naming: correlated assets have correlated volatility, so a vol spike cuts every position in the cluster at once. That is the mechanism doing its job, and it also means your entire book resizes on the same day, which is a liquidity event of your own making if the cluster is illiquid.

The small-account floor: where sizing down turns into negative expectancy

Volatility targeting says size down when volatility rises. Two separate constraints say there is a point past which you should stop.

Constraint one: cost scales with one over stop distance

Round-trip cost is a percentage of notional, and notional is your risk divided by stop distance. So cost measured in units of your risk is:

Round-trip cost as a share of 1R = (round-trip fee + slippage) ÷ stop distance

Binance's standard spot schedule is 0.100% maker and 0.100% taker for a Regular User, checked 6 August 2026. Entering and exiting as a taker is 0.20%, and assume a further 0.05% round trip in spread and slippage, for 0.25% of notional.

Set that against the strategy from earlier, with a gross expectancy of +0.20R. Break-even is where cost equals expectancy: 0.25% ÷ 0.20 = a 1.25% stop. Tighter than that and a genuinely profitable strategy is a losing one, on published fees alone, before anything goes wrong. Note what this does not depend on: account size. A tight stop is expensive at every account size, and this is the reason scalping strategies need either maker rebates or a much larger gross edge.

Constraint two: the venue's minimum position

This one does depend on account size. From Binance's public USDⓈ-M perpetual trading rules, read 6 August 2026:

• BTCUSDT: lot step 0.001 BTC, minimum notional 50 USDT. At a $100,000 BTC price the lot step is the binding constraint, at $100 of notional.
• ETHUSDT: lot step 0.001 ETH, minimum notional 20 USDT.
• SOLUSDT: lot step 0.01 SOL, minimum notional 5 USDT.

Your correct notional is (risk fraction × equity) ÷ stop distance. Setting that equal to the venue floor and solving gives the smallest account that can size the trade correctly:

Minimum viable equity = (venue minimum notional × stop distance) ÷ risk fraction

At a 1% risk on BTCUSDT perpetuals with a $100 floor: a 3% stop needs $300 of equity, and a 6% stop needs $600. Below that, the smallest position the venue will accept risks more than your rule allows. A $250 account forced into the $100 minimum with a 6% stop is risking $6, or 2.4%, not 1%.

What this actually tells you

The two constraints move in opposite directions, and that is the finding. When volatility rises, the vol-targeted stop widens, which lowers the correct notional, and simultaneously raises the minimum equity needed to trade it, because the formula above is proportional to stop distance. The small account gets squeezed from both sides at exactly the moment the model is telling it to be careful.

If discretionary sizing is not working at your account size, the honest alternative is not a smaller version of the same thing. It is a non-discretionary schedule with no stop and no sizing decision at all, which is what dollar-cost averaging is for.

Leverage tiers cap your model above a certain notional

Every model above outputs a notional. Above a certain notional the venue, not your model, decides what is possible.

Binance states the rule directly in its USDⓈ-M futures documentation: the maximum leverage available depends on the notional value of the position, and larger positions allow lower leverage. The maintenance margin rate rises in the same brackets. From Binance's published leverage and margin table, read 6 August 2026:

Two things follow, and neither is about the headline leverage number.

The tier ladder is much tighter on alts than on BTC. SOLUSDT drops out of its top bracket at 50,000 USDT of notional; BTCUSDT holds its top bracket to 300,000. A trader scaling a single alt idea meets the constraint six times earlier than the same dollar size in BTC would suggest.

The maintenance margin rate is the part that changes your risk, not the leverage cap. Doubling notional inside a rising maintenance schedule moves the liquidation price closer even if you never touch the leverage selector, which can put liquidation inside a stop that was correctly placed at the old size. The mechanics of that are in how crypto leverage works; the sizing consequence is what belongs here.

Checking this before you design a size rule takes three readings. Open the venue's leverage and margin page for the symbol you trade, not for BTC, and note the notional at which it drops a bracket. Compare that against the largest position your model can produce, which is your maximum risk budget divided by your tightest plausible stop distance. Then recompute the liquidation price at the maintenance margin rate of the bracket you would land in rather than the one you are in now, and confirm it still sits beyond your stop.

Why a sizing rule can never rescue a losing strategy

This is the most commonly misunderstood thing about position sizing, so it gets stated plainly: sizing changes the shape of the return distribution. It does not change the sign of the edge.

The Kelly formula proves it on its own terms. Take a 30% win rate with 2R winners: f* = (0.30 × 2 − 0.70) ÷ 2 = −0.05. A negative optimal fraction is the formula's way of saying there is no positive amount you should stake. Expectancy is 0.30 × 2 − 0.70 = −0.10R per trade, and multiplying a negative number by any positive size leaves it negative. Every sizing model in this article is a multiplier applied to per-trade expectancy. None of them is an addend.

What sizing genuinely controls is the path: how deep the drawdowns are, how long the recovery takes, whether the compounding is fast enough to matter, and whether you are still solvent when the edge shows up. Those are worth a great deal. They are not an edge.

Three more traps specific to backtesting sizing rules:

A sizing rule is fitted on the same sample that produced the edge estimate. The ATR multiple, the lookback, the Kelly fraction and the cluster budget are all parameters, and every one is chosen after seeing the outcomes. Figure 1 is why this is more dangerous here than elsewhere: the parameter you are most likely to overfit is the mean return, and its error is punished asymmetrically.

A rule tuned on one regime fails in the next. A volatility target fitted in a trending sample is fitted to a period where rising volatility and falling prices coincided, the leverage effect Harvey and co-authors document for equities. In a choppy market the same rule resizes constantly, pays the fee drag from the previous section on every turn, and has nothing to show for it.

Test the rule against a shaded edge, not the fitted one. Re-run it with the win rate cut by five points, or the mean return cut by a quarter, and see whether it still works. If it only works at the estimate you measured, you have fitted the estimate rather than validated the rule. The whole failure catalogue this sits inside is in why most crypto traders lose money.

Putting it together: the sizing decision in order

Conclusion

The 1% rule is a risk budget. Everything in this article is about what you do with that budget once you accept that it does not, by itself, produce a position size.

Fixed fractional and fixed dollar are a choice about compounding, and the arithmetic is unambiguous: fractional decays toward zero without arriving and recovers more slowly, flat sizing arrives at zero and recovers faster, and a symmetric round trip costs the fractional sizer 0.4% while costing the flat sizer nothing. Kelly answers a question worth asking, namely how large this edge really is, and returns a number nobody should trade, because a five-point error in a win rate moves the recommended size by 75% and cuts realized growth by more than half. Half Kelly gives up a quarter of the growth, halves the volatility, and takes the probability of ever halving the account from one in two to one in eight. That trade is the reason fractional Kelly exists.

The variable that actually breaks naive sizing in crypto is volatility. A flat percentage stop applied to a 2%-a-day asset and an 8%-a-day asset produces two positions with the same stated risk and completely different probabilities of realizing it, and correcting for that means measuring the stop in volatility units and letting the notional come out proportional to one over volatility. Recent realized volatility is genuinely predictive because volatility clusters, and the same clustering is why the model always sizes you one move late into a regime break.

Then three constraints bound the whole system. Correlated positions are one position when it counts, so budget at the cluster. Fees scale as one over stop distance, so below roughly a 1.25% stop on standard spot fees, a strategy earning 0.20R per trade earns nothing. Venue minimums and leverage tiers cap the model from below and above respectively.

And the constraint that outranks all of them: none of this creates an edge. Every model here is a multiplier on per-trade expectancy. Get the sign wrong and no amount of sizing sophistication changes the destination, only how long the trip takes.

Frequently asked questions

Sources and further reading

Primary sources for the mathematics and the venue parameters above:

• J. L. Kelly Jr., "A New Interpretation of Information Rate", Bell System Technical Journal 35 (1956), 917–926, the original derivation of the growth-optimal betting fraction, including the statement that maximizing expected wealth leaves the bettor broke with probability one
• Edward O. Thorp, "The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market", covering fractional Kelly growth and volatility results, and the drawdown probability formula
• Harvey, Hoyle, Korgaonkar, Rattray, Sargaison and Van Hemert, "The Impact of Volatility Targeting", Journal of Portfolio Management 45(1), 2018, covering volatility clustering, and the effect of volatility scaling on Sharpe ratios and tails across 60-plus assets
• Binance — leverage and margin of USDⓈ-M futures
• Binance — USDⓈ-M perpetual leverage and margin tiers
• Binance — trading fee schedule

Related CoinBeaver articles:

• Managing risk in crypto: a practical framework for traders
• Crypto stop-loss strategy: where to place your stop
• How crypto leverage trading works
• Dollar-cost averaging in crypto
• Why most crypto traders lose money

This article is educational and is not financial or trading advice. All account balances, win rates, ATR values, and price levels used in the worked examples are illustrative figures chosen to show the mechanism, not market data or a description of any real strategy's results. The Kelly and fractional-Kelly results are properties of the models stated, not predictions about any market. Exchange fee schedules, minimum order sizes, and leverage and margin tiers change without notice; the Binance figures above were read on 6 August 2026 and must be re-checked on your own venue before sizing a position.