Home › Learn › Lesson 39

Why risk comes before strategy

By Cuongboss · Updated 9 Oct 2026 · Published 9 Oct 2026

Almost every trading education starts with a method: a pattern, an indicator, a setup. Risk arrives later, as a chapter near the back, phrased as a warning. That order is backwards, and it is possible to show why with arithmetic rather than adjectives. Below, one strategy is held completely fixed while only the size of each bet changes — and the outcomes range from tripling the account to losing 99.4% of it. Then the harder question: if the maths has an optimal answer, why is that answer one almost nobody should use?

The dial is set before the tools are chosen — and where it is set is what greys two of them out. That is the whole lesson: the risk budget is not a safety wrapper around a strategy, it is the thing that decides which strategies you have.

What does “risk comes before strategy” actually mean?

It means the risk decision is made first in time and it constrains everything after it. Not “remember to use a stop”, which is advice about a tool. The claim is about sequence: you fix what one trade may cost you before you have any idea what the trade will be, and that number then decides which strategies are workable for you at all.

Most people run the sequence backwards without noticing. They find a method they like, take a position that feels right, and discover the risk afterwards — usually on the day it goes wrong. The course this site teaches from puts it the other way round in the definition itself: risk management is “a method for limiting and protecting your capital, keeping it safe from being wiped out by one or a handful of trades”, and its list of building blocks ends with the line that gives the game away — the trader “decides the maximum position size, based on the risk accepted for each individual trade”. Size is downstream. Risk is upstream.

That still sounds like housekeeping. The rest of this lesson is the argument that it is not: that the risk number can turn a winning strategy into a losing one without touching a single entry or exit.

Can a winning strategy still lose money?

Yes, and it does not take bad luck. It takes the wrong size.

Here is a strategy that is unambiguously good. It wins 40% of the time; each winner makes twice what each loser costs. Its expectancy is 0.4 × 2 − 0.6 × 1 = +0.20R per trade — twenty percent of one unit of risk, earned on average, every single trade. Any trader would take it.

Now run that same strategy 200 times at seven different risk settings, risking a fixed percentage of the current balance each time. Nothing about the trades changes: same win rate, same reward-to-risk, same order of wins and losses. Only the fraction of capital at stake moves.

Read the bars from the top and the first three behave the way intuition says they should: doubling the risk from 1% to 2% roughly doubles the outcome. Then the curve stops obeying. The peak arrives at 10%, and after it more risk buys less money — 15% returns ×4.42, worse than 10% for half again as much exposure. At 20.39% the account ends exactly where it began: two hundred trades of a strategy with positive expectancy, and nothing to show for any of them. Past that, the same good strategy destroys capital. At 30% it returns six tenths of one percent of what you started with.

The mechanism is not mysterious, and it is worth understanding rather than memorising. Losses compound against you in a way gains do not repay symmetrically. Lose 30% and you need +42.9% to get level; lose 50% and you need +100%; lose 80% and you need +400%. Every time the account takes a large percentage hit, the wins that follow are working on a smaller base and a bigger repair bill. Past a certain size of bet, the repair bill grows faster than the edge can pay it — and the edge itself never changed.

This is the whole argument in one sentence: the strategy sets the sign of your edge, and the risk setting decides whether that edge ever reaches your account. A decision that powerful cannot reasonably be made second.

The same arithmetic runs in reverse, and it is worth seeing from that side too. The most famous trading record ever published — W.D. Gann’s 264 winning trades out of 286, watched by a magazine’s representative in October 1909 — reports a win rate of 92.3% and never once states the size of the losing trades. With twelve wins for every loss, an average loss twelve times the average win leaves the month flat, at that same spectacular win rate. The profile of Gann in Wisdom works that number through in full. Read it as the practical test of this lesson: when somebody shows you a strategy, the win rate is the part they volunteer, and the risk side is the part you have to ask for.

So why not just risk 10%, since that grows fastest?

Because the number that grows an account fastest and the number a person can actually follow are not the same number, and the gap between them is enormous.

The bars above are outcomes at the end. They say nothing about the road. So the same experiment was re-run as 20,000 separate 200-trade paths, recording not just where each one finished but how far it fell along the way. You can watch a smaller version of it happen in the site’s simulator:

At 1% per trade, the median worst moment of a 200-trade run is a 11.8% drawdown — unpleasant, survivable, the sort of thing you can explain to yourself. At the growth-optimal 10%, the median worst moment is a 78.8% drawdown, and 99.4% of runs lost more than half the account at some point. Note the word median: that is the middle path, not the unlucky one. Losing three quarters of your money is the typical experience of running the mathematically best risk setting.

Nobody sits through that. They cut the size, or change the method, or stop — all of which are rational, and all of which mean the ×6.98 in the first table never arrives. A strategy you abandon in month four pays exactly nothing, no matter what its expectancy was. So the honest way to read the two figures together is this: the arithmetic can tell you the ceiling, but it cannot tell you your number. Your number is the largest one you will still be following after the worst month it produces.

The simulator’s 43.3% and −70% look kinder than 99.4% and 78.8%, and the gap is worth half a minute because it is a trap of its own. Despite its label, the simulator calls a run ruined when equity falls to half of the starting balance, and it stops measuring that run there. The 99.4% in this lesson counts a fall of half from the highest point the account reached. At 10% risk an account usually climbs before it breaks, so it can lose 70% from its peak and still sit above half of the deposit. Counted the simulator’s way, about 43% of the 20,000 paths hit half the starting balance — the same answer the tool gives. Counted from the peak, 99.4% lose half. And its median drawdown reads about −70% rather than 78.8% because a ruined run stops being measured: re-run frozen the same way, the 20,000 paths give 71%. Both views are correct; only the peak-based one is what you live through, because nobody experiences watching $30,000 shrink to $9,000 as “still near my deposit of $10,000”.

And there is a line further along the same ladder that this lesson stops short of: on this exact method, long-run growth crosses zero at 20.39% risk per trade, so above that point a genuine edge still loses money. The gambler’s mindset measures the rest of that ladder, and why the urge to go past it is a reading on whether you have an edge at all.

Does the same shape show up on real trades, or only in the model?

A model with a fixed 40% win rate is a clean argument and a weak witness. So the experiment was repeated on trades a rule actually produced. The rule was fixed before any result was looked at — it is the breakout “chase” from Lesson 38 with the target set to 2R so that it matches the model: buy the close of the first 4-hour candle that closes above the highest high of the previous 50 candles; put the stop at the low of that breakout candle; take profit at twice the risk; hold one position per coin at a time; ignore a new signal within five candles of the previous one; and if one candle touches both stop and target, count it as a loss. The candles are Binance spot, BTC and ETH from January 2020 and SOL from August 2020, to 26 September 2026, with a 0.1% round-trip fee charged on every trade.

Across the three coins the rule took 510 long trades and won 37.6% of them — close to the model’s 40%, with exits pinned at +2R and −1R. Then the fees take a slice the model never charged: 0.03R on the median trade, far more on tight stops. The average was +0.13R per trade before fees and +0.09R after — less than half the model’s +0.20R, and an honest number: its 95% bootstrap range runs from −0.03R to +0.22R, so this backtest does not prove the rule has an edge. What it shows is what the risk setting does to whatever edge there is, on real sequences of wins and losses.

Three things carried over from the model and one did not. The shape carried over: growth rises, peaks and falls. The ratio carried over: growth peaks at 4.4% and reaches zero at 8.9% — double again, because doubling is what any small edge does, not a quirk of the 40% model. The pain at the peak carried over: a 72.8% fall at 4.4%. What did not carry over is the location. The model’s best setting, 10%, lies beyond this rule’s breaking point: run these real trades at 10% and a strategy with a positive average ends at ×0.57 after losing 97.8% from its high. The thinner the edge, the closer the cliff, and real edges are thin. Look at the 1% row too: a 24.5% worst fall on this one real path, about double the model’s median of 11.8% — over 510 trades instead of 200 and on half the edge. (Both columns follow the trades in the order they actually closed; the worst fall is measured from the highest point reached.) Even the careful setting hurts more on real trades.

Split by coin, the same rule gives three different answers. On ETH growth peaked at 3.4% and reached zero at 7.0%; on SOL at 12.2% and 24.9%; on BTC there was no peak at all — after fees its 174 trades averaged −0.04R and every risk level shrank the account (×0.92 at 1%). The mirror-image rule, shorting breaks of the 50-candle low, lost 0.08R a trade after fees over 381 trades, and no setting rescues that: the strategy sets the sign, risk only scales what is there. If the “optimal” risk can be zero or 12% depending on which coin you happened to test, it is not a number you can trade. It is a fact about the sample.

The losing streaks were real too. In the order the trades actually closed, the longest run was 14 losses in a row. Shuffling the same 510 results 2,000 times gives a median longest run of 11 and a 95th percentile of 17, so 14 is ordinary, not bad luck. At a full 1R each, fourteen straight losses cost 13.1% of the account at 1% risk, 46.7% at 4.4%, and 77.1% at 10% — and with fees the real ones averaged a little over 1R, so a little more. That streak, not the average, is what a risk number has to survive — which is where Lesson 40 picks up.

Limits of this test: trades from the three coins are merged in the order they closed and compounded one after another, which understates risk whenever positions overlap; it is one rule on one timeframe; and the rule was tested on the same data it is reported on. Reproduce with de-cuong-bai/do-so-bai-39-thuc-do.py.

Which raises the obvious question.

Where does the right risk number come from, if not from the maths?

From your life outside the chart. And there is a way to measure that which takes about thirty seconds.

Convert one losing trade into days of income. Take your monthly take-home pay, divide by 22 working days, and see how many of those days a single stop costs you. The rule can be identical — 1% of capital in every row below — and the felt experience is not remotely the same.

Look at the fourth row. Same discipline, same 1%, same strategy — and every stop costs nearly three weeks of pay. Ask what a person in that row does at the moment price approaches the stop. They widen it “just this once”. Or they take profit at a third of the target because holding hurts. Or they skip the next valid signal entirely. All three are people breaking a rule they genuinely believe in, and none of them has a discipline problem. They have a size problem, and the discipline is downstream of it.

Our course treats this as a measurement rather than a personality trait. When a student says they are struggling psychologically, the first two questions are not about emotion at all — is the capital you are trading inside the range that feels comfortable, and is each stop about 1% of it? Most cases resolve there, and the remedy is to trade a smaller account and grow into a larger one, not to try harder. Fear at row four is the correct reading of a real situation. The instrument is working.

Two consequences follow, and both are uncomfortable.

First, the right risk number is personal and it moves. The same trader, same strategy, needs a different setting after a pay rise, after a baby, after quitting a job. Nothing on the chart changed; the thermometer did.

Second, the number has to be set before you look at anything. Decide it while a live position is open and you are no longer measuring your capacity for loss — you are negotiating with a specific trade that you want to work. Lesson 40 takes that tolerance and turns it into an actual percentage, because the streak your method produces is countable and the arithmetic runs backwards.

Why doesn’t depositing less money make you safer?

This is the most popular way to skip step one while feeling responsible, and it is worth ten seconds of arithmetic.

Say you have $100,000 in savings. You decide the exchange makes you nervous, so you fund the account with $10,000 and leave the rest in the bank. You keep setting stops at $1,000 a trade, because $1,000 is 1% of your capital and 1% is the rule.

Nothing about your risk has changed. You will still lose $1,000 on a losing trade, and you own the same $100,000 before and after. What has changed is that $1,000 is now 10% of the balance on the exchange — so ten stops in a row takes the visible account to zero, at which point you will either transfer more money in a bad mood or conclude the strategy failed. The comfort was real and the protection was not: you moved the timing of a bank transfer, not the size of a loss.

The test is simple. Reducing risk means reducing the amount one trade can cost you. If a change does not shrink that number, it is a change of scenery. Two things genuinely shrink it: a smaller position for the same stop, or the same position with a closer stop. Nothing else on the screen qualifies — not the deposit, not the leverage slider, not how certain the setup looks.

What order do the four decisions actually go in?

Four decisions, and the order is the lesson.

Step one is the only one that comes from outside the chart, and it is the one almost everyone leaves until last. Step two belongs entirely to the market: the stop goes where the idea would be proven wrong, which has nothing to do with the size of your account. Step three is arithmetic — money at risk divided by the distance to the stop — and it is worked through in full in Lesson 42 on position sizing.

Step four is the part that sounds strange until you have done it once: the first three steps eliminate strategies. A method whose stops sit 15% away from entry needs a very different position from one whose stops sit 1.5% away, and on a small account the second may be untradeable after fees while the first is fine. A style that requires watching a five-minute chart all day is not available to someone with a job, whatever its backtest says. None of that is discoverable from the strategy itself. It falls out of the budget you set in step one, which is precisely why the budget has to be set first.

The four steps on one real trade

Take the April 2026 BTC trade from the chart above and run it in order. Step one, away from any chart: say $5,000 a month take-home and a $20,000 account, so the thermometer reads 0.9 days of pay for a 1% stop — the budget is $200. Step two, from the chart: the stop goes under the breakout candle, 69,011.29, which is 4.05% below the 71,924.22 entry. The market chose that distance, not the account. Step three is the division:

Step four is where the budget starts deleting strategies, and the same backtest shows how. A round-trip fee is a fixed slice of price, so it costs more R the tighter the stop. On this trade 0.1% of fees came to 0.025R. On 13 June 2026 the same rule gave a BTC stop only 0.36% from entry, and there the fee alone was about 0.28R — more than a quarter of a full loss, paid on every trade before price moved. A method that lives on stops that tight needs a much larger edge just to stand still, and once you size by a fixed budget that cost is visible before the first trade rather than after the fiftieth.

Run the list bottom-up instead and something has to bend. The strategy demands a size; the size demands a stop; and the stop, being the only flexible thing left, gets pushed out to wherever the arithmetic permits — to a level that no longer means anything about the trade. That single reversal is how one ordinary loss becomes the one you remember.

What do people get wrong about this?

Treating risk as a chapter rather than a constraint. Reading the risk material, agreeing with it, and then choosing trades exactly as before. The test of whether risk really comes first is whether it has ever caused you to not take a setup you liked. If it never has, it is not first.

Believing bigger risk is simply bigger reward. The bars say otherwise beyond 10%: more exposure, less money, and eventually a losing system built from a winning one. “High risk, high return” is true over a range and false outside it, and almost nobody is told where the range ends.

Copying someone else’s percentage. A number that works for a full-time trader with two years of savings is a different instrument in the hands of someone whose stop costs three weeks of pay. The thermometer table is the reason percentages do not transfer between people.

Taking the risk number from a backtest’s optimum. The same breakout rule peaked at 3.4% risk on ETH, 12.2% on SOL and nowhere on BTC. An optimum that moves that much between three coins describes the sample, not your next year — and on a real, thin edge the breaking point sits much closer than any model with a fat edge suggests.

Setting the risk number with a position open. At that moment you are not deciding what you can afford to lose; you are deciding what you can bear to admit. The two answers differ, and the second one is always larger.

Calling the result a psychology problem. Moving stops, cutting winners early and skipping valid signals are the standard symptoms of a size that is too large for the person holding it. Discipline is much easier to keep at a size that does not hurt, which is why the fix usually starts with the account rather than the mindset. The lessons on why most new traders lose money in year one matter, but they cannot outvote arithmetic.

Assuming the stop guarantees the loss. A stop is a plan, not a promise: gaps, thin books and fast markets can fill you worse. That is an argument for sizing on the modest side, not for skipping the calculation.

When is this lesson wrong?

Four limits, stated plainly, because a model presented as complete stops being useful the first time reality disagrees with it.

The simulation is friendlier than trading. Every figure above assumes no fees, no funding, no slippage, independent trades and a fixed reward-to-risk ratio. Real results are worse on all five counts. That does not weaken the conclusion — it means the collapse arrives sooner in practice, not later. The real-trade test above shows how much sooner on one rule: with fees, real streaks and a thinner edge, the zero-growth point fell from 20.39% to 8.9%.

The real-trade test proves less than it seems to. Its average after fees, +0.09R, has a 95% range that includes zero, the rule was tested on the data it is reported on, and two of its three coins disagree with the third. Read it as evidence about what risk does to a thin edge — not as evidence that this rule has one, and not as a reason to trade it.

The exact thresholds belong to these assumptions, not to you. The 10% peak and the 20.39% break-even come from a 40% win rate with 2:1 winners. Change either input and both numbers move. What survives every set of inputs is the shape: there is always a peak, always a level past which a profitable strategy shrinks the account, and the second is always closer than people assume.

Per-trade risk is not the whole of risk. Four positions at 1% each are not four independent 1% risks if all four are long the same market on the same idea — correlated trades get stopped out together, which is a 4% day wearing four costumes. Sizing each trade properly is necessary and not sufficient, and portfolio-level exposure is its own subject.

There is also a case where none of this applies: if you are not trading yet, the correct risk budget is zero, and the sequence starts at knowledge before capital.

Educational content only — not financial advice, and not a trade recommendation. The definition of risk management, its three building blocks and the reasons for fixing the maximum loss in advance are from the slide course this site teaches from (Part 1). The two-question check on capital size and per-trade stop, and the point that reducing risk is a matter of structure rather than willpower, are from the course notes. Everything else is arithmetic computed for this article and reproducible from de-cuong-bai/do-so-bai-39.py: a 40% win rate with +2R winners and −1R losers gives an expectancy of +0.20R per trade; expected log growth per trade is 0.4 × ln(1+2f) + 0.6 × ln(1−f), which peaks at f = 10.00% and returns to zero at f = 20.39%; compounding that growth over 200 trades gives ×1.46, ×2.04, ×4.35, ×6.98, ×4.42, ×1.15 and ×0.006 at 1%, 2%, 5%, 10%, 15%, 20% and 30%. The drawdown figures come from 20,000 simulated 200-trade paths at each setting, seed 20260906, measuring the deepest fall from a running peak. Recovery percentages are 1÷(1−loss)−1. The income table divides monthly take-home pay by 22 working days and then divides one 1% stop by that figure. Those figures are before fees, funding and slippage. The real-trade section is reproducible from de-cuong-bai/do-so-bai-39-thuc-do.py: Binance spot 4-hour candles (BTC and ETH from January 2020, SOL from August 2020, to 26 September 2026); long on the close of the first candle closing above the prior 50-candle high, more than five candles after the previous signal candle; stop at that candle’s low; target 2R; one position per coin; a candle touching both counts as a loss; a trade still open after 260 candles closes at the close (none did); 0.1% round-trip fee; growth and drawdown computed on the trades in the order they closed; streak baseline from 2,000 shuffles. Screenshots captured 9 Oct 2026. Published 6 Sep 2026, updated 9 Oct 2026.