Swing trading strategy: the written specification

Ask a hundred swing traders to describe their strategy and most will describe an entry. They will tell you about the shape of the pullback, the moving average they wait for, the way the volume dries up before the move. Ask the same hundred where the trade is wrong, in advance, as a number, and the answers thin out immediately. Ask how the position ends when it neither works nor fails, and the answers stop. Ask what evidence they have that the whole thing does anything at all, and you will usually be shown a chart. One chart. The one the idea came from.

This is not a failure of discipline, and it is not laziness. It is a category error. An entry feels like the hard part because it is the part that requires a decision under uncertainty, and it is the part that reading and screen time actually improve. The other four questions feel like paperwork. They are not paperwork. They are the parts of the strategy that make the entry mean anything, and without them the entry is a guess that arrived with a justification attached.

Consider what a specification actually has to do. It has to be precise enough that it produces the same answer twice. It has to define its own failure, because a claim that cannot fail cannot be tested. It has to fix the size of the bet, because a trade without a size is not a trade. It has to end, in both directions, including the direction where nothing happens. And it has to have been counted, over a record rather than over a memory. Those five requirements are not a checklist bolted onto a setup. They are what turns a setup into a strategy.

The most useful way to see the dependency is to draw the five as a chain rather than a list. A list suggests the items are independent and that having three out of five is sixty per cent of a strategy. They are not independent. Question three, the stop, defines the risk unit. Question four, the exit, is stated in that unit. Question five, the evidence, counts results in that unit. Leave question three blank and the last two do not get harder, they become unanswerable, because there is no unit left to state them in.

The lower rail of that figure is not a caricature. It is close to a verbatim transcription of what a swing trading note usually contains, and the two clauses it does carry are both impressions rather than events. The third slot is empty, and everything after it is a dashed frame. Notice that the failure is silent. Nothing announces itself. A trader working from that note can place trades all year, feel busy and disciplined, and never once encounter the gap, because the gap only shows up when you try to compute something, and nothing in that workflow ever tries.

A strategy you cannot hand to somebody else is not a strategy. It is a habit you have not written down, and habits cannot be tested, taught, improved or trusted.

The first question is the definition of the setup, and the standard it has to meet is unusual: two competent people given the same chart and the same sentence should mark the same bar. Not similar bars. The same bar. If the sentence needs judgement in the moment to resolve, then it is neither testable nor repeatable, and those two failures are the same failure wearing different clothes. A rule that cannot be applied mechanically cannot be run over history to see what it does, and a rule that cannot be run over history is a rule you are trusting on the strength of how it feels.

Most setup descriptions fail this test on the first word. Consider the sentence a large fraction of swing traders would recognise as their own: buy the pullback in an uptrend when it looks ready to go again. Every clause in it is a judgement call. Uptrend by what measure and over what window. Pullback of what depth, and how do you know it is a pullback rather than the start of a decline. Ready to go again is not a condition at all, it is a feeling with a verb in it.

The consequences are not academic, and they are much larger than most people expect. Here is the same sentence and one precise clause applied to identical bars.

Three readers, all applying the same sentence honestly, entered at 494.08, 497.57 and 505.10. That is a spread of eleven rupees on a share trading near five hundred, which sounds tolerable until you follow it through. Each reader also placed a stop under a low they considered recent, and because they entered at different heights above roughly the same floor, the risk unit ranged from 3.84 to 15.17 rupees a share. At a fixed risk budget of five thousand rupees, that is 1,302 shares for one reader and 329 for another, which is 3.87 times the capital committed to what everybody involved would describe as the same trade.

That number is the real cost of a vague setup definition, and notice where it lands. It does not land on the entry, which is the part everyone worries about. It lands on the size, which is the part that decides how much the outcome matters. Two people can follow identical rules, be right at the same time, and have wildly different accounts at the end of the year purely because one sentence in the specification was written in English rather than in conditions.

The precise clause on the right of that figure is not more sophisticated than the vague one. It is not better analysis. It is the same idea, written so that it resolves: a twenty session high made within the last ten sessions, then three to eight pullback sessions, none of which closes below the twenty session exponential moving average. Run that over the twenty one sessions in the panel and exactly one qualifies. Run it over the whole drawn record and six qualify. Give it to a second person and they get six too, because there is nothing in the sentence for them to disagree with.

There is a cost to precision and it should be stated plainly, because a page that pretends otherwise is selling something. A clause tight enough to resolve mechanically will refuse trades that look obvious afterwards. On the drawn record, the clause above sits out several advances that any experienced eye would call clean, because the pullback ran nine sessions instead of eight, or because one close slipped under the moving average. That is not a defect being tolerated. It is the price of having a rule at all, and the alternative, loosening the clause until it catches everything, returns you immediately to a sentence that two people read differently.

The setup says a chart is interesting. The trigger says you are in. They are different questions and collapsing them is one of the more common structural errors in a retail plan, because a setup can persist for many sessions while a trigger is a single event with a timestamp. If your specification does not separate them, you will find yourself entering somewhere inside a multi-session window at a price determined by when you happened to look at the screen.

A trigger has to be an event that the market either produces or does not. A close beyond a level is an event. A print through a level is an event. A limit order resting at a price is an event. Looks ready, is coiling, feels like it wants to go, and appears to be holding are not events, they are descriptions of a state, and a state cannot be timestamped.

Which trigger you choose is a real decision with real trade-offs, and the specification should record not just the choice but the reason, because the reason is what you will interrogate later when you want to know whether the trigger or the setup is the weak part.

Two things follow from that table. First, the trigger and the stop are a matched pair. A trigger that fills you early demands a stop placed where early entries are still wrong, and a trigger that waits for the close demands a stop far enough back to survive the session it entered on. Choosing them independently produces a specification whose two halves disagree. Second, whichever row you pick, the specification must name the level numerically and the event exactly. A close above the previous session's high is a specification. A close above resistance is not, because resistance is a drawing, and two people draw it differently.

The third question is where the idea is proven wrong, and the answer has to exist before the order is sent. Not because deciding in advance is emotionally safer, though it is, but because the answer is load-bearing. The distance from entry to invalidation is the risk unit. Once you have it, the position size follows from your risk budget, the target follows in multiples of it, and every result you ever count is expressed in it. Leave it undefined and you have not deferred a decision, you have removed the unit that four other numbers are quoted in.

The failure mode here is specific and extremely common: the stop gets written as a rupee tolerance rather than as a level. I will cut it if I am down five thousand rupees sounds like discipline. It is the opposite, and the reason is worth seeing rather than asserting.

On the left, the clause names a level on the chart, one tick below the lowest low of the pullback sessions. That single sentence produces four numbers in sequence and none of them is an opinion. The level is 489.48. The risk unit is 8.09 rupees a share. At a five thousand rupee risk budget the size is 618 shares. The exits are 517.79 and a session count. Change nothing about your account and every one of those numbers stays where it is, because they are properties of the chart.

On the right, the same sentence in rupee form produces a different stop level for every position size. At 300 shares the five thousand rupee tolerance is 16.67 a share and puts the stop at 480.90, roughly eight and a half rupees below the pullback low, at a price nothing on the chart has an opinion about. At 1,000 shares the tolerance is five rupees a share and puts the stop at 492.57, which sits inside the range those same pullback sessions had already traded through days earlier. That stop is not tight. It is placed at a price the market demonstrably visits while the setup is still intact.

The deeper problem is circularity. The rupee stop's level depends on the size, and the size is supposed to depend on the risk unit, which is the distance to the stop. Written that way the specification defines each quantity in terms of the other and settles on whatever number you happened to type into the quantity box. Nothing decided it. The chart certainly did not.

Where the level should sit structurally, which of the several defensible reference points to use, and how to handle instruments whose ordinary daily range is wider than your intended risk are a separate and substantial subject, and this page deliberately does not re-open it. Our guide to where to place a stop loss works through the structural options and the four common misplacements. What matters for the specification is narrower and prior to all of it: whatever placement rule you adopt, it has to be written as a rule that produces a price, it has to be written before entry, and it has to be the thing your size is computed from rather than the thing your size implies.

Most specifications that get this far have half an exit. They have a target, usually a multiple of the risk unit or a level on the chart, and they have the stop from question three. Between them those two cover the case where the idea works and the case where it fails. What they do not cover is the case that actually happens most often, which is neither.

A swing thesis is a claim about the next few weeks. If those weeks pass and the claim has produced nothing, the claim has failed. It has just failed quietly, without touching the stop and without giving you the clean signal that a stop-out provides. Almost no retail specification has a rule for this, which means the position stays open by default, and default is not a decision.

That position was entered on a session that qualified under the clause, and then nothing happened. Best progress over the following ten sessions was 0.57 of the risk unit, reached on the third session and never improved on. The stop was never threatened. The target was never approached; it sat sixteen rupees above, off the top of the scale the chart is drawn on. On the eleventh session the written time stop closed it, with the idea neither proved nor disproved.

What those ten sessions cost is worth pricing, because it is invisible in a profit and loss statement that only records closed trades. Six hundred shares at 481.45 is 2,88,870 rupees of exposure, held for ten sessions. One five thousand rupee risk slot was occupied for the same period and was therefore unavailable to anything else. The statutory round trip on that notional comes to roughly 643 rupees, which is about an eighth of the risk budget the trade was sized against, paid regardless of what happened. That is the real shape of dead money: not a loss, but capital and attention consumed by a position that was never going to tell you anything.

Now the honest part, because the convenient version of this argument is not the true one. On this particular occasion the time stop was worth very little. Without it, the same position would have run two more sessions and been stopped out at the level anyway, so the clause bought back two sessions and part of a loss. Across all six qualifying trades on the drawn record, the rule spent forty sessions holding positions with the time stop and forty two without it. That is a marginal difference, and anyone telling you that time stops transform a strategy is not reading their own numbers.

The point of the time stop is not that exiting early is better. It is that a thesis with a horizon needs a rule for the end of that horizon, and a specification which says explicitly that there is no time stop is complete. Only the blank is indefensible.

What the clause reliably does is bound the dead-money problem. Nineteen of the forty sessions this rule spent in positions, roughly half, were spent in trades that had not yet reached one risk unit of progress. That is the pool the question addresses, and it is large enough to be worth a written answer in either direction. Whether you set the horizon at eight sessions or fifteen, whether you require one risk unit of progress or merely a higher low, whether you exit fully or halve the position, are all defensible answers. Having no answer is not one of them.

Everything that happens between the trigger and the exit, trailing, scaling, moving the stop to break-even, and the far harder question of when to do nothing at all, is the subject of our guide to what happens between entry and exit. The specification's job is narrower: it must state in advance which of those actions are permitted and on what condition, so that the middle of the trade is executed rather than improvised. A management action that is not in the specification is a new strategy authored while a position is open, and it is authored by somebody who currently has money on the outcome.

The last question is the one that separates a specification from a well-written wish. You have four clauses that resolve mechanically. Why do you believe the combination does anything at all?

The usual answer is a chart. Somebody shows you the instance the idea came from, marked up, with the entry and the run that followed. That is not evidence, and the reason is structural rather than statistical: the chart that produced the rule cannot also test it. The rule was fitted to that chart. Of course it works there.

Honest evidence requires three things that the marked-up chart does not have. It requires the whole record rather than the memorable part of it, which means counting the times the rule fired and went nowhere as carefully as the times it fired and ran. It requires costs deducted, because a per-trade edge smaller than the toll is not an edge. And it requires some part of the data to have been held back and never looked at while the rule was being written, because a rule adjusted until it fits the data it was adjusted on has learnt the data rather than the market.

Run the clause over the whole drawn record and this is what an honest evidence base looks like. Six hundred and forty three sessions, six qualifying signals, and a position open on forty of those sessions. Split the record two thirds to one third and four of the six signals sit in the part that was looked at, which leaves two observations in the part that was not. Two. That is the entire quantity of untainted evidence behind a rule that a trader would describe, without any dishonesty, as tested.

This is the ordinary situation, not a pathological one, and it follows directly from the arithmetic of the horizon. A selective swing rule on a single instrument fires a handful of times a year. To accumulate a hundred untainted observations you need either many years or many instruments, and both have costs: many years means the market regime has changed underneath you, many instruments means the rule has to survive being applied to things it was not designed on. There is no version of this where the evidence is cheap.

How to run the count so it does not lie to you, what a held-out sample actually protects against, why the cost line has to go in before you look at the result rather than after, and how a strategy can pass every test and still have been fitted, are the subject of our guide to back-testing integrity. The specification's obligation is smaller and comes first: question five has to be written as a claim you could be wrong about. Tested on three years of data with costs at the observed statutory rates and the final year held back is a claim. It has worked well for me is not, because there is nothing in it that could turn out to be false.

It is worth stating the population you are joining, once, plainly, and with its limits attached. About 93 percent of individual traders in equity derivatives made net losses over FY22 to FY24, aggregate net losses exceeding ₹1.8 lakh crore (SEBI, September 2024). That study examined the equity derivatives segment specifically, which is the fast, leveraged end of the market, and it does not describe cash-equity swing trading or predict any individual's outcome. It is quoted here for one narrow reason: the gap between people who believe they have a strategy and people who have a specification is very wide, and the aggregate numbers in adjacent segments are consistent with that gap being expensive.

The five questions are not country-specific, but the numbers you put into question five are. A swing trade in Indian cash equity is a delivery trade, and delivery carries a statutory bill that is fixed by law rather than negotiated, so it belongs in the specification as a constant rather than as an afterthought.

Two consequences for the specification. First, the cost line goes into question five before you look at the result, not after. A per-trade edge quoted gross is not a number about your account. Second, the toll scales with turnover rather than with holding period, which is the structural reason a swing horizon is cheaper per unit of opportunity than a shorter one; the trade-off you accept in exchange is overnight gap risk and a much slower feedback loop. That comparison, holding period against cost against the hours your life actually has, is worked out in our guide to swing trading versus intraday trading, and the case for stretching the horizon further still is in our guide to positional trading.

There is one more Indian specific that belongs in the specification rather than in the risk section of your imagination. A swing position is held overnight, so the price at which your stop level is tested may not be a price that ever traded. A gap through the level fills you below it, and on an instrument with a daily price band the position can be locked at a limit with no exit available at all. The specification cannot prevent that. What it can do is state the assumption explicitly, so that when it happens you are executing a plan that anticipated it rather than discovering a hole.

The output of all of this is not a folder of notes. It is one page, and it should be short enough that a stranger can read it in two minutes and long enough that they could execute it without asking you anything. Below is the specification used throughout this guide, written out in full, with each clause in the form that makes it resolve.

Notice what the page does not contain. It contains no market view, no forecast and no discretion. It also contains no promise, because a specification is a description of what you will do rather than a prediction of what will happen. That distinction is the reason a written strategy survives a bad run: nothing in it was contingent on the run being good.

A complete specification is not a permanent achievement. It decays, and it decays in recognisable ways, all of which look like reasonable adjustments at the time. These are worth naming because you will not catch them by reading the document; you catch them by comparing the document to what you actually did.

The common thread is that all four preserve the appearance of a system while removing the property that made it one. What made it a system was that it was fixed in advance and that its results were countable in a stable unit. An exception, a moved stop, a drifting risk budget and a stale count each break exactly one of those two properties, which is why none of them announces itself as a departure.

This is also the honest reason a trading journal exists, and why it is a component of the strategy rather than an accessory to it. The journal is the instrument that detects the drift, because it records what you did in the same fields the specification predicted, and the gap between the two columns is the whole diagnosis.

If you take one thing from this page, take the inversion. The setup is not the strategy and the pattern is not the edge. They are one line of a document, and the document is the thing that has value, because the document is what can be tested, taught, executed by somebody who is not you, and defended after a losing month.

It follows that the productive work is usually not looking for a better setup. It is finishing the specification you already have. Most traders reading this could write question one properly in an afternoon, question three in ten minutes, and question four in one sentence, and would come out the other side with something that is genuinely testable for the first time. Question five will take longer and will probably return an uncomfortable answer about how thin the evidence is. That discomfort is the point, and it is not a reason to stop; it is the first accurate piece of information you will have had about your own trading.

The five questions are the spine of the method we teach in our structured curriculum, where the specification is built before any chart is read, and where every later stage, sizing, execution, review and the psychology of holding through the middle, is defined against it rather than bolted on afterwards. The order matters more than any individual answer. A trader with a mediocre setup and a complete specification can find out what is wrong and fix it. A trader with an excellent setup and four blanks cannot, because there is nothing in their process capable of producing that information.

• SEBI study of individual traders in the equity derivatives segment, September 2024. The source of the aggregate loss finding quoted once in the section on evidence above. Two limits travel with it and are stated there. The study examined the equity derivatives segment, which is the leveraged, high frequency end of the market, so it does not describe cash equity swing trading. And it is an aggregate across a population, so it does not predict the outcome of any individual account. sebi.gov.in
• Securities Transaction Tax on delivery equity. 0.1 percent on the purchase and 0.1 percent on the sale, computed on the volume weighted average price. The charging provision is Section 98 of the Finance (No. 2) Act 2004; the current rate was set by the Finance Act 2012 with effect from 1 July 2012. The Finance Acts of 2024, 2025 and 2026 amended only the derivatives entries of that table, so delivery equity is unchanged.
• Stamp duty on securities transfer. 0.015 percent on a delivery transfer, payable by the buyer only for an exchange trade, under Schedule I Article 56A(b) of the Indian Stamp Act 1899 as inserted by the Finance Act 2019 and effective 1 July 2020. It is levied at the central rate and collected by the clearing corporation, which is why it does not vary between states.
• Exchange transaction charges, cash segment. NSE charges ₹307 per crore of turnover each side, all in, with effect from 1 March 2026 under circular NSE/FA/73061; the split between the transaction charge and the investor protection fund contribution changed on that date while the total did not. BSE charges ₹375 per crore for group A and B scrips on a flat rate basis with effect from 1 December 2022, under notice 20221109-7.
• SEBI turnover fee. ₹10 per crore of turnover, 0.0001 percent, on both purchase and sale in the cash segment, under Regulation 41(1) of the SEBI (Stock Brokers) Regulations 2026, gazetted 7 January 2026. Those regulations repeal and replace the 1992 regulations, so citations to Schedule III of the 1992 rules are out of date.
• GST on trading charges. 18 percent on brokerage, exchange transaction charges, the SEBI turnover fee and depository charges, under Notification 11/2017-Central Tax (Rate), Serial 15, Heading 9971. STT and stamp duty are outside it because the broker recovers them as a pure agent under Rule 33 of the CGST Rules, which is also the reason a contract note has to itemise them on separate lines.
• The figures on this page. Every price series here is synthetic, generated for this guide, and is not any real instrument. The specification is applied to it mechanically by code, so every level, count, quantity, session number and holding period printed on a figure is an output of that code rather than a drawing made to fit the argument. No outcome is scored anywhere, no success rate is stated or implied, and the risk unit and rupee figures are illustrative throughout.