How to build a trading system in India: the machinery around the strategy

Take the two words apart, because the industry uses them interchangeably and the confusion is expensive. A strategy is a decision rule pointed at the market: it names what qualifies as a candidate, what triggers an entry, where the stop sits, how the position is exited, and what evidence supports the whole thing. It is a claim about instruments and prices. A system is pointed the other way, at the person operating the rule and at the book the rule builds. It decides how large any single position may be relative to capital, how much risk may be open in total, whether today is a day on which the rule may be run at all, what is written down and when, and what evidence would end the arrangement.

The distinction becomes concrete the moment you try to hand each one to somebody else. Give two competent people a written strategy and they will take broadly similar trades, because the rule constrains entries and exits. Give those same two people a written strategy and nothing else, and within a month their accounts will look nothing alike. One will hold three positions and one will hold eleven. One will have stood down through a week they judged unfit and the other will have traded it. One will remember the losing month as a run of bad luck and the other as a run of overrides, and neither will have any evidence, because neither was writing anything down at the time. Every one of those differences sits outside the strategy, and together they dominate the outcome.

This is the central asymmetry: the strategy determines what your edge is, and the system determines whether you are still there when the edge arrives. A trader with a modest, well-operated rule generally survives long enough to find out whether it works. A trader with a stronger rule and no machinery frequently does not, because the ordinary losing streak that every real edge produces arrives before the average does, and there is nothing in the strategy layer that tells you what to do while it is happening. The regulator's numbers describe the population that runs on strategy alone: SEBI's study released on 7 July 2025 reported that 91 percent of individual traders in the equity derivatives segment incurred net losses in FY25, with an aggregate net loss of ₹1,05,603 crore; its earlier study, released in September 2024, reported 93 percent net loss-makers among individual traders in equity futures and options across FY22 to FY24.

The table below is the whole page in miniature. Read down the third column and notice how many of the questions that decide an account's fate have no answer at all in the strategy layer. That empty space is the machinery, and building it is the actual work of this guide.

The five components are not a checklist to be completed once. They are a cycle, and each one exists to feed the next. The specification produces signals. Sizing turns each signal into a position and keeps the book inside a risk budget. The operating rules decide whether the day is one on which any of this runs. The record captures what actually happened, decision by decision, while it is happening. Review consumes that record and either confirms the strategy, adjusts the machinery, or retires the rule, which returns you to the specification. Then it goes round again.

Drawing it as a loop is not decoration; it is the fastest way to see the failure mode. A cycle survives the removal of no component. Take any one out and the remaining four stop being a cycle and become an open line: a process that starts somewhere, runs forward, and terminates when something external stops it. In trading, the thing that eventually stops an open line is almost always the account balance, which is an expensive way to receive information you could have generated for yourself.

Notice which link is drawn cut. It is not an arbitrary choice. Of the five components, the record is both the most frequently absent and the one whose absence does the most structural damage, because it is the only component that manufactures evidence. Sizing errors show up in the equity curve eventually. A missing record does not show up anywhere, because the thing it would have measured is the gap between the system and the operator, and that gap is invisible to every other instrument you own. The rest of this guide walks the loop in order.

The specification is the component this site has covered most thoroughly, and repeating it here would help nobody. What matters for the machinery is narrower: the specification is not just an idea, it is an interface. The four components downstream of it each need a specific output, and a specification that fails to produce that output silently disables the component that depends on it. Sizing needs a stop distance that exists before the trade is placed, not a discretionary exit decided later. The operating rules need a signal that is generated at a known time, so that a day can be gated before the market is watched rather than after. The record needs a named rule so that a logged decision can be attributed to something. Review needs countable instances, which means the specification must define what a qualifying occurrence is, whether or not you traded it.

Judged that way, most retail specifications fail not because the idea is weak but because they are incomplete in ways that break the machinery downstream. A rule with an entry and no pre-defined stop cannot be sized, so the position size becomes a mood variable. A rule with no definition of a qualifying instance cannot be reviewed, because you can only count the trades you took, and the trades you took are a biased sample of the trades the rule generated. Building rules to that standard is the specific judgement the method we teach is organised around, because the precision of the reading determines whether any of the machinery has something real to grip.

This is the component retail skips most completely, and the one where the error is hardest to notice, because the arithmetic that hides it looks like prudence. The starting rule is sound and widely taught: risk a small fixed fraction of capital on any single trade, and derive the position size from the stop rather than the other way round. If equity is ₹10,00,000 and the per-trade risk budget is one percent, then ₹10,000 is at stake on each trade, and the number of units is that budget divided by the distance from entry to stop. Tight stop, large position; wide stop, small position. The risk is held constant while the position floats. Everything about that is correct, and it is where most retail sizing both begins and ends.

The problem is that the rule is a statement about one position, and you do not hold one position. Ask a trader running this rule what their risk is and the answer comes back instantly: one percent. But if six positions are open, the arithmetic total at risk is six percent, and the trader who answered "one percent" was answering a question about a single trade while holding a book. That number, the sum of what you would lose if every open stop were hit, is portfolio heat, and it is the quantity a per-trade rule is structurally incapable of seeing.

Then correlation makes it worse in a way that is genuinely counter-intuitive. Six independent bets of one percent each do not usually all lose together; independence is what makes the arithmetic total a remote tail rather than a normal Tuesday. But retail books are rarely independent. Positions get selected by the same screen, in the same conditions, often in the same handful of sectors, and they respond to the same drivers. When four of the six respond to one driver, those four stop behaving like four bets and start behaving like one larger one. The figure below computes what that does.

Read the bottom line of that figure carefully, because it is the point. The per-trade rule is not being broken anywhere on that scale. It reports one percent in every case, and in every case it is right about the thing it measures. The failure is that nobody is measuring the other thing. This is why a heat cap is a separate written rule rather than a consequence of the sizing rule: a ceiling on total open risk, checked before every new entry, that outputs a refusal when the book is already full. A cap that has never once prevented an entry is not a cap.

There is a second constraint hiding in the same arithmetic, and it surprises people who have got the risk side right. Sizing from the stop means a tight stop produces a large position. With stops around four and a half percent of price and a one percent risk budget, each position is roughly twenty-two percent of equity. Six of them is about 132 percent of equity, which is more capital than the account holds. The risk budget said yes and the capital said no, and whichever constraint binds first is the real limit on how many positions you can carry. The table works this through.

Two numbers in that last row matter and they fail in different directions. The 5.99 percent of heat is a risk statement: if the cluster resolves badly, that is what leaves the account, and against a cap of, say, three percent this book was never permissible. The 132.2 percent of capital is a feasibility statement: the book cannot be held at all without leverage, and a trader who reaches for leverage to make a sizing rule fit has inverted the rule's entire purpose. Set both limits in writing, check both before every entry, and let the tighter one decide. The broader discipline this sits inside, and how the per-trade fraction interacts with drawdown depth, is covered in risk management for traders.

A specification tells you what to do when a signal appears. The operating rules decide the prior question: whether the machine is running today at all, and in what state. This sounds like bureaucracy until you notice that the alternative is having that question answered implicitly, every morning, by whatever mood you woke up in. The operating rules are simply that decision moved out of the moment and into writing, where it can be consistent.

A workable set has three layers. The pre-market sequence runs before the session and establishes the facts the day depends on: yesterday reconciled against the contract note so the position list is real rather than remembered, the regime condition your strategy depends on checked and recorded, current portfolio heat computed and the remaining room under the cap known as a number, and a fitness check against a standard you wrote in advance. The in-session rules are deliberately narrow: act on what the scan produced, size by the rule, and do not add instruments intraday. The post-session routine closes the record for the day while the reasoning is still accurate.

The single test of whether this is a control or a ritual is whether any item can output a no. A checklist that has never once stopped you from trading is decoration, because it has no failure mode and therefore carries no information. The version below shows two days that are identical in every respect except one gate output, and the point is that the difference is decided by arithmetic rather than by how the trader feels about the setup.

The weekly layer is thinner but does something the daily layer cannot. Once a week, recompute heat from scratch rather than incrementally, since incremental tracking drifts. Re-examine the correlation structure of what you are holding, because a book that was diversified when it was built often is not a month later. Read the week's record and count the overrides. None of this requires long. What it requires is that it happens on a schedule rather than when you feel like reviewing, since the weeks you least want to look at the record are exactly the weeks it contains the most information.

Everything downstream of this component depends on evidence, and evidence has an expiry. The reason to write down a decision at the moment you make it is not diligence or self-improvement. It is that human memory of a decision is systematically rewritten by the outcome of that decision, and the rewriting is invisible from the inside. Once you know a trade lost, you do not merely remember the trade differently; you remember having had reservations you did not have, and you remember a conviction level you never held. The revised memory feels exactly like the original one. There is no internal signal that distinguishes them.

This matters because it destroys the one comparison that review depends on. To improve, you need to know whether a losing run happened because the rule performed as rules do, or because you did not run the rule. From memory, that question is unanswerable, and worse, it reliably resolves in a flattering direction: the trades you skipped get remembered as sensible caution, and the ones you took against the rule get remembered as justified reads. A record made before the outcome cannot be edited this way, because it was fixed while the future was still unknown.

The practical design follows from the failure mode. Capture must happen before the outcome is known, which in practice means before the order is placed, and it must be short enough that you actually do it on the bad mornings when it matters most. Six fields is generally enough: which rule fired, the level, the stop and why it is there, the size with the arithmetic shown, what would prove the trade wrong, and your state of mind in a handful of words. The arithmetic is worth writing out rather than just the answer, because a sizing error is visible in the working and invisible in the result. And record the instances you passed on, with the gate that stopped you, since a sample of only the trades you liked cannot tell you anything about the rule.

Every system needs a way to end, and the timing of that decision is the whole game. The criteria have to be written before capital is committed, because the moment they become relevant is precisely the moment you are least able to set them. Inside a drawdown, judgement is degraded, the urge to give the rule a little more room is at its strongest, and every argument for continuing sounds like patience rather than what it usually is. A line drawn in advance is not a smarter decision than one drawn later. It is a decision made by a version of you who had no stake in the answer.

Write the triggers as numbers, and express each one relative to your own tested distribution rather than a round figure. A drawdown deeper than the worst the test produced. A losing run longer than any the test generated. A live expectancy that sits below the tested figure across a stated number of trades. Or the cleanest of all, the disappearance of the structural reason the edge existed in the first place, which retires a system regardless of what the curve is doing. Each is falsifiable and each is checkable from the record. What follows shows why the timing matters more than the number.

The recovery arithmetic in that caption is the quiet argument for pre-commitment. Drawdowns are not symmetric: losing 18 percent requires a 22 percent gain to undo, while losing 32 percent requires 47 percent, and the gap widens with every point you let it run. This is why the second decision in the coral lane, increasing size to recover faster, is so destructive. It is the intuitive response and it is exactly backwards, because it raises the required recovery on a book that has already demonstrated it is in an unfavourable stretch. A written line removes that decision from the situation entirely.

One caution about what these criteria can prove. A drawdown breaching your worst tested figure is a valid trigger to stop and investigate, but it is not by itself proof that the edge is gone; a genuine edge produces losing runs, and a test covering a few hundred trades has simply not seen the worst run the rule is capable of. Triggering the criterion means the live evidence has left the range your test explored, which is a reason to stop rather than a verdict. The distinction between a rule that has decayed and a test that flattered you in the first place is what the backtesting integrity guide is about, and it is the first thing to check when a trigger fires.

Review fails in two opposite ways and both are common. Reviewing too often draws conclusions from noise, which produces constant tinkering and destroys the consistency the machinery exists to create. Reviewing too rarely lets a genuine problem, most often an operator problem, run for months. The fix is to separate the horizons and be explicit about what each one is entitled to conclude, because most damage comes from a daily observation being used to justify a change that only a quarterly sample could support.

The last row is the one that separates a system from a habit. If the edge had a named reason, and that reason disappears, the rule should be retired whether or not the equity curve has noticed yet. A curve that is still rising after its mechanism has gone is measuring something else, usually a favourable regime, and it will stop measuring it without warning. Conversely, the fourth row protects a sound rule from being abandoned during the ordinary bad stretch that every edge produces, provided the record shows it was actually being followed. Both protections depend entirely on the record.

The honest statement of purpose is narrower and less inspiring than the usual one, and it is worth saying plainly. None of this makes you better at reading a chart. Not one component here improves the strategy, finds an edge, or raises the quality of a single decision considered on its own. What the machinery does is remove the variance that you introduce, so that the outcome starts to reflect the rule rather than the operator. Its job is to make your behaviour independent of your mood.

A system does not stop you feeling terrible after three losses. It stops your feeling terrible from changing the size of the fourth trade.

Everything above is a mechanism serving that single end. Sizing rules exist because conviction inflates position size and conviction is highest after wins and lowest after losses, which is exactly backwards. Heat caps exist because the moment you most want another position is the moment you already have too many correlated ones. Operating rules exist because the question of whether to trade today should not be re-litigated every morning by whoever you happen to be that morning. The record exists because your account of what you did is unreliable in a specific, predictable direction. Retirement criteria exist because you cannot be trusted to draw a line while standing on the wrong side of it. Every component is a defence against a known, documented, universal tendency, and none of them is a defence against a bad strategy.

Which produces the discipline's most uncomfortable fact. Machinery does not manufacture an edge; it protects one. Built around a rule with no real edge, everything described here will do its job perfectly and you will lose money in an orderly, well-documented, precisely-sized fashion. That is not a wasted outcome, because you will find out relatively cheaply and with evidence, which is more than most retail traders ever get. But it is worth being clear that this page is about the half of the problem that comes after you have something worth running, and the other half is not optional.

• SEBI study on individual traders in the equity derivatives segment, released 7 July 2025. Reported that 91 percent of individual traders incurred net losses in the equity derivatives segment in FY25, with an aggregate net loss of ₹1,05,603 crore. Cited here to establish the base rate that operating machinery is built to resist, not as a claim about any strategy. sebi.gov.in
• SEBI study on participation and profit-loss of individual traders in the equity futures and options segment, released September 2024. Reported 93 percent net loss-makers among individual traders across FY22 to FY24. Corroborates the persistence of the base rate across an earlier and non-overlapping window. sebi.gov.in
• The portfolio heat aggregation in the second figure. Computed on this page using standard variance aggregation across positions of equal risk: the combined exposure is the square root of the sum of every pairwise product of position risks weighted by their correlation. The figures 2.45, 3.79, 4.24 and 6.00 percent follow arithmetically from six positions of one percent each under correlations of zero, a four-position cluster at 0.7, that cluster at 1.0, and all six at 1.0. The two boundary cases are self-checking: zero correlation gives six effective independent bets, and perfect correlation gives one, with the aggregate equal to the arithmetic sum.
• The equity path in the fifth figure. An authored illustrative sequence of 100 trade outcomes, deliberately irregular in run length and magnitude, used identically in both lanes so that the only difference between them is the decision rule. Drawdown, recovery and decomposition figures are computed from that sequence. It is an illustration of decision arithmetic and is not derived from, nor a prediction of, any market.
• The component framing. The treatment of a trading system as operating machinery around a strategy, comprising specification, sizing and heat, operating rules, a contemporaneous record and pre-committed review, reflects standard systematic-trading practice, presented here as an operator's guide. Specification and testing are deferred to the linked guides rather than repeated.