Advanced8-12 min readTopic 14 of 20

    Position Sizing & Portfolio Construction

    Rohit Singh

    Mr. Chartist · SEBI RA

    Module Progress
    0/20
    Module

    Two systematic traders can run the identical rulebook on the identical instruments and end up in completely different places, because sizing — not signal generation — is where most of the variation in a book's behaviour comes from. Position sizing answers one question per trade: how many units, given the distance to the point where this idea is wrong. Portfolio construction answers the harder question above it: how much of the account each strategy gets, and whether the positions that are open right now are genuinely separate bets or one bet wearing several names.

    The same rupee risk budget applied to three instruments with different stop distances. Equal rupees deployed would have meant three different amounts of risk; equal risk means three different quantities.A fixed risk budget of five thousand rupees is divided by three different stop distances of sixteen, thirty-two and forty-eight rupees, giving three hundred and twelve, one hundred and fifty-six and one hundred and four units respectively. The rupee risk is identical in all three cases while the capital deployed varies by a factor of three, because the stop distance and not the conviction sets the quantity.Same risk budget. Same entry. Three different sizes.risk budget 5,000 = 0.5% of 10,00,000 equity | entry 640.00IDEASTOP DISTANCEQUANTITYCAPITAL DEPLOYEDA16.003121,99,680B32.0015699,840C48.0010466,560Rupee risk is 5,000 in all three rows. Only the quantity moved. Illustrative arithmetic.
    The same rupee risk budget applied to three instruments with different stop distances. Equal rupees deployed would have meant three different amounts of risk; equal risk means three different quantities.

    What does position sizing actually decide?

    Position sizing is not a statement about conviction. It is a division sum performed after two other decisions have already been made.

    The first is the risk budget: the amount of capital you have decided this single idea may cost you if it is wrong. The second is the stop distance: the gap between where you enter and the price at which the idea is invalidated, which comes from the chart or the rule, never from how much you feel like losing.

    Once both exist, the quantity is arithmetic. There is no judgement left in it, which is precisely why it is the part of the process that survives automation best — and the part that most obviously exposes a system that never defined its invalidation price.

    Quantity = Risk budget (₹) ÷ Stop distance per unit (₹)
    • Risk budget — a fraction of account equity you set in advance, not per-trade improvisation.
    • Stop distance — |entry price − invalidation price|, before costs.
    • Quantity — rounded down, and further constrained by lot size where the instrument trades in lots.
    • This identity is definitional. It says nothing about whether the trade will work.

    Note — This page is education, not advice. Nothing here recommends a risk fraction, an instrument or a trade. Every figure is illustrative arithmetic on round numbers, not a real position in a real stock on a real date.

    What is fixed fractional position sizing?

    Fixed fractional sizing means the risk budget is a constant percentage of current equity rather than a constant rupee amount. It is the default in most systematic books for two structural reasons.

    It shrinks automatically. As equity falls through a losing run, the rupee risk per idea falls with it, so the sequence of losses gets arithmetically shallower rather than staying the same size against a smaller account. And it grows automatically, without a decision to grow it — which removes the moment where a trader chooses to 'size up because things are going well', historically one of the more expensive discretionary decisions available.

    The cost is that it recovers slowly. A reduced account risks smaller amounts, so it takes more work to climb back. That trade — slower recovery in exchange for a floor you are much less likely to hit — is the whole argument for the method.

    How does the arithmetic run in rupees?

    Take an account of ₹10,00,000 and a risk fraction of 0.5% per idea. The budget is ₹5,000 for this trade, whatever it turns out to be.

    If the entry is ₹640 and the invalidation sits at ₹608, the stop distance is ₹32. ₹5,000 ÷ ₹32 = 156 units, and the position is worth about ₹99,840 — roughly a tenth of the account, holding a twohundredth of it at risk.

    Now change only the chart. A second idea has an entry of ₹640 and an invalidation at ₹592, a stop distance of ₹48. The same ₹5,000 budget buys 104 units, a position worth about ₹66,560. Same account, same risk, same entry price — a third less capital deployed, purely because the stop is wider.

    Notice which number never appeared in either calculation: how good the setup looked.

    Example — Illustrative arithmetic with round figures. Costs — brokerage, exchange charges, statutory levies and slippage — are excluded here to keep the identity visible; in a live system they are subtracted before the budget, and they are covered in Costs, Slippage & Taxation.

    Why is equal rupees not equal risk?

    The most common sizing error in a systematic book is not a formula mistake. It is deploying the same rupee amount into every position and believing that constitutes a balanced portfolio.

    It does not, because instruments do not move by the same amount. Put ₹1,00,000 into an instrument whose typical candle range is a fraction of a per cent, and ₹1,00,000 into one whose typical candle range is several per cent, and you have made one bet several times the size of the other while looking at two identical numbers on your positions screen.

    This is why risk, not capital, is the unit a systematic book is measured in. Two positions are equal-weighted when a move to their respective invalidation prices costs the same, not when they cost the same to open.

    Two positions of identical rupee value. Because the instruments have different typical ranges, a move to each invalidation costs very different amounts — so the account is not balanced, it just looks balanced.Two positions are each worth one lakh rupees. The first is in an instrument whose typical candle range is under one per cent, so a move to its invalidation costs about two thousand seven hundred rupees. The second has a typical range several times larger, so the same one lakh position costs about twelve thousand six hundred rupees at its invalidation. The positions screen shows two identical numbers while the account carries one bet several times the size of the other.Two positions of 1,00,000 each. One book, two very different bets.Quiet instrumenttypical candle range 0.9%position 1,00,000cost at invalidation2,700Volatile instrumenttypical candle range 4.2%position 1,00,000cost at invalidation12,600A systematic book is measured in risk units, not in rupees deployed. Illustrative figures.
    Two positions of identical rupee value. Because the instruments have different typical ranges, a move to each invalidation costs very different amounts — so the account is not balanced, it just looks balanced.

    What is volatility-based position sizing?

    Volatility-based sizing replaces a stop distance taken from a chart level with one derived from how much the instrument typically moves. The usual measure is an average of the true range over a lookback window — the ATR family — and in a systematic rulebook that window is specified in candles, on a named timeframe, not in days.

    The stop is then placed some multiple of that range away from entry, and the same division sum runs unchanged. The effect is that a quiet instrument gets a tight stop and therefore a larger quantity, while a volatile one gets a wide stop and a smaller quantity. Every position in the book is then scaled to the same expected disturbance rather than to the same rupee amount.

    Quantity = (Equity × risk fraction) ÷ (k × ATR over N candles)
    • ATR over N candles — average true range on the timeframe the strategy trades, N specified in the rulebook.
    • k — the multiple of that range the stop sits at. A parameter, and therefore something you can overfit.
    • Equity × risk fraction — the same rupee risk budget as before; nothing about the identity changes.
    • Volatility measured over past candles is a description of the recent past, not a forecast of the next move.
    One risk budget, two instruments with different typical candle ranges. The volatility measure sets the stop distance, the stop distance sets the quantity, and both positions end up carrying the same rupee risk at very different position sizes.Two panels show the typical candle range of a quiet instrument and a volatile one over the same lookback in candles. The quiet instrument's average true range gives a sixteen rupee stop and three hundred and twelve units; the volatile one's gives a forty-eight rupee stop and one hundred and four units. Both positions carry the same five thousand rupee risk budget at very different sizes.The range sets the stop. The stop sets the quantity.lookback stated in candles on the traded timeframe, never in daysQUIETATR(14) = 8.00stop = 2 x ATR = 16.005,000 / stop =312VOLATILEATR(14) = 24.00stop = 2 x ATR = 48.005,000 / stop =104Same 5,000 at risk in both. Past range describes the past; it does not forecast the next move.
    One risk budget, two instruments with different typical candle ranges. The volatility measure sets the stop distance, the stop distance sets the quantity, and both positions end up carrying the same rupee risk at very different position sizes.

    Which sizing method should a systematic strategy use?

    MethodStop distance comes fromStrengthWhere it breaks
    Fixed quantityNothing — the quantity is fixedTrivial to implement and to reason aboutRisk varies wildly between instruments and over time; it is sizing in name only
    Fixed rupee valueNothing — the deployed capital is fixedPositions look tidy on the screenEqual capital is not equal risk; the volatile name silently dominates the book
    Fixed fractional, structural stopA chart level — the price that invalidates the ideaRisk is constant per idea and the stop means somethingNeeds a rule that produces an unambiguous invalidation price on every trade
    Volatility-based (ATR-style)A multiple of typical range over N candlesComparable across instruments; adapts as conditions changeThe lookback and the multiple are parameters you can tune until they fit history
    Optimal-fraction formulasAn assumed edge and payoff distributionMathematically tidy given the assumptionsThe assumptions are estimated from limited data; small estimation errors produce very aggressive sizes

    Watch out — Sizing formulas that take an estimated edge as an input inherit every error in that estimate and amplify it. Treat anything that computes a size from a historical performance statistic with more suspicion than the strategy that produced the statistic.

    What is portfolio heat?

    Portfolio heat is the total amount at risk across every open position at one moment — the sum of what it would cost if every position simultaneously reached its invalidation price.

    A rule that risks 0.5% per idea does not cap the account at 0.5%. Twelve open positions at 0.5% each is six per cent of equity at risk simultaneously, and nothing in the per-trade rule prevented that. Per-trade risk governs one decision; portfolio heat governs the account, and it needs its own limit and its own enforcement.

    In an automated system, heat is checked before an order is sent, alongside the other pre-trade limits. When a new signal would push total open risk past the cap, the correct behaviour is to skip the trade — not to squeeze it in at a reduced size, which quietly changes the strategy into a different one whenever the book is busy.

    Total open risk accumulating position by position toward a heat cap. The seventh signal is refused because it would breach the cap — a skip is a decision the rulebook made, not an opportunity missed.Six open positions each risking around half a per cent of equity stack up to three per cent of total open risk, which is the account's heat cap. A seventh signal arrives and is refused, because taking it would push total open risk beyond the cap. The refusal is a rule the book followed, not an opportunity lost.Per-trade risk does not cap the account. Heat does.TOTAL OPEN RISK, POSITION BY POSITION0.5%P10.5%P20.5%P30.5%P40.4%P50.6%P6HEAT CAP 3.0%7TH SIGNAL: SKIPTaking it would breach the cap,so the rulebook refuses the order.THE TEMPTING MISTAKEShrinking it in anyway quietlychanges the strategy when busy.Illustrative percentages. The cap itself is the reader's decision, not a prescribed number.
    Total open risk accumulating position by position toward a heat cap. The seventh signal is refused because it would breach the cap — a skip is a decision the rulebook made, not an opportunity missed.

    How is heat reduced as trades work?

    Heat is not static, because the risk in a position changes as the stop moves.

    When a stop is raised to the entry price, that position's contribution to open risk falls to roughly zero — the remaining exposure is to a gap through the level rather than to the level itself. When a trailing rule locks in a level above entry, the position contributes negative risk to the tally in the sense that its worst modelled outcome no longer costs the account anything.

    Systems that recompute heat continuously therefore free up capacity as trades mature, and systems that compute it only at entry do not. Both are defensible designs. What is not defensible is having no explicit choice: an unstated convention here is the difference between a book that carries three units of risk and one that carries ten.

    Note — Zero heat is not zero risk. A stop at entry still leaves you exposed to a gap through it on results, board meetings, block deals or regulatory news — and to the possibility that a price band freezes the instrument before your order can work.

    Why can five uncorrelated-looking strategies be one bet?

    This is the failure that turns a well-sized book into a badly-sized one without any single rule being broken.

    Five strategies with different names, different timeframes, different entry logic and different instrument lists can all be long the same underlying exposure at the same time. A momentum system, a breakout system, a pullback system and two sector rotations can converge on the same handful of large-cap names — or, more subtly, on the same broad direction, so that a single index-level move hits all five at once.

    Correlation is also not a constant. Strategies that behave independently in ordinary conditions have a tendency to move together precisely when it matters, because a large market-wide move is a shared input to all of them. A book sized on the assumption that the parts are independent is sized correctly only for the conditions in which independence held.

    Five separately-designed systems, each carrying its own permitted risk, all holding the same directional exposure. The book believes it holds five bets. It holds one, five times over.Five strategies with different names and different entry logic — momentum, breakout, pullback and two rotations — each carrying its own permitted risk, all converge on the same long directional exposure. The book believes it holds five separate bets when it holds one bet five times over, so a single market-wide move reaches all five at once.Five systems. Five risk budgets. One exposure.Momentum0.5% riskBreakout0.5% riskPullback0.5% riskRotation A0.5% riskRotation B0.5% riskNET LONG, ONE DIRECTIONone market-wide move reaches all fiveCompare the positions the systems held on the same day - not the logic that produced them.Correlation is not constant. It tends to rise on exactly the days the book can least absorb it.
    Five separately-designed systems, each carrying its own permitted risk, all holding the same directional exposure. The book believes it holds five bets. It holds one, five times over.

    How do you check whether your strategies are genuinely separate?

    1. 1

      Compare positions, not descriptions

      Take a period of your research or paper record and list which instruments each strategy held on the same day. Overlap in the holdings is the evidence; a difference in the entry logic is not.

    2. 2

      Aggregate exposure by direction

      Sum the net long and net short across all strategies at each point. A book that is persistently net long in one direction is running one macro position regardless of how many systems produced it.

    3. 3

      Group by shared driver, not by name

      Instruments in the same sector, with the same input cost, the same rate sensitivity or the same export destination will move together. Names in one sector are one exposure for risk purposes even when the strategies that bought them are unrelated.

    4. 4

      Look at the days that hurt, not the average

      Average relationships across a long period hide the concentration. Isolate the worst days for the book and check how many strategies contributed. If nearly all of them did, they are not independent when it counts.

    5. 5

      Cap exposure at the group level

      Set limits on total risk per sector and per direction, on top of the per-trade and per-strategy limits. This is the control that actually prevents the failure; measurement alone only detects it.

    How should capital be allocated between strategies?

    Allocation is a separate decision from sizing, and it is made far less often — quarterly or after a defined review, not in response to a bad week.

    The simplest defensible approach is to give each strategy an explicit share of the risk budget rather than of the capital, so that a strategy trading a volatile segment does not consume the whole book's risk while nominally holding a modest allocation. Each strategy then sizes its own positions inside its own budget and cannot borrow from another's.

    • Allocate risk, not rupees. Rupee allocations make a volatile strategy far larger than the number suggests.
    • Give each strategy its own heat cap inside the account-level cap, so one busy system cannot fill the whole book.
    • Change allocations on a schedule you set in advance. Re-allocating in response to the last few weeks is discretionary trading applied to your strategies instead of to stocks.
    • Include the cost of running each strategy — turnover, spread, and the levies covered in Costs, Slippage & Taxation — in the allocation decision, not just its research characteristics.
    • Hold back uncommitted capital deliberately. Margin requirements move, and a fully allocated book has no capacity to absorb a change in what the broker asks for.
    • Write down the conditions under which a strategy's allocation goes to zero before you fund it — this is what Strategy Decay & When to Switch Off is for.

    What Indian market constraints change the sizing arithmetic?

    The clean identity above is modified by real market plumbing, and a system that computes quantity without applying these will send orders that get rejected — or worse, accepted at a size you did not intend.

    • Lot sizes. Derivatives trade in fixed lots, so quantity is not continuous. The computed size is rounded down to whole lots, and if that rounds to zero the correct action is to skip the trade rather than to take one lot anyway — one lot may carry several times the intended risk.
    • Product type. Intraday and delivery products carry different margin treatment and different broker behaviour at cut-off. The same computed quantity is a different amount of leverage depending on which one the order specifies.
    • Margin, not just capital, is the binding constraint. A quantity your risk rule permits can still be an order your available margin will not support, and margin requirements are revised over time and by segment — read the current specification at the exchange and confirm with your broker rather than hard-coding it.
    • Price bands and circuit limits. An instrument that freezes cannot be exited at the stop, so the modelled risk of that position is a floor on the loss, not a ceiling.
    • Liquidity. A size that is fine on paper is not fine if it is a large share of the instrument's typical traded volume — the exit is the problem, not the entry. A practical universe filter is whether you could exit the whole position in one order without moving the price.
    • Corporate actions. Splits, bonuses and demergers change quantity and price on the broker's books. A sizing routine reading an unadjusted price series will compute confidently wrong numbers.

    What invalidates a sizing framework?

    The controls above stop working in identifiable ways. These are the signs the framework has drifted from what you designed.

    • A position whose size you cannot reconstruct from the rule. If you cannot rederive the quantity from equity, risk fraction and stop distance, something else set it.
    • A stop moved after entry to make a position size comfortable. That inverts the whole method — the stop sets the size, never the other way round.
    • Heat regularly at the cap. That is not efficiency; it means the cap is doing the work the strategy's selectivity was supposed to do.
    • A risk fraction raised after a good run. Fixed fractional sizing already scales with equity; anything on top is a discretionary bet on recent conditions continuing.
    • Several strategies stopped out on the same day by the same market move. The book was one position and the sizing assumed it was several.
    • Sizing computed from equity that includes unrealised profit on positions that are still open, so paper gains are being used to fund additional risk.

    Key points

    Quantity = risk budget ÷ stop distance. The stop sets the size; the size never sets the stop.
    Fixed fractional sizing shrinks the rupee risk automatically through a losing run.
    Equal rupees deployed is not equal risk — the volatile instrument silently dominates the book.
    Volatility-based sizing derives the stop from typical range over N candles, in candles not days.
    Sizing formulas fed by an estimated edge inherit and amplify every error in that estimate.
    Portfolio heat is total open risk; a per-trade limit does not cap the account.
    When heat would be breached, skip the signal rather than shrink it into the gap.
    Five differently-designed strategies holding the same direction are one bet, not five.
    Allocate risk between strategies, not capital, and change allocations on a set schedule.
    Lot size, product type, margin, price bands and liquidity all modify the clean arithmetic.

    Pro tip — Once a week, list every open position with the rupee amount it would cost at its invalidation price, and add the column up. That single number — not the value of the book, not the count of positions — is what your account is actually exposed to. Most traders who do this for the first time find the total is several times what they assumed, and none of it required breaking a per-trade rule.

    Frequently asked questions

    What is the position sizing formula?

    Quantity equals the risk budget in rupees divided by the stop distance per unit. The risk budget is a fraction of account equity fixed in advance; the stop distance is the gap between entry and the price that invalidates the idea. On an account of ₹10,00,000 risking 0.5%, the budget is ₹5,000, and a ₹32 stop distance gives 156 units. The identity is definitional — it determines size, not outcome.

    What is volatility-based position sizing?

    It sets the stop distance from how much the instrument typically moves rather than from a chart level, usually an average true range measured over a stated number of candles on the strategy's timeframe, multiplied by a chosen factor. The same division sum then runs unchanged. The result is that quiet instruments get larger quantities and volatile ones smaller, so every position carries comparable rupee risk instead of comparable rupee value.

    What is portfolio heat in trading?

    Portfolio heat is the total risk across all open positions at once — what it would cost if every position reached its invalidation price simultaneously. A 0.5% per-trade rule does not cap the account, because twelve open positions at 0.5% each is six per cent at risk together. Heat needs its own cap, checked before each order, with a new signal skipped rather than shrunk when the cap would be breached.

    Why is equal rupee allocation not equal risk?

    Because instruments move by different amounts. Two positions of ₹1,00,000 each look balanced on a positions screen, but if one instrument's typical range is several times the other's, a move to its invalidation price costs several times as much. A systematic book is measured in risk units rather than rupees deployed for exactly this reason.

    How do I know if my trading strategies are correlated?

    Compare the positions they actually held on the same days, not the logic that generated them. Aggregate net exposure by direction and by sector, then isolate the worst days for the book and count how many strategies contributed. Strategies that look independent in ordinary conditions often move together in a large market-wide move, because that move is a shared input to all of them.

    How much capital should each strategy in a portfolio get?

    Allocate risk rather than rupees, so a strategy trading a volatile segment does not consume the book's risk budget while holding a modest-looking capital share. Give each strategy its own heat cap inside the account-level cap, change allocations on a schedule set in advance rather than in response to recent weeks, and keep capital uncommitted because margin requirements change.

    Should position size increase after a winning streak?

    Fixed fractional sizing already increases the rupee risk as equity grows, without any decision being made. Raising the risk fraction on top of that is a separate discretionary bet that recent conditions will continue, and it is applied at exactly the point where the account has the most to lose. Whether to do it is a decision for the person whose capital it is, but it should be recognised as a change to the strategy rather than as an adjustment.