Order flow trading in India

A candlestick reports four numbers: the price at the start of the interval, the highest and lowest prices reached inside it, and the price at the end. That is a genuinely useful compression, and for most purposes it is enough. But compression is lossy by definition, and what a candle throws away is not random detail. It throws away the process. Two markets can produce a byte-for-byte identical candle through completely opposite processes, and the candle has no way of telling you which one you are looking at.

The figure below draws that claim literally. Both panels contain the same candle: open 1,240, high 1,252, low 1,236, close 1,248. Both panels contain the same total volume, 11,590 shares. If you saw either panel's candle on a chart you would read it the same way, because it is the same candle. What differs is the ladder either side of it, which splits that identical volume by which participant crossed the spread at each price.

In the left panel, buyers did nearly all of the crossing. They paid up for 7,630 shares' worth of aggression, and the heaviest of it landed at 1,252, the high of the session, where somebody sat passively selling into every bid that arrived. The buyers got their high print, and then the close came back four points. Effort was spent at the top and the result did not hold.

In the right panel, sellers did nearly all of the crossing. They hit bids for 7,630 shares, concentrated at 1,238 near the low, and price rose anyway, closing eight points above the open. Every one of those aggressive sells was taken by a passive buyer who did not flinch. Same candle. The opposite argument.

These two sessions do not deserve the same conclusion. The first shows demand exhausting into supply at a high. The second shows supply exhausting into demand at a low. A trader reading only the candle would treat them identically, which is not a subtle loss of nuance but a straightforward inversion of what happened.

This is the honest case for order flow, and it is a strong one. The question a candle answers is where did price end. The question order flow asks is who was in a hurry, and urgency is the more informative of the two, because urgency is expensive. A participant who crosses the spread has decided that the price available right now is worth more than the better price they might get by waiting. That decision costs real money on every single transaction, and people do not repeatedly pay a cost for no reason.

Two clarifications belong here, because both are routinely fudged. First, order flow is not a different kind of volume. Volume counts how much traded. Order flow starts from exactly the same transactions but adds one field that volume does not carry: which side initiated. That single field is the entire difference, and everything on this page depends on it being available. Second, order flow is not the order book. The book is resting intent, the queue of unfilled orders waiting at each price, and it belongs to the market-depth guide along with the impact cost of walking it. Order flow is the record of what actually traded against that book. The book is what people say they will do. The flow is what they did.

Hold on to that second distinction, because it becomes load-bearing later. A great deal of Indian order-flow marketing shows you a picture of the book, or a picture of volume, and calls it flow.

Every transaction that has ever printed on an exchange has two counterparties, and exactly one of them chose the moment. That is not a philosophical observation, it is a structural fact about how a matching engine works, and it is the foundation of everything that follows.

One side is passive. It posted a limit order at a stated price and then waited. It might wait a second, or an hour, or forever. In effect it has written the market an option: anybody may trade against it at that price until it is cancelled or filled. In return for granting that option, the passive side gets a better price than the market is currently offering, and it collects the spread when it is hit.

The other side is the aggressor. It sent an order that crossed the spread, a market order or a limit order priced through the touch, and took whatever was resting there. It did not wait, and it did not get to name its price. It paid the bid-ask spread as the fee for immediacy, plus whatever further slippage its size caused as it consumed successive price levels.

The market's own vocabulary is precise about this. An aggressive buyer lifts the offer. An aggressive seller hits the bid. The verbs carry the direction of the crossing, because the crossing is the thing being named.

The asymmetry between these two roles is what makes the aggressor side informative. A passive order tells you what somebody would like. An aggressive order tells you what somebody was willing to pay to have now, and that willingness has a price tag attached. Patience is free; impatience is not. When one side of a market consistently pays for immediacy and the other consistently collects for supplying it, that is a real and measurable asymmetry of behaviour, and it exists whether or not anybody is watching.

Now the awkward part. Nothing about the aggressor side is visible on a price chart. A chart shows that 11,590 shares traded and that the close was 1,248. It cannot show that 7,630 of those shares were bought by somebody crossing, because that field is simply not present in the data a chart is drawn from. Volume is a single unsigned number per interval. Order flow requires that number to be split in two, and the split is possible only if each individual transaction carries a tag naming its initiator.

Which gives the single most important sentence on this page, so it is worth stating flatly: order-flow analysis is volume analysis plus one extra field, and that field is the aggressor side. Remove the field and you are doing volume analysis, whatever the chart is called. Nearly every dispute about whether a given Indian retail tool is doing "real" order flow reduces to one question about that field: is it in the data, or is the tool working it out for itself? Hold that question. The rest of the page earns it.

Once each transaction carries an aggressor tag, arithmetic becomes available, and delta is the whole of it. Add up the volume that was buy-aggressed, add up the volume that was sell-aggressed, subtract the second from the first. Positive delta means buyers did more of the crossing over that window. Negative delta means sellers did. Run the subtraction continuously through a session and you get cumulative delta, a line that rises while aggressive buying dominates and falls while aggressive selling does.

The figure below performs it print by print over ten trades, so that nothing in the arithmetic is hidden.

Across those ten prints, 3,450 shares were bought by somebody lifting the ask and 1,650 were sold by somebody hitting the bid, so delta closed at plus 1,800 on 5,100 shares traded. Price moved forty paise. That ratio of urgency to result is exactly the sort of thing worth noticing, and it is not a thing the raw candle can support.

Now a correction that a great deal of published material gets wrong, including material that is otherwise careful. Positive delta does not mean there were more buyers than sellers. There cannot be more buyers than sellers. Every share bought is a share sold; total buying and total selling are equal by construction, in every market, in every interval, always. What delta measures is not the balance of buyers against sellers, which is fixed at zero, but the balance of urgency between them. It answers "which side was more willing to pay for immediacy", never "which side was bigger".

That distinction is practical, not merely semantic. If you believe delta counts buyers, a large positive delta reads as bullish on its own and you will trade it directly. If you understand that delta counts impatience, a large positive delta becomes a question rather than an answer: buyers were paying up, so what did they get for it? Sometimes the answer is plenty of price, and the read is unremarkable. Sometimes the answer is almost no price at all, and that is far more interesting, because it means somebody was standing on the other side of it.

Three ways cumulative delta is commonly read, in rough ascending order of usefulness:

Delta and price agreeing. Price rises and cumulative delta rises with it. This is the ordinary case and it is close to uninformative, because the chart had already told you price rose. A tool that mostly shows you this is mostly redrawing your chart in a second colour.

Delta at an extreme. A very large one-directional delta inside a short window marks urgency, which is genuine information about the last few seconds. Urgency is not the same as direction that persists. A panicked retail flush and a well-worked institutional exit can look similar on a delta reading, and only one of them implies anything at all about tomorrow.

Delta and price diverging. Cumulative delta falls hard while price refuses to break, or climbs steadily while price refuses to make new highs. This is where nearly all of the value sits, and it is the subject of the next section.

If order flow gives you one durable read, this is it, and much of the rest of this page is scaffolding for it.

Absorption is what happens when heavy aggression meets a passive participant who does not move. Sellers hit the bid, and hit it again, and keep hitting it. Cumulative delta falls hard. Volume piles up. And price does not go down, because every aggressive sell is being taken by a resting bid that keeps being replenished. The selling is not being accepted by the market at a lower price. It is being soaked up at this price by somebody who wants the shares.

Its opposite is acceptance. The same aggression arrives, the resting orders are consumed and not replaced, price moves through the level and then trades comfortably on the other side of it. Nobody was standing underneath.

The figure below sets the two side by side under a deliberately strict constraint: the aggression is identical in both panels, and so is the cumulative delta, which reads minus 4,200 in each. The only thing that differs between them is whether the passive bid comes back.

That constraint is the point of the figure. Delta on its own cannot separate absorption from acceptance, because delta is the same number in both. Delta against price separates them instantly. Which yields a rule worth carrying out of this page:

Order flow is interesting exactly where it disagrees with price, and close to worthless where it agrees. Agreement restates the chart. Disagreement means somebody is standing against the flow.

Readers who have worked through the Wyckoff method will recognise this immediately, because it is the same idea in a different vocabulary. Wyckoff's law of effort versus result asks whether the effort expended, which he read through volume, produced a proportionate result in price. Where large effort produces small result, something is absorbing it. Order flow is that law measured at the resolution of individual transactions rather than daily bars. It is not a new insight. It is an old insight with a finer instrument, which is a fair description of most of what microstructure has to offer.

Now the honest limits, because absorption is also the read most often oversold.

It is confirmed only afterwards. While it is happening, "a bid that keeps refilling" and "a bid that has not run out yet" are the same observation. What separates them is what happens next, which is precisely the thing you wanted to know in advance.

You cannot see who is absorbing, or why. A large passive buyer might be accumulating a position, or hedging a derivatives book, or working an index rebalance on a fixed schedule, or unwinding something entirely unrelated. Only the first of those implies anything about direction. The others are mechanical, and they stop when the mechanical task is finished, without warning and without regard to the level.

A refill can simply run out. Absorption that holds for twenty minutes and then fails looks, in hindsight, exactly like a level that was defended right up until it was not. There is no quantity of absorption that guarantees the level survives, and there is no threshold at which the read becomes safe.

None of that makes absorption useless. It makes it evidence rather than proof, which is the correct status for very nearly everything in this field.

Everything above is true of markets in general. What follows is specific to what an Indian retail trader actually receives, and it decides whether any of the above is usable by you.

Start from the requirement. To measure order flow rather than estimate it, you need a transaction-by-transaction record in which each trade carries the side that initiated it, or failing that, the exact quote in force at the instant of each trade so that the initiator can be established without ambiguity. That is the input. Delta, cumulative delta, footprint and absorption are all arithmetic performed on it.

Now the supply. The following describes the position as of 18 July 2026. Data products, tariffs and their terms change, so treat this as a description of the shape of the problem, and verify the specifics with the exchange and with whoever supplies your feed before you rely on any of it.

The National Stock Exchange publishes its real-time data in named tiers. Level 1 carries the best bid and ask. Level 2 carries market depth to the best five bid and ask prices, and it is Level 2 that your platform's depth window is showing you in continuous trading. Level 3 extends the same aggregated depth to the best twenty prices a side. Level 3 is not a recent innovation and it is not hidden, but it is a separately priced exchange product, offered for the cash segment, rather than anything a retail app includes by default. All three are views of the book, of resting intent, rather than of the transactions.

These tiers are also periodic snapshots rather than an event stream, and that difference is not a technicality. An event stream reports every order added, modified, cancelled and executed, in sequence, each with its own timestamp. A snapshot reports what things looked like at the moment of sampling and says nothing whatever about what happened between samples. NSE has described the five-depth feed in terms of a one-second refresh, and SEBI, describing the same free five-level quote snapshot, put it as being updated almost every second. In a liquid name during an active period, a great many transactions will occur inside that second, and the snapshot compresses all of them into one after-the-fact picture, in much the same way a candle compresses a minute.

Now the part that is genuinely surprising, and that most coverage of this subject gets backwards. The full order-by-order tick-by-tick feed, which does report every add, modify, cancel and trade as a separate timestamped event, is not locked away behind a privilege or a colocation contract. In 2018 SEBI required the exchanges to supply tick-by-tick feeds to all trading members free of cost, conditional on the member building the infrastructure needed to receive and process them. The gate is real, but it is architectural rather than regulatory. That feed is a firehose measured in events per second per symbol; consuming it needs machinery, and what a broker can practically push down to a phone is a small periodic packet. Retail does not lack this data because it is forbidden. Retail lacks it because nobody is redistributing a firehose to a mobile app, which is a different problem with different implications.

So look at what the retail packet actually contains: last traded price and quantity, an average price, cumulative volume, total buy and total sell quantity, the session's open, high, low and close, open interest where it applies, and five bid and five ask entries. There is no field anywhere in it naming the initiator of any individual trade.

Put all of that together and the consequence is unavoidable. A retail footprint or delta chart in India is not reading the aggressor side. It is inferring it, almost always by comparing each trade price against the most recently seen bid and ask: at or above the ask, call it buy-initiated; at or below the bid, call it sell-initiated; anywhere in between, guess, usually from the direction of the last price change.

It is worth being precise that there are three tiers of knowledge about the aggressor side, not two, because the distinction between the middle one and the bottom one is exactly what is being blurred when a retail tool borrows institutional vocabulary.

Measured. The side is stated. BSE's un-netted professional trade feed, in which every on-exchange trade is reported individually, carries an explicit aggressor-side field per trade, with its own timestamp. Nothing is being worked out; the exchange says which side crossed.

Derived. The side is not stated but follows with certainty. NSE's tick-by-tick feed carries no aggressor field, but each trade names both the buy and the sell order, against a stream containing every order's arrival. Whichever of the two was not already resting is the aggressor. That is a deduction, not a guess, and it is exactly as reliable as the stream itself.

Inferred. The side is estimated from a price comparison against a quote that was sampled at some earlier instant. This is the retail case, and it is the only one of the three that can be wrong.

That rule is not stupid. On calm data it is right most of the time. But it fails in exactly the conditions that make order flow interesting, and the figure below shows what the failure costs.

Ten trades. Three tags wrong, covering 1,470 of 4,770 shares. And the delta reading does not merely lose precision, it changes sign: the true reading is minus 670, meaning sellers were doing the crossing on net, while the inferred reading is plus 1,070, which says the opposite. Somebody acting on the second is not being slightly misled. They are being told the reverse of what happened.

Three further details make this worse rather than better.

The errors cluster exactly where you care. Mis-tagging happens when the quote moves between the trade and the sample, when a trade prints between the bid and the ask, and when the quote used for comparison is already stale. All three become far more likely during fast, heavy, one-sided activity, which is precisely the condition under which somebody decides to consult their delta chart in the first place. The instrument is least reliable at the moment it is most consulted.

The errors do not announce themselves. A footprint built on inferred sides looks exactly as crisp and authoritative as one built on measured sides. The numbers are rendered to the individual share. There is no confidence band, no shading, nothing that says "these three prints were guesses". The confidence of the display is entirely unrelated to the confidence of the data behind it, and that is a far more dangerous property than plain inaccuracy.

The errors compound in a cumulative measure. Delta over one minute absorbs a handful of mistakes. Cumulative delta over a session carries every one of them forward for the rest of the day. A divergence appearing on a cumulative delta line late in the afternoon may be a real divergence, or it may be the accumulated residue of a few hundred small mis-attributions since the open, and nothing on the chart distinguishes those two cases.

Note carefully what this argument is not. It is not that Indian markets are opaque, or that retail is being cheated, or that the exchanges are concealing something. The data tiers described here are published, ordinary and unremarkable, every major market has a version of the same ladder, and the deepest tier of all is required to be given to trading members for nothing. Nor is the argument that inference is worthless, because a mostly-right tag really is more informative than no tag at all. The argument is narrower and much harder to escape: inference is not measurement, a footprint built on inferred sides inherits the inference error silently, and any honest account of order flow for Indian retail has to say so on the way in rather than in a disclaimer at the bottom.

The commercial consequence follows from the architecture rather than from any conspiracy. Reading order flow properly means consuming an event stream, and consuming an event stream means infrastructure. That is a real threshold, and it is the actual reason most Indian retail traders are working with an estimate. It is also, usefully, a threshold you can ask about in plain language instead of guessing at.

The previous section is deflationary but it is not nihilistic. A good deal survives it, and the surviving material has the considerable advantage of being arithmetic rather than estimate. Three things are worth knowing about.

Volume at price. Take every transaction in a session, bin it by the price at which it happened, and you have a profile: a histogram of how much trading occurred at each price. This requires only the traded price and the traded quantity, both of which are published without ambiguity. No assumption about who crossed the spread enters anywhere. The heaviest bin is the point of control, the price at which the market did most of its business; the contiguous band around it holding the bulk of the session's volume is the value area. The figure below computes both from the price path drawn beside it, so the profile is an output of the session rather than a decoration next to it.

What a profile is actually telling you is where agreement happened. A heavy node is a price at which a lot of buying and a lot of selling both found the price acceptable, and such prices tend to attract trade again. A thin node is a price the market passed through quickly without much business, and prices tend to move through those regions quickly a second time as well. This is a real and stable observation about how markets distribute activity, and it needs no aggressor tag whatever.

What a profile is not telling you is who initiated any of it. A heavy node marks agreement on price; it cannot say whether that agreement was reached by buyers paying up or by sellers hitting down. That is a smaller claim than delta makes. It is also a claim that survives contact with the data, which delta on a snapshot feed does not.

The depth window. The five-level ladder your platform already shows is genuinely useful, provided you read it for what it is. It gives you the spread, the near-touch depth, and a workable estimate of what your own size would cost to execute immediately. The full treatment, including how to walk the book and compute the impact cost of a given order, belongs to the market-depth guide. Two cautions carry over here: depth is resting intent and can be cancelled in the time it takes you to click, and a large resting order is not support, it is one participant's current opinion.

Delivery percentage. This one is specific to the Indian cash market and it is badly underused. Both exchanges publish, per security and per trading day, the deliverable quantity alongside the quantity traded, and the ratio itself, in their daily delivery-position files. You do not have to compute it, and you do not have to wait for the next morning: the files appear the same trading day, after the close. It separates volume that resulted in an actual change of ownership from volume that was opened and closed inside the same session.

It is not order flow, and it should not be dressed up as such. It is daily rather than intraday, it says nothing about who crossed the spread, and it necessarily arrives only after the close, because deliverable quantity is a post-session netting figure that does not exist while the market is open. Any tool showing you a live intraday delivery percentage is showing you an estimate of a number that has not been calculated yet. But the published figure is measured, and it speaks to commitment, which is usually the very thing intraday delta is reached for in order to establish. A large move on heavy volume with a low delivery proportion tells a materially different story from the same move with a high one. Two practical notes: the ratio is blanked for securities in the trade-for-trade series, where every trade goes to delivery by definition, and the volume guide covers how to use the figure properly.

The pattern in that table is the practical takeaway of the whole page. The measured rows are modest and reliable. The inferred rows are exciting and unreliable, and they are the ones the marketing is built on. That ordering is not a coincidence: the further a claim sits from what the data can support, the more room there is to make it sound impressive.

Suppose the data problem vanished tomorrow and you received a clean, tagged, order-by-order stream. Order flow would still be a narrower tool than it is usually presented as, for reasons that have nothing to do with feeds.

Order flow is a microstructure lens. It reads the mechanics of execution over seconds and minutes. That is its natural horizon, and information decays out of it quickly, because the conditions it describes, who is currently impatient and who is currently absorbing, are conditions that change within minutes and frequently within seconds. A read about the next ninety seconds is a genuine read. It is not an opinion about next week, and it does not become one by being observed carefully.

This creates a timeframe mismatch that catches a large number of people. Most retail traders who adopt order flow hold positions for days. They are using an instrument calibrated to seconds to inform a decision measured in sessions, and the instrument does not carry that far. If your holding period is a week, the marginal value of knowing who crossed the spread at 11:42 is close to zero, and the cost in attention is not.

The second problem is competition. Microstructure information is contested by the participants best equipped to contest it: those already holding the event stream, sitting closest to the matching engine, and acting automatically. This does not make retail order-flow reading pointless, but it does set the realistic bar. Any read that is simple enough to be mechanised has, in all likelihood, already been mechanised by somebody faster. What survives longest is the part requiring judgement, which is the absorption read, precisely because it is a contextual judgement about whether a level is being defended rather than a pattern that can be matched.

Order flow is a real lens that is frequently sold dishonestly, and both halves of that sentence deserve to survive. The concepts are sound. The aggressor and the passive side are a genuine structural asymmetry. Delta is a coherent measure of urgency. Absorption is the direct descendant of a hundred-year-old idea about effort and result, measured with a finer instrument, and it remains the most useful single read available at this resolution. None of that is marketing.

What is marketing is the quiet omission of the data question. A footprint chart drawn from inferred aggressor sides on a snapshot feed is a real chart of an estimated quantity, and it is presented as a precise chart of a measured one. The estimation is usually reasonable. It is never disclosed. And in the fast, heavy, one-sided conditions under which anybody actually consults it, the estimate is at its worst.

So the first question is not "how do I read order flow". It is am I measuring it, or am I being shown an inference. Four questions settle it for any tool you are offered:

Where does the aggressor side come from? There are only three acceptable answers, and they are not equivalent. The record states it. Or it follows deterministically from an order-by-order stream. Or it is estimated from the trade price against a previously sampled quote. Only the third can be wrong, and the third is the retail case.

Is the feed an event stream or a snapshot? If it samples at an interval, everything that happened between samples has been reconstructed rather than observed.

What happens to an ambiguous print? Every derivation rule has to do something with a trade that prints between the bid and the ask. Ask what. A tool that cannot say has not thought about it.

Does the display distinguish certainty from estimate? Almost none do. Knowing that the display will look identical either way is most of the defence.

If the answers put you on the inferred side of that line, which for most Indian retail traders they will, the sensible response is neither to abandon the subject nor to pretend the problem away. It is to use the parts that survive: volume at price, which is arithmetic; the depth window, read as perishable intent; delivery percentage, read as a daily commitment tell; and the absorption idea itself, which you can watch for using price and volume together without needing a tagged delta at all. Heavy volume at a level with no resulting price movement is absorption's signature, and it is visible without any aggressor field whatever. That is the version of this subject that an Indian retail trader can actually stand behind.

And the deeper point sits upstream of all of it. Order flow, at its very best, refines the timing of an entry at a level you had already identified. It does not find the level, it does not decide whether the trade is worth taking, it does not set the invalidation, and it does not size the position. Those decisions are made before any of this becomes relevant, and they are what the method we teach is built around. A flawless absorption read on a trade that should never have been on the list is worth precisely nothing.

• The NSE data tiers, and the five and twenty level figures. NSE's published description of its real-time data subscription sets out the tiers used on this page: Level 1 as the best bid and ask, Level 2 as market depth to the best five bid and ask prices, Level 3 as depth to the best twenty, and a tick-by-tick feed carrying the full order book. The exchange's own capital-market technical specification confirms twenty levels on each side rather than twenty in total, and its published tariff lists Level 3 as a separately priced product in the cash segment. nseindia.com real-time data subscription
• The snapshot refresh, and the definition of tick-by-tick. NSE has described its five-depth dissemination in terms of a one-second refresh. A 2018 SEBI board paper describes the free five best bid and ask quote snapshot as being updated almost every second, and uses the word snapshot for it. SEBI circular SEBI/HO/MRD/DP/CIR/P/2018/62 defines the tick-by-tick feed as a detailed view of the entire order book covering the addition, modification and cancellation of orders as well as trades. sebi.gov.in
• Tick-by-tick is required to be provided to trading members free of cost. This is the correction that matters most on this page, because the common account of the subject has it backwards. The same 2018 SEBI circular requires stock exchanges to provide tick-by-tick feeds to all trading members without charge, subject to those members creating the infrastructure needed to receive and process the feed. The constraint on retail access to order-by-order data is therefore architectural and commercial rather than regulatory, and it is wrong to describe the feed as gated behind colocation. sebi.gov.in
• Where the aggressor side is stated, and where it is not. BSE's market and reference data interfaces manual documents an explicit aggressor-side field, with a companion timestamp, on trades in its un-netted professional feed, in which every on-exchange trade is reported individually. Its netted feed does not report trades individually. NSE's tick-by-tick trade message carries no aggressor field, but names both the buy and the sell order, so the initiating side follows deterministically from the order stream. Retail streaming packets carry neither: they publish last traded price and quantity, average price, volume, total buy and total sell quantity, session open, high, low and close, open interest, and five bid and five ask entries. This is the basis for the three-tier distinction between a measured, a derived and an inferred aggressor side. bseindia.com
• Delivery position data. Both exchanges publish daily security-wise delivery position files giving traded quantity, deliverable quantity and the ratio between them, on the same trading day after the close. NSE's file is titled as a security-wise delivery position for compulsory rolling settlement and states the percentage of deliverable quantity to traded quantity directly, so the figure is published rather than computed by the reader. The ratio is blank for trade-for-trade series. nseindia.com
• Market-access regulation. The Securities and Exchange Board of India frames the rules governing market-data dissemination, colocation and algorithmic access, which is the regulatory context for the tiering described here. Placing orders with no intention of executing them, in order to create a false impression of demand or supply, is prohibited market manipulation and is not a technique this page describes or endorses. sebi.gov.in
• Microstructure fundamentals. The passive and aggressive roles in a matched transaction, the definition of delta as buy-initiated minus sell-initiated volume, the identity that total buying equals total selling in every interval, and volume-at-price constructions including the point of control and the value area, follow standard definitions of exchange execution mechanics.
• All quantities on this page are illustrative. Every price, volume, delta and percentage in the figures and worked examples was constructed to demonstrate a mechanism. They are internally consistent, and each figure's totals are computed from the same series it draws, but they are not observations of any real security on any real day.