Does options flow predict price? An honest answer.

The short answer is no

No. Options flow does not predict price. Flow is a record of contracts that traded, and a record of what already happened carries no information about what happens next.

There is a narrower question worth asking, and the answer to that one is sometimes yes: does knowing that a contract traded far above its usual volume change what you should look at next? That is a filtering claim. It says flow can reorder your attention. It does not say flow knows where the stock is going.

Most of the damage in this category comes from collapsing those two claims into one. A scanner that narrows four hundred contracts to six is doing real work. A scanner presented as knowing which of the six will pay is doing fiction.

What people hope flow does, and what the data actually contains

The hope is straightforward. Somebody with better information — a fund, a desk, someone who did more work than you — is taking a position, and their footprint is visible in the option chain. Find the footprint, follow it.

Here is what the public chain actually publishes for a contract: ticker, expiration, strike, side, a last price and a bid/ask, the day's traded volume, an open-interest figure from the clearinghouse's reconciliation after the previous close, and an implied volatility with model Greeks derived from that price. The implied volatility and the Greeks are outputs of a pricing model applied to a quote, not observations of anything anyone did. Every claim about intent is an inference stacked on top of that list.

• It tells you that some number of contracts changed hands today on this exact strike and expiry.
• It tells you how many contracts were open as of the previous session's clearing reconciliation. During today's session that figure is a day old and contains none of today's trading.
• It does not tell you who traded, or whether that volume was one participant or two hundred.
• It does not tell you whether the trade opened a position or closed one.
• It does not tell you whether the print was standalone or one leg of a multi-leg structure.
• It does not tell you whether the trader had a directional view at all.

Why a large call print is not a bullish signal

This is the part most flow commentary skips, because it undermines the product. A large call print has at least half a dozen ordinary explanations that have nothing to do with someone betting on upside.

• One leg of a spread. What surfaced may be the long leg of a vertical, calendar, diagonal, or ratio. The other leg often fails the same volume or premium filter, so you see the bullish-looking half of a structure that is not net bullish.
• A roll. Closing a near-dated position and opening a further-dated one prints as two trades. Neither is a fresh opinion; together they are the same opinion with more time on it.
• Insurance against short stock. A trader short the equity buys calls to cap upside risk. Bought calls, bearish underlying position.
• Overwriting. Calls sold against long stock generate call volume from someone whose decision was about income and an exit price, not about direction.
• A closing trade. A buy can be someone buying back a short call they have carried for months. Volume rises exactly as it would for a new bullish position.
• A volatility trade. A delta-hedged long call is a position on how much the stock moves, not which way. It is re-hedged as delta changes and it pays only if realised movement beats what implied volatility already charged for. Direction is close to incidental.

The counterparty is not a second opinion

Every print has two sides, and the other one is usually a market maker managing inventory rather than anyone with a view. That has a consequence people get backwards: dealer hedging is not a vote, and it is not what produced the option volume you are looking at.

A dealer left short calls hedges by buying the underlying. A dealer left long calls hedges by selling it. Which way it runs depends on where the inventory ended up, and the hedge itself happens in the stock, not in the option — so the call volume on your screen was created by the customer order, while the hedging it triggers is mechanical and opinion-free. Reading a large print as "someone is bullish" quietly assumes the half of the trade you cannot see agreed to be wrong.

A hypothetical that shows the shape of the problem

Suppose a contract prints twelve thousand calls on a name that normally trades a few hundred. This example is hypothetical and illustrative — not a real trade, and not a claim about any outcome.

That single observation is identical whether it is a fund opening a directional position, a desk rolling an expiring hedge forward, or a long-time holder writing calls against stock they intend to sell at that strike. The scanner sees one number. The three cases have three different implications, and two of them are not bullish.

No amount of public chain data resolves which one it was. That is not a flaw in a particular tool. It is a property of the dataset.

The opening-versus-closing problem

Volume counts contracts traded. It does not say whether positions were created or retired. A large buy could be a new position or the close of a short, and those have opposite implications while looking identical in the volume figure.

There is one imperfect check available in public data. Compare the next day's open interest against today's volume: if open interest rises by roughly the traded amount, positions were opened on balance; if it falls, they were closed. That answer arrives a day late, which makes it useless for an intraday decision and genuinely useful for reviewing whether yesterday's flagged contracts represented new positioning.

It is not a complete check either. A position opened and closed inside the same session may never appear in open interest at all, and open interest nets everyone's activity — a day where one participant opened while another closed can show almost no change.

A better answer does exist, and it is worth being straight about: the options clearinghouse publishes an end-of-day dataset that separates opening from closing volume by buyer, seller, and account type. It is sold as a licensed product, it lands after the close rather than during the session, and ConvexRadar does not license it. Anything working from the public chain alone — here or anywhere else — is working without it.

Aggressor side, the familiar "bought at the ask" framing, is often presented as intent. Even a licensed tape carries no buy/sell flag on a trade; side is inferred by comparing the print against the prevailing quote. And it would only tell you who crossed the spread, not what the trade was for — multi-leg orders in particular get reported in ways that make leg-level side reading unreliable.

Where sweep and block labels actually come from

A sweep is an order routed across several exchanges at once to fill quickly against displayed size, usually read as urgency. A block is a large trade printed as a single transaction, often arranged before it reaches the exchange, usually read as planned institutional positioning. Both are real categories.

Getting close to either requires the consolidated options tape — exchange-level executions, timestamps, and trade condition codes — which is a licensed feed with real cost attached. Even with it, "sweep" remains a classification applied after the fact by clustering executions, not a field the tape hands you. A tool without the tape is working a step further back, deriving the label from chain aggregates: premium size, raw volume, volume relative to open interest.

ConvexRadar is in that second group and says so where it matters — the print type on each row carries a note that it is inferred from aggregated chain data rather than exchange-level tape. There is also no dark-pool options feed for anyone to buy. US listed options trade on registered options exchanges; the off-exchange venue structure that makes dark pools meaningful in equities does not exist for them, so "dark-pool options prints" does not name a dataset. When a label's provenance is inference rather than observation, it deserves less weight than a confident-looking badge implies.

The selection bias baked into flow marketing

Take a hypothetical stream that flags twenty contracts a day. Over a month that is roughly four hundred. Options are leveraged instruments, so in any market that moves at all, some of those four hundred will produce a large percentage gain without anything predictive having happened.

Those are the ones that get screenshotted. The rest do not. What you end up looking at is not a sample of performance — it is the right tail, selected after the outcome was already known, presented as though it were the process.

Survivorship compounds it, and not because anyone necessarily behaves badly. Nothing obliges a public record to keep existing. A stream going through a bad stretch tends to get posted about less, and whatever is still being promoted is by construction the thing that survived long enough to be promoted. That is exactly the population you cannot generalise from.

The only honest test is pre-registered: state what was flagged before the outcome is known, grade all of it, and write the method down. ConvexRadar publishes that ledger at /track-record. It grades each top-ranked contract against that same contract's price seven days later using retained chain snapshots, includes the losers, and states its own limits on the page: the prices are chain snapshots rather than fills, so spread and slippage are not modelled, and a contract that leaves the scan universe is dropped rather than counted as a loss — a survivorship gap in our own record, disclosed rather than buried. It measures whether the ranking carried information, not what a trader would have netted. It went up empty and fills in as snapshots accumulate, rather than being backdated into existence.

What flow is genuinely good for

None of this makes flow worthless. It makes it a filter rather than a forecast, and a good filter is worth having when the alternative is staring at a raw watchlist.

• Narrowing the universe. A liquid underlying can list several hundred contracts once you multiply expirations by strikes by side. Chain-pressure metrics get that down to a handful worth pulling a chart for.
• Flagging positioning that deserves a question. Volume many times a contract's standing open interest — on a contract whose open-interest base is large enough to mean anything — is a legitimate "why?", not an answer.
• Catalyst context. Whether a scheduled event falls inside the contract's life changes how you read the activity. Positioning into earnings is ordinary and common. The same activity with nothing on the calendar is a different observation.
• Pricing awareness. IV rank tells you where a contract's implied volatility sits against its own past year. Heavy activity on a contract whose IV rank is already high means long premium there is being paid at a rich level for that name — the wider range is partly in the price already, so a move has to be bigger than usual to pay for the option.
• Liquidity and size reality. Open interest, volume, and spread width tell you whether a position is one you can actually get out of.

Using flow without letting it make the decision

The practical fix is ordering. Flow belongs at the start of a process, never at the end of one.

• Filter to a short list on chain pressure, premium traded, and liquidity.
• Verify the underlying's structure on the chart independently. If the chart disagrees with the flow, the flow does not win.
• Check what the option costs relative to its own volatility history before deciding it is attractive.
• Check the calendar for events inside the expiry — including the ones that crush implied volatility while you are right about direction.
• Set position size and exits before entry, on the working assumption that this particular idea is wrong.
• Log the outcome, including the ones you would rather forget.

The limits, stated plainly

The risk rules matter more than the input. Defined premium at risk per position, a cap on total exposure, exits decided before entry, and no averaging into losers all hold whether or not a given flow read was any good. That is the whole point — a process that only survives when the read is right is not a process. Over time your own journal is a better guide than anyone's marketing, and it will tell you whether flow-sourced ideas work for you long before a vendor does.

ConvexRadar does not predict price and does not sell signals. It is research software, not financial advice, and nothing it produces is a recommendation to trade. What it does is rank contracts on volume against open interest, estimated premium traded, IV rank, call/put bias, and catalyst context from earnings, filings, analyst target changes and headlines; label each row's quality alongside the risk flags behind that label, including thin volume, very low open interest, high IV and earnings IV-crush risk; mark print typing as chain-derived rather than exchange-observed; and publish the graded ledger with losers and its own limits included.

What it cannot do, and does not claim to: identify who initiated a trade, tell you in real time whether a position was opened or closed, read a consolidated tape it does not license, see dark-pool options data that does not exist, or tell you which way a stock is going. Options involve substantial risk including total loss of premium.

If you want the mechanics behind the metrics, the /options-education center covers the Greeks, volatility and strategy structure. /options-flow-scanner describes how the ranking is actually built. /track-record is the graded ledger, published in full. /options-trading-journal is where your own closed trades turn into realised P/L and a win rate you did not have to take anyone's word for. /pricing lists the tiers — the free tier shows the top five ranked rows with every field intact, which is enough to judge whether the workflow fits how you already work.