Cross-Impact in Trading: How One Order Moves Another

An index future ticks higher on a burst of buying. A beat later, the best offer on one of its larger constituents lifts by a couple of cents, and the spread on that single stock widens for a few seconds before it settles. Nobody traded that stock in the moment the future moved. So what connected the two?

That question sits at the heart of cross-impact, and it is easy to answer badly. The lazy reading says the future “caused” the stock to move. The careful reading asks whether trading pressure in one instrument is genuinely linked to a response in another, or whether both simply reacted to the same piece of news at the same instant. I keep those two answers in separate columns of my notes, because collapsing them is how a trader talks themselves into a spillover that was never there.

What cross-impact actually measures

Start with the plain version. Cross-impact is the idea that trading activity in one instrument can be associated with a price or liquidity change in another, separate from the direct effect of trading that second instrument itself. To see the boundary, split it from its more familiar cousin. When you send a large buy order into a stock and its own price drifts up as you fill, that is market impact, the cost your own flow imposes on the thing you are trading. Cross-impact is the version that reaches sideways: order flow in instrument A lining up with a move or a liquidity shift in instrument B, which nobody was trading at that instant.

Researchers usually organise this as a matrix. The diagonal cells hold own-impact, the pressure a trade puts on its own price. The off-diagonal cells hold cross-impact, the pressure that appears to travel between instruments. That layout is worth picturing, because it makes the honest question obvious: how much of what you see off the diagonal is real transmission, and how much is just two prices answering the same call. Traders who watch this closely tend to read it alongside order-flow imbalance on each side, so the pressure and the response can be lined up in time rather than assumed.

One detail is worth holding onto here: the sign. Cross-impact can be positive, where buying pressure in one instrument lines up with an uptick in another, or negative, where buying in one lines up with selling pressure or a downtick in a substitute or a hedge. A trader who assumes every link runs positive will misread the ones that run the other way, and hedging relationships are full of the negative kind. When a desk buys the future and sells the cash basket to stay flat, the pressure it sends into those stocks points down even as the future goes up.

Cross-impact and correlation answer different questions

Correlation measures co-movement in returns. If two stocks tend to rise and fall together over a window, their returns are correlated, and that is the whole of what the number tells you. It stays silent on which one led, on whether any order flow passed between them, and on whether a third factor drove both. Cross-impact asks the sharper question: does pressure in one book, actual buying or selling, show up as a price or liquidity response in the other?

The gap between the two matters in practice. Two names can sit at 0.9 correlation for a year and carry no cross-impact at all, because both are simply tracking the same sector index and never touch each other’s order books. The reverse happens too. Correlation can slacken during a stress week while genuine pressure jumps between a hedge and the thing it hedges, which is one reason correlations break down under stress right when traders lean on them hardest. In my own notes I only tag a pair as cross-impact when I can point to the order flow on one side and the quote change on the other inside the same second. A correlation of 0.9 on its own never earns the tag.

The channels that link two instruments

Cross-impact is not magic, and it helps to name the specific pipes it can travel through. There are four common ones:

• Economically related companies share information. A supplier’s profit warning reprices its customers, and flow into one can precede a move in the other.
• An ETF, its futures, and its constituent stocks are bound together by arbitrage and hedging. When the basket and the wrapper drift apart, authorised participants and market makers trade the difference, and that activity ties the prices back together.
• Common investors adjust related positions together. A fund trimming risk sells several correlated names in the same session, so pressure in one is mechanically linked to pressure in the rest.
• A price move in one venue reshapes quotes and liquidity in another. Market makers who just got run over in the future widen their quotes in the cash names they use to hedge.

Those channels are the reason cross-impact belongs in the wider study of inter-market analysis. They are also where the price discovery process gets interesting, because the most liquid, lowest-cost venue often moves first and the linked instruments catch up. First to move is not the same as cause, though, and that distinction is the whole game in the next section.

A worked example, and the trap inside it

Take an illustrative sequence, with numbers invented to keep it clean. At 09:31:04 an S&P 500 future prints a fast three-tick up-move on heavy lots. Forty milliseconds later, the best offer on a large constituent lifts by two cents and its quoted size thins. A cross-impact study scanning that window would flag the future’s buy pressure as associated with the constituent’s quote shift, and the ordering looks decisive: future first, stock second.

Here is the part that catches people. That ordering does not prove the future’s trade moved the stock. A macro headline could have hit both feeds inside the same forty milliseconds, repricing the index and the single name together, with the future simply faster to react because it is the deeper, cheaper venue. The sequence is consistent with cross-impact. It is equally consistent with a common shock that owes nothing to the future’s order flow. I mark both explanations before I believe either, and over the years I have talked myself out of more of these spillovers than I have kept. The sequence alone is evidence of timing, never of cause.

Now flip the illustrative case to see what stronger evidence would look like. Suppose the same future prints its three-tick move, but this time a thinly covered constituent with no news of its own reacts, and it reacts in the exact direction the arbitrage relationship predicts, repeatedly, across dozens of separate bursts that each land on a quiet tape. That pattern is harder to blame on shared news, because there was little shared news to blame, and the repetition argues against a single coincidence. Even then I would call it evidence of a link rather than proof of a mechanism, but it is the kind of accumulation that moves a pair from “maybe” to “probably” in my notes.

Why cross-impact is hard to measure

If the worked example felt slippery, that is honest, because measuring cross-impact cleanly runs into four problems at once.

• Event timing. You have to fix the exact window in which one instrument’s flow is allowed to “explain” the other’s move, and a window that is too wide will swallow unrelated events.
• A benchmark for normal co-movement. Before you can call a response unusual, you need a baseline for how these two instruments move together on a quiet day.
• Market-wide news. A broad shock lifts everything at once and masquerades as transmission between any two names you care to pair.
• Simultaneous reaction versus directional influence. Two prices answering the same information look almost identical to one price pushing the other.

The omitted-information problem is the one that quietly wrecks estimates. If you fail to control for a common driver, a factor move, an index rebalance, a rate headline, its footprint gets misread as cross-impact and the number comes out too large. Frequency compounds it. The same pair can show strong linkage measured at the millisecond and almost none measured minute by minute, because you are capturing different mechanisms at each scale. It is the same lead-lag puzzle that shows up when one group of stocks turns before another. Knowing which moved first rarely tells you why, and a cross-impact estimate inherits that ambiguity rather than resolving it.

Where the research can take you next

If you want to push further, the academic work on market impact now includes methods for the nonparametric estimation of self- and cross-impact, which means models that estimate both the diagonal own-impact and the off-diagonal cross terms without forcing a rigid functional shape onto them. I will not hand you a headline result from that literature here, because a coefficient lifted out of its dataset, its universe, and its sampling frequency is worse than no number at all. Treat those studies as a map of what has been tried and where the potholes are, and read them with the measurement problems above already in mind.

Traders who study these links for a living, in the tradition of chartists like JC Parets and inter-market relative strength, tend to use them as context for where strength is rotating, not as a mechanical trigger. That posture matches what the research keeps confirming. Cross-impact estimates are sensitive to data frequency, to how you define the linkage between the two instruments, to the liquidity of each, and to whatever common information you failed to control for. They are descriptive tools for reading market structure, not standalone forecasts, and not instructions to act.

Read the link, not the coincidence

Cross-impact is worth understanding precisely because it is so easy to over-read. A trader who watches an index future jump and then dumps a lagging constituent on the assumption that the move “must” transmit is trading a story, not a measured relationship. The disciplined use is quieter. Know which instruments are genuinely linked, know through which channel, and keep alive the possibility that a shared shock rather than a spillover produced the sequence you just watched.

Get that habit right and cross-impact turns into a lens for reading market structure instead of a shortcut into a trade. Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.