Order Flow Persistence: Reading Runs of Signed Flow

Watch a fast tape long enough and you’ll see it: a stream of prints that keep lifting the offer, one after another, twenty or thirty in a row, and not one of them large. Nothing on the chart has broken. The spread has barely moved. Yet the flow keeps leaning the same way. That lopsided run is what traders mean by order flow persistence, and reading it well starts with knowing what it is and, just as important, what it can’t tell you.

Persistence describes serial dependence in the direction of trades. Buying or selling pressure can stay one-sided across many transactions even when each individual trade is small. It’s a property of the sequence of flow, not a forecast that price has to keep moving. I keep those two ideas in separate boxes when I read a tape, because collapsing them is the quickest way to talk yourself into a position the data never supported.

What order flow persistence actually measures

Start with signing. Every trade prints against one side of the market: it either pays the offer or hits the bid. Sign the offer-side trades positive and the bid-side trades negative and you’ve got signed flow. Persistence is the tendency of that sign to repeat. A long run of positive signs means buyers kept initiating, print after print, regardless of how far price actually traveled.

The interval you choose decides what you’re even looking at. When I mark up a tape, I bucket flow into fixed one-minute windows and sign each print by whether it paid the offer or hit the bid. Shrink the window to ten seconds and a burst of aggressive buying looks like a wall of persistence. Stretch it to fifteen minutes and the same burst blends into the noise around it. Neither view is wrong. They answer different questions, and a persistence number quoted without its interval is close to meaningless.

There’s a second measurement trap that catches people constantly. Trade count and signed volume aren’t the same series, and they can point in opposite directions. I keep them in two separate columns for exactly that reason. Forty small buys can run consecutively on the tape while a single large sell quietly swamps them in size. Count prints and you see persistent buying. Weight by volume and you see net selling. The order-flow-imbalance measure already gives TaB readers a way to net aggressor volume over a window. Persistence is the companion idea: it describes how the sign behaves through the sequence, before you ever net anything.

Why it isn’t the same as momentum

Price momentum is a statement about returns. Return autocorrelation is a statement about how one bar’s return relates to the next. Persistence is a statement about flow, and flow and price can decouple for long stretches. A deep, resilient book can absorb a run of same-direction trades with almost no net move, because fresh limit orders keep refilling the level the aggressors are eating. You get persistent flow and a flat chart at the same time.

This is the first misread to guard against. A run of same-direction trades tells you one thing: demand was one-sided over your chosen window. The reach of that observation ends there. The move, if it comes, depends on how much resting liquidity stood in the way, which is a separate question from how the flow was signed. Treating persistence as a price prediction skips that entire middle step, and the middle step is where most of the outcome lives. I’ve watched long buy runs go nowhere against a heavy offer, and short, quiet runs move price hard when the book was thin. The flow looked similar both times. The liquidity behind it did not.

Core orders and reaction flow

A 2026 order-flow study draws a distinction that sharpens all of this: core orders versus reaction flow. Core flow is the initiating trading demand, the decision that sets something in motion. Reaction flow is the activity that responds to it: market makers adjusting quotes, other participants chasing or fading, algorithms firing off follow-on child orders. Both streams generate signed trades, and on a raw tape they look identical. You can’t see intent in a print.

The study models both streams with Hawkes processes. A Hawkes process is a self-exciting point process, which is a formal way of saying each event lifts the near-term probability of more events. One aggressive buy makes the next aggressive buy a little more likely for a short while, and the effect decays as time passes. Let the core stream excite the reaction stream and the reaction stream feed back, and you get clustering that lingers. That interaction is the mechanism the model uses to explain persistent signed flow, the ragged low-regularity paths of volume and volatility it calls rough, and market impact that scales as a power of size rather than in a straight line. The point for a reader is narrower than the mathematics. A single decision can echo through the book as many small prints, and the echo stays measurable even when the original decision is invisible.

How one decision gets sliced over time

Here’s the everyday reason persistence shows up at all. A participant with a large decision rarely fires it as one order. Doing so would walk straight up the limit order book, clearing level after level and paying an ugly average price. So the decision gets divided into a schedule of smaller child orders released over minutes or hours. Each child is modest. The sequence they form is persistent by construction.

Now the reaction begins. Displayed liquidity at the touch gets consumed, and how fast it refills sets the tone. If makers requote quickly and the bid-ask spread stays tight, the schedule can run for a long time with little drama. If liquidity thins and the spread gaps, the same schedule starts to move price and draws in momentum traders who pile on their own same-direction prints. That feedback is why execution desks lean on schedules anchored to a volume benchmark such as VWAP. The goal is to hide a persistent parent order inside the market’s own rhythm so it doesn’t stand out as one lopsided run. When it’s done well, the persistence you’d measure on the tape is deliberately smeared thin.

What a run of same-direction trades does not reveal

This is the limitation to state plainly, because it’s the one that gets glossed over. Public trade and quote data don’t identify the investor behind a print, the parent order it belongs to, or the reason it fired. You see size, price, side, and time. You don’t see motive. A run of thirty offer-lifting trades is equally consistent with one institution working a patient buy schedule, a dozen retail orders arriving by coincidence, or a market maker hedging a position it took on somewhere else entirely.

Reading the tape for intent has a long history, and the honest practitioners always respected its ceiling. The lessons collected from Jesse Livermore describe reading order flow as gauging pressure and conviction, never as decoding a specific actor’s plan. A model that reproduces clustering does the same thing at higher resolution. It tells you the flow was self-exciting. It doesn’t prove that a particular source of demand drove any single episode. Persistence is a description of behavior, and a description is not a confession.

A reading checklist for the tape

When I sit down with signed flow, I run the same short list before I let a run mean anything. It keeps the observation honest and stops a clustering statistic from turning into a trade instruction on its own.

• Name the interval first. State the window in seconds or minutes and hold it fixed across the comparison, because persistence measured at two different intervals is two different measurements.
• Separate trade count from signed volume. Read both columns and note when they disagree, because that gap is often the whole story. Tools like tick imbalance bars formalize the count-versus-size split if you want it built into the chart.
• Check the clock against the schedule. Before I read anything into a run, I look at whether it sits on top of an economic release, an open, a close, or an index rebalance, because event-driven flow clusters for reasons that have nothing to do with a hidden buyer.
• Refuse the standalone signal. A persistence number is context for a decision, never the decision itself. Pair it with the book, the spread behavior, and the level structure before it earns any weight.

None of these steps needs a special data feed. They need discipline, which is the part that actually gets skipped.

Read the flow, respect the fog

Order flow persistence is one of the more useful lenses on a tape once you strip the false promises off it. It measures how one-sided the sequence of trades was over a window you define. It doesn’t tell you where price must go, who was behind the prints, or why they traded. The 2026 core-and-reaction framing is a clean way to hold both halves of that at once: the clustering is real and measurable, and the motive stays behind the glass. Read the flow for what it says about pressure, then let the book, the spread, and the calendar decide how much that pressure is worth. Learn the pattern. Ride the trend. Keep the gains.

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