Blog
Blog
Visual temporarily unavailable
This section's visual is under review and a replacement is being prepared.
Microstructure measurements are not verdicts. Treating them as trading calls is how useful data becomes another expensive superstition.
Order Flow Imbalance, VPIN, depth, spread, and price impact all describe parts of the live auction. They answer better questions than a candle can answer. But none of them says "buy," "sell," or "hold." That language belongs to advice, not measurement.
The disciplined use of microstructure is risk-aware interpretation: what does this measurement describe, where does it fail, and what uncertainty remains after reading it?
Every measurement should answer one question.
OFI asks which side is more aggressive right now. VPIN asks whether recent flow looks toxic or unusually imbalanced. Depth asks how much resting liquidity is available near price. Spread asks how expensive immediacy has become. Kyle's Lambda asks how much price moves per unit of signed flow.
These are different questions. Combining them without naming the question creates a fake sense of confirmation. Positive OFI and high VPIN do not say the same thing. Positive OFI is directional pressure. High VPIN is flow toxicity. A thin book is fragility. A widening spread is liquidity provider caution.
Good interpretation starts by refusing to blur those meanings. If a metric cannot be stated in one plain sentence, the trader is not ready to use it.
OFI is the easiest metric to overclaim.
Positive OFI means buyers have been more aggressive than sellers over the measured window. It does not mean price must rise. Sellers may absorb the pressure. A larger passive seller may be using aggressive buyers as exit liquidity. Other venues may show the opposite flow. A news event may overwhelm the orderbook entirely.
The useful reading is conditional. Positive OFI says the current auction contains buying pressure. If depth on the ask is thinning and other venues confirm, the pressure is more meaningful. If the ask keeps replenishing and cross-venue flow is neutral, the pressure may be getting absorbed.
That is the difference between a reading and a rule. Order Flow Imbalance raises a question about pressure. It does not answer the whole market setup.
VPIN is often misread as a directional tool. That is the wrong category.
VPIN estimates whether recent volume is unusually imbalanced in a way associated with informed or toxic flow. It is about adverse selection risk. It warns that one side of the market may know more, or that liquidity providers may face a worse selection problem than usual.
High VPIN does not tell you whether the next move is up or down by itself. It tells you the market may be less safe to provide liquidity into, and that volatility risk may be higher. Direction needs another input: order flow, price structure, catalyst context, or cross-venue confirmation.
This limitation makes VPIN more useful, not less. A metric that clearly says "market-structure uncertainty is elevated, direction unresolved" is better than a metric marketed as certainty. VPIN explained belongs in market-structure context before it belongs in any action-oriented workflow.
The same reading means different things in different books.
A negative OFI reading in a deep, stable book may produce little price movement because buyers absorb the selling. The same reading in a thin book can move price sharply. A breakout on tight spread and replenishing depth has a different risk profile from a breakout on widening spread and vanishing liquidity.
This is where price impact matters. Kyle's Lambda frames the question cleanly: how much does the market move for a unit of flow? If Lambda is elevated, every directional reading deserves more caution because less flow is needed to displace price.
Risk-aware interpretation treats liquidity as a multiplier. It changes how much a reading matters, how fragile the market state may be, and how quickly the context can change.
A reading on one venue is not the market.
Fragmented markets make local readings dangerous. A strong imbalance on one exchange can be genuine local pressure, the start of broader repricing, or noise that arbitrage will erase. Without cross-venue context, the trader cannot tell which case they are looking at.
This does not mean every trade requires perfect multi-venue data. It means single-venue conclusions need single-venue language. "This exchange shows buying pressure" is precise. "The market is being bought" is a larger claim and needs broader evidence.
Exchange fragmentation matters because it forces humility into measurement reading. The market may be larger than the feed.
No interpretation survives bad inputs.
If an orderbook feed dropped updates, depth is unreliable. If timestamps are misaligned, lead-lag analysis is unreliable. If volume is dirty, volume-derived measurements are unreliable. A risk-aware system must be able to say "no usable reading" when data quality fails.
This is not operational neatness. It is data-integrity logic. A stale value is not a weak reading. It is not a usable measurement. A reconstructed historical value is not the same as the value available in live time. A backtest that uses cleaned data without modelling live feed failures gives the analysis information the live reader would not have had.
Data quality belongs beside market-structure context because it determines whether the measurement is real enough to interpret.
The safest way to avoid overclaiming is to rank claims by strength.
Weak claim: buyers were more aggressive on this venue over this window. Stronger claim: buyers were more aggressive across several venues while ask depth thinned. Stronger still: the same condition has historically preceded upward short-horizon price movement in a held-out sample, after fees and realistic execution assumptions.
Most trading commentary jumps from the first claim to the last. That is where the damage starts. Each step requires evidence. Without that evidence, the stronger claim is just a story attached to a number.
This ladder is useful because it keeps language honest. "Pressure is positive" is not the same as "future return is favorable." "Liquidity is thin" is not the same as "price will break." "Toxic flow is elevated" is not the same as "direction is known."
DepthSignal is a data layer, not a promise machine.
The product value is that traders and teams can read microstructure measurements as market context rather than trying to force a chart-only view to answer every question. It does not remove market uncertainty.
At DepthSignal, we treat microstructure features as context that improves the quality of a market-structure question. The correct use is not "the reading says act." The correct use is "the market state makes this setup more or less fragile than the chart alone suggests."
That distinction protects the user and the product. Measurements deserve precise language. Anything stronger becomes marketing fiction.
Risk-aware interpretation means keeping every reading in its proper size.
OFI measures pressure. VPIN measures toxic flow context. Depth measures available resistance. Spread measures immediacy cost. Lambda measures impact. Together, they describe the auction more honestly than a price chart alone. They still do not guarantee the next market move.
The trader who overclaims microstructure turns better data into a worse belief. The trader who reads it precisely gets something more useful than certainty: a clearer map of where uncertainty actually sits.
DepthSignal is a market-data platform, not a source of financial advice, trading signals, or investment recommendations.
It means keeping the claim proportional to the evidence. Pressure is not the same as future return. Thin depth is not the same as confirmed breakout.
Because the jump from a narrow measurement to a broad forecast usually happens without new evidence. That is where a context tool turns into a false promise machine.
As context that improves the question, not as certainty that removes doubt.
It is the discipline of moving from smaller claims to larger ones only when the evidence expands. Venue-local pressure is a smaller claim than market-wide direction. A measured reading is smaller than a forecast.
It means the data, venue scope, or market state is too compromised to justify a clean interpretation. Refusing to overstate a weak reading is part of the product discipline.