By TickDistill — order-flow microstructure signals. Educational content, not financial advice.
Microstructure signals — VPIN, order-flow imbalance, big-order pressure — decay as measures of edge, but never as measures of state. A VPIN reading is always current; a strategy built on a naive VPIN threshold from two years ago may already be crowded out. TickDistill sells the measurement — clean, computed, point-in-time-correct order-flow state — not a strategy that depends on the signal staying unexploited. That distinction is the product’s entire foundation.
The shovels-not-gold principle is the observation that selling the instrument of discovery is more durable than selling the discovery itself. During the California Gold Rush, hardware merchants selling picks and shovels earned steadier, longer-lived returns than most gold miners. The shovel seller’s revenue did not depend on any single miner finding gold.
In financial data, the equivalent is selling computed inputs and measured state rather than pre-packaged directional calls. Bloomberg terminals sell price feeds and analytics; they do not tell you whether to buy or sell. TickDistill occupies the same structural position for order-flow microstructure: the measurement layer, not the alpha layer.
A microstructure signal decays as alpha — as a source of excess return — when enough participants discover and trade on it. The academic literature documents this clearly.
Crowding compresses signal alpha over time. McLean & Pontiff (2016), “Does Academic Research Destroy Stock Return Predictability?” (Journal of Finance), found that return predictability declines by roughly 58% post-publication as capital floods the documented anomaly. The mechanism applies to any systematically exploitable pattern: once widely known, arbitrage closes the gap.
Microstructure signals face particularly fast decay. Order-flow patterns involve information that market makers and algorithmic participants actively counteract. Informed-flow detection (the mechanism underlying VPIN, formalized by Easley, López de Prado & O’Hara, 2012, in “Flow Toxicity and Liquidity in a High-Frequency World,” Review of Financial Studies) is itself a target of adversarial adaptation: when enough market makers use a signal to detect toxicity, the behavior of informed traders adjusts.
The practical consequence: a naive “trade when VPIN crosses 0.7” rule documented in 2012 faces a fundamentally different competitive landscape today. But the VPIN reading — the measured probability that recent order-flow is informed — is just as informative as ever. The measurement did not decay. The naive strategy built on it did.
| Layer | What it is | Decays? |
|---|---|---|
| Raw market data (prices, sizes) | Quotes and trades as reported | Never (factual record) |
| Computed state (VPIN, OFI, CVD, imbalance) | A derived measurement of current order-flow conditions | No — always current |
| Simple threshold rule (“buy when X > 0.7”) | A fixed strategy using the signal | Yes — crowded, adapted against |
| Composite multi-signal strategy with regime filter | A more complex but still public-ish strategy | Yes, but slower |
| Truly proprietary combination + execution | Full systematic strategy with edge from execution and selection | Decays slowest — but this is not what we sell |
The computed-state layer — the measurement — sits in the stable middle. It does not promise a direction; it describes a condition. Conditions are facts. Facts do not expire.
The confusion is semantic but consequential. In common quant usage, “signal” means an input to a model. In marketing — and in the claims of many data vendors — “signal” slides toward meaning “edge” or “alpha.” That conflation is the whitespace TickDistill occupies.
A VPIN is a signal in the technical sense: it is a measured scalar that characterizes current order-flow toxicity. The Easley-López de Prado-O’Hara paper defines VPIN as:
VPIN = (1/n) * Σ |V_b^τ - V_s^τ| / V^τ
where V_b^τ and V_s^τ are the buy- and sell-initiated volumes in bucket τ, and n is the number of buckets in the estimation window. Because VPIN uses fixed equal-volume bucketing, V^τ is the same constant volume V for every bucket by construction, so this reduces to the canonical Σ |V_b − V_s| / (n · V) (50 volume buckets per day and a 50-bucket averaging window are the public ELO/PINstimation defaults). This quantity is a measurement of imbalance. It is not a prediction of where price goes next.
Similarly, Order Flow Imbalance as defined by Cont, Kukanov & Stoikov (2014, “The Price Impact of Order Book Events,” Journal of Financial Econometrics 12(1), 47–88) nets bid-side pressure against ask-side pressure over a window:
OFI_t = Σ ( OF^b_n − OF^a_n )
where each book transition n contributes a three-case order-flow term — when the bid price rises the full new bid size adds to OF^b_n, when the price is unchanged the size delta q^b_n − q^b_{n-1} applies, and when the bid price falls the new size enters negative (with the symmetric construction on the ask side). Netting bid minus ask makes positive OFI = buy-side pressure, matching the ofi catalog contract below. OFI is a state variable, not a forecast. Cont, Cucuringu & Zhang (2023, “Cross-impact of order flow imbalance in equity markets,” Quantitative Finance 23(10)) extend OFI to multi-level and cross-asset settings — the same measurement principle, wider scope. See What Is Order-Flow Imbalance (OFI), and Why Does It Need the Order Book? for the full three-case construction.
Vendors who call these quantities “alpha signals” are borrowing the authority of the literature while quietly implying durability the literature does not promise.
TickDistill computes each signal as a causal z-score of the underlying microstructure quantity: the raw measurement is normalized against its own rolling historical distribution, using only past data, so the output expresses rarity — how unusual this condition is right now for this market.
Three properties follow from this design:
1. Non-directional output. A high VPIN z-score says informed order-flow is elevated. It does not say whether informed traders are long or short. A trader who uses this to size their risk, filter their entries, or monitor for regime stress is using a measurement. A trader who blindly buys every high-VPIN event is building a strategy — and that strategy’s decay is their responsibility, not ours.
2. Always current, never stale. The measurement updates with each new data window. The “alpha” question — does it work? — is separate and belongs to the strategy layer. We answer the measurement question; the client answers the strategy question.
3. Point-in-time correctness as a hard rule. Every baseline TickDistill uses is computed from past data only, with mechanical windows excluded (funding settlements, auction periods) that would otherwise distort the local distribution. A backtest built on a look-ahead-contaminated baseline is not measuring past signal quality — it is measuring future knowledge. We treat that as a disqualifying defect, not a calibration choice. See Sigma-Normalization: Why Order-Flow Signals Should Be Measured in Standard Deviations, Not Raw Numbers for the full treatment.
The concept of VPIN, OFI, and order-flow imbalance is public. The public formula is the starting point, not the product. What TickDistill adds — and does not publish — is the calibration: the per-market profile that sets the normalization window, the floor threshold below which events carry insufficient statistical weight, the exclusion calendar that strips mechanical noise, and the z-score construction that makes the output comparable across regimes.
This calibration is proprietary because it is the labor-intensive, error-prone work that determines whether a signal is trustworthy. A miscalibrated z-score that drifts during high-volatility regimes, or that inflates during funding windows, is worse than no signal — it is a confident wrong measurement. The moat is not the formula; the moat is getting the calibration right and verifiable.
Every package in the TickDistill catalog follows the same non-directional contract:
vpin measures order-flow toxicity (per Easley-López de Prado-O’Hara 2012). High z-score = elevated informed flow. No directional claim.ofi measures net order-book pressure (per Cont-Kukanov-Stoikov 2014). Positive = buy-side pressure. Negative = sell-side pressure. A fact, not a forecast.cvd (Cumulative Volume Delta) accumulates net taker-initiated volume. It measures what buyers and sellers are doing, not what price will do next.big_order_base detects prints above a market-specific sigma floor. It identifies that something large happened. Whether it was informed or mechanical, whether it will move price, is context the client brings.density_separated_big_orders classifies co-directional big-order clusters. It measures conviction zone structure. Interpreting that structure is the trader’s job.Regime context — for example, what kind of market day is likely given recent volatility structure — is the subject of Why Market Days Cluster: Volatility Clustering, Regime Persistence, and What “Bull Follows Bull” Really Means. That piece makes the same honesty argument: TickDistill sells the regime state; the directional call is the client’s.
Bloomberg LP does not tell you whether to buy IBM. It sells you the price, the fundamentals, the news, and the analytics — cleaned, normalized, and delivered. You build the strategy. Bloomberg’s revenue is durable because the measurement service is always needed, regardless of whether any specific strategy works.
TickDistill is the Bloomberg of order-flow microstructure inputs for the crypto and futures markets that Bloomberg does not cover at this depth. The data product is the derived state (z-scored, point-in-time, masked, normalized) — delivered via REST API, bulk Parquet download, MCP for AI-native agents (see Why MCP-First? Agent-Native Distribution for Order-Flow Signals), and an on-chart visualization that keeps the same measurement-not-advice line (see How Do We Render Order-Flow Signals on a Chart?). The strategy is yours.
Selling derived, non-reconstructible state rather than raw data is not only the correct product philosophy — it is the correct legal structure. Raw redistribution of exchange market data (prices, sizes, timestamps in reconstructible form) is governed by exchange licensing agreements. A derived signal — a z-scored imbalance, a classified regime flag, a normalized big-order magnitude — is a TickDistill work product, not a redistribution of the underlying feed.
This legal discipline and the product philosophy are the same choice, arrived at from two directions. See Why Derived, Not Raw Data? The Legal and Architectural Case for the full legal and engineering argument.
Q: If you publish the VPIN formula, isn’t the signal compromised? No. The public formula is the starting point, not the product. The formula is in the Easley-López de Prado-O’Hara (2012) paper, freely available. What is proprietary is the calibration — the normalization windows, the per-market floor, the exclusion calendar, and the z-score construction — which determines whether the output is trustworthy. Formula + wrong calibration = bad signal. Formula + correct calibration = the product.
Q: How is a VPIN “always current” if microstructure signals decay? Decay means that a naive fixed-threshold strategy built on VPIN may stop working as competitors adapt. It does not mean the VPIN reading becomes inaccurate. Today’s VPIN reflects today’s order-flow — that information is always fresh. What you do with it (the strategy) is what can become crowded.
Q: Does TickDistill offer any directional signals at all? No. Every signal expresses a measured state: toxicity level, imbalance, conviction zone, regime. Directional signals — “go long” — are strategies, not measurements. Strategy construction is the client’s domain and is explicitly out of scope.
Q: What is the difference between a signal and alpha? A signal is a measured input. Alpha is the excess return generated by a strategy that uses that input. Signals are facts; alpha is a strategy property that decays as it is discovered and traded. TickDistill sells signals. Whether they generate alpha depends on how you combine them, your execution quality, and whether the combination is already crowded — none of which we control or claim.
Q: Why should I trust a measurement that does not tell me what to do with it? Because a measurement that over-promises is useless — it fails when it matters most. An honest measurement of order-flow state lets you build, test, and adapt your own strategy. It tells you what is happening in the market right now, correctly and reproducibly. That is the permanent value: accurate, point-in-time, non-directional market state. The strategy is where the creativity lives — and that is yours to keep.
TickDistill sells clean, computed order-flow inputs — not trading advice or guaranteed alpha. Backtests are illustrative and not a promise of future results.