Volume Integrity Assessment
The decision: whether a venue's reported activity can be relied on — for due diligence on a listing or partnership, verifying a market maker's activity, an examination, or admitting the venue's data into indices and models.
The logic: reported volume is a claim. Real volume leaves an economic footprint: it moves price, and it comes from a heterogeneous crowd. Fabricated volume has to imitate both, and imitating the first one costs money, so the footprint is tested first.
Data to pull
For the pair under review, across every venue that trades it — the cross-venue comparison is the method, so a single-venue pull will not answer the question.
| # | Granularity | Fields | Purpose |
|---|---|---|---|
| 1 | 1m, 7 days | ohlcv, vwap, trades | Footprint and volume share |
| 2 | 1m, 7 days | buysell | Order-flow balance |
| 3 | 1m, 7 days | vdist | Trade-size uniformity |
Seven days survives a weekend and a couple of news events; a month is better where feasible.
Step 1 — Size up the claim
Per venue, over the window: quote volume (volume * vwap), share of total quote volume, true average trade size (volume / trades), and trades per active minute.
The objective at this stage is shape, not a verdict. A large volume share carried by very few trades raises the question of whether the implied trade size is plausible for the venue's clientele. A large share carried by millions of tiny prints raises the question of whether average size sits at the venue's minimum order size, where machine-generated flow tends to concentrate. Neither is evidence; both indicate where step 2 should be pointed.
Step 2 — Test the economic footprint
This is the step that does the work.
impact_per_1M = ln(high/low) * 10000 / (volume * vwap / 1000000)
impact_ratio = impact_per_1M(venue) / median(impact_per_1M of other venues)
Median the ratio over the window.
| Observation | Meaning |
|---|---|
| Ratio near 1 | Moves price about as much per dollar as its peers; the depth claim is consistent with the market. |
| Ratio far below 1 | Claims to absorb far more volume per unit of price movement than any peer. Either it genuinely is the deepest market for the pair, or many of its prints do not correspond to real order flow. |
| Ratio far above 1 | Genuinely thin. A market-quality finding, not an integrity one. |
Three sanity checks: exclude intervals where high equals low, since zero range carries no information; use a median rather than a mean, or one violent minute dominates; and require at least three peer venues for a stable peer median.
Step 3 — Corroborate
Two inexpensive checks that move confidence without originating a conclusion.
Order-flow balance. The standard deviation of buysell across the window's minutes, and the share of minutes within 0.02 of 0.50. Organic markets wander as buy pressure builds, exhausts and reverses; a market holding 0.50 minute after minute is being held there.
Trade-size uniformity. The entropy of the vdist bucket counts. Real flow mixes retail clips, algo slices and occasional blocks; a flat, even spread of sizes points at a single automated source.
Step 4 — Consolidate into one table
| Check | Venue A | Venue B | Venue C |
|---|---|---|---|
| Quote volume share | 8% | 41% | 51% |
| Trades per active minute | 190 | 24 | 610 |
| True avg trade size | 0.013 | 0.240 | 0.011 |
| Median impact ratio | 1.1 | 0.06 | 0.9 |
buysell dispersion | 0.089 | 0.005 | 0.081 |
| Size entropy | 0.08 | 0.34 | 0.07 |
| Verdict | consistent | inconsistent | consistent |
The venues and numbers above are illustrative, constructed to show the pattern.
What matters is not any single row. It is that Venue B claims 41% of the market's volume while moving price a sixteenth as much per dollar as its peers, holding its buy/sell balance almost perfectly steady, and trading at an almost invariant size. Each of those has an innocent explanation on its own; all three at once do not. The convergence is the finding.
Explanations to rule out before acting
The feed, not the venue. Aggregated-trade feeds merge same-price fills from one taker order into a single large print, inflating average size and distorting the size distribution — several of these signatures from an entirely honest venue.
A dominant market maker. One large provider quoting both sides tightly produces low buysell dispersion, uniform sizes and genuinely low impact. The discriminator is usually price discovery: a real market maker's venue still participates in setting price.
A compromised peer set. If several venues in the comparison group inflate volume the same way, the peer median absorbs it. Choose peers on structural grounds (supervision, institutional participation, independent corroboration), not because they scored well.
Too short a window. One day containing a macro event can make any venue look strange.