Market impact depends on venue, liquidity and execution method

Identify the execution venue before naming impact

A central limit order book exposes displayed price levels that can support a static calculation of an aggressive order walking available depth. Institutional FX is fragmented across venues and internalization, while a retail OTC or CFD provider can be the counterparty or price source. These structures do not provide one interchangeable market book.

A small client fill at a dealer does not by itself show that the order reached or moved a central market. That causal claim would require routing, hedging, and external market evidence. The correct record can report the provider quote and fill difference while marking the external path unobserved.

Venue identity includes provider, market model, instrument, session, order type, and access method. A broad market label such as FX does not resolve them. Capacity analysis must begin with this state because observable liquidity and the meaning of a displayed lot depend on where and how execution occurs.

A venue taxonomy should be versioned because routing and internalization policies can change without the public symbol changing. Historical orders remain tied to the access model then in force. A new classification starts a new capacity cohort unless equivalence is demonstrated from actual execution records.

Separate book walk, slippage, and causal market impact

Book-walk cost compares the VWAP across consumed displayed levels with the initial executable best quote. Slippage compares a decision or order benchmark with realized fill. Market impact asserts that the order changed the later market-price path. The first two can be calculated from appropriate records; the third needs causal evidence.

Calling every adverse fill market impact overstates what the data show. A quote can move because of unrelated order flow, latency, volatility, or dealer pricing. Conversely, a fill at the best quote does not prove zero causal effect after execution. The label should match the measurement rather than the desired sophistication.

The equation first aggregates all fills against the arrival benchmark, so an individually improved fill can offset an adverse fill within that net difference. If the aggregate result is favorable, the nonnegative cost is zero and the improvement should be stored separately. The benchmark timestamp and side must remain explicit for either interpretation.

A benchmark family can include arrival quote, decision price, and interval VWAP, but each yields a different descriptive cost. The report should not select the most favorable benchmark after the fill. Declaring the primary benchmark before execution limits hindsight and makes comparisons across orders reproducible.

Reconstruct VWAP from the displayed levels actually consumed

For the hypothetical 12-unit buy, five units fill at 100.00, four at 100.01, and three at 100.03. The total is USD 1,200.13 and VWAP is USD 100.010833. Quantity weighting is essential; a simple average of the three prices ignores that the levels contribute different volumes.

Twelve units at the arrival best ask of 100.00 would cost USD 1,200.00, so the static incremental difference is USD 0.13. The expression assigns side = +1 to a buy and floors adverse cost at zero. A sell requires side = −1 and an arrival best bid. The benchmark must be timestamped.

The book is assumed to remain static and every displayed unit to fill. That assumption excludes cancellations, hidden liquidity, queue position, latency, and dynamic response. The calculation is a transparent teaching snapshot, not evidence about a real venue or retail OTC execution.

A replay engine can test alternative static depths, yet those rows remain scenarios until sourced from timestamped books. It should retain the exact levels and quantities rather than only VWAP. That permits independent reconstruction and reveals whether a single shallow level drives the result.

Use the within-depth cases as genuine counterexamples

Four units fit entirely at the displayed best ask and have zero static book-walk difference. Five units use all displayed best-ask depth and also have zero under the assumed snapshot. These cases show that a larger order does not mechanically create positive walk cost until it crosses the available level.

Zero static difference does not mean zero fees, spread relative to another benchmark, or causal impact. It means only that the hypothetical aggressive buy fills at the initial best ask under the frozen displayed book. The conclusion should remain as narrow as the calculation.

Likewise, the 12-unit USD 0.13 difference should not be extrapolated linearly to every larger order. Deeper levels are unspecified, and the book can change. A capacity range requires repeated venue-specific observations across relevant order sizes and states, not one illustrative staircase.

The boundary beyond observed depth should be marked unavailable rather than extended at the last price increment. Linear continuation can understate a thin tail or overstate a replenishing book. Either error creates a false capacity curve whose precision comes from an assumption that was never recorded.

Collect the state variables needed for a capacity study

A useful record includes parent order, child fills, quantity, side, arrival benchmark, book snapshot, displayed and known hidden-liquidity limits, market volume, participation, duration, volatility, session, order urgency, venue, and execution method. Missing fields constrain which claim can be made.

Split orders must link to one parent decision. Treating each child as an independent small order can hide aggregate participation and understate implementation cost. Conversely, attributing every market move during a long execution to the parent can overstate causality without a benchmark model.

Data quality should preserve canceled and unfilled child orders where they inform completion risk. A fill-only dataset can make capacity look better by omitting orders that failed to execute. The analysis should state whether its objective is price, completion, time, or a combination.

Parent-level summaries should reconcile requested, filled, canceled, and still-live quantities. Price cost on the filled portion can look small when most of the order failed to complete. Reporting completion beside VWAP prevents that favorable-looking subset from being mistaken for full capacity.

Control side, benchmark time, and quantity reconciliation

The arrival best price must be executable for the order side at the stated time. A midpoint or stale last trade is a different benchmark. For a buy use the ask; for a sell use the bid. Side errors can turn adverse cost into apparent improvement or vice versa. This sign check precedes cost aggregation.

Total filled quantity in the benchmark equation must equal the sum of unique fill quantities. Partial execution can be analyzed for the filled portion, but completion shortfall should remain visible. Duplicate events or a submitted quantity used in place of fills can distort both VWAP and incremental cost.

Clock alignment is central. Book snapshot, decision time, order arrival, and fill times can differ by milliseconds or longer depending on the context. A benchmark chosen after the first fill is outcome-dependent. Preserve source clocks and latency assumptions rather than presenting one timestamp as exact without evidence.

Do not transport coefficients across markets without validation

Empirical institutional equity impact can depend on order size, trading rate, volatility, turnover, and market volume. A coefficient estimated on that population is not a default model for retail FX lots, another venue, or a different order-size range. Units and market structure change the meaning of every input.

A square-root or linear relationship can fit one dataset and fail elsewhere. Extrapolation beyond observed participation is particularly weak. The report should retain estimation period, sample, model form, parameter uncertainty, and validation performance before using a coefficient to define a capacity ceiling.

When evidence is insufficient, state that capacity has not been estimated. A universal lot threshold is less informative than an explicit data gap. The calculator can still report product loss sizing while keeping execution capacity as an independent unavailable constraint.

Treat OTC and internalized execution with bounded language

In retail OTC forex, the customer commonly trades against a dealer rather than one live central exchange. The provider can internalize, hedge, or route according to its model, much of which may not be visible to the client. A fill record therefore supports provider-benchmark analysis, not a claim about moving the entire currency market.

The absence of routing evidence does not prove zero external effect. It makes the causal path unobserved. The correct conclusion is epistemic: report what can be measured and avoid both exaggerated impact and unsupported zero-impact assurances. Unknown is a valid result here.

Provider quotes, rejected orders, requotes where applicable, execution time, and fill differences can still support a useful service-quality analysis. That analysis should not be relabeled central-book capacity. Its cohort and benchmarks need to match the provider contract and order method.

Define capacity as a monitored range with a decision criterion

Capacity requires a predeclared objective such as maximum VWAP difference, completion rate, participation, or time. Different objectives can produce different acceptable quantities. A single largest lot without its criterion is not reproducible and can be chosen after outcomes to make the execution look favorable.

The range should be conditioned on venue, instrument, session, volatility, and execution style. Step changes in liquidity can make yesterday’s estimate stale. Updating on a declared schedule and after structural changes preserves a versioned decision rather than reacting to every isolated fill.

A monitored range is not a guarantee. Hidden liquidity, cancellations, and market response can make realized execution better or worse. The system should retain uncertainty and an unavailable state rather than convert a sparse sample into a hard universal product limit.

Attack causal language with counterfactual questions

Ask what the market path would have been without the order. A single realized path cannot answer that counterfactual. Comparing pre- and post-trade prices alone confounds the order with all information and flow arriving during execution. Stronger designs need an explicit benchmark and assumptions.

An adversarial test labels the USD 0.13 static difference permanent impact; the review should reject the word permanent because the snapshot says nothing about later prices. Another applies the CLOB staircase to a dealer fill and should fail venue compatibility. A third uses midpoint instead of arrival ask for a buy.

Boundary tests include zero total filled quantity, negative fill quantity, missing side, stale benchmark, inconsistent total Q, and price improvement. VWAP is undefined at zero quantity, while the aggregate net cost floors at zero and an aggregate improvement is recorded separately. Invalid data should not be repaired by absolute value.

Separate capacity from stop-loss position sizing

Stop-based sizing limits modeled account loss for a price distance. Execution capacity estimates how order size affects price, completion, or both under a venue state. The smaller ceiling can bind, but one cannot be substituted for the other. A small stop-risk lot can still be difficult to execute in a thin market.

Costs inferred from capacity can enter per-unit loss or a rounded all-in replay, but causal impact should not be invented when only book walk is observed. The account-currency budget, product value, and execution estimate need compatible units and timestamps before they are combined.

Margin and daily risk are further independent constraints. An order can pass venue capacity and fail account funding or aggregate exposure. Showing all ceilings with binding labels prevents an execution study from being interpreted as permission to use the largest quantity the book appears to show.

Build an execution audit trail that preserves the snapshot

The record needs venue model, provider, instrument, parent and child IDs, side, quantity, order instruction, arrival benchmark and time, book levels with timestamps, fills, cancellations, fees, VWAP, nonnegative book-walk cost, improvement, and completion status.

Model versions should retain cohort filters, objective, parameters, uncertainty, and validation period. Earlier orders remain linked to the version available at decision time. Replacing the predicted cost with realized fills would erase the evidence needed to assess whether the capacity model worked.

Corrections are appended as lifecycle events. A busted fill, revised timestamp, or duplicate should trigger recalculation with both versions traceable. The audit should distinguish observed book data from inferred hidden liquidity and causal claims, so a derived field never acquires stronger provenance than its inputs.

Interpret market-structure sources without overreach

BIS material supports fragmentation and internalization in FX execution. CFTC material supports the dealer-counterparty nature of retail OTC forex. Kyle provides a market-microstructure model linking order flow, depth, and price response, while Almgren and coauthors show empirical equity impact depends on market and order variables.

Those sources do not provide the hypothetical 100.00 book, 12-unit order, or USD 0.13 cost, and they do not establish a capacity threshold for an SG Group user. Institutional model coefficients are context-dependent. Applying their qualitative lessons to retail sizing is an article-level inference.

The evidence supports distinguishing venues and claims, not asserting that every order has measurable permanent impact. Observable VWAP and benchmark differences can be reported precisely within their data. External routing and causal market movement remain unknown unless separately evidenced.

State the static result without calling it universal impact

In the frozen hypothetical CLOB, four and five units fill at the best ask with zero static book-walk cost, while 12 units consume three levels and add USD 0.13 versus the arrival ask. VWAP is USD 100.010833 for the largest row. These are arithmetic properties of the stated snapshot.

The result does not prove permanent price impact, predict a real fill, or transfer to OTC execution. Hidden liquidity, order-book changes, fees, latency, and incomplete fills are outside the example. A real capacity estimate needs venue-specific observations and an objective defined before the records are inspected.

The decision consequence is disciplined language and data collection. Report provider fill difference when that is all that is observable; report static walk when a valid book supports it; reserve causal impact for evidence capable of answering that claim. No universal lot follows from one display.