Deutsch
日本語
العربية
한국어
Español
Português
Indonesia
Tiếng Việt
中文
Single-broker triangular arbitrage no longer produces consistent profit — brokers equalize cross-rate quotes in real time. The opportunity has migrated to multi-broker execution, where genuine cross-rate discrepancies exist between independent liquidity providers. This migration introduces latency-contaminated signals, multiplied commission exposure, and new toxic flow classification risk. This guide covers the full technical architecture.
Triangular arbitrage exploits temporary inconsistencies between three currency pairs where the implied cross rate — derived from two direct quotes — differs from the actual quoted cross rate. Starting with currency A, converting to B, then C, then back to A should return exactly the starting amount in an efficient market. When it returns more, the discrepancy is the arbitrage profit.
Given three pairs EUR/USD, GBP/USD, and EUR/GBP, the no-arbitrage condition requires:
Example:
EUR/USD = 1.08500 GBP/USD = 1.27000
EUR/GBP implied = 1.08500 ÷ 1.27000 = 0.85433
If EUR/GBP quoted = 0.85200 (23 pips below implied):
Step 1: Sell EUR → buy USD at 1.08500 (1,000 EUR → 1,085 USD)
Step 2: Buy GBP with USD at 1.27000 (1,085 USD → 854.33 GBP)
Step 3: Sell GBP → buy EUR at 0.85200 (854.33 GBP → 1,002.74 EUR)
Net: 1,000 EUR → 1,002.74 EUR = +2.74 EUR before costs
The cycle must execute simultaneously or near-simultaneously — any delay between legs introduces directional exposure. The profit window exists only while the mispricing persists: typically milliseconds to seconds depending on market liquidity and the source of the discrepancy.
Modern retail brokers enforce cross-rate consistency in real time through automated cross-rate synthesis. When EUR/USD or GBP/USD updates, EUR/GBP is recalculated and re-quoted simultaneously — within microseconds. Any triangular mispricing on a single broker’s book is corrected by the pricing engine faster than any retail-level execution system can detect and act on it.
The broker’s pricing engine derives cross-rate quotes mathematically from direct pairs in real time. When EUR/USD updates, EUR/GBP is recalculated and re-quoted simultaneously — within microseconds. Implied and quoted cross rates are always aligned.
On single-account ECN/STP brokers, all three pairs source from the same underlying liquidity pool. A price event that creates a temporary EUR/GBP mispricing simultaneously updates EUR/USD and GBP/USD. By the time the update propagates, all three quotes are consistent again.
In the rare cases where a genuine discrepancy does appear on a single broker’s book, institutional participants with co-located sub-100 microsecond systems close it before any retail-level software can respond. Retail VPS-to-broker latency of 1–20ms has no chance against this.
Backtests of single-broker triangular arbitrage consistently show apparent profit opportunities. Live execution shows near-zero profitable trades. The gap: historical tick data captures pricing states that existed briefly; the execution engine sees the corrected price before the fill is processed.
Single-broker triangular arbitrage is not a viable strategy in 2026. Profitable triangular arbitrage requires genuine pricing discrepancies between independent liquidity sources — which means multi-broker execution.
| Role | Provides | Function |
|---|---|---|
| Broker A | EUR/USD + GBP/USD | Software monitors both direct pairs and computes the implied EUR/GBP in real time. |
| Broker B | EUR/GBP (cross rate) | Provides the actual cross-rate quote. When the gap between implied and quoted exceeds the threshold (net of all costs), the trade fires. |
| Execution | Simultaneous 3-leg | EUR/USD and GBP/USD open on Broker A; EUR/GBP opens on Broker B in the opposing direction. All three legs must fill — a partial fill creates directional exposure. |
The three-broker setup distributes one pair per broker, maximising inter-broker mispricing probability. But it also maximises partial-fill risk — if one of three legs fails to fill within the timeout, the other two create an unhedged directional position that must be explicitly closed.
| Broker | Pair | Execution requirement |
|---|---|---|
| Broker A | EUR/USD | Fast FIX API preferred — first leg |
| Broker B | GBP/USD | Fast FIX API preferred — second leg |
| Broker C | EUR/GBP | Implied rate from A and B compared to C’s quoted rate in real time |
Moving from single-broker to multi-broker multiplies the cost structure. The minimum profitable discrepancy is not fixed — it must be computed dynamically from live spread and commission inputs:
Where:
S(x) = half-spread at broker x (pips)
C(x) = commission per lot at broker x (account currency)
PipValue = pip value for EUR/GBP at configured lot size
SlippageBuffer = conservative slippage estimate per leg
Example (2-broker, 1 standard lot, EUR/GBP pip = $10.78):
S(A)=0.15, S(B)=0.25, C(A)=$3, C(B)=$4, Slippage=0.2 pips
MinDiscrepancy = 0.15 + 0.25 + [(3+4)/10.78] + 0.2
= 0.40 + 0.65 + 0.20 = 1.25 pips minimum to break even
Any detected discrepancy below this dynamically computed threshold must be discarded. SharpTrader’s triangular arbitrage module computes this threshold at each tick event from live spread and commission inputs — not from a fixed value set at configuration time.
In a multi-broker setup, a detected triangular discrepancy can arise from one of two fundamentally different causes: (1) a genuine cross-rate mispricing between independent LPs, or (2) a temporary apparent mispricing created solely because one feed updated and the others have not yet. These two causes are indistinguishable at the moment of signal generation — but they produce completely different outcomes on execution.
Consider the following 200ms sequence:
| Time | Event | Apparent discrepancy |
|---|---|---|
| T = 0ms | EUR/USD moves 3 pips on all LPs. Broker A’s feed updates immediately (fast connection). | Appears: 2.8 pip triangular discrepancy |
| T = 0–80ms | Broker A shows new EUR/USD. Broker C’s EUR/GBP has not yet updated. System detects “discrepancy”. | Discrepancy exists only in feed timing |
| T = 80ms | Broker C’s EUR/GBP feed receives the same event and updates. Discrepancy disappears. | Collapses to zero |
| T = 100ms | If the system fired at T=0ms, orders now arrive at Broker C — which has already updated its quote. | Trade fills at corrected (unfavourable) price — loss |
A system executing on latency-contaminated signals produces a specific order flow pattern that brokers classify as toxic: rapid three-way position opening concentrated at price feed update events, with high rates of immediate closure after unfavourable fills. This is the signature of a system exploiting infrastructure rather than trading on genuine market structure. Brokers identify this pattern and restrict the account — degrading even the profitable genuine-discrepancy trades.
Filtering latency-contaminated triangular signals requires combining four independent tests applied in sequence. No single test is sufficient.
(1) MaxFeedTimestampDelta < SynchronisationThreshold
— All three feeds have updated within ~60ms of each other
— No single feed is running ahead of the others
(2) Discrepancy.Duration >= PersistenceWindow (e.g. 80ms)
— Genuine LP discrepancies persist; latency artifacts collapse in 30–100ms
— Applied before the more expensive checks below
(3) abs(FastFeed.RateOfChange, 100ms) < DirectionalEventThreshold
— No significant price movement in the preceding 100ms window
— A directional event is the most common cause of feed latency artifacts
(4) CurrentDiscrepancy > LiveMinThreshold × (1 + SafetyMargin)
— Computed dynamically from live spreads + commissions + slippage
— Safety margin (typically 20–30%) covers execution slippage variability
Failure of ANY condition → signal discarded, no order placed.
If feeds are not approximately synchronised at signal detection time, one feed has likely updated on a market event before the others. Typical threshold: 60ms between the most recent update on any feed.
Latency artifacts collapse within 30–100ms. Genuine LP discrepancies persist for 200ms to several seconds. Require the discrepancy to survive a minimum window before executing — fast-moving artifacts self-filter.
If any direct pair (EUR/USD or GBP/USD) has moved more than the configured threshold in the preceding 100ms, this is a strong indicator that the discrepancy is a latency artifact from a recent market event propagating through feeds at different speeds.
The minimum profitable discrepancy is recomputed at every tick from live spread and commission inputs. During news events when spreads widen 3–5x, signals that would be profitable in normal conditions automatically fail this condition and are discarded.
In well-calibrated deployments, this four-condition filter stack reduces latency-induced signal contamination from the typical 60–80% of unfiltered signals to under 15%.
Multi-broker triangular arbitrage produces characteristic order flow patterns that broker risk systems identify regardless of individual trade profitability:
| Pattern | What brokers see | Mitigation |
|---|---|---|
| Correlated 3-pair opening | Three positions in correlated instruments open within milliseconds. Flagged as technical pattern — not consistent with discretionary trading. | Parallel execution + lot size variation (±10–20%) |
| Short consistent hold times | Positions closed within seconds to minutes, consistently. Narrow hold-time distribution is a primary statistical flag. | Configured minimum hold time (30–120 seconds) |
| Win rate vs spread correlation | Win rate drops exactly when spreads widen. Inverse correlation between P&L and spread width is a technical flow signature. | Dynamic threshold filter eliminates low-margin signals during wide-spread periods |
| Cross-account correlation | If the same entity holds accounts at multiple brokers, negatively correlated P&L at signal events is detectable if brokers share risk data. | Diversify across multiple triangle combinations and sessions |
Target under 5ms round-trip to each broker from the VPS. Asymmetric latency between the VPS and each broker creates leg-level timing inconsistency — the “simultaneous” execution becomes sequential, increasing partial-fill risk.
Each broker’s price feed must run on an independent network socket with hardware-timestamped tick receipt at the socket level — not the processing level. Shared processing queues introduce artificial timestamp correlation that distorts the synchronisation filter.
All three legs must be submitted simultaneously via parallel execution threads. Sequential order placement introduces leg-level timing gaps that create detectable patterns in order logs and increase partial-fill risk on the last leg.
All feed timestamp comparisons require clock synchronisation to a reliable time source. A 10ms clock drift can cause the synchronisation filter to misclassify signals. Use PTP (Precision Time Protocol) if available, NTP with under 5ms jitter otherwise.
No. Retail brokers’ pricing engines enforce cross-rate consistency in real time through automated cross-rate synthesis. Any triangular mispricing within a single broker’s book is corrected microseconds after it appears — faster than any retail execution system can respond. Profitable triangular arbitrage in 2026 requires genuine cross-rate discrepancies between independent brokers sourcing quotes from different liquidity providers.
In a multi-broker setup, price feeds from different brokers update at different speeds. When one broker’s feed updates before another’s, a temporary apparent triangular discrepancy appears — not because of genuine LP mispricing, but because one feed is ahead of the others. Executing on this signal is unprofitable: by the time orders arrive, the slower broker’s feed has updated and the apparent discrepancy has vanished.
Four independent conditions must all be true simultaneously: (1) feed update timestamps are approximately synchronised across all brokers; (2) the discrepancy has persisted above the minimum threshold for a minimum window (e.g. 80ms); (3) no significant directional price event has occurred on any direct pair in the preceding 100ms; (4) the discrepancy exceeds the dynamically-computed minimum profitable threshold including live spreads, commissions, and slippage.
Minimum two. One broker provides the two direct pair quotes (EUR/USD and GBP/USD); the second provides the cross-rate quote (EUR/GBP). Three brokers maximise genuine mispricing opportunity but multiply commission cost and execution complexity. Three-broker setups also increase partial-fill risk: if one of three legs fails to fill within the configured timeout, the other two create an unhedged directional position that must be explicitly closed.
It produces a recognisable pattern: three correlated positions opening within milliseconds, consistent short hold times, and win rate inversely correlated with spread width. Brokers’ risk systems identify this as technical order flow rather than discretionary trading. Mitigation: configured minimum hold times, lot size variation across legs and trades, and diversification across multiple triangle combinations and trading sessions.
A VPS targeting under 5ms round-trip to each execution broker; independent feed connections per broker on separate network sockets with hardware timestamps at socket receipt level; parallel execution engine that submits all legs simultaneously rather than sequentially; FIX API connection preferred for the cross-rate broker; and clock synchronisation with under 5ms jitter (NTP or PTP). Granular per-signal logging is essential for ongoing filter calibration.
Multi-broker execution · 4-condition signal filter · Parallel order submission · FIX API compatible · 25 years of arbitrage development