Traders spend enormous effort trying to improve signals. They test indicators, refine entry conditions, add alternative data and search for patterns that might survive changing markets.
But a trading strategy can lose money even when its signal still works. The problem can sit between the model and the market: execution drift.
Execution drift occurs when the prices, timing, fills, costs or routing experienced in live trading gradually move away from the assumptions used when the strategy was designed or backtested. The signal has not necessarily decayed. The way the strategy reaches the market has changed.
Backtests Usually Assume Cleaner Execution Than Reality
Historical tests often rely on simplified assumptions. They may use closing prices, midpoint prices or fixed transaction-cost estimates. Live markets add spreads, queue position, latency, partial fills, market impact and periods when liquidity disappears.
Regulators and market-structure authorities emphasize the importance of execution quality. FINRA Rule 5310, for example, requires firms to use reasonable diligence to seek favourable terms for customer orders. FINRA Rule 5310 on Best Execution
For systematic traders, the same principle applies analytically even where no comparable regulatory duty applies: the strategy should be evaluated using the execution it actually receives, not the execution it was assumed to receive.
Small Slippage Can Compound Into a Different Strategy
A few basis points of slippage may appear trivial on one trade. Across thousands of transactions, the effect can transform the economics of a strategy.
This is especially important for short-horizon systems where expected edge per trade is already small. A model with a five-basis-point expected advantage can become unprofitable if execution deteriorates by three or four basis points.
The signal may still rank opportunities correctly. The realised return disappears because implementation consumes the edge.
Latency Changes More Than Speed
Latency is often discussed as a race to be faster. Its deeper importance is consistency. If a system reaches the market unpredictably, the trader cannot reliably know which price environment the signal will encounter.
During calm periods, a delay of milliseconds or seconds may not matter. Around macroeconomic releases, market opens or sudden volatility, the same delay can result in a materially different fill.
Liquidity Regimes Can Change Under the Same Price Pattern
Two identical chart patterns can have very different execution characteristics. One occurs in deep liquidity with narrow spreads. The other appears during a thin session when spreads widen and market depth falls.
The SEC has published investor material explaining how orders are routed and executed and why execution mechanics can affect the price an investor receives. SEC material on trade execution
A signal-only model may treat the situations as equivalent. An execution-aware model does not. It considers whether the market can absorb the intended order without materially changing the price.
Order Type Choice Can Create Drift
Execution drift can also come from order-type behaviour. A strategy tested using marketable assumptions may be deployed with passive limit orders to reduce cost. Fill rates fall and the system misses profitable moves. Alternatively, a strategy designed around passive execution may be switched toward aggressive orders, raising transaction costs and market impact.
Neither change looks like signal decay. The trading rule may be unchanged while the implementation logic has altered the return distribution.
Broker and Venue Changes Matter
A change in broker, liquidity provider, venue routing or smart-order-routing logic can create execution drift. The trader may still receive fills, but at different prices, with different rejection rates or different latency patterns.
This is why execution analysis should be segmented. Average slippage across an entire strategy can hide deterioration in one instrument, one session or one venue.
Execution Quality Needs Its Own Monitoring
Traders often monitor P&L, win rate, drawdown and signal accuracy. Execution deserves its own dashboard.
Useful measures can include implementation shortfall, effective spread, fill rate, rejection rate, time-to-fill, adverse selection after execution and realised cost by instrument or time of day.
The objective is not to optimise every metric independently. It is to detect whether the relationship between signal and realised trade is changing.
AI Can Diagnose Drift but Can Also Hide It
Machine-learning systems can help classify execution outcomes, detect abnormal slippage and identify conditions in which a strategy performs poorly. They can compare realised fills with expected distributions and surface patterns that manual review may miss.
But more adaptive execution can also make attribution harder. If an AI execution layer changes order behaviour dynamically, traders need records explaining why the system chose one route or order type over another. Otherwise, poor performance may be difficult to distinguish between signal weakness and execution behaviour.
Capacity Is Part of Execution Drift
A strategy can also drift simply because it becomes larger. A system tested with small orders may perform well until position sizes increase. The signal remains valid, but the strategy begins moving the market against itself.
This is a capacity problem. The more capital deployed, the more execution assumptions need to be revisited. Strategies that scale in theory may not scale economically once market impact is included.
Signal Decay and Execution Drift Need Different Fixes
The distinction matters because the remedies are different. Signal decay may require new research, feature changes or a different model. Execution drift may require better routing, smaller order sizes, different timing, revised cost assumptions or tighter liquidity filters.
If traders diagnose the wrong problem, they can damage a strategy that was still fundamentally sound.
The Trade Is Not Finished When the Signal Fires
A trading system is not only a forecast. It is a chain from signal generation to order construction, routing, fill, position management and exit. Any part of that chain can change.
As electronic markets become faster and more fragmented, the implementation layer deserves as much attention as the model itself. For many strategies, the next source of performance deterioration may not be that the market stopped rewarding the signal. It may be that the market stopped filling the signal the way the strategy expected.
References
FINRA Rule 5310 — Best Execution and Interpositioning — https://www.finra.org/rules-guidance/rulebooks/finra-rules/5310
U.S. SEC — Trade Execution — https://www.sec.gov/about/reports-publications/investorpubstradexec
CFA Institute — Market Organization and Structure — https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2025/market-organization-structure