Backtest
A simulation of a strategy using historical data and fixed rules.
Maximum drawdown
The largest peak-to-trough decline in an equity curve.
Slippage
The difference between an expected execution price and the actual fill price.
Request for quote (RFQ)
An execution workflow (common in OTC / block-style trading) where a taker requests price quotes from one or more liquidity providers, then chooses whether to trade. For research, RFQ data can be useful — but field semantics, participant visibility, and “trade executed elsewhere” outcomes can bias labels if not modeled explicitly.
Traded away
A term used in some RFQ / block-trade workflows to indicate that, even if a quote was provided or an RFQ reached a “filled” state, the final execution occurred with a different counterparty or venue. Treat it as a separate label dimension from “filled” when measuring win rates or execution quality.
Schema drift (exchange API)
When an exchange changes, clarifies, or expands API fields/enums over time (new states, renamed values, new access rules). In backtests and research pipelines, schema drift can silently change your dataset definitions, so it should be tracked like a versioned dependency.
Order book checksum (market data integrity)
A field (when provided by a venue) intended to help clients detect corrupted or out-of-sync order book state after applying incremental updates. In research pipelines, treat checksums as a helpful signal — but not a guarantee. Always complement them with sequence-gap tracking, periodic snapshots, and explicit “out-of-sync” masking so backtests don’t rely on accidental data cleanliness.
Sequence gap (market data)
A discontinuity in a sequence number / nonce stream for incremental market-data updates (for example, when deltas are dropped, delayed, or delivered out of order). Sequence gaps are a major source of silent data corruption in order-book features; robust adapters detect gaps, resync from a snapshot, and label the recovery window so downstream research can exclude it.
Depth cap (order book)
A limit on how many price levels (or aggregated depth) an order book feed includes per update. Depth caps can change over time or differ by channel/subscription. For research, always record the realized depth per snapshot so imbalance/liquidity features remain comparable.
API contract test (market data adapters)
A small automated check that validates expected fields, enums, and semantics for an exchange API/channel (for example: required fields in WebSocket pushes, sequence rules, and depth levels). Contract tests help catch breaking changes early so backtests don’t silently drift.
Market impact (price impact)
The tendency for your own trades to move the market price, especially when trading size relative to available liquidity. Impact is a core microstructure cost and often increases in stressed or thin markets.
Impact function (impact model)
A mapping from signed order flow (how aggressively you buy/sell) to expected price movement. In research, impact functions are used to stress-test slippage assumptions and to avoid mistaking “free liquidity” backtests for implementable strategies.
Signed order flow
A proxy for buy versus sell pressure, often constructed from trades (e.g., buyer-initiated volume) or quote changes. Signed order flow is used in impact studies and short-horizon price response analysis.
Order attribution (execution metadata)
Extra identifiers that help you segment and audit execution results (for example: strategy ID, order origin, session ID, or venue-side tags when available). Attribution fields are not “alpha” by themselves, but they can prevent accidental mixing of incomparable order flows when you measure fills, slippage, and latency.
Broker identifier (exchange API)
An optional exchange-side field (when supported and permitted) that associates an order with a broker/partner channel. For research, treat it as audit metadata: use it to segment execution-quality reports and detect adapter/instrumentation drift, not as a trading signal.
Tick size (minimum price increment)
The smallest price step an order can be placed at on a specific venue or contract. Tick size affects spread, queue position, and whether limit prices are valid.
Partially observable market (POMDP)
A setting where the “true” market state is not directly observable, so a model or agent must infer it from imperfect observations (price, volume, order book snapshots, funding, on-chain data). In practice, your observation window (what history you include) is a tunable assumption that can change backtest behavior.
Observation window (lookback horizon)
How much past information a model uses at decision time (for example, the last N bars or N minutes of events). Longer windows can capture regimes but increase feature drift and overfitting risk; shorter windows can miss context and amplify noise.
Walk-forward validation
A time-ordered validation process that repeatedly trains on past data and tests on later data.
Overfitting
When a model or strategy learns historical noise instead of durable structure.
Feature engineering
The process of converting raw data into model inputs.
Sharpe ratio
A return-to-volatility metric often used to compare risk-adjusted performance.
Turnover
How much of a portfolio is traded over a period.