Order Book Replay

Unlike simple bar-based backtests, this technique replays the entire limit order book (LOB) snapshot for each event. This includes the bid and ask queues at every price level, not just the best bid and offer. An execution algorithm can thus simulate walking the book to fill a large order, experiencing the exact price impact that would have occurred historically. This is critical for testing iceberg orders and liquidity-seeking algorithms that depend on hidden depth.

A core feature is the simulation of time-price priority. When a simulated limit order is placed, it enters a specific position in the FIFO queue at its price level. The fill simulation logic must track cancellations and trades ahead of it in the queue. This allows for precise modeling of adverse selection—the probability that your resting order gets filled only when the market moves against you—which is essential for market-making strategy validation.

The replay engine processes raw market data messages (add, modify, cancel, execute) sequentially. This event-driven architecture ensures that the strategy logic is triggered by the exact same information flow a live trading system would receive. It captures fleeting arbitrage opportunities and quote stuffing events that exist for only milliseconds, making it the gold standard for high-frequency trading (HFT) backtesting.

Fill probability is determined by the interplay of order entry latency and queue dynamics. The simulator can inject configurable delays between the strategy's decision signal and the order's arrival in the market. This exposes latency arbitrage vulnerabilities, where a strategy that appears profitable on paper fails in practice because faster competitors consistently deplete the intended liquidity before the simulated order arrives.

Order book replay enables accurate modeling of maker-taker fee schedules. Because the engine knows whether a simulated order added or removed liquidity based on its queue position and execution type, it can apply the correct exchange fees or rebates. This granularity is vital for strategies with thin profit margins, where the difference between paying a taker fee and earning a maker rebate determines net profitability.

Advanced implementations model self-impact, where the simulated strategy's own orders affect the market. If a large simulated buy order consumes multiple price levels, the engine updates the LOB state to reflect this removed liquidity. Subsequent decisions within the same replay step see the modified book, preventing the unrealistic assumption of infinite liquidity and revealing negative feedback loops in aggressive execution algorithms.