Market replay versus backtesting

Market replay versus backtesting comes down to who makes the decisions. Market replay plays a recorded session back in time order, and you watch it and place simulated orders. Backtesting runs written rules over historical data and reports what the rules would have done, with no person involved. Replay tests how you read the market and handle orders. Backtesting tests whether a written rule holds over many sessions.

What market replay does

A replay platform reads recorded trades, and sometimes recorded order book updates, and plays them back at a speed you choose. At 1x, a regular ES session of 6.75 hours takes 6.75 hours. At 12x it takes about 34 minutes. You see the chart, the DOM and the tape build as they did on the day, and you do not know what comes next.

In replay you practise reading a heatmap or a footprint at speed. You also place and cancel orders before you know how the session ends. A backtest cannot train either skill. Replay costs screen time. Five sessions at 4x take about 8.5 hours of screen time.

What backtesting does

A backtest takes a rule written in code or settings and applies it to every bar or trade in a historical data set. It reports counts and totals. A backtest over 60 sessions finishes in seconds, so you can test one rule across hundreds of days that replay would take months to cover.

A backtest only tests what you can write down. Its results also depend on how the engine fills simulated orders. The example below shows how much one fill model can change a count.

A worked example with one limit order

Take an invented test in ES. The question is how often a resting limit buy order at the session's opening price would fill if price came back to that level within 30 minutes. This is a fill question, not a trading rule.

The backtest uses one-minute bars from 60 sessions. Its fill model says a limit order fills whenever a bar's low touches the order price.

On the bar data, every touch counts as a fill. Replay with recorded depth shows the queue. Suppose on a typical day 350 contracts already rested on the bid at the opening price when the order joined. CME matches most futures orders first in, first out, so 350 contracts had to trade at that price before the order filled. In 18 of the 41 touches, fewer than 350 contracts traded there before price moved away. Those 18 orders never filled.

The backtest reports 41 fills. Replay shows 23. The gap comes from the fill model, not the market. The note on market, limit and stop orders on a futures DOM shows the same queue with a smaller order.

What each method needs in data

Backtesting on bars needs only bar history. Backtesting on tick data needs every trade. A queue-aware backtest needs recorded order book data, and many retail platforms do not keep it.

Replay has the same split. A trade replay rebuilds the chart, the tape and the footprint. Full depth replay also rebuilds the DOM and the heatmap, and it needs depth that someone recorded on the day. A later note covers full depth replay and why it needs recorded data.

What the platforms on this site list

NinjaTrader lists Market Replay and free simulation. Quantower lists Market Replay that uses provider history at tick, minute or day resolution. Its documentation does not support historical full-depth or market by order replay. Bookmap lists replay from its Digital+ tier, using recorded data. Liquivue lists imported-recording replay. On the backtesting side, Investor/RT sells a Trading Systems package for backtesting and optimization at US$10 a month on top of its core plan. The rankings page sets out replay coverage per platform with sources.

Using both

The two methods check each other. A backtest gives a count over many sessions. Replay shows how many of those fills a real queue would allow, at your own reaction time. If a backtest says 41 fills and replay of the same sessions gives 23, the backtest's fill model needs fixing before you rely on any number it produces.

A check you can run

Pick a backtest result that depends on limit orders. Choose five of its sessions and replay each one with depth, if your platform has it. For each simulated fill, note the displayed size ahead of the order when it joined and the volume that traded at that price before price left. Count how many backtest fills the queue would have allowed. If the replay count is lower, use it.