Quant Memo
Foundational

Paper Trading and Shadow Mode

Two ways to run a strategy against real, live market data without risking real money — paper trading simulates fills in a separate account, shadow mode runs the real production system alongside the live one purely to compare.

Prerequisites: Production-Research Parity

A backtest is convincing but it's still a simulation of the past. The next natural step before trusting a strategy with real capital isn't necessarily to trade it live — it's to let it run against live, unfolding market data first, where nothing about the future is known in advance the way it is in a backtest, but without yet putting real money on the line.

Paper trading does this by routing a strategy's orders to a simulated account: the strategy generates real signals off real live prices, and a paper-trading layer pretends to fill those orders — usually at the observed market price, sometimes with a modeled fill probability and slippage — and tracks a simulated P&L, all without touching an actual brokerage account. It's a live rehearsal for a strategy in isolation.

Shadow mode is a related but distinct check, more often used for a system upgrade than a brand-new strategy: the new version of a production system runs in parallel with the currently live one, seeing the same real-time data and computing what it would have done, but its outputs are only logged and compared, never actually sent as orders. Where paper trading asks "does this strategy work against live data," shadow mode asks "does this new system produce the same (or better) decisions as the system it's about to replace."

Worked example. A firm develops a new signal and paper-trades it for six weeks: it generates simulated orders against live prices, and the paper P&L broadly matches what the backtest predicted for a similar six-week stretch, giving some confidence the strategy's edge is showing up in live conditions and not just in-sample. Separately, the firm rewrites its order-execution engine for speed, and rather than paper-trading the strategy again, it runs the new execution engine in shadow mode: both the old (live) and new (shadow) engines receive the same real-time signals and market data, the old engine's orders actually execute, and the new engine's would-be orders are only logged. After two weeks, an engineer compares the two logs and finds the new engine agrees with the old one on 99.8% of decisions — the remaining 0.2% trace to a rounding difference in a tick-size calculation, caught and fixed before the new engine ever handles real capital.

What this means in practice

Paper trading and shadow mode both exist to surface problems — a stale data feed, an unhandled edge case, a parity gap between research and production — while the cost of being wrong is zero, rather than after real money is already at risk. Paper trading is the right check for "does this new idea work"; shadow mode is the right check for "does this replacement system behave like the one it's replacing." Neither substitutes for the other, and neither is a perfect predictor of live behavior — a paper account's simulated fills are still an approximation, and a strategy that trades in genuinely large size can move the market in ways paper trading, which has no real market impact, will never reveal.

Paper trading simulates a strategy's fills against live market data with no real money at risk, testing whether an idea works outside of backtesting. Shadow mode runs a new system in parallel with the live one purely to compare their decisions, testing whether a replacement is safe to deploy. Both are cheap insurance before capital is on the line.

Related concepts

Practice in interviews

Further reading

  • Chan, Ernest P. Algorithmic Trading, ch. 8
ShareTwitterLinkedIn