Comparison, in Backtesting & Validation
Paper vs Live Trading
Both run the same strategy against live market data, but only one gets filled at real prices with real consequences. Paper trading uses a simulated account: orders are matched against quoted prices with assumed fills, so it proves the pipeline works without proving the edge survives execution. Live trading routes real orders into the book, where you discover slippage, queue position, partial fills, rejects, and the behavioral cost of watching real money move. The gap between the two is where overfit backtests and naive cost assumptions get exposed. This matrix compares what each can and cannot tell you.
6 criteria. Published May 26, 2026. By AI Fin Hub Research.
Runs the strategy against live data in a simulated account with no real capital. Orders are filled by the broker's simulator at or near quoted prices.
Pros
- Zero capital at risk while you validate that signals, routing, and risk controls work end to end
- Catches integration bugs, data-feed gaps, and timing errors before any money is involved
- Lets you run forward in real time on unseen data, which is genuinely out-of-sample
- Free and repeatable, so you can iterate on plumbing without cost
Cons
- Simulated fills are optimistic: you usually get the quoted price with no market impact
- Ignores partial fills, queue position, rejects, and the cost of crossing the spread under stress
- No financing, borrow, or realistic commission drag unless explicitly modeled
- Removes the psychology entirely, so it cannot reveal how you behave during a real drawdown
Debugging the system, confirming a strategy behaves as designed, and forward-testing logic before any capital is committed
Routes real orders into the market with real, deliberately small capital, exposing the strategy to actual fills, costs, and consequences.
Pros
- Reveals true slippage, market impact, and partial fills that decide whether the edge survives
- Exposes broker-side realities: rejects, latency, borrow availability, and real commissions
- Forward, fully out-of-sample evidence with the only fills that count
- Surfaces the behavioral cost of real losses, which often changes how a strategy is run
Cons
- Real capital at risk, even at small size, so bugs now cost money
- Small size can itself mask impact that will appear when the strategy is scaled up
- Slower to gather statistically significant evidence than a fast backtest
- Emotionally and operationally demanding, with monitoring and reconciliation overhead
Measuring real execution costs and impact, validating the edge survives reality, and the final gate before scaling capital
Decision table
See the tradeoffs side by side
| Criterion | Paper Trading | Live Trading (small size) |
|---|---|---|
| Capital at risk | None | Real, kept small |
| Fill realism | Optimistic, quoted prices | Real fills, real impact |
| Slippage and impact | Not captured | Measured directly |
| Partial fills and rejects | Usually ignored | Real |
| Tests psychology | No | Yes |
| What it proves | The plumbing works | The edge survives execution |
Verdict
These are sequential stages, not alternatives. Paper trading is the right place to prove the system runs, the signals fire, and the risk controls trigger, all without spending a cent, and its forward run on unseen data is genuinely out-of-sample for the logic. But paper fills are optimistic and silent about slippage, partial fills, and market impact, which is exactly where a lot of backtested edges evaporate. So the moment the plumbing is clean, move to deliberately small live size to measure those execution costs and to feel the behavioral pressure that paper removes. Treat small-size live as the real validation gate before scaling, and remember that even small live size can hide the impact that appears when you add capital, so model impact explicitly before you size up.
Run the numbers next
- Playgrounds Execution Simulator
Estimate execution cost in closed form — square-root permanent impact, linear temporary impact, half-spread, and a latency-drift band. See the slippage a naive backtest hides.
- Playgrounds Order Book Replay Visualizer
Drop a Level-2 snapshot CSV and replay the book snapshot by snapshot. Animated depth bars, best bid/ask, spread and imbalance. Understand microstructure before.
- Generators Trading System Blueprinter
Pick your data source, LLM, broker, storage, risk engine, and logger. Get a Mermaid architecture diagram and a copyable starter file tree — the full stack before you write code.
Questions people ask next
Why do strategies that work on paper fail live?
Paper fills are typically granted at the quoted price with no market impact, partial fills, or queue competition. Live, you cross the spread, sit behind others in the queue, get partially filled in fast markets, and occasionally get rejected. For high-frequency or thin-liquidity strategies these costs can exceed the entire backtested edge. Paper trading therefore validates logic but systematically overstates returns by omitting the execution costs that decide profitability.
How small should initial live size be?
Small enough that a worst-case bug or a string of losses is an affordable tuition payment, but large enough that real fills and costs are representative rather than dominated by fixed fees. The goal is to measure slippage and impact per trade, not to make money, so size to learn. Once execution costs match expectations and the system has run clean for a meaningful sample, scale up gradually while re-checking that impact has not grown with size.
Is paper trading out-of-sample?
For the strategy logic, yes: running forward on live data the strategy was not fitted on is genuinely out-of-sample and far more honest than re-running a backtest. What it is not is a realistic execution test, because the fills are simulated. So paper trading gives you trustworthy evidence about whether the signal generalizes, and untrustworthy evidence about whether the net-of-cost edge survives. Live small-size trading is needed for the second question.
Sources
- Advances in Financial Machine Learning , Marcos Lopez de Prado, Wiley (2018)
- Trading and Exchanges: Market Microstructure for Practitioners , Larry Harris, Oxford University Press (2003)
Keep the topic connected
- Learn Slippage
Slippage as the gap between expected and executed price: the components (spread, market impact, latency), and how to model each in a backtest.
- Learn Maker-Taker
Maker-taker fee model: makers get a rebate, takers pay. Why the model exists, what it incentivizes, and how to size up real net cost.
- Learn Look-Ahead Bias
Look-ahead bias: when a backtest accidentally uses data the strategy wouldn't have had at decision time. The most common variants and how to catch them.
- Checklist Trading Strategy Validation Checklist
A sign-off checklist for validating a trading strategy before risking capital: data hygiene, out-of-sample testing, trial accounting, deflated Sharpe, and risk backtests.