Every systematic desk learns the same lesson in the same order: the backtest looks one way, the paper record looks a little worse, and the live record looks worse again for a while. People who have not seen this treat it as evidence of fraud. People who have seen it treat it as evidence of reality. The useful skill is knowing where the gap comes from, so you can tell an ordinary gap from a damning one.
Five sources of the gap
Costs. A backtest with no fees, spreads or funding has a free lunch the live account does not. On a strategy that trades often, this alone can be most of the difference.
Slippage. The backtest buys at the price in the data. The live account buys at the price it can get, which in a fast market is worse, and in a thin market is much worse. Entry zones rather than single prices exist partly to make this honest: the record can show where the entry actually happened.
Hindsight in the rules. Even careful researchers know, while writing the rules, what the market did next. Walk-forward testing reduces this; nothing removes it. The live record is the first one written without knowing the ending.
Selection. The backtest you are shown is the one that survived. The live record is the only one there is.
Behaviour. A backtest never hesitates, never skips a signal after three losses, never adds size after three wins. A live record is run by people and software with caps and daily stops, which is correct, and which costs something.
How big a gap is normal
There is no universal number, but there is a universal shape: the live record should be worse than the backtest by roughly the costs and slippage the backtest ignored, and no more. If the backtest showed a good year and the live record shows a somewhat smaller gain with the same drawdown profile, that is a gap with a reason. If the backtest showed a good year and the live record is losing money with twice the drawdown, the backtest was not describing the same strategy.
You can only make that comparison if the two records are kept apart. This is the whole reason the desk labels every figure backtest, paper or live and never blends them. A blended record, backtest for the early years and live for the recent ones, is the most common way the gap is hidden: the long smooth curve is mostly hypothesis, and the short real stretch at the end is too short to see.
The middle step: paper
Paper trading sits between the two. It runs in real time on a simulated account, so it has no hindsight and real slippage on entries, but no real capital and therefore none of the behaviour that capital produces. It is where a desk finds out whether its rules can be executed as written. The desk publishes its paper record as its own thing, labeled, with the number of signals and the result in R, and does not add it to the backtest to make the curve longer.
Reading across the gap
When you look at a desk's three records, read them in this order:
- The backtest, for the shape: the drawdowns, the behaviour by regime, which sleeves carry it. Treat the return as an upper bound.
- The paper record, to see whether the rules survived contact with real-time execution: did the entries happen where the rules said, and did the results in R look like the backtest's?
- The live record, judged by its length before its slope. A live record of three weeks tells you almost nothing either way; the desk says so on its own record page.
The gap between 1 and 3 should be explainable by costs and slippage. If it is, you have learned something real about the strategy. If it is not, you have learned something real about the backtest.
What this means for you
Do not pay for a backtest. Pay, if you pay at all, for a desk that shows you the backtest, the paper record and the live record separately, tells you the live one is young when it is young, and lets you measure the gap yourself. A desk that will not keep the labels apart is asking you to take the gap on faith.
Backtested results are hypothetical and have inherent limitations; paper results are from a simulated account; live results are from real capital and are shorter and separate. Past performance of any kind does not predict future results. Nothing here is personalised investment advice.