Paper trading versus backtesting
Backtesting asks whether a rule would have worked on past data. Paper trading asks whether you can follow a rule in real time. They answer different questions, and each is dangerous when used to answer the other's.
The short version
| Backtesting | Paper trading | |
|---|---|---|
| Question it answers | Would this rule have worked historically? | Can I follow this rule now? |
| Sample size | Potentially thousands of trades | However many you take, slowly |
| Speed | Years of data in seconds | Real time, no faster |
| Captures your behaviour | No | Yes — that is the point |
| Main failure mode | Fitting the rule to the past | Too few trades to conclude anything |
| Needs a fully mechanical rule | Yes | No |
Where backtesting misleads
- Curve fitting. Enough parameters will fit any history. The tuned version performs beautifully on the data it was tuned on and ordinarily everywhere else.
- Lookahead bias. Quietly using information that was not available at the decision point — a closing price during the session, a figure later revised, an index constituent list that already knows who survived.
- Survivorship. Testing on assets that still exist today excludes the ones that failed, which is exactly the population that would have hurt you.
- Costs assumed away. Spreads and fees are frequently modelled optimistically or not at all, which flatters frequent trading most.
- It cannot model you. A backtest never panics, never skips a signal because the last three lost, and never doubles up to get even.
Where paper trading misleads
- Sample size. Practice accumulates trades at real-world speed. Reaching a number that means anything takes months, and most people conclude long before that.
- No real money. The discipline is untested in the way that matters most.
- Perfect fills. Simulated execution is kinder than real execution.
- One market regime. Three months of practice samples whatever conditions those three months had. A trending quarter tells you little about a choppy one.
Neither establishes an edge on its own. A good backtest is a reason to investigate further, not evidence that money should follow. A good practice run is a sign you can execute, not proof the approach is profitable.
How they fit together
The sensible order is roughly:
- Form an idea and write it as a rule precise enough to argue about.
- Test it against history, if the rule is mechanical enough to test, keeping parameters few and costs pessimistic. Treat the result as a filter for obviously bad ideas, not a forecast.
- Practise it forward, logging every trade, to find out whether the rule is followable and what you actually do under pressure.
- Review the process before the profit for the first several dozen trades.
- Decide about real money separately, knowing neither step proved profitability.
A rule that backtests well and that you cannot follow is worthless to you specifically. That gap — between what the rule requires and what you reliably do — is the thing forward practice measures and history never will.
Which to start with
If you are new, start with forward practice. Backtesting rewards programming and statistics skills, and it is easy to produce an impressive, meaningless result before you have the experience to tell the difference. A journalled practice run is harder to fool yourself with, because the failures are yours and dated.
RATTLE does forward practice, not backtesting: you log closed trades and review them. It does not simulate a strategy over historical data, and it is not trying to.
Start a forward practice logNext
See what paper trading can honestly teach you for the broader picture, or the mistakes that make practice results misleading for the specific ways a forward log goes wrong.