Paper trading a bot

What paper trading genuinely tests

A backtest replays history through your rules. Paper trading runs those rules against a market that has not happened yet, in real time, through the same connection real orders would take. That difference catches a specific set of problems.

• Bugs that only appear on live data — a malformed order, a reconnect that loses track of an open position, a timezone that shifts everything by an hour.
• Latency between the signal and the fill, which no historical test models.
• Spread widening around news and at session boundaries, when the quoted spread is not the spread you get.
• Whether the bot handles a dropped connection by reconciling with the broker or by assuming its own memory is still correct.
• How the strategy behaves in a market condition your historical sample happened not to contain.

What it does not test

Demo fills are usually better than live ones. There is no real counterparty, no order queue and often no requoting, so a demo account tends to fill at the quoted price where a live account might not. Expect live execution to be slightly worse and do not treat the demo result as a ceiling.

And it does not test you. This is the larger omission. Watching a losing run on demo is not the same experience as watching one with real money, and the difference is exactly where most people's plans fail. Demo tests the bot; only a small live account tests the operator.

How long to run it

Weeks, not days, and the length should be set by trades rather than by the calendar. A bot placing two trades a week needs months to produce a meaningful sample; one placing twenty a day needs less.

The specific thing to wait for is a losing run. A demo period that contained only good conditions has told you nothing about the case that matters. If the strategy has not yet had a bad week, the test is not finished.

The mistake almost everyone makes

Interfering. Increasing the stake after a win, switching the bot off during a drawdown, adjusting a parameter mid-test because a trade looked wrong.

Every one of those invalidates the test, and each is a rehearsal for the behaviour that will cost real money later. The value of a demo period is precisely that it costs nothing to discover you cannot leave it alone — which is genuinely useful information about yourself, available for free.

Set the parameters, write them down, and change nothing until the period ends. If you cannot, that is the finding.

Moving to real money

• Start with an amount whose total loss would not change your month.
• Keep the same settings the demo ran on. Changing them at the moment of funding discards the test you just did.
• Compare live results against the demo period as they accumulate. A large divergence points at execution or at a market change, and both are worth knowing.
• Expect the first losing run to feel different with real money. It will. That is the part demo could not teach you.

Common questions