Paper Trading

What paper trading is

A paper account follows the live market but fills your orders in a simulation. The bot sends an order, the platform pretends to fill it, and the account balance moves. It is also called a demo account or simulated trading. Unlike a forward test on a small live account, nothing real is at risk, and nothing real is learned about how the market treats your orders.

What it is good for

• Finding bugs. A bot that sends the wrong size, loops on an error, or misreads a price feed shows it on paper, for free.
• Checking the plumbing. Connection drops, restarts, missed candles and time zone mistakes all appear when a bot runs for days.
• Checking that it matches the backtest. If the paper trades do not look like the backtest on the same days, something is off. A common cause is look-ahead bias in the backtest.
• Learning the platform. You can find out how stops, partial fills and order types behave before they cost anything.

What it misses

A simulated fill is usually optimistic. The platform may fill a limit order the moment the price touches it, when in real life you would be behind other orders in the queue and might not fill at all. It may fill a market order at the quoted price with no slippage, and it may use a tight or fixed spread. Some demo accounts do not charge fees, or charge less than a live account. Each of these makes the paper result better than the live one.

Take a made-up bot with a gross edge of 0.30 percent per trade. On paper the costs are 0.05 percent, so it nets 0.25 percent. Live, with a wider spread, slippage and fees of 0.20 percent, it nets 0.10 percent. Over 200 trades that is 50.0 percent of the traded amount on paper and 20.0 percent live, less than half. The strategy is the same. Only the fills changed.

There are two other gaps. Paper trading has no pressure, so you may leave a bot alone on paper that you would switch off at a live loss. And a bot that adds to losing trades, like a martingale or a grid, can look fine on paper because the paper account never faces a real margin call.

How many trades does it take

Suppose a paper run shows a 55 percent win rate. The true rate behind it is not known. A rough 95 percent range around the result is 1.96 x the square root of (p x (1 - p) / n), where n is the number of trades. At 30 trades the range is plus or minus 17.8 points, so the true rate could be anywhere from 37.2 to 72.8 percent. At 300 trades it is plus or minus 5.6 points. The win rate calculator gives the range for your own numbers.

The win rate is only half the picture. The average win and loss matter too, and they vary more. See expectancy for the full measure.

How to run a paper test of a bot

• Write down what you expect before you start: the number of trades, the win rate range and the drawdown. Then compare, so the result cannot be bent to fit afterwards.
• Run it long enough to see more than one type of market, not one quiet week.
• Set the paper costs to the pessimistic live costs: the full spread, a slippage allowance and the real fee schedule.
• Check each trade against what the live order book could have done. If a limit fill needed the price to trade through, count it only if it did.
• Then move to a small live test. Paper results do not replace it.

What it does not tell you

A good paper result does not mean the bot will earn live, and a poor one does not always mean it will lose. It is evidence that the bot runs as designed. It is weak evidence about profit. The overfitting guide explains why even a clean forward test can be a coincidence.