Trade Simulator

What a trade simulator is

A trade simulator is a program that imitates trading. You, or a bot, send an order. The simulator decides whether it would have filled, at what price and when. It then keeps a pretend account: positions, cash, profit and loss.

The idea is simple. Real money teaches fast and expensively. A simulator lets you see how an order type works, how a rule behaves and how an account moves, before any money is exposed. That is a real use. It is also all it is. A simulator does not predict anything. It only replays rules you give it against prices you give it.

Simulators differ in two ways that matter. The first is where the prices come from: stored past data, or a live feed. The second is the fill model, the set of rules that turns an order into a trade. The second one usually decides whether the result means anything.

Simulator, paper trading and backtest

People use these three words loosely. Here is a clean way to separate them.

• Backtest. A rule is run over stored past prices, as fast as the computer can go. It answers one question: what would this rule have done in that data? See how to backtest.
• Paper trading. Pretend orders are placed against live prices as they arrive, with a pretend balance. The market moves in real time, so you cannot peek ahead. See paper trading.
• Trade simulator. The engine behind either of them, or a tool that bundles both. Some replay past data bar by bar, so you can practise as if it were live. Some are live only. A demo account is a simulator that a broker runs for you.

So the question is never "simulator or backtest?" It is "which prices, and which fill model?" A backtest can run inside a simulator. Paper trading always does.

What each one can and cannot teach

A backtest is fast and covers years, but it is easy to bend. You can change the settings until the past looks good, which is overfitting. Paper trading is honest about time, because you cannot see tomorrow, but it is slow, and a few weeks of it covers one kind of market at best. That is why the usual route is both, in order: backtest, then forward testing on live prices.

No pretend method teaches the feel of risking your own money. That is a plain fact about the method and not a reason to rush past it. It is a reason to treat simulated results as the most optimistic version of what a rule could do.

Fill assumptions: where simulations go wrong

A fill is the trade that results from an order. Three things shape it.

• Spread. The gap between the price you can buy at and the price you can sell at. A buy pays the higher one and a sell receives the lower one. A simulator that fills at the middle price skips this cost. See spread.
• Slippage. The difference between the price you expected and the price you got. It happens when the market moves before the order is filled, or when the order is larger than what is offered at the best price. See slippage.
• Latency. The time between the signal and the order arriving. If a bot reacts to a bar that has just closed, the order is filled on a later price, not the one it saw.

How big each is varies by market, by time of day, by order size and by broker or exchange. Check the numbers for the one you would use. Do not borrow figures from this page.

Two more fill traps are worth knowing. A simulator may fill a limit order the moment price touches it, when in a real queue you might not be filled at all. And a simulator may let a stop order fill exactly at its price, when a fast market can fill it further away. Both make a result look better than it would have been.

A worked example with made-up numbers

Take a made-up rule that makes 10 round trips of 100 units each. The moves on the middle price, per unit, are 0.50, -0.30, 0.40, -0.30, 0.20, 0.50, -0.30, 0.10, -0.20, 0.40. They add up to 1.00 per unit, so with ideal fills the total is 1.00 x 100 = $100.00.

Now change only the fill assumptions, one at a time. The spread costs 0.04 per unit per round trip. Slippage costs 0.02 per unit on each side, so 0.04 per round trip. Latency costs 0.03 per unit on entry, because the order arrives a bar late and the price has already moved. Every figure is made up.

The entries and exits never changed. With spread added, the total falls to $60.00. With slippage too it falls to $20.00. With latency the total is -$10.00, a loss, and only 5 of 10 trades are winners instead of 6. The average gross move is 0.10 per unit, so any total cost above 0.10 per unit turns the result negative. Here the full cost is 0.11. The rule did not change. The simulator's kindness did.

Notice the shape of the problem. The smaller each gross move is, the larger a fixed cost looms. A rule that trades often for small moves is the most exposed to the fill model. That is true of many bot types, and it is why a backtest with no costs is close to useless for them. All numbers here are made up and calculated by us on 31 Oct 2026, exact.

How to run a fair test

1. Write the rule down first. Entry, exit, stop, size and the data you will use. Do not change it halfway because the first result looked poor or good.
2. Use costs at least as bad as you expect. Include spread, fees and a slippage allowance. Then run it again with the costs doubled. If the result only survives the gentlest costs, it is fragile.
3. Delay the signal. Fill on the next bar, not the one the signal came from. Using information the bot could not have had yet is look-ahead bias.
4. Count enough trades. A handful of trades says very little, because luck can dominate. More trades make the average result steadier, though they never make it certain.
5. Test more than one kind of market. A range, an uptrend and a downtrend behave differently. A rule that fits only one of them should say so.
6. Look at the worst stretch, not only the total. The deepest fall from a peak and the longest losing run decide whether a rule can be lived with.
7. Shake the order of the trades. Reshuffling the same trades shows how much of the result depended on sequence. The Monte Carlo simulator does this.
8. Keep a final stretch of data untouched. Tune on the rest, then test on it once. Peeking more than once turns it into more training data.
9. Move to live prices with pretend money. Run forward testing or paper trading and compare the fills you get with the fills the backtest assumed.

What a simulator does not tell you

It does not say a rule will work next month. It does not include your nerves, a platform outage, a rejected order or a change of rules at your broker or exchange. A good simulated run means the rule survived a test. It does not mean the rule is safe. Treat it as the first filter and not as proof.