How to test a grid bot before using real money

Why test at all?

Grid bot settings interact: range, number of levels, margin, leverage and the market all change each other's effect. A setting that worked in a quiet week can liquidate in a trending one. Testing is the cheapest way to find that out, before the market teaches it with real money.

Three ways to test, and what each one shows

1. Calculator (instant): checks the setup's arithmetic without any price history: step size, net profit per grid after fees, order size, worst-case liquidation price and how much an exit from the range would cost. Use it first; it catches impossible setups in seconds. Try the grid bot calculator.
2. Backtest (minutes): replays past prices with your settings. Fast and good for comparing settings, but it only shows how the bot would have done in that particular period, and it is easy to make optimistic.
3. Virtual forward test (days): runs the bot on live market prices with a virtual balance. It is the closest to real use because nobody knows the next price yet, but it takes time.

Each layer catches different problems. Skipping the calculator wastes test time on broken setups; stopping at the backtest leaves the most important question unanswered: what happens on prices the setup has never seen?

What a realistic test must include

A test is only as good as its rules. If any of these is missing or simplified, the result is usually better than reality:

• Both fees: maker and taker fees on every buy and every sell.
• Funding: paid or received on the full position size for as long as inventory is held.
• Liquidation: with the exchange's maintenance margin, at the moment the position is largest.
• A strict fill rule: an order fills when price trades through the level, not when it merely touches it.
• Exchange limits: minimum order size and quantity step; tiny orders that the exchange would reject must not count.
• Liquidity: on thin pairs, virtual fills are more optimistic than a real order book would allow.

A step-by-step test plan

1. Pick a liquid pair and write down why you expect it to range.
2. Check the math in the calculator: the step should be several times the round-trip fee, and the worst-case liquidation price well outside the range.
3. Backtest over different conditions: at least one rising, one falling and one sideways period, not only the latest week.
4. Compare setups fairly: same pair, same period, same margin; for example a long grid against a neutral grid on the same range.
5. Write the exit rule before the forward test: a time limit, maximum inventory, a price outside the range or a loss you accept.
6. Run a virtual forward test on live prices for one to two weeks, without changing settings mid-test.
7. Read the result with the metrics below, then decide.

Which numbers to look at

• Total P/L, not grid profit alone: completed round trips plus the open position, minus fees and funding.
• The worst moment (largest drawdown) during the test: could you have held through it with real money?
• Distance to liquidation at that worst moment.
• Time in range: how long price stayed between the lower and upper limits.
• Open inventory at the end: a bot that ends holding a large losing position has not finished its test.
• Fee share: how much of the gross grid profit went to fees.

Common testing mistakes

• Generalising from one good week, usually a sideways one.
• Trying many settings and keeping the one with the best past result: it fits the past, not the future.
• Leaving out fees or funding because they look small.
• Testing with different margin or leverage than you will use for real.
• Testing with an amount you would not actually risk, so the drawdowns feel unimportant.
• Changing settings in the middle of a test and still reading it as one result.

From test to real money

If the test result is acceptable, including its worst moment, move to the exchange with the same settings and a smaller amount than you tested with, and keep the exit rule. In Botrader, the “Take to exchange” button lists a bot's settings in the target exchange's form order, ready to copy; Botrader itself never sends orders.

Keep comparing: the first real weeks are also a test, this time of how close the simulation was to reality.