How to Build a Trading Bot

The seven steps

1. Write the rule in full

Say exactly when to enter, when to exit, how big to trade, and when to do nothing. Each line must be something a program can check with numbers. "Buy when it looks strong" is not a rule. "Buy at the next open when the 10-bar average closes above the 30-bar average" is.

2. Get clean data

Get prices for the market you will trade, with the time zone and the bar size written down. Look for gaps, repeated bars, bad ticks and changes that are not real prices, such as a split. Bad data makes a bot do the wrong thing quickly. See survivorship bias for a data problem that hides losers.

3. Code the rule

Turn each line of the rule into code and keep it small and readable. Decide each time which bar a decision uses, and never use a bar that had not closed. A short working example in plain Python is on the Python trading bot page.

4. Backtest it, with costs

Run the rule on past data you did not tune it on, and count the spread, the fees and slippage. Look at the average result per trade, the worst fall and the number of trades, not only the total. See backtesting, overfitting and look-ahead bias.

5. Add the risk rules

Set the size of each trade from how much you are willing to lose if the stop is hit, a daily loss limit, and a maximum number of open positions. Add what happens on a lost connection, a rejected order and a restart. See position sizing and daily drawdown rules.

6. Forward test

Freeze the settings and run the bot on data that arrives after you froze them. Compare the results with the backtest. A big gap means the backtest was too kind. See forward testing.

7. Paper trade

Run the whole bot, orders and all, against a practice account or simulated fills for weeks. This tests the plumbing: order handling, timing, logging and the limits. See paper trading. A practice account fills more kindly than a real market, so it does not prove real results.

A worked example: the rules written out

This is a made-up example to show how complete a rule must be. It is not a strategy to trade. The account is a made-up $10,000 and the price is a made-up 40.00.

Calculated by us on 29 Oct 2026, exact arithmetic on made-up numbers. Note what the table forces. The size comes from the stop, not from a feeling, so a wider stop means fewer units. The daily limit is reached after 2 stopped trades, which is a figure you can check against how often the rule loses. And a stop does not fix the price: a gap or a fast market can fill it worse, so the real loss can be bigger than $100. Some markets also close, which the rule must say what to do about. Rules for stops vary by broker and exchange, so check yours. See stop loss orders.

How bots break at each step

• Rule: vague wording gets filled in by the code, and the code fills it in differently from what you meant.
• Data: gaps and bad ticks create trades that never existed.
• Code: using the current bar to decide a trade at that bar's price. Test results then look far better than anything achievable.
• Backtest: tuning settings until the curve looks good fits noise. Leaving out costs removes the difference between a small edge and none.
• Risk rules: no limit on size or daily loss lets one bad day end the account. A martingale-style size rule ends this way in the long run.
• Forward test and paper trade: skipped because the backtest looked good. This is the step that finds the plumbing faults.

What this page does not tell you

It does not give a strategy, and the moving average rule above is an example of how to write a rule, not a suggestion. The numbers are made up. Nothing here says a bot built this way will make money, and many do not. Keep keys small and unable to withdraw, and keep the code out of any public place. This is education, not advice.