Building a Mechanical Trading System

A discretionary trader decides each trade by judgment: "the chart looks strong today". A rule-based (systematic) trader decides by a written rule: "buy only if the price closes above yesterday's high and the risk is within budget". The rule is the same every day, whatever the mood.

Rules have a clear benefit. After three losses in a row, a person may hesitate on a valid setup or force a poor one. A written rule does not care. Rules can also be tested on past data, which judgment cannot.

They also have costs. Rules cannot read a surprise that no rule anticipated, such as a sudden policy announcement. A system needs upkeep: someone has to notice when the market no longer behaves as the rules assume. Neither style is better for everyone. Many traders use rules for the parts where mood does the most damage, such as size and exits, and keep judgment for the rest.

A useful rule set is like a recipe with quantities. Vague steps ("cook until done") give different results each time. The parts are these: a filter that decides if the day is tradable; an entry trigger; a position size calculated from your risk budget; a stop and target set at entry; an exit or roll rule; and a record of every trade.

Use price action for the triggers, that is, what price itself does: a close above a marked level, a break and retest of a range, a rejection from a level. Keep the trigger simple enough that two people reading it would mark the same trade. Add time rules too. For example, close single-stock option positions before the expiry week, because stock derivatives end in delivery and margin rises through the final sessions.

The flow below shows how the pieces link. The last steps matter as much as the first: the log, the weekly review and the rule-change step. Rules are edited only in the review, never in the middle of a live trade.

A backtest applies your rules to old price data to see what would have happened. It gives figures such as number of trades, average win and loss, and the largest drawdown. It is useful for rejecting weak ideas. It is not proof that a good one will work.

The main danger is curve fitting. If you keep changing numbers until the past looks perfect, you have fitted the rules to that one history, and they will likely fail on new data. One protection is to split the data. You design and tune on the first block, then lock the rules and test once on a block you have not touched. If it fails there, you rethink the idea instead of tuning it to pass.

The last step is live testing with very small size. Live trading shows what data cannot: bid-ask spreads, slippage, brokerage, STT (raised for F&O from 1 April 2026) and your own behaviour. A system that looks good before costs can be a loser after them, and options with wide spreads suffer most.