Designing a rules-based trading system

The answer

A trading system is a written specification with 6 parts: setup, entry, filter, exit, stop, size. Every part must be stated so exactly that a stranger, or a computer, could execute it without asking you a question.

The value of writing it down is not automation. It is that a rule you cannot write cannot be tested, and a rule that cannot be tested can only be defended.

Why this costs you money

You have been trading for 3 years. Ask yourself a simple question: what is your win rate?

Most people do not know. They have an impression, and the impression is usually wrong in a specific direction — they remember the large wins and the recent losses, and they forget the long middle.

Here is what that costs. Without a written system, you cannot tell the difference between these 2 situations:

• The rule is sound and you are in a normal losing streak.
• The rule is broken and you are watching it fail.

They feel identical from inside. Without a written specification and a recorded expectation, you have no way to separate them, so you do the only thing available: you decide based on how you feel this week. That means you abandon sound rules during ordinary drawdowns, and you persist with broken ones during lucky runs. Both errors are expensive and both are guaranteed by the absence of a document.

There is a second cost, and it is larger. An unwritten rule silently changes. You start with "I buy breakouts above the 50-day high." Two losses later you are buying breakouts above the 50-day high when the market feels strong. Three months later you are buying whatever looks like it is going up. You never decided to change anything. There was nothing to change, because there was nothing written.

How it works

Here are the 6 components. A system missing any of them is not a system.

1. Setup — what is eligible. Which universe of instruments this applies to at all. Which market. What minimum average daily traded value, so you can actually get in and out. Any exclusions: no companies under a certain size, nothing in the first 6 months after listing, nothing under a trading restriction. The setup is the filter you apply before you look at anything.

2. Entry — the exact trigger and the exact execution. The condition that makes you buy, stated with numbers. Then: at what price, with what order type, at what time of day. "Buy on a breakout" is not an entry. "Buy with a limit order at the open of the next session, if today's close is above the highest close of the last 50 sessions" is an entry.

3. Filter — what stops you from taking a valid signal. This is the component people leave out and it is often the one that matters most. A filter is a condition that must also be true. Common ones: only take long signals when the index is above its own long-term average; do not enter within 2 days of a scheduled results announcement; no more than 2 open positions in the same sector.

4. Exit — how a winning position ends. A target, a trailing rule, a time limit, or a condition that says the reason for the trade has gone. Most people specify the entry in great detail and leave the exit to judgement, which means the entry is being tested and the exit is not.

5. Stop — how a losing position ends. Where it sits when you enter, whether it moves, and what happens if the price opens past it. That last part is not a detail. On a bad day the stop does not execute at your level, and a system that has not stated what to do in that case has a hole exactly where the largest losses live.

6. Size — how much. Expressed as a rule, not a feeling. The most common professional form is: risk a fixed small percentage of capital on each position, and let the distance to the stop determine the number of shares. A position with a wide stop is smaller. A position with a tight stop is larger. This is the component that determines whether you survive, and it is the one retail traders spend the least time on.

Now the design principle that governs all 6.

Every number in your system is a parameter, and every parameter is a chance to fit the past.

Count them. "The highest close of the last 50 sessions" is 1 parameter. "Exit after 20 days" is another. "Only when the index is above its 200-day average" is a third. A system with 3 parameters is testable. A system with 12 is a description of history.

Fewer parameters are more robust for a reason you can feel: with 12 dials, you can make almost any set of past prices look profitable, so a profitable result tells you almost nothing. With 2 dials, you cannot, so a profitable result means something.

A useful discipline: each parameter should be justified by a mechanism you can say out loud, not by a test result. "50 days because that is roughly a quarter, and quarterly reporting drives repricing" is a justification. "50 days because 50 tested better than 45" is a fitted number, and article 6 in this cluster explains why that is worse than useless.

Discretionary against mechanical. This is a real trade-off, not a moral question.

Skilled discretionary traders exist and some are excellent. But a discretionary approach cannot be tested, which means it cannot be improved except slowly and expensively, and it cannot be distinguished from luck for many years. If you are going to be discretionary, be discretionary about a written system rather than instead of one.

What it tells you, and what it does not

A written system tells you what you are doing. That is more valuable than it sounds and it is genuinely rare.

It tells you what a normal losing streak looks like, so you can tell one from a failure.

It does not tell you the rules are good. Writing a bad rule precisely produces a precisely bad rule. The specification is what makes testing possible; it is not the test.

It does not make you follow it. Every serious study of mechanical trading finds that the gap between the system's results and the trader's results is caused by the trader overriding the system, almost always after a loss.

And it does not remove judgement. Judgement moved. It now lives in the design decisions, made calmly, in advance, instead of in the execution decisions, made under pressure, during a drawdown. That is the entire benefit and it is enough.

The decision rule

A rule is finished when you can hand it to somebody who has never traded, and they can execute it for a week without contacting you.

Until then it is a preference.

Before any money is committed, the document must state 5 things: what makes a trade eligible; what triggers it; what ends it in profit; what ends it in loss; and how large it is.

And 1 more, which almost nobody writes: what would make me stop trading this system entirely. State the loss, or the period without a new high, that ends it. Write it before you start, because you will not be able to write it honestly later.

Try this now

Take the sheet you started in article 3, where you wrote your entry rule. If you have not done that, do it now — it takes 2 minutes. Then complete the other 5 components.

1. Setup. Write the eligible universe in 1 sentence, with a minimum daily traded value. Example shape: "Companies in the [index], with an average daily traded value above [amount], excluding anything listed less than 1 year."
2. Filter. Write 1 condition that must also be true. If you cannot think of one, write "none" — that is an honest answer and it is also a parameter you have chosen.
3. Exit. Write the exact condition that closes a winning position.
4. Stop. Write where the stop sits at entry, whether it moves, and what you do if the price opens past it.
5. Size. Write the rule that determines how many shares. If your honest answer is "about a fifth of my capital, more if I like it", write that down too. It is a real rule and now you can see it.
6. Now count every number in the document. Every threshold, every period, every percentage. Write the total at the top of the sheet.
7. Finally, read the whole thing back and circle every word that requires judgement: strong, clean, confirmed, holding, looks, seems.

What you should see. Two things, and both are the point.

Your parameter count from step 6 is probably between 4 and 10. Anything above 6 means you will need substantially more data to test it honestly, and article 6 explains why.

And you will almost certainly have circled words in step 7 — usually in the exit and the stop, rarely in the entry. That asymmetry is the finding. Most traders have a precise entry and a vague exit, which means the part they have thought hardest about is the part that matters least. The entry decides whether you are in. The exit and the size decide what you make.

Keep this document. It is the input to every remaining article in this cluster.

Three real cases

1. The Turtle rules, 1983 to 1984 (United States) — a complete written system, handed to beginners Two commodity traders settled a disagreement about whether trading could be taught by recruiting people with no market background and giving them a fully specified rule set. The rules covered the entire 6 components: which markets were eligible, a breakout entry with a stated lookback, a filter on recent trades, a trailing exit, a volatility-based stop, and a position size derived from volatility so that each position risked a similar amount. Several participants went on to long careers. Reported outcomes within the group varied a great deal, and the variation is attributed to how closely each person followed the rules rather than to differences in the rules themselves. The rules were later published openly, which is the second lesson: a real edge is usually not destroyed by publication, because most people cannot follow it.

2. The factor zoo, documented 2016 (United States) — what happens when everybody adds parameters Researchers surveyed the academic literature and counted several hundred published factors claimed to predict share returns. They argued that given how many had been tested, the usual statistical threshold for calling a result significant was far too weak, and proposed a substantially higher bar. The lesson transfers directly to system design. Every extra condition you add is another search, and the published academic literature has the same problem as your spreadsheet, at a larger scale and with better mathematics.

3. SEBI's studies of individual derivatives traders (India) — what unwritten systems produce at scale India's regulator has published studies examining the profit and loss of individual traders in the equity derivatives segment. The published findings report that a large majority lost money over the periods studied, with substantial aggregate losses. This is not a backtest. It is the recorded outcome of millions of real accounts. Very few of those accounts were operating a written specification with a stated size rule and a stated stopping condition, and that is the closest thing available to a controlled experiment on the value of writing it down.

The question that resolves it

A novice asks: what are my rules?

An expert asks: where is the document, and when did I last change it?

The first question can be answered from memory, and memory rewrites itself after every trade. The second requires a file with a date on it. The difference between a trader who improves over 10 years and one who does not is very often just that file.

What would make this wrong

If written systems were unnecessary, then experienced discretionary traders would show more stable results than mechanical ones over long periods. Where comparable data exists, the opposite is generally reported. If you find good evidence the other way, this article is too confident.

The honest limits are 3.

First, writing it down does not make it profitable. A precisely written bad system loses money precisely. Most of the value here is diagnostic, not generative.

Second, some genuine skill does not reduce to rules. Reading a company's management, judging whether an industry is changing, sensing when a market's character has shifted — these resist specification and they are real. Discretionary investors who work this way are not deluded. But they should be honest that they cannot test their approach and must therefore be much more conservative about size.

Third, an over-specified system is brittle. A rule with 15 conditions will stop working when 1 of them becomes irrelevant, and you will not know which. The target is not maximum precision. It is enough precision to be executable and few enough parameters to be testable.

Carry this

• 6 components: setup, entry, filter, exit, stop, size. Missing 1 is missing a system.
• Count your parameters. Every number is a chance to fit the past.
• The stranger test: could somebody execute this for a week without calling you?
• Write the stopping condition before you start, because you cannot write it honestly during a drawdown.

Knowledge check

Related

• Backtesting and system design
• Optimisation, robustness and Monte Carlo
• Performance metrics: Sharpe, Sortino, expectancy and drawdown
• ← Sentiment, cycles and system testing