Turning an idea into precise, testable rules

What you will learn

• How to write a trading idea as a hypothesis with a reason behind it.
• A rule template that covers every decision, including ties and missing data.
• How to time signals and orders so that no future information is used.
• How to count the free parameters in a rule set.

Explanation

The basic yes/no test for rules was covered in the course "Technical Analysis in Depth". Here you write a full specification: a document so complete that two people who never talk to each other would produce the same list of trades from the same data.

Step 1: the hypothesis

A hypothesis is a testable statement about market behaviour plus a reason. Example: "Broad equity indices that are above their long-term average tend to keep rising more often than not, because trends in investor flows and economic news persist for months." The reason matters: it tells you what should not work if the idea is true, and it protects you from testing random patterns.

Step 2: the rule template

Fill in every line. "Not needed" is an allowed answer; an empty line is not.

Step 3: timing without looking ahead

Look-ahead bias means using information that was not yet known at the decision time. The safest timing convention is: calculate the signal with data up to the close of bar t, and trade at the open (or close) of bar t+1. A rule such as "buy at today's close if today's close is above the average" is technically possible only if you can place the order in the last seconds of trading; most backtests should not assume that.

Step 4: count the parameters

A parameter is any number you could change: lookback lengths, thresholds, stop distances, check days. Write them in a list. Each extra parameter gives the test another way to fit past noise (lesson 6). A useful habit is to fix parameters from the hypothesis before looking at any results, for example "about one year, so 10 to 12 months".

Vague words to replace

• "Strong", "weak", "oversold" → a number and a formula.
• "Breaks out" → close above the highest close of the last N bars, by how much, on which bar.
• "Usually" or "if it looks good" → remove; a rule cannot contain judgement.

Worked example

Hypothetical example: the idea above becomes this specification.

• Universe: one broad equity index ETF. Data: monthly closes, adjusted for distributions.
• Signal: 10-month simple moving average (SMA), the average of the last 10 monthly closes including the current one.
• Schedule: last trading day of each month, after the close.
• Rule: if close > SMA, hold the ETF next month; if close < SMA, hold cash. If close = SMA exactly, keep the current state.
• Entry/exit: at the open of the first trading day of the next month. Sizing: 100% of the account. Parameters: 1 (the length 10).

Hypothetical closes for months 1–10: €100, €102, €99, €104, €107, €105, €110, €108, €112, €113.

• Sum = €1,060. SMA = €1,060 ÷ 10 = €106.00.
• Close €113 > €106.00 → buy at the next month's open.

Month 11 closes at €103. The new window is months 2–11: €1,060 − €100 + €103 = €1,063. SMA = €106.30. Close €103 < €106.30 → sell at the open of month 12. Costs, spreads and taxes are excluded here.

Common mistake

Practice activity

Rewrite this vague idea as a complete rule set on paper: "Buy a stock index when it is oversold in an uptrend and sell when it has recovered." Use the template above and count your parameters.

Key takeaways

• Start with a hypothesis and a reason, not with a pattern found on a chart.
• A complete specification lets two people produce identical trades from the same data.
• Calculate signals on bar t and trade on bar t+1 to avoid look-ahead bias.
• List every parameter and fix it before you see any results.

Next step

In the next lesson you will check the data that your rules depend on, because a perfect rule set tested on flawed data still gives misleading results.