Building a strategy: what to decide before you write a single rule

What a strategy is

A strategy is four decisions: when to enter, when to exit, how much to risk, and what stops the bleeding when you are wrong. Remove any one and the other three cannot save you. Most beginners spend all their effort on the entry. The entry is the least important of the four. Read that again, because nobody believes it the first time.

A strategy exists to harvest an edge. An edge is a small statistical tilt that shows up over hundreds of trades. One good trade is not an edge. Ten good trades are barely a hint. The trader's equation only pays across a sample, and a strategy is the machine that produces the sample. See the trader's equation.

This help center is about systematic trading. Every rule written down, every decision repeatable. If a rule lives in your head, it cannot be tested, and an untested rule is a guess with confidence.

The five questions

Answer these before writing any rule. If you cannot, you are not behind. Answering them is the actual work.

1. What is the edge hypothesis? Why should this make money? What inefficiency does it harvest? “Buy when RSI is below 30” is a rule, not a hypothesis. The hypothesis underneath might be “this market mean-reverts after panic selling.” That can be tested, and it tells you when the rule should fail. A rule without a hypothesis fails silently, and you will not know why.

2. Which asset, venue, and timeframe? Edges are local. BTC on 4-hour bars and SOL on 1-minute bars are different games with different players and different costs. A pattern that is real in one is noise in the other. Pick deliberately. Let the hypothesis pick.

3. Which regime? Trending, ranging, high volatility, low volatility. Pick one. A strategy for all markets is a strategy for none. The trend strategy bleeds in chop. The mean-reverter dies in a trend. This is not a flaw to fix. It is a fact to plan around. See market regimes.

4. How do you exit? Most retail strategies are 80 percent entry and 20 percent guesswork. The ratio is backwards. The exit decides how much of each correct call you keep and how much of each wrong call you eat. The exit should come from the hypothesis. If the thesis is “panic mean-reverts within a day,” the exit is the reversion completing or the day ending. Not a 2 percent take-profit you picked because 2 is a round number.

5. How do you size? Sizing does not create edge. It decides whether the edge survives its worst drawdown, and whether you do. Fixed notional, fixed fractional, volatility-targeted. The options are in risk. Pick one on purpose.

The traps

Each of these feels like diligence while you do it. That is what makes them traps.

• Indicator stacking. Add a filter, the backtest improves, add another. Each condition fits the rules tighter to one stretch of history. You are not refining a strategy. You are memorizing a chart. See overfitting.
• No exit, just a stop. A stop is survival, not a thesis. If trades only end at the stop or when you get bored, the strategy has no opinion about when its edge is spent.
• Asset hopping. Twenty pairs, report the best. The best of twenty random results is always good. That is selection, not validation. See single-pair backtests.
• Timeframe shopping. Same logic, six timeframes, pick the green one. Same trap, different axis.

What good looks like

Before the first backtest, a real strategy has:

• A hypothesis in one sentence that a stranger could repeat back.
• A named regime, venue, and timeframe, and a clear answer to “when should this not be running?”
• An exit that follows from the hypothesis.
• A sizing rule that survives a drawdown worse than the one you imagined, because the real one will be worse.

With those four, a backtest can teach you something. Without them, the backtest will tell you what you want to hear, and you will believe it.

How Edgecraft handles this

Describe an idea to Edgecraft, and the agent asks for these answers first. Hypothesis, venue, timeframe, regime. Then it generates the code, with all four parts present from the first draft. The strategy library carries reference strategies with every part written out and properly described so you can see a complete strategy before building one. The system allows you to start as vague as possible and will guide you through the process to reach a fully defined architecture, but it also allows you to start with a pre-built personal strategy as complex as you’d like. The agent will translate natural language, Pine code, or Python and will build it for you in seconds.