Everything in this series stays inert knowledge until you test your first hypothesis yourself. The four steps below are not a course — they are one complete work cycle you can run on the platform today: from a question you write down, to a result that holds or fails. Either way you have learned something measurable, which is the only thing that compounds into experience.
1. Write the hypothesis first
What to do. Write one sentence saying why the return exists and who pays it, then turn it into a numeric condition with no room for interpretation. "Trends persist because information is priced in gradually" is testable; "the stock looks strong" is an impression.
The common mistake. Starting from an indicator you liked and then hunting for a reason to justify it — that is thinking in reverse, which is why the structure collapses at the first honest test outside its own sample.
Start with one asset and one idea, not a basket of stacked indicators.
2. Test it on enough history
What to do. Choose a period containing both a bull and a bear market, run it with real costs, then read the whole dashboard — compound growth, volatility, maximum drawdown, profit factor — not the first shiny number alone.
The common mistake. Settling for a short rising window that flatters everything, then being surprised when the system collapses at the first real reversal.
Use the backtest lab — the same dashboard whose numbers ran through the earlier parts of this series.
3. Build a portfolio
What to do. Combine assets with low correlation, fix the weights and the rebalancing cadence in advance with a written rule, then read beta and the diversification ratio before the return. A portfolio is judged by its worst quarter, not its best year.
The common mistake. Diversifying by name rather than by correlation — five assets that move together on the day of stress are really one position with five labels, and five separate illusions of safety.
Use the cross-asset risk page — the correlation 0.15, diversification 1.50 example from the risk part.
4. Monitor, don't react
What to do. Compare live performance with what the backtest projected, and decide in advance what warrants a review — a breach of the expected drawdown, a persistent shift in correlations, or a change in market structure. Not one red day.
The common mistake. Editing the rule after every loss and re-fitting it to whatever just happened, so the strategy slowly becomes a discretionary decision wearing a numeric costume, and loses the discipline that defined it.
Do it by re-running the backtest periodically on new data: compare, don't patch immediately.
Three rules that outlast any strategy
- Start small. One hypothesis and one asset beat ten indicators combined.
- Always measure. What is not measured is not improved, and what is not recorded is not measured.
- Accept rejection. A hypothesis that fails in testing just saved you from losing on it live.
You don't need a brilliant model
You do not need a brilliant model to begin. You need a written rule, an honest test, and the discipline to execute what you proved. The return is a consequence, not a goal.
That closes the loop — and this series. It began with a single idea: decide by rule, not by impression. Everything since has been the machinery for doing exactly that. The only step left is the one that compounds — write your first hypothesis down, and run the cycle.