Improvement work starts with attribution, not ideas. Before touching a rule we split the trade history by time of day, day of week, regime, holding period, instrument and signal type, and find out where the strategy actually earns and where it quietly bleeds.
Typical findings
- A filter that removes three quarters of the trades, and most of the profit with them.
- A stop set so tight it converts winning days into a fee-generation machine.
- Entries clustered in the first fifteen minutes, where measured slippage is three times the daily average.
- A pyramiding rule that only helps in one volatility regime and costs money in the other.
- A carry or overnight rule whose entire edge came from two events in four years.
How changes are accepted
Each proposed change is a hypothesis with a test attached and a fold it must survive. If it does not, it is rejected and written down as rejected, with the numbers — so nobody re-proposes it next quarter. A typical improvement engagement rejects two thirds of the ideas it generates, including ours.
What we will not do
We will not tune parameters until the equity curve looks pretty. Curve-fitting is easy, fast, and the reason many strategies die three weeks after they go live.