ALMAZOV TRADE
ALMAZOV TRADE
AccountDocumentationContact the managerEN/RU
← Research & Development

Model research

Choosing settings for a backtest

When tuning hyperparameters, an isolated record that vanishes with a small change can signal overfitting. We refine the search where neighbouring settings score well and test the chosen settings on fresh data.

Compare at the same risk

In May 2025 we scaled our equity curves to the same historical drawdown. This let us compare what each configuration earned at that level of risk.

Move a setting by one bar

In this example, A sets the moving average period and B sets the position holding period. Select the peak or the plateau to see what happens when either setting moves.

2,665 parameter pairs
Score · height and colour
04080120
A narrow spike reaches 120. To its right, a lime outline marks the plateau selected for further testing.
Selected parameter score82
Point on the chart
Synthetic illustration · scores, not returns.

Change the period

In March 2026 we found that the model behaved differently before and after the period used for tuning. That led us to revisit how we selected the parameters. Holding out the next period before the search gives the chosen settings an independent test.