Understand what the software is optimizing
A tool can implement target weights faithfully while producing a return below a concentrated holding. It may also fail to keep the intended exposure because of minimum orders, unavailable assets or stale data. Inspect its objective and execution record separately. An automatic action is not evidence of a profitable rule.
Absolute and relative drift are different
At a 20% target, a current weight of 25% is five percentage points above target. Relative to the target, the deviation is 25%, calculated as (25−20)/20. A setting labelled “5%” is ambiguous until the documentation defines it. A five-percentage-point band and a 5%-relative band would trigger at different weights.
Check turnover and the cost budget
In a hypothetical month, total one-way traded notional of 8,000 at an entered 0.2% commission costs 16. Add only spread, slippage or other costs not already counted in the execution result. Do not label a quoted commission as the complete cost. A minimum order or rounding rule can leave residual drift even after a completed rebalance.
Test the order lifecycle
- Preview the proposed holdings and cash before approval.
- Test insufficient funds, partial fills and rejected orders.
- Confirm whether pending orders are included in exposure.
- Restart during a test and check for duplicate instructions.
- Verify what happens when a symbol is unavailable or a price is stale.
- Distinguish stopping new orders from closing existing exposure.
Use a test environment where supported, retain logs and compare the actual ending positions with the intended positions. A demo exercise checks behavior under its assumptions; it does not establish live fill quality or future returns.
Keep selection and evaluation apart
If you choose a band because it performed best in one historical window, that window is development data. Freeze the rule before testing another period. Compare risk, turnover and costs alongside return. No universal band or momentum setting is recommended here.
Sources and scope
References support the stated definitions or research context. The worked examples on this page use invented inputs and do not reproduce a market study. Source checks: 16 October 2026.
Continue with a practical tool
Revision published . Historical author labels are explained on the About page.