AITOUZI / QUANT BACKTEST METHODOLOGY
Backtests: inspect bias before returns
A backtest shows what would have happened under specified data, rules and execution assumptions. It is not a live-performance promise. Inspect point-in-time availability, universe selection, execution timing, costs, drawdowns and benchmark consistency.
Three common sources of bias
Using unpublished filings creates look-ahead bias. Selecting today's successful companies creates survivorship or hindsight bias. Repeated tuning creates overfitting. A fixed six-stock basket needs an explicit hindsight-selection warning, not a claim of unbiased market-wide validation.
Use the same ruler for strategy and benchmark
Align dates, adjustment and dividend treatment; include turnover costs, slippage and cash handling. A signal known after close cannot assume frictionless execution at that same close. Read annualized return with drawdown, trade count and market regimes. Use the assumptions shown in each result.
Reading checklist
| What to check | How to interpret it |
|---|---|
| Annualized return | Use start/end equity and elapsed time, not a sum of daily returns. |
| Maximum drawdown | Worst decline from a prior equity peak, not worst daily return. |
| Out of sample | Freeze rules before validation; do not silently change a poor test window. |
Limits & counterexamples
Beating an index in one period does not establish durability. Missing delistings, filing timestamps, feasible execution or licensed coverage should weaken conclusions, not be filled with invented data.
Sources & verification
For research and education only, not a securities recommendation or personalized investment advice. Historical results do not predict future performance.