advanced practice · system design prompt
Design an honest momentum backtest
Outline a 12-month momentum rule on 3 ETFs: train/holdout split, benchmark (SPY), transaction cost assumption, metrics (CAGR, Sharpe, max DD), and three ways your backtest could lie to you.
The mechanism
Backtest = hypothesis on history. Pre-register rule, holdout/walk-forward, benchmark, document costs. In-sample-only edge is curve-fit.
Expected evidence
Written protocol before code; explicit holdout; limitations section mandatory.
Try this drill in the practice workspace →
Hints when you need them
- Pre-register lookback and rebalance frequency.
- Survivorship and snooping are top lies.
Watch for these mistakes
- Optimizing lookback on full history
- Survivorship bias
- No benchmark