# Clenow Weekly Momentum Rotation — System Test

A survivorship-bias-free evaluation of Andreas Clenow's weekly momentum rotation
(*Stocks on the Move*, 2015; *Trading Evolved*, 2019) on the Nasdaq-100 and S&P 500,
on split- and dividend-adjusted prices. The full write-up — methodology, the
random-portfolio (monkey) baseline, the sizing ablation, metrics-by-lookback, the
sizing out-of-sample test, exposure analysis, and the annotated sample chart — is the
[**Deep Dive (`index.html`)**](index.html), which is the authoritative version.
This file is the headline summary.

## What it is

Each week, rank the large-cap members by the annualized slope × R² of a 90-day log-price
regression and hold the top 20; require each name above its 100-day SMA with a positive
slope and no recent >15% gap; open new positions only when the index ETF is above its
200-day SMA; size inverse to ATR(14); exit on a broken trend, a gap, or a rank drop.

## Setup

- **Universe:** point-in-time, survivorship-bias-free index membership (Nasdaq-100
  ~286 ever-members; S&P 500 ~959), resolved per trading date.
- **Window:** 2005-01-03 → 2026-05-15 (21.4 years). Starting capital $100,000.
- **Prices:** split- and dividend-adjusted (ratio reconstruction from adjusted close)
  before any indicator, filter, fill price, or P&L — so a stock split is not mistaken
  for an overnight gap.
- **Disclosure / OOS pivot:** published 2019; conservative OOS pivot 2021.
- Commissions, slippage, and interest on idle cash are not modeled.

## Headline results

| Metric | Nasdaq-100 system | Buy & hold QQQ | S&P 500 system | Buy & hold SPY |
|---|---|---|---|---|
| CAGR | 11.43% | 15.34% | 9.31% | 10.88% |
| Max drawdown | −29.81% | −53.40% | −22.74% | −55.19% |
| Sharpe | 0.77 | 0.77 | 0.65 | 0.64 |
| Rolling 36m Calmar (median) | 0.64 | 0.90 | 0.48 | 0.62 |
| Final equity ($100k) | $1,009,709 | $2,109,756 | $669,487 | $908,276 |
| N trades / win rate / PF | 1,356 / 46.9% / 1.42 | — | 1,996 / 48.6% / 1.23 | — |

## Findings

- **Near-index return at roughly half the drawdown.** The system captured most of the
  index's return while cutting the worst-case drawdown by 24–32 points, with Sharpe equal
  to buy-and-hold on the Nasdaq-100 and slightly above it on the S&P 500. Worst 12-month
  stretch: −26.1% / −15.8% vs buy-and-hold −48.0% / −47.4%.
- **The stock-picking beats a random portfolio.** Against 200 random portfolios run
  through the identical rules, the momentum ranking lands at the 98th–99th percentile on
  CAGR (a +3.5 pp/yr increment over the random median) on both universes; it carries a
  deeper-than-typical drawdown (the cost of concentrating into names that trend together).
- **The edge is concentrated in bear markets.** In 100% of rolling 12-month windows where
  the index fell more than 20%, the system returned more than buy-and-hold — by ~15 points
  (Nasdaq-100) to ~29 points (S&P 500). It captures most of the index's gain in strong
  bulls.
- **Sizing vs. selection (ablation).** Swapping ATR sizing for equal weight lifts CAGR
  (NDX 12.07%, SPX 11.04%) and deepens drawdown; under both weightings the ranked book's
  drawdown is a low-percentile outlier, so the deep drawdown follows the selection, not
  the sizing.
- **Sizing out-of-sample.** The Bandy safe-f fraction fitted pre-2021 is stable across the
  pivot and its drawdown-95 stays within the −20% target out-of-sample — the sizing is not
  overfit.
- **Exposure.** The book holds a position on ~100% of trading days (dollar deployment
  ~85–86%), so its exposure-adjusted return equals its CAGR; deployment is cut to ~38–52%
  in 2008 and 2022.

## Reproduce

```
python prototype.py --index QQQ_Nasdaq-100 --regime-etf QQQ --start 2005-01-01 --end 2026-05-15
python prototype.py --index SPY_SandP_500  --regime-etf SPY --start 2005-01-01 --end 2026-05-15
python monkey_baseline.py --index QQQ_Nasdaq-100 --regime-etf QQQ --k 200
python ablation_weight.py --index QQQ_Nasdaq-100 --regime-etf QQQ --k 200
python refinements.py
```

The data loader and the survivorship-bias-free dataset are internal to the research
platform and not redistributed; the engine and analysis code are complete in `code/`.

Results are historical simulations, not investment advice.