Rolling-window study AMZN, CSCO, NFLX, GOOG, META, NVDA, GM, TSLA, AAPL, COST

2018-01-01 → 2026-06-14 · fmp · 1-year window, weekly step · scale · A=10,000 · N=8 · X=2.0% · C levels 70% / 90%

Δ = strategy CAGR − buy&hold CAGR, per 1-year window. Positive = the ladder beat buy&hold that year. Overlapping windows are autocorrelated, so read these as intuition, not statistics.

Summary (pooled across tickers)

C = 70%   (3880 windows)

C = 90%   (3880 windows)

What environment? (Δ vs the window's own buy&hold return)

Each dot is one 1-year window. Down-sloping cloud ⇒ the strategy gives up ground in rallies and protects in declines.

C = 70%

C = 90%

Regime buckets (median Δ in the down / flat / up thirds)

Windows split into thirds by their buy&hold return. If "Down" is green and "Up" is red, the strategy is a defensive, mean-reversion play.

When? (Δ over calendar time, per ticker)

Each point is one 1-year window's Δ (left axis). Solid lines are Δ vs the ticker's own buy&hold; dashed coloured lines are Δ vs S&P 500 / Nasdaq-100 CAGR (toggle any via the legend). By default a point sits at the window's start, so it reflects the year that follows it; switch a chart to mid/end to move its points. Price is the faint dashed line (right axis).