Rolling-window study

Rolling-window regime study

For each ticker, slides a 1-year window forward one week at a time across the chosen span, runs the strategy in every window, and plots strategy CAGR − buy&hold CAGR against the window's own return. The goal is to see in what environment (rising / flat / falling) and in what periods the ladder helps or hurts.

Tickers comma-separated Span start Span end blank = start + 3 years Data source Interval Strategy mode

Strategy parameters

Total capital (A) First-step C levels (%) e.g. 70, 90 Steps (N) Threshold % (X)

Reserve in QQQ hold each window's un-deployed reserve in QQQ instead of cash (needs QQQ data over the span)

Each C level is run in every window. Overlapping windows are highly autocorrelated — read the cloud as illustrative, not as an independent sample.

Rolling studies

+ New study

#	Tickers	Span	Mode	Windows	C levels	Median Δ	Corr(Δ,mkt)
3	AMZN, CSCO, NFLX, GOOG, META, NVDA, GM, TSLA, AAPL, COST	2018-01-01 → 2026-06-14	scale	3880	70, 90%	+0.88 pp	-0.93
2	AMZN, CSCO, NFLX, GOOG, META, NVDA, GM, TSLA, AAPL, COST	2018-01-01 → 2026-06-14	scale	3880	70, 90%	-4.29 pp	-0.96
1	AAPL, ORCL, COST, META	2023-06-04 → 2026-06-04	lifo	420	70, 90%	-6.49 pp	-0.99

Median Δ & correlation are for the first C level. Negative correlation ⇒ the strategy is defensive (helps when the market falls, lags when it rises).