20-day return: what it was followed by

Data as of 2026-10-16

What an extreme 20-day return reading was historically followed by, measured across the stocks this site covers.

How to read this. These figures are measured only on instruments currently covered by this site — a set of companies that survived. Signals that appear to have ‘worked’ on the downside are conditioned on these names having recovered; a company that fell and never came back would not be here to measure. Read every number below as a property of this surviving sample, not a universal rule, and never as advice.

All figures are ‘excess’ returns — measured relative to each instrument’s own historical drift, not relative to the market. This is not alpha and not a market-beating claim; it isolates the signal from the instrument’s background trend. Measured from the first day each signal appeared (not its most extreme point), so it reflects what was knowable in real time.

Overbought / high extreme

Across 2839 past episodes where a stock was at a high extreme on this signal, the 2836 with a complete 20-session forward window lagged the stock’s own trailing drift by 0.01 percentage points on average (48% of those 2836 beat drift).

the stock forward excess return vs its own drift after high-extreme 20-day return readings, by horizon.
horizon	excess vs own drift	beat-drift rate	episodes with a complete horizon
1 session	-0.18%	41%	2839
5 sessions	-0.37%	44%	2839
20 sessions	-0.01%	48%	2836

Oversold / low extreme

Across 2768 past episodes where a stock was at a low extreme on this signal, the 2763 with a complete 20-session forward window beat the stock’s own trailing drift by 0.14 percentage points on average (53% of those 2763 beat drift).

the stock forward excess return vs its own drift after low-extreme 20-day return readings, by horizon.
horizon	excess vs own drift	beat-drift rate	episodes with a complete horizon
1 session	+0.26%	50%	2768
5 sessions	+0.31%	52%	2768
20 sessions	+0.14%	53%	2763

By instrument

Each instrument’s own history on this signal (where it has enough episodes to characterise):