Markets have
memories.
Find familiar conditions.
Explore unfamiliar outcomes.
A pullback. A quiet rally. A sudden shift. Find earlier moments with similar conditions and see the different paths they took.
Looking through the archive.
Loading daily adjusted prices for SPY and its market benchmark.
Does resemblance
actually help?
A convincing chart is only the beginning. Revisit up to 60 historical dates and compare each group’s median outcome with a simple baseline.
Let the evidence answer.
Both estimates are judged on the same dates. The check reports when the model has too few matches, and when the baseline does just as well.
Same lens · Earlier data only · Visible failuresNothing up our sleeve.
Market Memory is a nearest-neighbor experiment. It finds resemblance in a few measurable conditions; it cannot establish that history will repeat.
How are matches chosen?
We compare 20-session momentum, distance from the 60-session average, 20-session annualized volatility, and 20-session return minus SPY’s return. Inputs are scaled using only eligible earlier history. Trend’s weight is split across its two inputs. Constant inputs are ignored.
Up to 12 episodes qualify at a weighted RMS distance of 0.75 or less. That fixed research cutoff is not a confidence score. Each match’s entire 60-session lead-in and 20-session outcome is separate from every other selected match and from the selected situation’s lead-in.
What is the baseline?
A chronological sample of the same asset’s eligible history, spaced 81 sessions apart without selecting for resemblance or returns. Its middle 80% is the 10th to 90th percentile. The sample sizes differ: — baseline episodes and — matches. These are descriptive distributions, not statistical significance tests.
What can this experiment tell us?
The historical check compares median-return estimation error. It does not backtest an executable strategy, include trading costs, or demonstrate a profitable edge. Repeatedly adjusting the lens while inspecting results can overfit the experiment. A fixed, untouched future test is still needed before making predictive claims.
Where do the prices come from?
Daily adjusted closes from Yahoo Finance, using SPY’s observed sessions as the calendar. Adjusted prices reflect provider adjustments for distributions and splits. Missing values are never filled, and windows with missing observations are excluded. Today’s potentially incomplete bar is excluded.
Historical queries use a current revised data snapshot, not a point-in-time vendor archive. The 25 ETF quick picks are starting points; ticker search also accepts supported USD-priced stocks and ETFs with at least 500 shared sessions. Each asset is compared only with its own history. Currently available assets are not a survivorship-free sample of all securities.