SHAPE COMPARISON
EXPERIMENT 001 / BITCOIN
Time repeats.
Does the pattern?
Eleven years. One calendar.
Find the shape inside the signal.
Loading 4,018 daily observations…
JAN 1 → DEC 31 / aligned by month & day
01 / ISOLATE
Same frame. Different years.
Each line sits in its own box: January 1 to December 31,
annual low to annual high. Price magnitude falls away.
02 / COMPARE
The anatomy of a year.
OBSERVATIONS COMPUTED FROM DATA
2017 & 2020
Both end with a steep late-year ascent. Their shared outline hides very different returns and drawdowns. Test whether acceleration before October predicts stronger fourth-quarter returns in later, untouched years.
2018 & 2022
Both move from high early prices toward low year-end prices, with interrupted declines. Four years apart is a hypothesis-generating observation; two examples cannot establish a four-year cycle.
2019 & 2025
Both rise toward a high region, then retreat before year end. Peak timing differs. Their price-shape resemblance does not carry through to daily-return correlation. Compare turning-point timing as a separate feature.
How to read this experiment
Daily PriceUSD observations from Coin Metrics Community Data. Returns use the January 1 observation as the starting value, not the previous December 31. Extremes are daily sampled prices, not intraday highs and lows. The chart aligns month and day on a leap-year calendar; non-leap years connect February 28 to March 1.
Isolated y = (price − annual minimum) / (annual maximum − annual minimum). No rotation, time shifting, or reversal is applied. Shape comparisons use Pearson correlation of 365 matching calendar dates (February 29 excluded). Correlation of price levels can be high simply because both years trend upward. It does not establish periodicity, causation, or forecasting ability. The correlation of daily log returns is reported separately on 363 matching dates, excluding February 29 and March 1 to keep all intervals one day long across leap and non-leap years.
Annual min–max normalization uses information from the entire year, including its future. This is an exploratory retrospective view, not a usable real-time predictor. A predictive test needs a rule frozen on earlier data, later untouched years, a baseline, and a record of all attempted rules. Eleven annual samples are limited evidence.
03 / RECORD
From resemblance to a test.
Notes save in this browser. Export to keep a portable copy.