90-day forecasts using only earlier observations
At each January 1 with at least 1,095 daily observations, we fit the
site's power-law curve to data through that date and extrapolate 90
days. We compare that price with the recorded price 90 days later.
The simple baseline predicts that the price will remain unchanged
from January 1. Neither forecast uses observations after the origin
date during fitting.
Median absolute percentage error; lower is better
| Asset |
Data through |
Forecasts |
Power-law error |
Unchanged-price error |
Model wins |
| Bitcoin |
2026-10-20 |
13 |
42.9% |
27.9% |
4/13 |
| Ethereum |
2026-10-20 |
8 |
62.5% |
34.8% |
2/8 |
| Solana |
2026-10-20 |
3 |
67.6% |
53.0% |
1/3 |
| XRP |
2026-10-20 |
8 |
42.3% |
14.7% |
3/8 |
| Dogecoin |
2026-10-20 |
9 |
84.3% |
37.4% |
2/9 |
Error is 100 × |forecast − actual| / actual. We report the median
across each asset's eligible annual forecasts. Model wins count
cases where its absolute error was strictly smaller than the
baseline's; ties are not wins. Missing origin or target observations
are skipped, and evaluation windows do not overlap.
Download every forecast and error (JSON)
·
Read the evaluation code ·
Why time-series evaluation must exclude future data
What this comparison does and does not establish
This tests a 90-day extrapolation of the fitted central trend, not
the profitability of trading rainbow bands. Large errors or losses
against the simple baseline are evidence against treating that curve
as a reliable price forecast in this sample. A low median error
alone does not establish future accuracy.
The five assets were selected today, so this comparison has survivor
selection bias. It uses the current archive, which can contain
revised data and mixed providers, rather than a preserved copy of
what every provider published at each historical date. There are few
annual observations per asset, and results depend on the horizon and
origin schedule. We have not searched across horizons to select a
favorable result.
The chart pages' year-end examples instead use the full-history fit
and include later data. Those illustrations are intentionally
separate from this evaluation. Neither page presents a live trading
record, includes transaction costs, or promises future returns.
What is tracked
cryptorainbowcharts.com tracks historical daily price observations,
fitted power-law trend values, current rainbow bands, data ranges,
and model diagnostics for each supported asset.
The site records model context rather than trade calls. A rainbow
band is a historical valuation zone, not a buy signal, sell signal,
forecast, or guarantee.
How to judge it
A useful track record for this site is whether chart data is updated
consistently, whether pages disclose their data range, and whether
the model remains transparent enough to inspect.
Short histories, stablecoins, wrapped assets, tokenized funds, and
changing market regimes can make the fitted bands less informative.