CryptoChemistry.net builds analytical models on top of complete daily price histories - for Bitcoin from its earliest traded prices, for other assets from their first reliable market data, and for the S&P 500 back to 1927. This page explains, at a high level, how the main model families work, how they are calibrated, and - just as importantly - what they can and cannot tell you.

1. Data foundation

Every model runs on daily candles (open, high, low, close, volume) normalised to UTC days and stored in our own database. Series are checked for continuity and validity before models consume them: crypto series cover every calendar day, while stock-market series deliberately contain trading days only (no weekends or exchange holidays), and models built on them respect that calendar.

2. Quantile logarithmic regression

The core long-horizon model is an asymmetric quantile regression fan fitted in logarithmic price space over the asset’s full history. Instead of one “fair value” line, it estimates a family of percentile bands (from the 5th to the 95th percentile), so the question it answers is not “what should the price be?” but “where does today’s price sit inside the asset’s own long-term growth channel?” The fit is refreshed daily as new data arrives, and the same core is reused across assets - each asset is calibrated on its own history, never on Bitcoin’s.

Derived views include the fair-value (median) curve, valuation “rainbow” zones, and the percentile position readings used across the dashboard.

3. Risk metrics

Risk metrics compress a signal into a 0–1 score, where values near 0 have historically corresponded to undervaluation and values near 1 to overheated conditions. The normalisation uses the asset’s own historical distribution (quantile clamping followed by a smooth non-linear mapping), so a reading is always relative to that asset’s past - not an absolute prediction.

The Universal Risk Metric blends several independent normalised components (price-based risk, regression position, cycle timing, long-term moving-average extension, and others) into a single cycle thermometer. Components can be toggled individually, and the blend is a plain average of the enabled components - no hidden weighting.

4. Cycle analytics

Cycle definitions: market cycles are anchored to historically observed intraday extremes - a cycle is named after the peak that ends it, tops use the daily candle high and bottoms the daily candle low. The current cycle’s bottom remains provisional until a new cycle confirms it.
Cycle ROI: return profiles are measured from each cycle’s bottom (recovery shape) and from each peak (drawdown depth and time to reclaim the prior high), so cycles of different eras can be compared on a “days since anchor” axis.
Cycle projection: projections are built from historical cycle analogues (four- and eight-year rhymes; for the S&P 500, the US election cycle) replayed in the percentile space of the regression fan rather than in raw prices - this respects the long-term trend of diminishing returns. Projected paths are rendered as deliberately faded “ghost” candles: low visual weight encodes low certainty.

5. Long-term valuation models

Alongside the regression fan we maintain a set of widely known long-horizon models, each implemented and refitted on our own data: the 200-week moving-average heatmap (price coloured by its ratio to the 200-week average), moving-average support bands, mining-economics multiples, scarcity models, and long-term corridor models. Where a published model has demonstrably broken down (for example the stock-to-flow model after 2021), we show the divergence honestly instead of hiding it.

6. Technical indicators

Indicator pages (moving averages, RSI, MACD, Bollinger Bands, bull market support band and others) use the standard textbook definitions - for example Wilder’s 14-period RSI - computed on our daily series. They are provided as context, not as trading signals.

7. Calibration and validation

Model parameters are chosen by comparing variants against real historical data, not by taste. Before a model ships, it must pass landmark validation: reproduce the well-known historical extremes of the asset it models (for Bitcoin, the 2013/2017/2021/2025 peaks and the 2015/2018/2020/2022 bottoms) from data alone. A model that cannot describe the past is not allowed to draw the future.

8. Limitations - what these models are not

Not forecasts. Every projection assumes that the future rhymes with the past. It may not. Structural breaks - new market participants, regulation, macro shocks - can and do invalidate historical analogues.
Descriptive, not prescriptive. A risk reading of 0.9 says the price is high relative to its own history; it does not say the price will fall, or when.
Data-bound. Model outputs inherit the cadence and latency of the underlying data - most series update once per day, and some readings are cached, so a panel can briefly show a slightly older value than a live chart next to it.
Survivorship and regime caveats. Younger assets have fewer completed cycles, so their models rest on thinner evidence than Bitcoin’s or the S&P 500’s.

9. Disclaimer

Data and model outputs are provided for informational purposes and may be delayed. They do not constitute investment advice. See also our Terms of Use and Privacy Policy.