Bitcoin Price Prediction Models Explained

How power-law, stock-to-flow, machine-learning, and sentiment models try to predict Bitcoin's price — what each claims, its method, and its documented limits.

Updated June 2026 · Reviewed by the PipeFlare team

Bitcoin price models — power law, stock-to-flow, ML/LSTM, and sentiment analysis — each fit a pattern to past price data, and every one has well-documented failure points

Every cycle, a new chart claims to have 'solved' Bitcoin's price — understanding each model's actual method (and its public failures) is how you avoid false certainty

About bitcoin price prediction models explained

Every Bitcoin cycle produces a new chart claiming to have found the underlying formula for its price. The four recurring approaches are power-law growth models, the stock-to-flow model and its 'rainbow chart' variants, machine-learning models like LSTMs, and sentiment-analysis models. Each fits a mathematical or statistical pattern to Bitcoin's past price data and projects it forward. This page explains what each model actually claims, its stated method, and its well-documented limitations — it does not predict a price, and you should treat any page that does with real skepticism.

How it actually works

Power-law models, most prominently associated with physicist Giovanni Santostasi, plot the logarithm of Bitcoin's price against the logarithm of time since genesis. The resulting relationship looks close to a straight line, which the model's proponents argue reflects the same type of scaling behavior seen in network growth, city populations, and other systems that grow by compounding adoption. A 2026 paper describing the power law was published in the peer-reviewed journal Nonlinear Science, giving the approach more academic scrutiny than most crypto price models receive — though peer review of a historical curve fit is not the same as a validated predictive claim about the future. Stock-to-flow, popularized by the pseudonymous analyst PlanB starting in 2019, models Bitcoin's price as a function of its 'stock' (existing supply) divided by its 'flow' (new supply issued per year) — a ratio that jumps at each halving. The 'rainbow chart' is a popular visual variant that bands historical price into colored zones from 'fire sale' to 'maximum bubble.' Stock-to-flow's most cited real-world test came in 2021: PlanB's own worst-case model floor predicted Bitcoin would be at or above $98,000 for November and $135,000 for December 2021, but the price actually closed 2021 around $47,000 — well below either stated floor, prompting significant public criticism, including a well-known rebuke from Ethereum co-founder Vitalik Buterin about the danger of models that give false certainty. Machine-learning models, most commonly LSTMs (Long Short-Term Memory neural networks) trained on historical price and volume data, try to learn nonlinear patterns statistically rather than assume a fixed formula. Academic research on these models consistently documents a core weakness: they tend to perform very well on the historical data they were trained on and meaningfully worse on new, unseen data — a failure mode called overfitting, where the model has essentially memorized the past rather than learned a generalizable pattern. Sentiment-analysis models take a different angle entirely, scoring social media chatter, search trends, and news tone to gauge crowd psychology rather than fit a price curve — the crypto Fear and Greed Index (covered in a separate guide on this site) is the best-known public example of this category, though it is explicitly a sentiment gauge rather than a price-target model.

Start here

1Treat any single model as one lens, not a forecast — every model listed here has a well-documented period where it diverged sharply from actual price.
2Check whether a model has been tested against a real out-of-sample period (data it wasn't fit to) — this is the single best filter for separating a genuine attempt at rigor from a chart drawn to fit the past.
3Be specifically skeptical of models with a hard price 'floor' or 'target' by a named date — stock-to-flow's most public failure was exactly this kind of specific, falsifiable claim.
4If you want sentiment context rather than a price target, use a transparent, published tool like the crypto Fear and Greed Index rather than an opaque proprietary model.

Strengths

Power-law and stock-to-flow models are genuinely testable against history, which is more rigorous than a purely narrative-based price take.
The 2026 peer-reviewed publication of power-law research subjected the approach to more academic scrutiny than most crypto price frameworks receive.
Machine-learning approaches can, in principle, pick up nonlinear relationships a fixed formula would miss — useful for research even when the output isn't a reliable price forecast.

Common misunderstandings

Stock-to-flow's most specific, testable prediction publicly failed in 2021, when Bitcoin closed the year well below the model's own stated floor.
No model in this category has reliably predicted black-swan events — exchange collapses, regulatory shocks, or macro shifts — because none of them are designed to.
LSTM and other ML models are well documented in academic research to overfit historical data, performing far better in backtests than on genuinely new price action.

Common questions

What is the Bitcoin power law model?

The Bitcoin power law model, most associated with physicist Giovanni Santostasi, plots the logarithm of Bitcoin's price against the logarithm of time since its creation. The resulting pattern is close to a straight line across Bitcoin's full history, which proponents argue reflects the same scaling behavior seen in other systems that grow through compounding network adoption. A study describing the model was published in the peer-reviewed journal Nonlinear Science in 2026, though peer review of a historical curve fit does not validate it as a guaranteed predictor of future price.

Did the stock-to-flow model actually fail?

Its most specific, public prediction did. PlanB's stock-to-flow model projected a worst-case price floor of $98,000 for November 2021 and $135,000 for December 2021. Bitcoin actually closed 2021 around $47,000, well below either floor. PlanB had said he would consider the model invalidated if that floor wasn't met, but later walked that commitment back, which drew significant public criticism, including from Ethereum co-founder Vitalik Buterin.

Can machine learning actually predict Bitcoin's price?

Academic research on LSTM and other machine-learning models applied to Bitcoin shows genuinely mixed results, with a well-documented recurring problem: models that perform strongly on historical training data often perform noticeably worse on new, unseen data — a pattern called overfitting, where the model has essentially memorized past price movements rather than learned something that generalizes. No published ML model has demonstrated reliable, repeatable predictive power on genuinely new market conditions.

What is the rainbow chart?

The rainbow chart is a popular visual variant of long-term Bitcoin price models, typically built on a logarithmic regression of historical price, that bands the price history into colored zones from 'fire sale' (undervalued) at the bottom to 'maximum bubble' (overvalued) at the top. It's a visualization style rather than an independent predictive method — the underlying regression is a different model than power-law or stock-to-flow, though all three get referenced together in casual crypto discourse.

Are sentiment-analysis models different from these price models?

Yes — sentiment-analysis approaches, like the crypto Fear and Greed Index, don't try to fit a price curve at all. Instead they score inputs like social media chatter, search trends, and market volatility to produce a read on crowd psychology, which some traders use as a contrarian signal (buying during extreme fear, taking profit during extreme greed). It is explicitly a sentiment gauge, not a price-target model, and the distinction matters — the Fear and Greed Index has never claimed to predict a specific price.

Why do so many Bitcoin price models eventually get criticized?

Because every one of them fits a pattern to a finite amount of historical data, and Bitcoin's actual price is driven by a mix of adoption trends, macro conditions, regulatory shocks, and market psychology that no historical curve fully captures going forward. A model that fit the past well can look impressive right up until a new kind of event — a major exchange collapse, a regulatory action, or a macro shift — breaks the pattern it was built on. This is a general critique well documented across the crypto-research and academic-finance discourse, not a claim specific to any one model.

Sources

Ready to put this into practice?

Exchange sign-up bonuses pay both you and a referrer after a qualifying trade.

See bonuses →