Satoshi Institute · price prediction
Bitcoin price prediction model
What will Bitcoin be worth? Honestly: nobody knows a single number. What you can do is state your assumptions about trend, positioning and the macro backdrop, and read off the range those assumptions imply. This model starts from the long-term power-law trend, then tilts expected return up or down for each technical, sentiment and macro condition you set, and shows the result as a probability range around a median — never as one price. Background reading: can you predict the Bitcoin price?
Technical
Ranges widen with the square root of time.
Above ~2.4× has marked late-cycle conditions; below ~0.8× deep bear markets.
Trailing one-year price change.
Sentiment & positioning
Read contrarian: euphoria lowers expected return and widens the range.
−1 heavy distribution to exchanges, +1 sustained accumulation off exchanges.
Macro
−1 tightening, +1 expanding global M2.
Policy rate minus inflation.
A stronger dollar has generally been a headwind.
Implied Bitcoin price in 12 months
Downside · P10
Low · P25
Median
High · P75
Upside · P90
What each input contributes
| Family | Input | Effect on expected return |
|---|---|---|
| Trend | Power-law baseline | +28.6% |
| Technical | Price vs 200-week MA | +6.2% |
| Technical | 12-month momentum | +2.0% |
| Sentiment | Fear & greed (contrarian) | +0.0% |
| Sentiment | ETF / exchange flows | +0.0% |
| Macro | Global liquidity direction | +0.0% |
| Macro | Real policy rate | +0.0% |
| Macro | Dollar (12-month change) | +0.0% |
| Total | Expected return used | +36.8% |
Every number above is a modelling choice, shown so you can disagree with it. Zero all the tilts and the model reduces to the power-law trend with historical Bitcoin volatility.
Where the inputs come from
Not sure what to set? The Cycle Position tool reads Bitcoin against its own history and has a macro backdrop panel for money supply, real rates and the dollar. Set the conditions there and it hands them straight to this model.
BTC ≈ $62,958
What this is not. Not a prediction, not advice, and not a claim that these relationships hold. It is a disciplined way to convert your own assumptions into a range, with the arithmetic on display. Volatility is assumed constant over the horizon and returns lognormal, both of which understate Bitcoin's real tail risk. The power-law baseline extrapolates a 2013-onward fit and grows less reliable the further out it runs. Use the width of the range for position sizing, and revisit the inputs as conditions change.
FAQ
Can a model actually predict the Bitcoin price?
No — and this one doesn't claim to. What it does is convert assumptions into a distribution.
Every honest forecast has two parts: a central expectation and a width. The central expectation here comes from the long-term power-law trend, adjusted by the conditions you set. The width comes from Bitcoin's historical volatility, scaled by the square root of your horizon. The median is the least interesting output on the page; the distance between the P10 and P90 is the number that should change your behaviour.
What the model can and cannot do:
- Can: show how much your assumptions actually matter, and how wide the outcome range stays even under confident inputs
- Can: make disagreement precise — every input's contribution is listed separately
- Cannot: know the future, capture regime change, or price in news that hasn't happened
Where do the coefficients on each input come from?
They are stated modelling choices, calibrated to the direction and rough magnitude of well-documented historical relationships — not parameters fitted to a held-out sample.
That is a deliberate design decision. A model fitted tightly to 2013–2021 Bitcoin would look impressive and mislead you, because the investor base, market structure and liquidity regime have all changed. Modest, transparent coefficients that you can see and argue with are more useful than precise-looking ones you cannot inspect.
Set every condition to neutral and the model reduces to the power-law trend with Bitcoin's historical volatility — the most assumption-light version of the forecast available.
Why is high fear & greed treated as bearish?
Because sentiment measures positioning, not value.
When sentiment readings are euphoric, leverage is usually crowded on the long side and the marginal buyer has already bought. Historically that has coincided with lower forward returns and higher realised volatility, which is why extreme readings both lower the expected return and widen the range in this model. Deep fear does the reverse: it has often marked points where forced selling was closer to exhaustion than to beginning.
This is a tendency, not a rule. Sentiment extremes can persist for months, which is exactly why the output is a range.
How should I use the range rather than the median?
Size positions so the P10 outcome is survivable, and plan so the P90 outcome doesn't force a decision you'd regret.
- If P10 would break your plan, the position is too large regardless of how attractive the median looks
- If P90 would leave you wishing you'd bought more, you're probably underexposed relative to your own conviction
- If the range spans a 5× spread, that is information: the honest conclusion is that the horizon is too long for a point view
For treasuries, pair this with the drawdown survival tool — that one answers what happens at the bottom of the range, which is the question a board actually asks.
Why does the range widen so much at longer horizons?
Uncertainty in price compounds with time: the width of the band grows with the square root of the horizon, so a four-year view is roughly twice as wide as a one-year view.
At Bitcoin's ~55% annual volatility, an 80% range over one year already spans a large multiple; over five years it spans an order of magnitude. Anyone quoting a precise five-year Bitcoin price is quoting the midpoint of something very wide and omitting the width.
Methodology
Starts from the power-law trend drift over the chosen horizon, adds a transparent tilt to expected return for each technical, sentiment and macro condition, then converts the result into a lognormal price distribution and reports its quantiles. The output is a range, never a single target.
- Trend Baseline (annualised log drift)
- μ_base = ln(fair(t + H) ÷ fair(t)) ÷ H
- Condition Tilts
- Δ = −0.20·ln(M ÷ 1.5) + 0.10·m + −0.22·(FG−50)/50 + 0.12·f + 0.22·L + −0.05·(r−1) + −0.011·d
- Expected Return Used
- μ = clamp(μ_base + Δ, −65%, +125%)
- Annual Volatility
- σ = clamp(0.52 + 0.14·max(0,(FG−55)/45) + 0.10·max(0, ln(M ÷ 1.6)), 0.42, 0.95)
- Price Quantile at Horizon
- P_q = P_spot · exp(μ·H + z_q · σ · √H)
- Reported Quantiles
- z = −1.2816 (P10), −0.6745 (P25), 0 (median), +0.6745 (P75), +1.2816 (P90)
fair(t) is the power-law fit log₁₀(P) = 5.2169 · log₁₀(days) − 14.8056, the same fit used by the corridor tool.
Variables
- H
- Horizon in years (selected in months)
- M
- Price ÷ 200-week moving average (Mayer-style trend stretch)
- m
- Trailing 12-month price change, as a decimal
- FG
- Fear & greed reading, 0–100, applied contrarian
- f
- ETF / exchange flow regime, −1 distribution to +1 accumulation
- L
- Global liquidity direction, −1 tightening to +1 expanding
- r
- Real policy rate in per cent (policy rate minus inflation)
- d
- Dollar index 12-month change in per cent
Each coefficient is a stated modelling choice, not an estimate fitted to a held-out sample — the per-input contribution table on the page exists so those choices can be audited and overridden. Set every tilt to neutral and the model collapses to the power-law trend with Bitcoin's historical volatility. Constant volatility and lognormal returns understate real tail risk: Bitcoin has repeatedly moved further than a lognormal allows. The spot price is live from CoinGecko; the trend baseline uses a 2013-onward fit whose reliability decays with horizon length.