Research · Context
Where Things Stand
Where Bitcoin sits in its own history across three valuation lenses, and what historically followed from similar readings. The page shows the historical outcomes, their sample sizes, and where the evidence fails to establish a reliable direction.
Current readings
Computed from the daily public snapshot for 14 October 2026, published 2026-10-15 11:22 UTC. The charts and tables below are drawn in the browser from the same file; these figures are the ones they show.
| Valuation composite | 42.3 | reads as average |
|---|---|---|
| MVRV percentile | 41st percentile | component |
| Mayer percentile | 56th percentile | component |
| Power-law percentile | 31st percentile | component |
| Band cuts | 10 / 30 / 70 / 90 | very cheap · cheap · average · expensive · very expensive |
Regulatory notice
Crypto Exponentials is not a registered investment adviser, broker-dealer or financial planner, and nothing on this page is investment advice, a recommendation, a solicitation, or an offer to buy or sell any asset. Everything below is a description of historical data. Historical outcomes are not forecasts; Bitcoin has fallen more than eighty per cent from a high on several occasions and may do so again, including from any state described here. You are responsible for your own decisions and should consult a qualified professional who knows your circumstances.
If you came here to decide whether to buy, sell or hold
- Read the evidence brief below, literally. It answers the three questions you actually have — is there evidence to add, to reduce, to hedge — each with a strength. Today: evidence to add moderate, to reduce none, to hedge none.
- Know what a fall of half would do to you. Every cycle on record contained one; the Cycle Monitor shows them. Whether such a fall would force a sale is a fact about your position, not about the market, and no reading on this site changes it.
- Check whatever brought you here against its record. If it was an RSI reading, a Pi Cycle chart, a bubble call or a moving-average cross, the Technical Signals and Indicator Autopsy pages show what it did after every previous trigger, against a random day. Most did nothing. The record of that signal against a random day is on those pages.
- Then decide — and it will still be yours. Buy, sell or hold is your situation multiplied by the evidence: horizon, size, tax, and what a fall would do to you. This site can only supply the evidence. It treats holding as a decision with evidence for and against it, not as what you do when nothing fires.
The current state, and the three lenses behind it
In plain terms: Bitcoin is currently average by these measures — it has been more expensive than this on 58% of days since 2013, and cheaper on the other 42%. All three measures agree. What has historically followed from this state, including how wide and how uncertain that record is, is set out below.
The five states are bands on the same historical scale. Very cheap is the lowest 10% of readings, cheap the next 20%, average the middle 40%, expensive the next 20%, and very expensive the highest 10%. These labels describe where today's reading sits in history. They do not predict what happens next.
| Measure | What it compares | Where it sits | Reads as |
|---|---|---|---|
| MVRV | market value against what holders paid | dearer than 41% of past days | average |
| Mayer multiple | price against its 200-day average | dearer than 55% of past days | average |
| Power-law deviation | price against the trend in time since genesis | dearer than 31% of past days | average |
| Composite | an equal-weight descriptive composite of the three, not a model | dearer than 42% of past days | average |
Evidence brief: what the tests support today
| Question | What the evidence says today | Strength |
|---|---|---|
| Is there evidence to add? | valuation reads average, a state whose forward record carries no reliable direction; price is in a drawdown, and drawdown days were followed by a median +112% at one year against +14% for advances — on a label that is late and unstable; a hash-ribbon recovery fired 49 days ago; across twenty completed episodes the median year after was +105%, 85% positive, with three named losers | moderate |
| Is there evidence to reduce? | valuation reads average, not the very-expensive state, which is the only one with evidence of poor forward outcomes; options skew at +2.77 points shows no fear premium; the LPPL bubble model finds no qualifying fit: the advance is not super-exponential; no tested sell rule on the technical-signals page shows an edge, and the textbook “overbought” reading ran the opposite way | none |
| Is there evidence to hedge? | skew at 2.77 points, neither cheap nor expensive | none |
| Is there evidence to rotate into gold or other crypto? | rotation rules tested on the relative-value page win at one lookback and lose at the others; mean reversion mostly loses, and its few wins depend on the lookback and the start date. A rule that depends on its parameter is not evidence | none |
The evidence today is mixed: a moderate historical difference in favour of adding, and no interval on this site excludes zero in either direction.
Strength is assigned by a rule, not by judgment: strong means the site's bootstrap interval excludes zero in that direction; moderate means the historical difference is large but the sample is small or overlapping; weak means a directional result with a known failure record; none means the tests establish no reliable direction. Which question applies to you depends on horizon, position size, tax, and whether a fall of half would force you out — none of which this page can see.
Market phase, and what it can and cannot tell you
By the definition below, Bitcoin is in a drawdown — 33% below its highest close, which was on Oct 7, 2025, and this phase has run 331 days so far.
- The label runs backwardsdays labelled drawdown were followed by a median +112% over the next year (n=4,180); days labelled advance by +14% (n=1,084). On this record the comfortable-sounding phase has been the worse one to buy
- It is late by constructiona drawdown can only be declared after price has already fallen a quarter from its high, so the label arrives well into the fall rather than at the top
- It is unstablethe label differs from where it stood 30 days earlier on 13% of days, even with the hysteresis gap in place
- It does not describe the phase it names6 of the 17 advances ended lower than they began, and the market has spent 79% of all days in drawdown, because the benchmark is the highest price ever reached
| Phase | From | To | Days | Price change over the phase |
|---|---|---|---|---|
| drawdown ← now | Nov 18, 2025 | ongoing | 331 | -10% |
| advance | Jun 12, 2025 | Nov 17, 2025 | 159 | -13% |
| sideways | May 20, 2025 | Jun 11, 2025 | 23 | +4% |
| advance | May 2, 2025 | May 19, 2025 | 18 | +10% |
| drawdown | Mar 11, 2025 | May 1, 2025 | 52 | +20% |
| advance | Nov 9, 2024 | Mar 10, 2025 | 122 | +5% |
| sideways | Sep 29, 2024 | Nov 8, 2024 | 41 | +15% |
| drawdown | Aug 6, 2024 | Sep 28, 2024 | 54 | +22% |
| advance | Feb 29, 2024 | Aug 5, 2024 | 159 | -7% |
| drawdown | Dec 5, 2021 | Feb 28, 2024 | 816 | +16% |
A drawdown is declared when price closes more than 25% below its highest close to date, and released only once it recovers to within 10% of that high; between those two levels the previous label stands. Outside a drawdown the phase is an advance when the 200-day trend in log price is rising and sideways when it is not. The two thresholds are round numbers fixed once, not values searched for; the gap between them exists because a single threshold reclassifies the market every few days as price crosses it, and the episode count with and without that gap is recorded in the methods page.
What historically followed each state
| State | Observations | 90 days after | 180 days after | 365 days after | Worst year from this state |
|---|---|---|---|---|---|
| very cheap | 291 | -12% spread -33% to +34% · n=291 90% CI on the median -22% to +28% | -22% spread -42% to +46% · n=291 90% CI on the median -35% to +49% | +23% spread -50% to +99% · n=291 90% CI on the median -45% to +117% | -58% from Aug 25, 2014 |
| cheap | 1,020 | +9% spread -27% to +81% · n=985 90% CI on the median -5% to +33% | +38% spread -43% to +167% · n=930 90% CI on the median -8% to +80% | +89% spread -52% to +185% · n=851 90% CI on the median +36% to +143% excludes zero | -65% from Jun 11, 2014 |
| average ← today | 2,638 | +9% spread -26% to +61% · n=2583 90% CI on the median -1% to +21% | +29% spread -40% to +152% · n=2548 90% CI on the median -6% to +44% | +83% spread -46% to +441% · n=2442 90% CI on the median -11% to +150% | -75% from Jan 18, 2014 |
| expensive | 663 | +12% spread -35% to +171% · n=663 90% CI on the median -15% to +94% | +19% spread -36% to +317% · n=663 90% CI on the median -13% to +166% | +18% spread -68% to +536% · n=663 90% CI on the median -22% to +145% | -79% from Jan 15, 2014 |
| very expensive | 173 | -25% spread -48% to +57% · n=173 90% CI on the median -38% to +26% | -17% spread -57% to +8% · n=173 90% CI on the median -38% to -8% excludes zero | -28% spread -75% to +14% · n=173 90% CI on the median -66% to -18% excludes zero | -83% from Dec 17, 2017 |
- Historical asymmetry, with its caveatvery-expensive readings have been followed by negative median returns at every horizon, on a small and heavily overlapping sample; one year later the ninetieth percentile of those outcomes is +14%, so even the better outcomes were poor. This is a description of a handful of episodes, not a rule
- The finding people expect, and do not getthe deepest discounts have a median one-year return of +23%, below the ordinary cheap state at +89%; buying maximum weakness has not been reliably better
- Overlapthese windows share days, so the effective sample is roughly four cycles, not the thousands of days in the count column
The composite through time
What would change the reading
| Lens | Now | Would read discount below | Would read premium above |
|---|---|---|---|
| MVRV (market value over realised value) | 1.54 | 1.36 | 2.06 |
| Mayer multiple (price / 200-day average) | 1.15× | 0.91× | 1.28× |
| Power-law deviation (log points) | -23 | -23 | 8 |
What this page cannot tell you
- It cannot time anythingthe longest unbroken discount or deep-discount spell in the record ran 396 days from Jun 10, 2014; a reader acting on the state at its start waited most of a year while it stayed there
- It cannot bound the downsidethe worst one-year outcome anywhere in this sample is -83%, and losses of that scale have begun from states this page would describe as ordinary
- It is not a probabilitythe columns are historical frequencies over four cycles of a single asset. They are not the chance of anything happening next; no honest method produces that number from this data
- It says nothing about youposition size, time horizon, tax, and whether a drawdown would force a sale are not in this data and matter more than any reading above
Method
Three lenses are ranked as percentiles against their own history: MVRV from Coin Metrics, the Mayer multiple (price over its 200-day average) and the deviation of price from a power law in time since the genesis block. Each percentile is computed as of that day against prior observations only, with at least 250 prior days required, so no reading uses information that was unavailable at the time. The composite is an equal-weight descriptive average of the three, with no weighting chosen by fitting and no optimisation; the state cuts sit at 10, 30, 70 and 90.
The forward-return columns are simple price changes measured from each qualifying day. Windows overlap heavily, which means the sample is far less independent than the day counts suggest: roughly four cycles of information, not thousands. The bootstrap interval on each median comes from a moving-block bootstrap of the daily calendar series: contiguous stretches of days are drawn from the original time index, the state is applied to the resampled index, and the median is recomputed from whatever qualifying days each draw produced. The number of observations therefore differs between replicates, which is correct, since how often a state occurs is itself uncertain and fixing it would make the interval look tighter than the evidence allows; the block length is the forward horizon itself, capped at a quarter of the sample so that short-lived states still resample meaningfully. The procedure is run under three independent seeds; the interval shown is the widest of the three, and a state is marked as excluding zero only when all three agree that it does. That caution is not decorative: at the one-year horizon the longest-lived state contains only about seven independent blocks, and a mark that moved with the random seed would be worse than no mark at all. This treats the overlap honestly rather than assuming it away, and it is why several states that look decisive on their medians turn out to have intervals straddling zero. The power-law model is fitted as of each day on an expanding window of the data available up to that day, so no classification uses coefficients estimated from later prices; a fit is produced only once five hundred qualifying observations exist. The threshold table takes the quantiles of the same reference distribution the ranking uses, so the levels it prints are the values today's reading would have to reach, not historical observations that happened to rank nearby.
How to read it, and how not to
The useful finding in the table, as of 14 October 2026, is narrow and specific: the very-expensive state has been followed by negative median returns at every horizon, while every other state has produced wide, overlapping outcomes. A reader who takes anything from this page should take that asymmetry, not a signal.
Note in particular that the deepest-discount state has a lower median one-year return than the ordinary cheap state. Nothing here says buying weakness has worked reliably; it says the record is noisy and the sample is small. Sources: the Crypto Exponentials daily snapshot (Blockchain.com price, Coin Metrics MVRV). Nothing on this page is investment advice.