Bitcoin Market Structure Dashboard

Robustness: the gap is measured under every Nasdaq beta in use, the fixed, the rolling and the regime-aware, at the 90 and 30 day views and over the ETF-anchor horizon.

Bitcoin trades around the clock but the Nasdaq only in session, so weekend moves land in the gap automatically. When the in-session-only and all-days measures agree, the divergence holds inside the trading session.

For each coin, a regression of its daily return on the Nasdaq return interacted with the prior day's VIX gives a beta that varies with the stress level. The prior-day VIX is used so the reading does not look ahead. The upward slope is clearest in the stress-rich years 2019 to 2023, where there is enough high-VIX variation to measure it; full-sample it is marginal for the alts and significant for Bitcoin. The rise is consistent across the seven stress episodes on record, 2010 to 2026: a heterogeneity test finds no difference between episodes (Cochran Q p = 0.33). The calm 2024-26 stretch alone has too few stress days to re-estimate it in isolation. Its level shifts over time as the trend beta drifts. The displayed curve therefore rides that locked shape at each coin's trailing 2-year structural beta, with the VIX increment taken relative to its window average, so it sits at the current level, not a whole-sample average.

Beta is the slope of Bitcoin's daily returns regressed on the Nasdaq's, over rolling windows (30, 90, 180 days). Before 2020 it averaged near zero (β≈…, 90-day): stretches of positive and negative coupling offset each other, with no persistent direction. The 2020 COVID break is where the coupling turned persistent; post-2020 the 90-day beta has averaged around one (~… across windows), split roughly evenly above and below 1.0. The shift holds at every window, so the equity gap and rolling beta here are measured from 2020 onward. Beyond direction, the Nasdaq explains under a quarter of Bitcoin's daily variance, the rest crypto-wide; on macro-data days, once the equity move is removed, Bitcoin's own reaction to the number is indistinguishable from zero (p = 0.69 and 0.54).

On the gold-versus-real-rates safe-haven axis, Bitcoin loads negative and gold positive. Through the 2010s Bitcoin had no loading on that axis; the negative loading has held since 2022. The decomposition runs over eight cross-asset series. Its other large factor is the common risk-on tide, the days when risk assets rise together and the dollar falls. Bitcoin's loading on it sat near zero through the 2010s and has climbed since the 2020 break to near its highest level today (the same post-2020 coupling the equity-coupling panel traces).

Inputs (8 daily series, 2010→today): Bitcoin, Nasdaq-100, gold, US dollar, high-yield credit (HYG), oil (Brent), EM equities (EEM) as daily log returns, plus the 10-year real yield as its daily change; standardised, correlation PCA, 2-year rolling windows. Monthly prints (CPI, payrolls) are excluded as non-daily.

Each coin's gap to the Nasdaq uses its own two-year rolling beta, the same rolling-beta convention as the trailing equity-gap series (the anchored wedge views pin the fixed structural beta as their baseline). The shared crypto factor is the first principal component of the four alts' gaps (ETH, LTC, XRP, BCH), computed with Bitcoin excluded. The crypto-wide share is how much of Bitcoin's gap that factor accounts for; the remainder is Bitcoin-specific. Read the split as where the gap comes from. Trust the ordering over the exact figure: the 95 percent interval is wide, but the crypto-wide majority and the substantial Bitcoin-specific minority hold up throughout. Because Bitcoin and the altcoins move together, the shared factor absorbs a little of Bitcoin itself, so the Bitcoin-specific share is a conservative floor. No tested public series accounts for that Bitcoin-specific part.

The top 10 is an equal-weight basket of the 9 other top-10 coins by dollar volume (ETH, SOL, XRP, etc.). Altcoins is everything outside the top 10: a basket that grows over time (~130 coins in the current window, 130-200 in recent years; far fewer clear the filter in the early years, so the left side of each chart is a smaller universe). The basket holds coins that survived, so early-year correlations and betas carry survivorship bias; trust the trend and discount the early levels. Both measures use the same 90-day rolling window on the same basket returns, so the identity beta = corr × (basket σ ÷ BTC σ) holds. All readings are same-day co-movement, not lead-lag.

The trend and the sigma are the PRESS model's own published numbers. The power-law trend is refit daily on an expanding window (log price on log age), so no reading uses future data. Sigma is the deviation from that trend divided by the volatility of all deviations to date, which keeps readings comparable as cycle amplitude compresses. The bands are the PRESS regime cuts: deep value at or below sigma −0.5, hot at or above +1.0.

The dated extremes in sigma: bottoms 2015 −1.70, 2018 −0.87, 2022 −1.36, all in the deep-value band; tops 2013 +2.53, 2017 +2.20, 2021 +1.09, all hot; the 2024 and 2025 highs +0.34 and +0.17, the normal band. The same deep-value reading is what the forward-risk panel below counts as low risk; one signal feeds both panels.

Bitcoin drawdown (zones). The DEEP / MID / SHALLOW zones split the full drawdown history into thirds. The three mature-cycle bottoms clustered at 77-85% below the prior peak; the first cycle bottomed at −92.7% in 2011.

All three duration charts. A major episode is any 20%+ fall from a record high, timed from the first day below the peak until price regains it, split into decline (peak to low) and recovery (low back to peak). Bitcoin has spent …% of its life in one.

Corrections and recoveries: the dotted fit lines. For each leg the dotted line is a least-squares fit of log-days on trough depth across the … majors begun before the January 2024 spot-ETF launch, with a one-standard-deviation band. Both legs scale with depth (decline R² …, recovery R² …). The fit stops at the ETF launch because every ETF-era episode has run a multiple of its depth-implied time.

The open episode (orange path). It is compared with past falls at the same age, and each path traces the running low: a flat stretch means no new low, not sideways price. Days can only grow and depth can only deepen, so its position is not settled.

Corrections and round trip: the bear dates. The dates under the bear labels map each past bear's duration onto this episode's start date: scenario arithmetic on four historical durations, not a forecast.

Bitcoin's own turbulence: the annualised volatility of its daily returns. The average steps down each halving epoch, from well over 100% early on to under 50% now. The current reading sits near the bottom of the whole history.

The 2-state hidden Markov model splits Bitcoin's power-law sigma (the standardised trend deviation) into two regimes, without user-chosen windows or epoch boundaries. One is an above-trend (speculative) regime, the signal; the other is a below-trend null regime, the baseline. The above-trend state works as a top indicator: it fires during speculative runs and fades when the market cools. A below-trend reading marks the absence of that signal; it is not a deep-value call. Since 2012, above-trend episodes have become less extreme. Inspired by G. Santostasi's bimodal regimes approach.

MVRV is the market-to-realised-cap ratio: market value against the aggregate on-chain cost basis of all coins. MVRV-z is the z-score of log(MVRV) over a leak-free two-year rolling window, lagged one day so it uses only prior data. This is a rolling own-history z, not the widely quoted full-history MVRV Z-Score; the two are not on the same scale. The percentile ranks today's z within the post-2013 series, since pre-2013 on-chain data is thin. Read the z-score rather than the realized-price level, which lost and dormant pre-2014 coins bias downward; the cost basis is sticky, so the series is slow-moving.

How often a deep fall followed each zone. A ≥30% drawdown over the next 90 days followed … of deep-value days · … of normal · … of hot (base rate …). 30-day horizon: … / … / … (base …).

How the periods are cut. A period runs from the day the power-law sigma crosses below −0.5 to the day it crosses back above; a break of a week or less counts as the same period. That gives … periods since 2012, … of them with red days. Red marks days followed by a ≥30% drawdown within 90 days, the same event the band frequencies count.

Tested out of sample, the gauge ranks risky quarters ahead of safe ones …% of the time (AUC …, 95% CI […, …]) and …% over a month (AUC …, […, …]). The out-of-sample test rests on just three completed Bitcoin cycles (the series starts 2012), so read the AUC as indicative. It is a full-history figure: the gauge held its ranking skill across the three completed post-2012 epochs, and the post-2024 regime has seen too few completed deep-fall episodes for a retest, so its skill there remains untested.

DVOL is Deribit's 30-day annualised implied-volatility index; the percentile ranks today's value within the post-2021-04 series, since DVOL launched. The variance risk premium is DVOL minus 30-day realised volatility (from daily log returns, annualised, quoted in vol points, the practitioner convention; academic work quotes variance units). The forward read is tested on a held-out sample: fitted through 2022 and evaluated on 2023 onward, DVOL ranks the next 30-day realised volatility at rank correlation about +0.42 (block-bootstrap CI excludes zero), beating a trailing-realised-vol baseline by about +0.22 on that holdout. Over the full DVOL era it also beats a leak-free HAR model (edge about +0.14, CI excludes zero), and the full-era rank correlation is higher than the holdout's, so the holdout figure is the conservative one. The overshoot of realised vol, roughly nine points on average with wide variation, is the premium. Bitcoin's premium is positive most of the time (implied above realised on about four days in five) but flips negative when a crash sends realised vol above implied. In the literature, it widens around large moves in either direction, unlike equity's downside-tied premium (Alexander and Imeraj).

The term basis is the annualised gap between the front Deribit quarterly futures and spot, (futures minus spot) divided by spot times 365 divided by days to expiry, rolling the front contract at 21 days to expiry. The percentile ranks today's basis within the post-2020 series. Positive (contango) means futures above spot; negative (inversion) means futures below spot, stress priced at a horizon. Funding can turn negative in stress too, but it is an instantaneous rate and carries no horizon.