Crypto Cycle Framework CCF

Generated from the model output of 2026-10-25 00:51 UTC. On-chain data through 2026-10-23; price through 2026-10-24.

1. Purpose

The Crypto Cycle Framework (CCF) answers three questions about bitcoin and ether at any point in their cycles:

1. How much is priced in? For each coin, several different valuation lenses, each ranked within a comparison window and averaged into a 0–100 composite cycle score, plus a model fair value in dollars with bands. Bitcoin uses four lenses; ether uses two (section 7).
2. What is the backdrop? Eighteen yes/no conditions that turn the popular bullish arguments into testable signals, in six groups: the business-cycle phase read from the Business Cycle Regime Engine, the business cycle (including Tom Lee's ISM thesis), global liquidity, real rates and risk sentiment, adoption and network, and flows and positioning. They are tested against bitcoin's returns.
3. How much to hold? A fixed rule turns each coin's cycle score into an exposure between 0% and 100%, simulated point-in-time and tracked live from October 2026 against holding the coin, a constant 50% and cash (section 8). An allocation model splits a crypto sleeve between bitcoin, ether and cash (section 9).

It is built to inform judgment, not replace it, with the same intent as the Global Multi-Asset Allocation Framework (GMAAF). It is a personal research project; views are the author's own and not those of any employer. Nothing here is a trading rule, a price forecast or investment advice.

2. How to read the page

Header. A generated one-paragraph summary, the data dates, and a strip of key figures (composite score, zone, price, realized price, MVRV, model fair value, halving epoch). The Score against switch (trailing four years or all history) changes every valuation score, the fair value and every valuation table; backdrop indicators do not depend on it.

Tabs, in order: Bitcoin, Ethereum, Allocation, Risk & returns, Macro & liquidity, Method & data. Each coin tab has sub-pages:

• Overview: a How to read this page guide; Today's read (valuation and, for bitcoin, the backdrop); the cycle-position gauge; the lens tiles; Backdrop at a glance (bitcoin only); price versus model fair value; premium or discount to fair value; fair value by lens; price against its valuation anchors; the composite score history; what followed similar readings; readings at past peaks and troughs.
• Valuation lenses: each lens's raw metric, its score, and all scores together.
• Evidence: forward returns by score and by premium to fair value, rank correlations by halving epoch (bitcoin) or period (ether), and for bitcoin a one-table record of every backdrop indicator.
• Live results: the zone-based exposure rule, its simulated record against holding the coin, a constant 50% and cash, sensitivity to the start date, calendar-year returns, and the frozen live record.
• Adoption & flows (bitcoin only): adoption-and-network and flows-and-positioning indicators, plus the spot ETF series being built.

Allocation: today's bitcoin/ether/cash split step by step, the ETH/BTC relative-value score and its evidence, the simulated record against the market-cap index, 100% bitcoin, a constant 50% and cash, and the live record. Risk & returns: rolling and trailing return, volatility, Sharpe ratio, maximum drawdown, correlation and beta for bitcoin and ether against broad markets, over any window (section 10). Macro & liquidity: the ISM deep dive (Tom Lee's three claims tested), then the business-cycle, global-liquidity and real-rates-and-risk indicators. Method & data: this document.

Reading the numbers.

• Composite cycle score (0–100). 0 is the cheapest reading in the comparison window, 100 the richest. Zones: deep value below 15, value 15–35, neutral 35–65, expensive 65–85, euphoric 85 and above. A low score means the coin is cheap relative to its own past, which has been the better starting point; it is not a forecast of a rally.
• Model fair value. Each lens implies the price at which it would sit at the median of its window, holding its own anchor fixed; the model fair value is the geometric average of those prices (four for bitcoin, two for ether). It moves with the market's own footing, so it is a reference point, not a target. The 35th–65th and 15th–85th bands are close to, but not identical with, composite scores of 35/65 and 15/85.
• Backdrop indicators are yes/no bullish conditions. For each one the page shows the reading the condition uses today, whether it is met, and bitcoin's median forward return on past days when it was met versus not met. Supported: met beat not met by more than 15 percentage points with a higher share of days up. Reversed: the opposite. No clear edge: neither. Cells where one side has fewer than 60 days show — or "too few days".
• Favorable or unfavorable. An indicator is favorable today if its condition is met and its record is supported, or it is not met and its record is reversed; unfavorable in the two opposite cases. The headline backdrop read counts only indicators whose verdict agrees in both periods; the 2018-on-only count is shown beside it. More conditions met has not, by itself, meant better returns.
• Periods. Valuation results: 2012–17 and 2018 on. Backdrop results: 2013–17 and 2018 on. Signal dates do not overlap, though forward windows from late-2017 days run into 2018–19.
• Evidence strength. Daily windows overlap heavily and there are only about four cycles; hundreds of days can be a handful of distinct episodes. Read direction and consistency across periods, not decimals.

3. Today

Valuation. The composite cycle score is 45 (neutral) and price is 10% below the model fair value of $90k (neutral band $74k–$104k). Since 2018, days with a score within 5 points of today's were followed by a median −10% over 6 months (48% up) and +71% over 1 year (85% up, 12 distinct episodes); for any day in that period the medians were +13% and +40%, so the 1-year result is better than typical. Days when price was 0–20% below fair value, as now, were followed by +93% over 1 year (83% up).

Backdrop. 9 of 18 conventional bullish conditions are met today. Counting only indicators whose record agrees in both periods (the stronger evidence), 2 point to stronger returns (Fed funds rate and Equity volatility (VIX)) and 2 to weaker (ISM manufacturing and Hash rate (hash ribbon)): neutral. On the 2018-on record alone the count is 4 to 5 (mildly unfavorable). (Favorable = condition met with a supported record, or not met with a reversed record; unfavorable = the two opposite cases.)

This summary describes what has historically followed readings like today's. It is not a forecast or a recommendation.

4. Design principles

• Point-in-time. Every score, fair value and backdrop signal uses only data that existed on that day. Percentiles use expanding or trailing windows; the power-law fit is re-estimated daily from data to date; the ISM counts from its release date; monthly market data counts from the first day of the next month and US M2 from the second month; daily and weekly series count from the next day.
• Nothing fitted to returns. Lens definitions, equal weights, zone cut-offs, fair-value percentiles and every backdrop condition were set from first principles or the conventional reading of the indicator. Two choices were made after looking at results and are disclosed: the default scoring window (section 6.1; both windows are shown) and the Fed-funds "flat" tolerance (+0.10 point, so that a Fed on hold counts as flat). Further caveats: the lenses themselves (MVRV, the Puell multiple, the power law, hash ribbons) were popularized with pre-2018 data in view, so 2018 on is out of sample for this model's parameters but not for the ideas; and the ISM history begins in 2013, so backdrop tests start then for comparability.
• One test for every backdrop indicator. Met versus not met, the same horizons, the same non-overlapping periods and the same verdict rule, so indicators can be compared.
• Different valuation evidence. The four lenses look at different things: holders' cost basis, the long-run growth path, miners' economics and the multi-year trend (the last two price-based lenses overlap somewhat).
• Percentiles, not fixed thresholds. The highest MVRV reading of each cycle fell from about 5.9× (2013) to 4.7× (2017), 4.0× (2021) and 2.8× (2024–25); a fixed rule such as "sell at MVRV 3.5" would have missed the last top.
• Valuation and backdrop kept apart. Valuation measures how much is priced in; the backdrop describes conditions. Blending them would hide which is driving a reading, and the evidence shows the backdrop is often already priced.

5. The valuation lenses

5.1 MVRV — cost-basis valuation

• Definition: market value ÷ realized value. Realized value prices every coin at the price it last moved on-chain (Coin Metrics CapMVRVCur). Realized price = price ÷ MVRV.
• Rationale: it is the market's aggregate unrealized gain, the closest analogue to price-to-book. Above roughly 2–3× many holders sit on large gains and profit-taking supply rises; below 1× the average holder is under water and selling dries up.
• Record: the 2015, 2018 and 2022 lows were all below 1× (0.56×, 0.69×, 0.75×). Cycle-high readings (which can come before the price top) were about 5.9×, 4.7×, 4.0× and 2.8×; on the price-peak days the readings were lower (see 18.4).
• Weaknesses: lost coins stay in realized value at old prices; custody and ETF structures change how coins move.
• Today: 1.49×, score 39; realized price $54,064.

5.2 Power-law residual — long-run growth path

• Definition: log₁₀ price minus an OLS fit of log₁₀ price on log₁₀ days since the genesis block (3 January 2009), re-estimated each day from data to that day.
• Full-sample fit (display only): log₁₀ P = -16.3328 + 5.6362 × log₁₀ days; residual σ = 0.3018 (a factor of 2.00×).
• Rationale: adoption has grown roughly as a power of time, and price has swung around that path; the residual measures how far ahead or behind price has run.
• Weaknesses: says nothing about timing; early fits are unstable; the exponent may not persist.
• Today: price is 0.56× today's trend ($144,094); score 29. The historical trend line on the charts is the full-sample fit, so earlier points on that line use later data; the scores never do. The power-law fair value ($105,626) sits below the trend because over the scoring window price has mostly traded below it.

5.3 Puell multiple — miner economics

• Definition: daily miner revenue in dollars (issuance plus fees, Coin Metrics IssTotNtv + FeeTotNtv, times price) ÷ its 365-day average.
• Rationale: miners are a steady structural source of selling, since they must sell new coins to cover costs. When revenue falls well below its trailing average, high-cost miners sell reserves or shut down, and that forced supply tends to clear near lows.
• Weaknesses: halvings cut revenue mechanically and depress the multiple for about a year; the noisiest lens.
• Today: 1.26×, score 75.

5.4 200-week multiple — multi-year trend

• Definition: price ÷ its 1,400-day simple average.
• Rationale: four years roughly spans one halving cycle. Bear markets have bottomed at 0.66–1.0×. Cycle-high readings were about 29× in 2013 (on a shorter, two-year average, since history begins in 2010), 16× in 2017, 6× in early 2021 and 2.5× in 2024–25.
• Weaknesses: price-only and overlaps with the power-law lens; it earns its place mainly at floors.
• Today: 1.21×, score 35.

6. Scoring, composite and fair value

6.1 Scores

Each lens is put in logs and ranked as a percentile (0–100) of its own values in a window ending that day: the trailing four years (default, 1,460 days) or all history (expanding). Two years of history are required before a score appears. The composite is the equal-weight mean of the lens scores, requiring at least three of four.

Why the trailing window is the default. In the all-history setting today's market is ranked against 2013's extremes, so the 2024–25 run never scored higher than 73; in the trailing window it reached 87 on 2025-08-13, about two months before the price top (where it read 77). The composite can peak before price does. Both settings sort recent-era forward returns about equally well (rank correlation -0.63 trailing, -0.61 all-history), so the choice is about which comparison is more meaningful for a maturing asset. It was made after seeing both, which is why both stay one click apart.

6.2 Model fair value

For each lens, the model asks at what price that lens would sit at the median of its scoring window, holding its own anchor fixed (realized price, the point-in-time power-law trend, the one-year average of miner revenue, the 200-week average). For a lens L (in natural logs) with window quantile Q:

fair value = price × exp(Q₅₀(L) − L)

The model fair value is the geometric average of the four lens fair values (at least three required); there is no single price at which all four lenses sit at their medians at once. The bands use the 15th, 35th, 65th and 85th percentiles the same way; they are close to, but not identical with, composite scores of 15/35/65/85.

How to read it. Fair value is not a price target. Its anchors rise in bull markets and fall in bear markets, so it says where price is relative to the market's own footing, not where price will go. Versus fair value, price stood +388% (2013 peak), −75% (2015 trough), +528% (2017 peak), −49% (2018 trough), +67% (2021 peak), −58% (2022 trough), +24% (2025 peak) at past turning points. The 2025 top was the least stretched top in the sample, a reminder that valuation measures can be early or late as the market matures.

7. Ethereum

Ether uses two lenses, ranked and combined exactly as bitcoin's (sections 5–6); both are required for a composite:

• MVRV (Coin Metrics CapMVRVCur for ETH): holders' aggregate unrealized gain. Lows at 0.30× (2018), 0.60× (2022) and 0.79× (2025); cycle highs fell from about 3.1× to 2.3× and 1.6×. Staking and smart-contract deposits move coins in ways that can blur the cost basis. Today 1.05×, score 29.
• 200-week multiple: price ÷ its 1,400-day average. Ether has traded below it in every bear market (0.39×, 0.83×, 0.60× at the lows). Scores begin in 2019. Today 0.93×, score 20.

Dropped: the fee multiple. Market cap ÷ trailing-year fees (Ethereum's nearest analogue to price-to-sales) was a third lens until October 2026. After the March 2024 Dencun upgrade deliberately moved most activity, and its fees, to cheaper layer-2 networks, the multiple rose from 12–45× in 2022 to 100–190× in 2024 and 600–1,700× in 2025–26, so it read expensive almost regardless of price. Its rank correlation with 1-year forward returns had the wrong sign before 2023 and was 0.03 from 2021 on. At the time it was dropped it sat at the 98th percentile and pulled ether's model fair value to about half the price. It was removed before the live record began; with it, the simulated 2021-on record was somewhat better (about 29% a year rather than 26%), because its near-permanent "expensive" reading after 2024 happened to keep exposure low while ether lagged. That is a broken input getting lucky, not a valuation signal, so the better backtest was not a reason to keep it. A fee measure that includes layer-2 activity could replace it if a reliable free source appears.

The bitcoin-specific lenses are not used: the Puell multiple lost its meaning when Ethereum moved to proof of stake in September 2022, and a power-law fit on ether's shorter history is unstable. Vertical lines on ether's charts mark network upgrades (EIP-1559 fee burn, The Merge, Dencun) rather than halvings.

Today: composite 24 (value); model fair value $2,747, price 14% below it:

Periods: MVRV scores start in 2017 and the 200-week multiple in 2019; the composite needs both, so it starts in mid-2019. Results are shown for before 2021 (from late 2017 for MVRV, mid-2019 for the composite) and 2021 on.

Composite score buckets — 1-year forward return

Rank correlation with 1-year forward return

Readings at past peaks and troughs

How to read it. Ether's composite has about one and a half cycles of scored history. Both lenses have had the expected negative correlation within each epoch and since 2021 (the figure for the whole pre-2021 period is positive because levels shifted between that period's two cycles). Two lenses that both track price against a slow-moving anchor are a thin basis, so treat ether's composite with more caution than bitcoin's.

8. Zone-based exposure and live results

The rule (fixed in advance, not optimized). Each coin's exposure comes from its composite cycle score on the trailing four-year window: deep value (below 15) 100%, value (15–35) 75%, neutral (35–65) 50%, expensive (65–85) 25%, euphoric (85 and above) 0%. The backdrop is not used. Decisions are made every Monday from that Monday's score (published Tuesday) and traded at Tuesday's close; a trade happens only when the target differs from the drifted position by at least 10 percentage points; trading costs are 0.2% of the amount traded; uninvested money earns the 3-month Treasury bill rate (FRED DTB3, 4.04% today). The simulation starts fully positioned at the first day's target.

Benchmarks. 100% in the coin (buy and hold, no costs); a constant 50% in the coin and 50% in T-bills, rebalanced daily without costs; and cash in T-bills. The constant 50% is the fair test of timing: the model's average exposure is close to 50%, so beating 100% on risk or beating cash on return is not by itself evidence of skill.

Live record. From 2026-10-24 the rules are frozen. A day is logged only after its UTC close, starting with the close of that day. Each daily run appends the completed days' positions, any pending Monday target, and the model's and benchmarks' values to the end of data/live/live_btc.csv, live_eth.csv and live_alloc.csv; earlier lines are never touched, so the repository's history would show any revision. If a Monday score is missing, no trade is scheduled that week, as in the simulation. The simulated history is context; the live record is the honest test.

Bitcoin

Today: composite 45, target exposure 50%. Simulated from 2013-05-11; results below are for 2018 on (from 2018-01-01).

Full simulation (from 2013-05-11): model +26.8% a year (maximum drawdown −63%), holding +62.7% (−85%), constant 50% +37.8% (−56%).

Start-date sensitivity. From each January 1 to today, the model beat holding in 3 of 12 start years and beat the constant 50% in 7 of 12:

Ethereum

Today: composite 24, target exposure 75%. Simulated from 2019-07-06; results below are for 2021 on (from 2021-01-01).

Full simulation (from 2019-07-06): model +33.2% a year (maximum drawdown −53%), holding +33.4% (−79%), constant 50% +27.2% (−52%).

Start-date sensitivity. From each January 1 to today, the model beat holding in 5 of 6 start years and beat the constant 50% in 6 of 6:

Alternative rules, tracked for comparison. Three alternatives were fixed on 2026-10-24, before any live results, and are logged daily beside the main rule (data/live/live_<coin>_<rule>.csv) with the same weekly mechanics, band and costs. None drives the main rule; they exist to learn, from data that did not exist when they were chosen, whether trend or volatility scaling improves on valuation alone. Valuation + trend: the average of the zone exposure and a trend exposure (100% at or above the 200-day average, 0% below). Risk-scaled valuation: zone exposure × (median 90-day volatility over the trailing four years ÷ the last 90 days' volatility), the ratio held between 0.5 and 1.5, capped at 100%. Trend only: 100% at or above the 200-day average, 0% below. Parameters are conventional and were not tuned, but these are ideas known to have worked, so their simulated records flatter them; the live record decides.

How to read it. The rule is a sizing discipline. It tends to trail the coin in strong uptrends, because it scales down as valuations rise, and to win from starting points near tops, because it is already scaled down when the decline comes. Results depend heavily on the start date, which is why the start-year tables matter more than any single headline number.

9. Allocation between bitcoin, ether and cash

The rule. (1) Start from market-cap weights. (2) Compute a relative-value score for ether against bitcoin: the average of the ETH/BTC price ratio's percentile within its trailing four years and 50 + half the gap between ether's and bitcoin's composites (low = ether relatively cheap). (3) Multiply ether's market-cap share by ×1.5 (below 15), ×1.25 (15–35), ×1 (35–65), ×0.75 (65–85) or ×0.5 (85 and above), capped at 80%; bitcoin gets the rest. (4) Scale each coin's share by its own zone exposure; the remainder is cash in T-bills. Same weekly mechanics and costs as section 8, trading when a coin's weight is 5+ points from target.

Today: market-cap split 85% / 15%; ETH/BTC at its 18th percentile; composite gap −21 points; relative score 29 (×1.25); exposures 50% bitcoin and 75% ether; result 41% bitcoin, 14% ether, 45% cash.

Does relative value predict relative returns? (since 2019)

Simulated record, 2021 on:

From each January 1 to today, the model beat the market-cap index in 2 of 6 start years. Ether's history is short and the extreme relative-value zones have not occurred, so the largest tilts are untested; the tilt is deliberately modest.

10. Risk and returns

This tab describes how bitcoin and ether have behaved; it makes no forecast and is not part of the valuation or the exposure rule. Every statistic is computed in the browser from daily prices, so any window can be chosen.

Assets. Bitcoin and ether daily closes (every calendar day). Broad markets are represented by liquid ETFs, using dividend-adjusted closes so that their returns are total returns: US stocks (SPY), US tech stocks (QQQ), global stocks (ACWI), emerging-market stocks (EEM), US investment-grade bonds (AGG), 20+ year Treasuries (TLT), gold (GLD), a commodity index (DBC) and the US dollar (UUP). ETF data begin in January 2010.

Measures.

• Return: price change over the window, including reinvested dividends for ETFs. Windows of a year or less show the total return; longer windows show the annualized return.
• Volatility: standard deviation of daily returns, annualized with 365 days for crypto (which trades every day) and 252 for ETFs.
• Sharpe ratio: average daily return minus the 3-month T-bill return for the same days, divided by the standard deviation of daily returns, annualized.
• Maximum drawdown: the largest fall from a peak to a later trough within the window. The drawdown chart instead measures each asset's distance below its highest close to date.
• Correlation and beta: from log returns of the two assets on common dates. Weekly (default) uses Friday-to-Friday returns, which are less affected by the different trading hours and days of crypto and US markets; daily uses US trading days, so a crypto Monday return includes the weekend. Beta = covariance ÷ the market's variance: a beta of 1.5 means the coin has typically moved about 1.5% for a 1% move in the market. Weekly correlations need at least 10 observations, so they are not shown for one-month windows.

Views.

• Rolling explorer: each point is the chosen measure over the window ending that day. The table beneath gives the latest reading, its percentile within the series' whole history, the 10th, 50th and 90th percentiles, and the extremes with their years. Up to three markets can be compared; for correlation and beta, choose the market to measure against (bitcoin can be chosen, to see ether against bitcoin).
• Snapshot: every asset over the same fixed window (1 month to 10 years, or year to date), ending on the last day with a close for every asset, with correlation and beta to the S&P 500 and correlation with bitcoin, plus a return-against-volatility chart.
• Sizing a crypto sleeve: for a base portfolio of global stocks (ACWI) and US bonds (AGG), the size of a crypto sleeve (the allocation model's bitcoin/ether split, the market-cap split, or either coin alone) at which crypto accounts for a chosen share of portfolio risk, with portfolio volatility, return and worst drawdown at several sleeve sizes. Estimated from weekly returns over 1, 3 or 5 years; share of risk = weight × covariance with the portfolio ÷ portfolio variance. Because crypto is several times as volatile as a balanced portfolio, a small sleeve takes a large share of its risk.
• Correlation matrix: pairwise correlations over a chosen window.
• Drawdown from all-time high and calendar-year returns.

How to read it. Rolling statistics overlap heavily from one day to the next, so a long run of high or low readings is one episode, not many. Bitcoin's one-year weekly correlation with US stocks averaged close to zero before 2020 and mostly 0.2–0.4 since (with a dip to about zero in 2024), as institutional ownership grew; long-history percentiles mix those regimes. Volatility has trended down across cycles, so a reading that is low against all history may be ordinary against recent years; the Show range and window choices make that visible.

11. The backdrop: how each indicator is built and judged

Each indicator turns a widely made bullish argument into a yes/no condition checked every day with only the data published by then: the ISM from its release date, monthly market data from the first day of the following month (US M2 from the second month, because it is published about four weeks late), weekly and daily series from the next day. On the page, reading used is the value the condition uses today; a newer month-to-date value is shown separately.

The test compares bitcoin's median forward return (6 and 12 months) on days the condition was met with days it was not, in two non-overlapping periods: 2013–17 and 2018 on. Supported = met beat not met by more than 15 percentage points with a higher share of days up; reversed = the opposite; no clear edge = neither; too little data = one side had fewer than 60 days (for example a condition that was almost always met in 2013–17). Each group also shows forward returns by the number of its conditions met.

Overall record (2018 on, 12 months): 1 supported, 8 reversed, 9 no clear edge. Supported: Equity volatility (VIX). Reversed: ISM manufacturing, Copper ÷ gold, Fed funds rate, G3 central-bank balance sheets, Corporate credit spread, Hash rate (hash ribbon), Net exchange flow and Stablecoin supply. Reversed in both periods: ISM manufacturing, Fed funds rate and Hash rate (hash ribbon).

Interpretation. Most conventional bullish readings have not helped, and several have come before weaker returns: by the time the backdrop looks good, bitcoin has usually moved. For indicators reversed in both periods, the opposite reading (for example manufacturing weakness) has been the more reliable entry point; single-period reversals deserve less weight. Because many indicators are tested, one or two will look good or bad by chance.

12. Business cycle

Thesis. A strengthening economy and easier money lift risk assets: the ISM manufacturing index at or above 50 (or rising), a weaker dollar, copper outperforming gold, and a Fed that has stopped tightening.

The business-cycle phase (shared source)

From October 2026 this framework reads its cycle state from the Business Cycle Regime Engine (BCRE) instead of keeping its own. BCRE sorts each month since 1970 into recovery, expansion, slowdown or contraction from a ten-indicator US growth composite (level against trend, and its 3-month direction), using only data published by that month-end. Its regime.json is fetched daily; each month's phase counts here from the first day of the next month. The phase is tested against bitcoin's forward returns like the ISM regimes, and it enters the backdrop as one condition (met in recovery or expansion). Both use BCRE's unrevised-only phase (its six never-revised inputs): the headline phase uses today's revised payroll, hours, production and permit data, which would put revision look-ahead into the tests. The panel shows the headline phase as the current reading. The ISM work below stays as a test of one specific thesis. Current reading: expansion for 2026-10 (provisional).

From 2018 on, the best 1-year medians followed recovery and contraction and the worst followed expansion and slowdown; read against the ISM results, a pattern of bitcoin doing best from the weaker phases would fit bitcoin leading the business cycle rather than following it. The sample covers only a few cycles, and 2013–17 rose strongly in every phase.

The ISM thesis

Tom Lee of Fundstrat has argued that bitcoin increasingly trades on monetary policy and the business cycle rather than the four-year halving clock. Three of his claims, as reported, are testable (paraphrased):

1. Bitcoin's strongest advances come when the ISM manufacturing index is above 50 (CNBC interview, December 2025, as reported by Crypto Briefing).
2. Bitcoin has not made a lasting cycle top while manufacturing was contracting, ISM below 50 (Fundstrat on X, January 2026, as reported by Asianet Newsable).
3. Bitcoin's distance from trend is almost perfectly correlated with the ISM (same CNBC interview).

ISM history from FMP's economic calendar begins with the March 2013 release; each reading counts from its actual release date. December readings are stored as revised in the following January release while keeping the original release date, a minor look-ahead in one month a year. Latest: 54.5 for 2026-09 (released 2026-10-01), 9 consecutive months above 50; regime: expanding, rising.

Claim 1 — strongest with ISM above 50. From 2018 on, the median 1-year return was +16% on days when the ISM was at or above 50 and +95% when it was below (2013–17: +123% and +166%). By level and direction (direction = 3-month change, so this table starts in July 2013):

Claim 2 — no tops during contraction. ISM at each cycle top (latest release before the peak day): 2013 57.3, 2017 58.2, 2021 60.8, 2025 49.1; at troughs: 2015 55.1, 2018 59.3, 2022 50.2. It held through 2021 but fails on the current dating of the 2025 peak, which came with the ISM at 49.1. That dating is provisional: Lee argued in January 2026 that this was not the cycle top, and it is confirmed only when the cycle completes.

Claim 3 — detrended bitcoin tracks the ISM. Same-month correlation of the ISM with bitcoin's distance from its 200-day average is 0.05; with bitcoin's 1-year log return, 0.30. The strongest relationship is bitcoin's 1-year return with bitcoin leading the ISM by 6 months (0.53).

Verdict. (1) not supported: returns were stronger with the ISM below 50; (2) fails on the current, provisional dating of the 2025 top; (3) not supported as stated: bitcoin leads the ISM rather than tracking it. Bitcoin has behaved like an asset that discounts the manufacturing upturn early, so the best entries came while the ISM was still weak. The business-cycle condition below uses a broader test (ISM at or above 50, or rising), so its figures differ from claim 1's.

The four business-cycle conditions

• ISM manufacturing: US ISM manufacturing PMI, counted from its release date. Above 50 means manufacturing is expanding. Bullish condition: at or above 50, or up over 3 months.
• US dollar index: DXY (DX-Y.NYB) at month end. Bullish condition: lower than 6 months earlier.
• Copper ÷ gold: Copper (HGUSD) ÷ gold (GCUSD) × 1,000 at month end: a gauge of growth expectations versus safe-haven demand. Bullish condition: higher than 6 months earlier.
• Fed funds rate: Fed funds rate, monthly (FMP federalFunds). "Flat" allows up to +0.10 percentage point of drift so that a Fed on hold counts as flat. Bullish condition: flat (within +0.10 pt) or lower than 6 months earlier.

Number of conditions met (if more met meant better returns, medians would rise toward the top):

13. Global liquidity

Thesis. Bitcoin is a liquidity asset. When central banks expand their balance sheets, when the Fed's balance sheet net of the Treasury's cash account and reverse repo grows, and when money supply accelerates, more money chases scarce assets.

The global measure uses the Fed, ECB and Bank of Japan (G3) balance sheets converted to dollars; current M2 for the euro area, China and Japan is not available on free sources.

• G3 central-bank balance sheets: Fed + ECB + Bank of Japan total assets in US dollars ($ trillion), month end. Bullish condition: rising over 6 months.
• US net liquidity: Fed balance sheet minus the Treasury's cash account and overnight reverse repo ($ trillion), weekly. Bullish condition: rising over 13 weeks.
• US M2 growth: Year-over-year growth of US M2 money stock; usable from the start of the second month after the reference month. Bullish condition: positive and accelerating over 3 months.

Number of conditions met (if more met meant better returns, medians would rise toward the top):

14. Real rates and risk sentiment

Thesis. Bitcoin pays no yield, so falling real yields lower its opportunity cost; and it trades as a high-beta risk asset, so calm equity markets, narrowing credit spreads and an equity uptrend should help.

• 10-year real yield: 10-year TIPS yield, month end. Bullish condition: lower than 6 months earlier.
• Corporate credit spread: Moody's Baa corporate yield minus the 10-year Treasury, month end. Bullish condition: narrower than 6 months earlier.
• Equity volatility (VIX): VIX at month end versus its trailing 12-month median. Bullish condition: below its 12-month median.
• S&P 500 trend: S&P 500 at month end versus its 10-month average. Bullish condition: above its 10-month average.

Number of conditions met (if more met meant better returns, medians would rise toward the top):

15. Adoption and network

Thesis. More users and more security make the network more valuable: active addresses, addresses holding coins, and hash rate (miners' investment in security). Address counts are imperfect, because exchanges batch transactions and one holder can use many addresses.

• Active addresses: Addresses sending or receiving bitcoin, 30-day average (Coin Metrics). Bullish condition: higher than a year earlier.
• Addresses holding bitcoin: Addresses with a non-zero balance, 7-day average (Coin Metrics). Bullish condition: 1-year growth above its own 1-year median.
• Hash rate (hash ribbon): Network hash rate: 30-day average versus 60-day average. A 30-day average below the 60-day marks miner capitulation. Bullish condition: 30-day above 60-day average.

Number of conditions met (if more met meant better returns, medians would rise toward the top):

16. Flows and positioning

Thesis. Coins leaving exchanges for long-term custody, and dollars building up inside crypto as stablecoins, are the fuel for the next leg up.

• Bitcoin on exchanges: Share of supply held on exchange wallets (Coin Metrics). Falling means coins are moving into longer-term custody. Bullish condition: lower than 90 days earlier.
• Net exchange flow: Coins sent to exchanges minus coins withdrawn, 30-day sum (BTC). Bullish condition: net outflow over 30 days.
• Stablecoin supply: USDT + USDC outstanding ($ billion): dollars parked inside crypto, ready to buy. Bullish condition: higher than 90 days earlier.
• Spot ETF holdings (building history): US spot bitcoin ETFs (IBIT, FBTC, ARKB, BITB, GBTC), BTC-equivalent held. FMP has no history of ETF shares outstanding, so the series is built forward from a daily snapshot. Collected since 2026-10-23 (1 day); latest about 1201.6k BTC ($97.4 bn). The condition will be shown after 60 days; a forward-return test needs at least two years of history.

Number of conditions met (if more met meant better returns, medians would rise toward the top):

17. Backtest design

• Question: when a score or condition was in a given state, what happened to bitcoin over the next 6 months, 1 year and 2 years (close to close)?
• Periods: valuation results for 2012–17 and 2018 on; backdrop results for 2013–17 and 2018 on. Signal dates do not overlap, but forward windows from late-2017 days run into 2018–19 (12 months) and 2019 (24 months). Agreement between the periods is stronger evidence than either alone, though 2013–17 is only five years and some cells are small. The early years are dominated by enormous returns in every state.
• Valuation buckets: composite and lens scores in five bands of 20 points; premium to fair value in six bands.
• Rank correlation: Spearman correlation of each day's score with the next 365 days' return, by period and by halving epoch. Negative is what a valuation measure should show. The 2024– epoch has less than two years of completed forward returns.
• Analogs: every past day with a composite within 5 points of today's, with the number of distinct episodes (runs separated by more than 60 days). Horizons can share the same days when no similar reading occurred in the most recent year.
• Baseline: the median 1-year return of all days since 2018 was +40% (64% up); compare any conditional result with that.
• Hindsight in event tables: cycle peaks and troughs are dated after the fact, so readings on those days are naturally near the extremes. The forward-return tests, not the event table, are the evidence.
• Not done on purpose: the valuation and backdrop evidence uses no trading rules, Sharpe ratios or optimized thresholds. With about four cycles and overlapping windows, such numbers would look precise and mean little. The one trading rule in the framework (sections 8–9) was fixed before its results were seen, and its results are shown by start year so that their sensitivity is visible.

18. Valuation results

18.1 Composite score buckets — 1-year forward return

18.2 Premium to model fair value — 1-year forward return

Rank correlation of the premium with 1-year return: -0.01 (2012–17), -0.64 (2018 on).

18.3 Rank correlation with 1-year forward return, by halving epoch

• 2024– epoch: less than two years of completed forward returns.

18.4 Readings at past peaks and troughs

18.5 What followed readings like today's (45)

18.6 What the valuation evidence does and doesn't say

• 2018 on: the composite and the fair-value premium sorted forward returns cleanly. Low scores and discounts to fair value were followed by gains on most days; high scores and large premiums mostly by losses. This rests on about one and a half cycles: one complete trough-to-trough cycle (2018–22) and one in progress.
• 2012–17: everything rose so much that rank order barely mattered; treat these numbers as context, not a guide to magnitudes now.
• Tops have been getting less extreme. The trailing-window composite read 96 in 2013, 100 in 2017, 81 in 2021, 77 in 2025 on the four price-top days. That is consistent with a maturing market, but four observations are too few to project.
• Bottoms have registered clearly. Every bear-market low since 2015 scored between 1 and 12 and sat 49–75% below fair value; these dates are known only in hindsight, so the forward-return tests are the stronger evidence.
• Sample size is the binding constraint. About four full cycles; day counts overstate the independent evidence because consecutive days share most of their forward window.

19. Data and pipeline

Checks. FMP and Coin Metrics closes agree within about 0.2% on overlapping days. Realized price, the 200-week average, the Puell multiple and the trailing MVRV percentile were recomputed independently from the raw files and matched the model. The ISM history was cross-checked for gaps (none from March 2013) and December revisions.

Automation. A GitHub Actions job runs daily at 14:07 UTC and on every push to the main branch. refresh/fetch.py pulls new data from Coin Metrics, FRED and FMP (a failed source keeps its last good file); refresh/model.py (with macro.py, backdrop.py, strategy.py and markets.py) rebuilds every score, fair value, signal, test and strategy, and appends the day to the live logs; tools/docs.py regenerates this document; tools/build.py builds the page; the job commits the data and deploys to GitHub Pages. FMP series need the FMP_API_KEY secret; Coin Metrics and FRED need no key.

20. Limitations

• About four cycles is a small sample, and the regime (spot ETFs, corporate treasuries, lower volatility) may be changing both amplitude and timing.
• Fair value moves with its anchors and can trail a fast market.
• Many backdrop indicators are tested, so some verdicts will reflect chance; favor verdicts that agree across both periods.
• On-chain address and exchange measures are estimates; batching, custodians and exchange consolidation change how coins move.
• Monthly market data is month-end; the latest month on the page can be month to date, while signals use only completed, published data.
• Spot ETF flows have no free history; the forward-built series is shown after 60 days and can be tested only once at least two years exist.
• December ISM readings are stored as revised in January while keeping the original release date (a minor look-ahead).
• Coin Metrics Community data is licensed CC BY-NC 4.0 (non-commercial use with attribution).
• The next halving date is estimated from block times.
• Ether's composite starts only in mid-2019 and rests on two lenses; a fee-revenue lens was dropped after its structural break (section 7).
• Strategy simulations use point-in-time scores, but the lenses and zone cut-offs were chosen by people who had seen earlier cycles; results depend heavily on the start date.
• A personal research project; views are the author's own and not those of any employer. For research and discussion; not investment advice.

21. Possible extensions

• Spot ETF flow history: issuer daily holdings files or a flow aggregator would add January 2024 onward.
• China liquidity: PBoC balance sheet and China M2 (not current on free sources).
• Holder cohorts: short- and long-term holder cost basis and SOPR (paid on-chain data, such as Glassnode).
• Derivatives positioning: futures basis, funding rates and open interest.
• GMAAF link: a bitcoin sleeve in GMAAF's real-assets model driven by the composite and the fair-value premium.

22. Glossary

• Realized price: the average price at which every bitcoin last moved on-chain.
• MVRV: market value ÷ realized value.
• Puell multiple: daily miner revenue ÷ its one-year average.
• Power law: a straight line in log price versus log time.
• Percentile score: where today's reading ranks within the comparison window, 0 (lowest) to 100 (highest).
• Model fair value: the geometric average of the four prices at which each lens would sit at its median, each holding its own anchor fixed.
• ISM manufacturing PMI: monthly US purchasing managers' survey; above 50 means activity is expanding.
• Net liquidity: the Fed's balance sheet minus the Treasury General Account (the government's cash at the Fed) and overnight reverse repo (cash parked at the Fed by money funds).
• TIPS real yield: the yield on inflation-protected Treasuries, a market measure of real interest rates.
• Baa spread: Moody's Baa corporate bond yield minus the 10-year Treasury yield, a gauge of credit risk appetite.
• Hash ribbon: the 30-day versus 60-day average of hash rate; the 30-day dropping below the 60-day marks miner capitulation.
• Exchange balance: bitcoin held in exchange wallets, as a share of supply.
• Stablecoin supply: USDT plus USDC outstanding, dollars held inside the crypto system.
• Supported / reversed / no clear edge: whether bitcoin did meaningfully better (supported) or worse (reversed) on days a backdrop condition was met.
• Spearman correlation: correlation of ranks, robust to bitcoin's extreme returns.

23. References

• Coin Metrics, Community Network Data and metric definitions.
• Federal Reserve Bank of St. Louis, FRED.
• Financial Modeling Prep, market, economic and ETF data.
• Mahmudov, M. and Puell, D. (2018), the MVRV ratio; Puell, D. (2019), the Puell multiple.
• Santostasi, G., the bitcoin power-law model.
• Edwards, C. (Capriole Investments), hash ribbons.
• Tom Lee (Fundstrat): CNBC interview, December 2025, on the ISM and bitcoin; Fundstrat on X, January 2026, on bitcoin tops and manufacturing contraction.