Market Intelligence · Cycle Models

Ben Cowen's Logarithmic Regression Bands and Diminishing Returns, Tested

What a log regression fair-value curve is, how Cowen's bands and risk metric work, what is not public, and a reproducible test on every Bitcoin cycle, ETH and ETH/BTC, including the case against it.

15.3x → 1.2x
Peaks above fair value
39x → 1.85x
Peak-to-peak multiple
0.40x
BTC vs curve, Nov 20
5,970
Daily closes fitted

Published 2026-11-21 · Updated 2026-11-21 · Deep Blue Alpha

Not Financial Advice. This article explains a public analysis framework and tests it against historical prices. It does not forecast prices and does not recommend any trade. Curves fitted to past data describe the past; they do not tell anyone what happens next. Deep Blue Alpha has no connection to Benjamin Cowen or Into The Cryptoverse, and nothing here implies his endorsement. Always do your own research.
Live on Deep Blue Alpha: today’s whale moves →
Learn the basics: Trend and regression models (free lesson)
Quick Answer · TL;DR

Benjamin Cowen of Into The Cryptoverse draws a logarithmic regression “fair value” curve through Bitcoin’s price history, with bands above and below it. His framework holds that each cycle peak reaches a smaller multiple above fair value than the one before. That is the diminishing-returns thesis. His exact coefficients and his 0–1 risk metric formula are not public.

We rebuilt a simple version from 5,970 Coin Metrics daily closes: log10(price) = −16.3206 + 5.6321·log10(days since Jan 3, 2009). On that line the five cycle peaks sat 15.3x, 11.1x, 6.6x, 3.0x and 1.2x above fair value. Peak-to-peak gains fell from 39.1x to 17.3x to 3.4x to 1.85x. On November 20, 2026 BTC closed at $58,851, about 0.40x the line, its lowest reading on this fit since August 2015. The pattern is real in the data. It also rests on five peaks, and a different fit gives different levels. Updated November 2026.

Who Ben Cowen is, and what is his versus what is borrowed

Benjamin Cowen runs Into The Cryptoverse, a YouTube channel and research site (intothecryptoverse.com) known for long-horizon Bitcoin cycle analysis. Over several years he has published recurring videos on Bitcoin and Ethereum logarithmic regression, a proprietary risk metric, the bull market support band, and Bitcoin dominance.

Some credit lines matter here. Log-growth curves for Bitcoin were already circulating before Cowen’s version, which he has said he built in 2019. His distinct contribution is the framing: a fair-value line fitted to non-bubble data, bands stepping lower each cycle, and cycles that lengthen. The Pi Cycle Top and 2-Year MA Multiplier indicators that often appear next to his name were created by Philip Swift, not Cowen. MVRV came from Murad Mahmudov and David Puell.

A sourcing note. Cowen does not publish a methodology page for these bands, and much of what circulates about them comes secondhand: transcript summaries of his videos and news write-ups. Where a claim below comes from one of those, we say so. We do not repeat his price targets as facts. Where a past view matters for testing the framework, we describe it as his past stated view and say whether it matched what happened.

What a log regression fair-value curve is, with the math

Plot Bitcoin’s price on a normal chart and early history disappears into a flat line at zero. Plot it on a log price axis and you see a curve that bends over: growth was fastest when Bitcoin was young and slowed as it aged. Put time on a log axis too, and much of that bend straightens out. A straight line on a log-log chart has a simple formula:

log10(price) = a + b · log10(days since genesis)
fair value = 10a × daysb

Here “days since genesis” counts from Bitcoin’s first block on January 3, 2009. The slope b says how fast fair value grows as Bitcoin ages. The intercept a sets the level. Fitting it is ordinary least squares, the same method behind any trend line in a spreadsheet.

Our fit, on public data

We fitted every Coin Metrics daily close from July 18, 2010 to November 20, 2026 (5,970 days). The result is a = −16.3206 and b = 5.6321. The typical miss (standard deviation of the residuals) is 0.30 in log10 terms, so price has usually sat within roughly half to double the line. Because b is about 5.63, each doubling of Bitcoin’s age multiplies this curve’s fair value by about 25.63 ≈ 50. At Bitcoin’s current age, the fitted line climbs about 36% per year. That is a property of the line, not a statement about future prices.

Worked example: the October 2025 peak

  1. October 6, 2025 is day 6,120 after January 3, 2009.
  2. log10(6,120) = 3.7868.
  3. log10(fair value) = −16.3206 + 5.6321 × 3.7868 = 5.0068.
  4. Fair value = 105.0068 ≈ $101,570.
  5. BTC closed at $124,824 that day, so the peak sat 1.23x above the line.

The same arithmetic for November 20, 2026 (day 6,530) gives a fair value of about $146,345. BTC closed at $58,851, which is 0.40x the line.

Our coefficients are not Cowen’s. His fit window, data source and bubble filter are not public. This is a transparent stand-in that anyone can reproduce, so the framework’s logic can be tested in the open.

How Cowen uses bands above and below the curve

A single line is not the whole framework. Cowen’s charts show a fair-value band plus parallel bands above it, often drawn as a color-coded “rainbow.” According to transcript summaries of his 2021 and 2022 videos, he describes the fair-value band as traced through “non-bubble data.” In other words, the bubble peaks are left out so they don’t pull the line upward.

Two uses come up repeatedly in his public work, per those summaries:

  • Peaks reach lower bands each cycle. His framework holds that each bubble tops out at a band below the previous cycle’s top band.
  • Fair value as a magnet in bear markets. He has argued that price tends to return to fair value before any sustained move back toward the prior peak band.

A filter changes the answer. When we refit the line using only days that sat no more than 0.15 log10 above the curve (3,014 of 5,970 days, iterated five times), the coefficients become a = −17.4165, b = 5.8793. That lower “non-bubble” line puts the peaks much higher above fair value, and today’s price closer to it (0.57x). Same data, same method, different levels. Keep that in mind for the criticism section.

The diminishing-returns thesis, tested on every cycle

Cowen’s framework holds that Bitcoin’s cycles show diminishing returns and lengthening cycles: each peak is a smaller multiple than the one before. Here is the raw record from one consistent daily-close series.

Bitcoin cycle peaks, lows and drawdowns (daily closes)

CyclePeak closePeak-to-peakLow closeDrawdownDays to low
2011$29.03 (Jun 8, 2011)—$2.11 (Nov 18, 2011)−92.7%163
2013$1,134.93 (Dec 4, 2013)39.1x$175.64 (Jan 14, 2015)−84.5%406
2017$19,640.51 (Dec 16, 2017)17.3x$3,185.07 (Dec 15, 2018)−83.8%364
2021$67,541.76 (Nov 8, 2021)3.44x$15,758.29 (Nov 9, 2022)−76.7%366
2025$124,824.45 (Oct 6, 2025)1.85x$58,525.08 (Jun 30, 2026)*−53.1%267

Source: Coin Metrics community data, daily PriceUSD, pulled November 20, 2026. Daily closes sit below intraday highs: CoinGecko lists the 2025 intraday ATH at about $126,080 on Oct 6, 2025. *Lowest close since the 2025 peak as of Nov 20, 2026; not confirmed as a final cycle low. Peak-to-peak spacing: 910, 1,473, 1,423 and 1,428 days.

The gain from each cycle low to the next peak fell even faster: 539x (2011 low to 2013 peak), 112x, 21.2x, and 7.9x (2022 low to 2025 peak). Drawdowns shrank too, from 92.7% to 53.1% so far.

Now the same peaks measured against the fair-value curve, which is the comparison Cowen’s bands are built for.

Each BTC peak and low as a multiple of fair value (two fits)

CycleAll-data line at peakPeak ÷ linePeak ÷ “non-bubble” lineLow ÷ line
2011$1.9015.3x35.5x0.43x
2013$101.8511.1x21.8x0.55x
2017$2,9716.6x11.2x0.59x
2021$22,7443.0x4.6x0.45x
2025$101,5701.23x1.8x0.45x

All-data line: log10(P) = −16.3206 + 5.6321·log10(days). “Non-bubble” line: log10(P) = −17.4165 + 5.8793·log10(days), fitted only on days within +0.15 log10 of the curve. Both are Deep Blue Alpha fits for illustration, not Cowen’s. Low ÷ line uses the all-data line on the low date.

Three things stand out. First, the peak multiples fall every single cycle on both fits. That is the diminishing-returns pattern, and it is plainly in the data. Second, the levels depend on the fit: 2021 was 3.0x or 4.6x above fair value depending on which line you trust. Third, the lows did not shrink the same way. They sat between 0.43x and 0.59x the line in every cycle, and the 2026 low (0.45x) matched the 2022 low (0.45x) on this fit.

An out-of-sample check

The fair test of any fitted curve is how it handles data it never saw. We refit the line using only data through December 31, 2020. That earlier line put the 2021 peak at 2.7x fair value and the 2025 peak at 1.09x, and both later lows at about 0.40–0.41x. A fit through 2016 put the 2017, 2021 and 2025 peaks at 7.2x, 3.2x and 1.3x. So the falling-peak pattern survived out of sample. The exact fair-value numbers did not hold still: the 2020 fit’s line for November 20, 2026 is about $166,900, versus $146,300 on the full fit and $103,000 on the non-bubble fit.

Cowen’s risk metric: what is public and what is not

Cowen’s best-known tool after the bands is his risk metric, a single number from 0 to 1 shown as a color scale. His August 29, 2021 video, “Bitcoin Risk Metric: How I Navigate Crypto,” is the main public explanation. According to a transcript summary of that video:

  • Blue readings mark “historically low risk” and red readings mark historically high risk.
  • He says the metric accounts for diminishing returns and diminishing volatility, and that its levels adapt over time.
  • He cites readings of about 0.12 in March 2020 and about 0.94 near $60K in 2021.
  • He frames it as his own personal framework, not financial advice and not a day-trading tool.

What is not public: the formula, the inputs, the normalization, and whether historical readings are recalculated when the model updates. According to the same summary, more detail goes to members of his paid service; the general public does not get it. No independent backtest of the risk metric turned up in our research. Because the levels adapt over time, past readings shown today may not match what the metric displayed in real time. That makes track-record claims hard to check from outside. The honest description is a closed, normalized oscillator with a stated design goal. Anyone presenting a formula for it is presenting their own approximation, not his method.

How the 2025 cycle top compared with his framework

This is where a framework can be tested against a view recorded in advance. In November 2023, with BTC at $37,109, The Daily Hodl reported Cowen’s stated view: each cycle top had sat about two and a half log regression bands below the previous one. He said that pattern pointed to a late-2025 top of “just north of $100,000.” The same article mentioned a rainbow-based figure near $111,000, so the report itself carried two numbers.

Cowen’s past stated view (Nov 2023) versus what happened

ElementPast stated view (as reported)What happenedMatched?
TimingLate 2025, Q4 of the post-halving yearPeak close Oct 6, 2025Yes
Level“Just north of $100,000” (~$111K also cited)$124,824 close; ~$126,080 intradayAbove the stated level
Smaller multiple than 2021Diminishing returns1.85x prior peak; 1.2x our fair-value lineYes

Sources: The Daily Hodl, Nov 13, 2023; Cointelegraph Magazine, Mar 6, 2026; Coin Metrics daily closes; CoinGecko intraday ATH.

Cointelegraph Magazine (March 2026) quoted Cowen saying Bitcoin tops in the fourth quarter of every cycle, and October 2025 fit that. The same piece reports he named his own biggest miss: in Q4 2023 he had thought BTC would stall below $35,000, and it broke out instead. Credit where due: he said so publicly. Taken together, the record shows the framework getting the direction of returns and the timing right in 2025 while the price level ran above his stated view.

After the peak, BTC lost about 53% to a June 30, 2026 close of $58,525 and closed at $58,851 on November 20, 2026. On our all-data fit that is 0.40x the line, the lowest reading since August 2015 (0.38x). On the non-bubble fit it is 0.57x. Which reading you accept depends entirely on which curve you draw, which is the core weakness of the method.

ETH’s fair-value curve and the ETH/BTC ratio

Cowen applies the same log regression approach to Ethereum, according to transcript summaries of a video series running from 2021 to 2023. Those summaries show his quoted ETH fair-value figures moving up between videos, from about $339 in an early update to about $600 in mid-2022 and about $875 in mid-2023. That rising number is what a refit on new data produces. In August 2023, The Daily Hodl reported his view that ETH was returning to its regression fair value.

Our own ETH fit uses daily closes from August 8, 2015 to November 20, 2026 (4,123 days), counted from the July 30, 2015 genesis block: log10(price) = −4.0483 + 2.1068·log10(days), with a residual spread of 0.36 log10.

ETH peaks and lows against our ETH fair-value line

PointDateETH closeLine valueMultiple
PeakJan 13, 2018$1,380.14$1499.3x
LowDec 15, 2018$82.83—0.28x
PeakNov 8, 2021$4,811.16$1,0754.5x
LowJun 18, 2022$992.79—0.76x
PeakAug 22, 2025$4,831.10$2,9061.66x
LowJun 6, 2026$1,564.52—0.46x
LatestNov 20, 2026$1,618.83$3,7160.44x

Source: Coin Metrics daily PriceUSD. ETH drawdowns: −94.0% (2018), −79.4% (2021–22), −67.6% (Aug 2025 to Jun 2026 low). Our fit, not Cowen’s.

The ETH peaks also shrink relative to the line: 9.3x, 4.5x, 1.66x. But ETH has three cycles, not five. Its slope (2.1) is far gentler than Bitcoin’s (5.6), and with about 11 years of history the fit leans heavily on the first year of trading. Change that first year and the line moves a lot. Treat the ETH curve as a rough sketch.

ETH/BTC

Cowen has described ETH/BTC as trending along a long-run log regression with a base that rises each cycle, and he long watched a 0.04 level, according to news reports. Here is the ratio from the same daily-close data:

ETH/BTC ratio highs and lows (daily closes)

PointDateETH/BTC
All-time highJun 12, 20170.1460
Cycle lowSep 6, 20190.0164
Cycle highDec 8, 20210.0877
Cycle lowApr 21, 20250.0181
Rebound highAug 24, 20250.0422
LatestNov 20, 20260.0275

ETH close ÷ BTC close, Coin Metrics daily PriceUSD.

The April 2025 low (0.0181) sat slightly above the 2019 low (0.0164), which fits the “rising base” description, though only barely. The ratio did reach 0.04 in August 2025, then gave back much of the move. The highs have fallen each cycle (0.146, 0.088, 0.042), which is diminishing returns applied to ETH against BTC.

What tracked Ethereum whales did this week

Curves describe history; on-chain data describes what large holders are doing now. Over the seven days to November 20–21, 2026, Deep Blue Alpha tracked 547 whale wallets moving ETH. In real DEX swaps, those wallets bought about $21.0M of ETH and sold about $46.7M, so the swap side leaned to selling. On the exchange side, $374.4M of ETH was withdrawn from exchanges against $327.3M deposited, a mild accumulation-side read. Swaps were under 10% of that week’s whale volume, so the exchange numbers describe transfers, not purchases. The live numbers are on the ETH whale page.

The criticism: curve fitting, small samples and moving targets

A fair look at log regression bands has to include the case against them. Most published critiques target log-growth and power-law Bitcoin models in general rather than Cowen by name, but they apply directly.

Curve fitting on very few cycles

Bitcoin has five cycle peaks. Any rule of the form “each peak sits N bands lower” is fitted on four or five points that are not independent of each other. Five points can fit many shapes. The drawdown series (92.7%, 84.5%, 83.8%, 76.7%, 53.1%) looks steady too, and it is the same small-sample problem.

Refitting moves fair value

Every new month of data shifts the line a little, and choosing which days count as “bubble” shifts it a lot. In our tests, fair value for the same day ranged from about $103,000 to $167,000 depending on the window and filter. A curve that is refitted after every miss can look accurate in hindsight. The meaningful test is a projection written down in advance, such as the November 2023 view above.

Time is not a cause

The x-axis is days since genesis. Time itself does not set prices; adoption, liquidity, leverage and regulation do. Critics of power-law models make this point often. A time-only fit summarizes the path. It does not explain it.

Survivorship

Bitcoin is the asset that survived and grew. The same method applied to failed coins would have shown a smooth curve right up to the collapse. Fitting the winner makes the method look better than it is.

Power law and stock-to-flow comparisons

A log-log regression is a power law. The best-known version, promoted by Giovanni Santostasi, uses the same shape. A May 2026 arXiv preprint by Carlos Baquero and Raquel Menezes (not peer-reviewed) measured an exponent of about 5.7 for 2010–2026 with a residual spread of 0.30 log10. That is very close to our 5.63 and 0.30, which is a useful cross-check. The authors found the fit beat simple benchmarks at 12–24 month horizons but failed structural tests, including sensitivity to where the time count starts. Stock-to-flow, the supply-based model associated with PlanB, is a different model. It projected roughly $100K for 2022, and BTC fell to about $17,600 that June, below its lower band. The log regression framework, with its diminishing returns built in, aged better than stock-to-flow over that stretch. That still does not make it a forecasting tool.

The fair summary: the diminishing-returns pattern is clearly present in Bitcoin’s price history and held out of sample. The specific fair-value number, the band levels and any target derived from them depend on choices that are not public in Cowen’s case and arbitrary in everyone’s.

Frequently asked questions

What are Ben Cowen’s logarithmic regression bands?

A fair-value curve fitted to Bitcoin’s price on log scales, plus parallel bands above and below it. Cowen describes the fair-value band as fitted to non-bubble data and says he built his version in 2019. His coefficients are not published.

What is the formula?

log(price) = a + b·log(days since Jan 3, 2009). Our fit on 5,970 daily closes gives a = −16.3206 and b = 5.6321 in log10 terms.

What does diminishing returns mean here?

Each peak was a smaller multiple of the one before (39.1x, 17.3x, 3.4x, 1.85x) and of the fair-value line (15.3x down to 1.2x on our fit).

What is the risk metric?

A proprietary 0–1 indicator from Cowen. He says it accounts for diminishing returns and volatility. Its formula is not public and no independent backtest was found.

Did the 2025 top fit his framework?

The Q4 timing and the smaller multiple matched his November 2023 view. The level ($124,824 close) came in above the “just north of $100,000” figure reported then.

Is this the same as the power law?

Mathematically, yes: a straight line on log-log axes is a power law. Versions differ in data window, bubble filter and bands.

Bottom line

Ben Cowen’s log regression bands rest on a simple idea: draw a fair-value line through Bitcoin’s history on log scales, and measure each bubble against it. Rebuilt in the open on 5,970 daily closes, that idea holds up on its central claim. Every Bitcoin peak from 2011 to 2025 sat a smaller multiple above the line than the one before, on two different fits and out of sample. Peak-to-peak gains fell from 39.1x to 1.85x. The 2025 top arrived in Q4 as his framework described, at a level above his stated 2023 view.

The weaknesses are just as real. Five peaks is a tiny sample. Fair value moves with every refit and every choice of bubble filter, so today’s reading ranges from 0.40x to 0.57x of the line depending on how it is drawn. The risk metric is closed, so it cannot be independently checked. Used as a way to describe where price sits against its own history, the bands are clear and useful. Used as a target machine, they promise more than five data points can deliver.

Disclaimer. Nothing in this article is financial, investment, tax or legal advice. All prices come from public historical data; all regression fits are Deep Blue Alpha illustrations, not Benjamin Cowen’s models. Descriptions of Cowen’s views are based on his public videos and secondhand reports cited inline and may not reflect his current thinking. Past patterns do not predict future results. Deep Blue Alpha is a data platform, not a financial advisor.

See what whales are doing, not what a curve says

Deep Blue Alpha tracks 20,000+ Ethereum whale wallets in real time, with real DEX buys and sells shown separately from exchange transfers.

Open the live whale feed →

Related reading

ETH whale activity → Whale Sentiment Index → Live whale feed → BTC/ETH bottom signals guide →