Bitcoin bear market bottoms: what the data supports (and what it doesn’t)

People ask the same questions every cycle.

Is the Bitcoin bottom in?
What marks a bear market bottom?
Which indicators actually work?

I ran the numbers on a few popular “bottom signals” and pulled published reporting on a few more.

Here is the honest answer.

Most indicators do not find bottoms. They confirm zones. They confirm regime shifts. They tell you when risk is changing. They rarely tell you “buy here.”

That is less exciting. It is also what the data supports.

---

TL;DR

• The cleanest repeating zone marker I can point to is Bitcoin supply held at an unrealized loss around ~10.5M coins. It has shown up near every major low, but it has arrived before a final leg lower in prior cycles.
• Weekly oversold plus a reclaim of the 50-week EMA has confirmed major lows in my backtest. It is confirmation, not discovery. It triggers months late and often 30% to 65% above the low.
• Fear & Greed is not contrarian in my backtest. It correlates positively with forward BTC returns at 7, 30, and 90 days. The popular “fear means buy” story does not hold in the general case.
• The live test in the current cycle is real. Supply-in-loss crossed the threshold in early June 2026. BTC then put in its lows between early June and the first days of July, bottoming around $58,000 to $58,600 depending on the source. BTC is just over $75k late on Sep 15.

Here is where BTC sits today.

---

What you are trying to measure when you say “bottom”

A bear market low is not one thing.

Sometimes it is a liquidation event. Sometimes it is slow exhaustion. Sometimes it is macro.

So the only honest way to talk about “bottom signals” is to say what each signal can do.

Does it flag stress?
Does it flag exhaustion?
Does it confirm recovery?
Does it only look good in hindsight?

I’ll keep that framing throughout.

---

Indicator 1: Bitcoin supply in loss (~10.5M) marks the zone

Claim: At every major bear market low, BTC supply held at an unrealized loss has crossed roughly 10.5 million coins.

This is the cleanest repeating zone marker in this piece.

Instances (historical)

Published reporting that cites Glassnode has documented the same zone around prior major lows (2018, 2020, 2022). One example is this CoinDesk piece from June 2026: https://www.coindesk.com/markets/2026/06/04/this-bitcoin-metric-has-marked-every-bear-market-bottom-and-it-s-just-flashed-again/

2026: what I see on the Glassnode chart

These readings are from Glassnode's BTC Percent Supply in Profit chart, daily resolution. Other providers run a few points lower on the same dates.

It shows a threshold crossing, then a second and deeper dip.

• Early June: supply in profit dipped to roughly 47% to 48%, meaning 52% to 53% in loss, with BTC around $61,000.
• Later in the low period: supply in profit dipped to roughly 45% to 46%, meaning 54% to 55% in loss, with BTC bottoming around $58,000 to $58,600 depending on the source.

At roughly 20 million BTC circulating, 54% to 55% in loss implies about 10.8 to 11.0 million coins in loss.

CoinDesk’s June 4 reporting lines up with the early threshold crossing and is useful corroboration: https://www.coindesk.com/markets/2026/06/04/this-bitcoin-metric-has-marked-every-bear-market-bottom-and-it-s-just-flashed-again/

Why it works (this is the part that matters)

This is not just a pattern someone noticed.

K33’s head of research (as quoted in published reporting) explains there is a mechanical ceiling. A large share of old coins never move. They are lost, or held by people with no intention of selling. Those coins never register as “in loss,” because their cost basis never updates.

That caps “supply in loss” in the rough 50% to 56% range across prior bear markets. https://nftplazas.com/bitcoin-underwater-supply-bottom-signal/

The caveat, and why the June sequence matters

In prior cycles, crossing the threshold has often arrived before one final leg lower.

That is exactly what the June 2026 sequence shows.

The threshold crossed in early June. BTC then made one more leg lower into the lows around $58,000 to $58,600 depending on the source.

So this is not a day-picker. It marks the zone.

How you use it

When supply in loss approaches this zone, stop assuming downside is infinite. Start asking what would force sellers to keep selling from here.

That does not mean price cannot fall. It means the supply dynamics are changing.

---

Get the daily brief

One email each morning: the scores that moved overnight, the coins that crossed a rating band, and the next big token unlock.

Email address Send me the brief

Double opt-in. Unsubscribe in one tap.

Indicator 2: Weekly oversold plus reclaim of the 50-week EMA (confirmation only)

This is my backtest. It is not a published benchmark.

I tested BTC weekly candles from Sep 2013 to Sep 2026, about 700 weeks. I used Wilder RSI(14) and a 50-week EMA.

Claim: After BTC goes oversold on the weekly and then reclaims its 50-week EMA, it has not made a lower low in the completed bear markets in the sample.

Results (RSI < 30 cutoff)

• 2015-01-18: low $209. Reclaimed 50-week EMA on 2015-07-12 at $310. +48%, 175 days later.
• 2022-11-27: low $16,465. Reclaimed on 2023-03-19 at $27,125. +65%, 112 days later.

2026 is the live case.

• 2026 low period: BTC printed its lows between early June and the first days of July, bottoming around $58,000 to $58,600 depending on the source. BTC reclaimed the 50-week EMA on 2026-08-30 at $78,225.

2018 depends on a cutoff

If I widen the RSI cutoff to below 32 instead of 30, 2018 qualifies:

• 2018-12-16: low $3,216. Reclaimed on 2019-04-21 at $5,304. +65%, 126 days later.

What weakens this signal (and why you should care)

• It confirms a low that is already 4 to 6 months old. It triggers at +30% to +65% above the low. It is confirmation, not bottom-finding.
• 2018 is sensitive to a cutoff. Weekly RSI bottomed at 30.2. Whether it “counts” turns on 0.2 in an arbitrary threshold.
• The EMA cross alone is worthless. There have been 18 upward crosses since 2013. Only 3 or 4 were preceded by oversold. The RSI filter carries the claim.
• Bitcoin has had three completed bear markets, ever. The sample is small.

Backtesting this since 2014

This is my backtest. It is not a published benchmark.

Related reading on the same question, with a more negative result on a different indicator over the same period: Does the Crypto Fear and Greed Index Work?.

Setup: Weekly RSI(14) below 30 while price is under the 50-week EMA, followed by the first weekly close back above that EMA.

Data: CoinGecko BTC daily closes, 2013-04-28 to 2026-09-22. Weekly closes are the last daily close of each week ending Sunday, 701 weeks. Both the simple and exponential 50-week averages become available on 2014-04-06. The EMA is seeded with the simple average of the first fifty weeks so the two start on the same week. Weekly Wilder RSI(14) is available from 2013-08-04, so the test starts 2014-04-06.

Definition: An occurrence is an oversold week below the 50-week EMA followed by the first weekly close back above it. Several oversold weeks feeding one cross count as one occurrence.

A lower low means any daily close after the cross below the lowest daily close between the oversold week and the cross.

Results

Three occurrences since 2014:

• Oversold 2015-01-18, crossed back above the 50-week EMA on 2015-07-12 at $310. No lower low. 90d -20.9%, 365d +109.2%.
• Oversold 2022-06-19, crossed back above on 2023-03-19 at $27,125. No lower low. 90d -2.9%, 365d +152.3%.
• Oversold 2026-02-08, crossed back above on 2026-08-30 at $78,225. Unresolved, 23 days of elapsed data at the time of writing. Mark it pending, do not count it as a success.

So two occurrences have meaningful history and neither was followed by a lower low.

Why I trust this more (and what I rejected)

I tested a rival definition of oversold and rejected it.

The rival: a weekly close more than 20% below the 50-week moving average. This rival rule, and the worked example below, use the simple 50-week average rather than the EMA. A percentage gap to a fixed 50-week average is a simple-average rule by construction. The argument about how many weeks in the window sat above the tested price also only exists for a simple average.

That definition produces six occurrences, one of which failed badly. Oversold 2022-01-23, crossed above 2022-04-03 at $45,847, then fell to a low 55.2% beneath its prior low.

That rival definition is not measuring oversold. It measures distance from a lagging average, and that gap is widest shortly after a fast fall from a high, because the average still carries the prior bull market's prices.

Concretely, on 2022-01-23 the weekly close was $35,219 against a 50-week MA of $48,395. Of the 50 weeks in that average, 43 closed above the week being tested, and the window's high was $64,654. The all-time high was only 75 days earlier and the cycle low, $15,794, was still 123% below that week's close.

Compare the RSI firing five months later (still using the simple 50-week average for the comparison): close $19,032 against a 50-week MA of $43,684, a 56.4% gap, with 49 of the 50 window weeks above it.

The separation shows up in price, not time. Measured against each cycle's eventual low, using the close of the oversold week:

• 2015: low $172 (2015-01-14), oversold week close $209.
• 2022: low $15,794, oversold week close $19,032.
• 2026: low $58,566, oversold week close $69,246.

Against those lows:

• RSI below 30 fired 21.4%, 20.5% and 18.2% above the low.
• The 20%-below rule fired 34.3%, 61.8%, 91.9%, 123.0%, 125.1% and 130.4% above it.

Those ranges do not overlap.

By time the two are barely distinguishable, a median 222 days since the all-time high against 185, with heavily overlapping ranges. So the difference is not that one fires earlier on the calendar, it is that one fires much further from the bottom in price.

Caveats that matter

• Three occurrences, two resolved. That is not a sample anyone should trade.
• The base rate for "the low held" is 60%. The denominator is every weekly bucket that has both a 50-week average and at least 90 days of subsequent daily data. Prior low is the lowest daily close in the 26 weeks up to and including that week. A lower low is any daily close after that week below that prior low, checked to the end of the series rather than a fixed window. On that definition, 385 of 638 weeks were never followed by a lower low. In a series that went from roughly $135 to roughly $86,600, the low holding is the default outcome. Two out of two is indistinguishable from chance against that baseline.
• The 18 to 21% clustering rests on three observations and should be treated as partly luck.
• In the 2015 case the low printed on 2015-01-14, four days before the oversold week. I treat that as near the low, not ahead of it.
• The signal is late. At the 2023-03-19 cross, BTC was 72% above the low it was confirming. I treat it as a regime marker, not an entry.
• The same test on a simple 50-week average produces the same three episodes with the same oversold weeks and the same outcome, but confirms them on different dates. The 2015 cross comes 105 days later on the simple average, at $285 rather than $310, and its 90-day return is +36.8% rather than -20.9%. Same event, opposite sign. The 2026 cross comes 21 days later. So the signal identifies the same episodes whichever average you use, and the date it confirms them is definition-dependent by up to fifteen weeks. That reinforces the existing framing that this is confirmation rather than discovery, and that it arrives late. It arrives late and the lateness is not well defined.
• The sample is one structural uptrend with the all-time high set 2025-10-07, near the end of the window. Any buy-after-weakness rule is flattered by that.

2026: the sequence fired, time is the missing piece

The framework has done what it did in prior cycles.

Supply-in-loss crossed the threshold. Price made its final leg lower into the lows around $58,000 to $58,600 depending on the source. Weekly conditions went oversold. Then BTC reclaimed the 50-week EMA on Aug 30.

BTC is just over $75k late on Sep 15. That is roughly 30% above the lows.

That is the same “stress then confirmation” story this marker is supposed to tell. The only thing missing is time.

I still will not label it “no lower low” yet. That is a future claim, and this article is built around checkable work.

---

Indicator 3: Fear and Greed is not contrarian in my backtest

This is the finding that made me remove Fear and Greed from per-coin scoring in the product.

Corrected 2026-09-22. The drawdown figures in this section were originally computed against a truncated all-time high — my BTC price series began on the same day as the Fear and Greed series, 2018-02-01, so the running ATH never included the December 2017 peak of $19,665. Every pre-2021 drawdown was understated, and one conclusion below was backwards as a result. The numbers here are now from a single corrected run. The full working is in Does the Crypto Fear and Greed Index actually work?.

I backtested Fear and Greed data from 2018-02-01 to 2026-09-22. That is 3,152 readings. I tested correlation of Fear and Greed versus forward BTC return.

Correlation vs forward BTC return

• 7 days: +0.085 (n=3,145)
• 30 days: +0.134 (n=3,122)
• 90 days: +0.216 (n=3,062)
• 365 days: -0.095 (n=2,787)

Under a year, it is positive at every horizon. The strongest positive is 90 days.

That is mild momentum. It is the opposite of “fear means buy.”

Only at 365 days does it go negative, and -0.095 is noise.

So the general contrarian reading is not supported by this dataset.

This part of the finding did not change with the ATH correction. The correlations never depended on drawdown.

The narrow setup that does something

Fear and Greed at or below 10, while BTC is more than 50% below its true all-time high.

Grouping consecutive readings into episodes — runs separated by 30 days or fewer count as one event — there are 11 sub-10 episodes since 2018, and 6 of them qualify as deep, meaning BTC was more than 50% below its all-time high at some point during the episode:

Five have completed a year. Two of those five were large gains, one was a large loss, and two were small moves either side of flat.

I am not going to reduce that to a hit rate. With five closed observations the ratio moves depending on where you anchor the return: measured from each episode's first day rather than averaged across it, one of the five flips sign. A number that changes with a formatting decision is not a finding.

The 2018-02-06 episode is new to this list. Under the truncated ATH it read as 31.8% off the high and was excluded by the filter; against the real December 2017 peak it was 64.8% off, and it is the worst outcome in the set.

Weaknesses that matter:

• Six episodes, five closed, and one of those is COVID. That is a liquidity crash, not a normal cycle bottom.
• The >50% drawdown filter does real work — this is the conclusion the ATH error reversed. I previously wrote that it was “nearly decorative,” on the basis that almost every sub-10 day already met it. Against the true all-time high, only 19 of the 42 days at Fear and Greed ≤ 9, and 34 of the 69 days at ≤ 10, were more than 50% below the high. Roughly half of extreme-fear days are not deep drawdowns at all, so the filter is doing most of the selecting.
• Sub-15 is loose and there is no gradient. Sub-15 ~61%, sub-20 50%, sub-25 58%, sub-30 60%, sub-40 57%. Flat. Only sub-10 separates — and “sub-10” needs stating precisely: over the full window to 2026-09-22, Fear and Greed printed ≤ 9 on 42 days and ≤ 10 on 69 days. An earlier version of this line said 19 days, which was the ≤ 9 count stopping at the end of 2025 and omitting the 2026 readings.

---

Other published bottom-adjacent signals

These are widely-cited indicators. I’m listing them as context and linking to the reporting I used. I’m not presenting them as guaranteed bottom callers.

• MVRV Z-Score: https://www.coindesk.com/markets/2026/06/08/a-crucial-bitcoin-market-indicator-is-signaling-that-the-worst-of-the-crypto-crash-might-be-over

• Realized price: https://www.coindesk.com/markets/2026/06/23/bitcoin-may-need-to-plunge-15-or-more-to-mark-bottom-according-to-this-long-time-indicator

• 50-week / 100-week MA crossover (archived): https://web.archive.org/web/20260418014539/https://www.coindesk.com/markets/2026/04/17/this-simple-indicator-has-called-every-bottom-since-2015-it-hasn-t-triggered-yet

---

The live test (June to early July 2026)

This is the part I care about.

Supply-in-loss crossed the threshold in early June. BTC then put in its lows between early June and the first days of July, bottoming around $58,000 to $58,600 depending on the source.

BTC is just over $75,000 late on Sep 15.

Either the zone mattered this cycle, or it didn’t. We get to find out in real time.

---

So how do you know when a bear market bottoms?

You don’t, in the sense people mean.

You can do something more realistic.

You can identify zones where stress is reaching a structural ceiling. You can recognize confirmation markers months later. You can stop using popular sentiment widgets as contrarian signals just because the story is convenient.

That is the edge here. It is not prediction. It is refusing to lie to yourself.

---

Related reads

• Is it altcoin season? — with BTC 32.4% below its high and the median top-50 alt 75.4% below its own, this is a drawdown rather than a rotation.
• Does the Crypto Fear and Greed Index work? — the episode-level test that supersedes the Fear & Greed figure quoted above.

---

Where Cyclesolved fits

We score coins across technicals, on-chain, fundamentals, sentiment, and macro. You get one 0 to 100 score, and you can open it and see the inputs behind it.

Start here:

• Live scanner
• How scoring works

No ads. No calls. No hype. No paid coin promotions, ever.