How accurate is this bull market tracker?
Rebuilt day by day from free historical data back to March 2018, the original version of this site’s verdict matched the hindsight bull and bear phases on 72% of days. Always answering yes would have scored 58%. But it changed its answer about 20 times a year, and one of its own inputs, price against the 200-day moving average, scored 78% on its own.
Three changes made in September 2026 lifted the rebuilt score to 80% and cut the changes to about four a year. Those changes were chosen after seeing this data, which flatters them, so read this as a verdict that got better, not one that will be right 80% of the time from here.
Right now
Yes, it is a bull marketHow the rebuild works
Every morning from 3 March 2018 to 25 September 2026 was replayed with the site’s own rules, using each data source as the site would have seen it that morning, publication delays included. Fourteen of the original fifteen signals can be rebuilt from free history. Google Trends cannot, so it is left out, the same way the live site leaves out a source that fails.
Some inputs come from stand-in sources with longer free history: Coinbase daily prices, Binance funding rates, CoinMarketCap dominance and Farside ETF flows. Checked against the live site on 17 and 25 September 2026, every rebuilt vote matched. Earlier years have fewer signals, because ETF flows only exist from January 2024 and the free realized price history starts in late 2022.
What counts as right
Bull and bear markets are only obvious afterwards, so the rebuild scores each day against a fixed hindsight definition. A peak followed by a fall of 50% or more starts a bear market, and a low followed by a doubling starts a bull market. On daily closes that gives five bear markets in the period: December 2017 to December 2018, June 2019 to March 2020, April to July 2021, November 2021 to November 2022, and October 2025 to at least June 2026. The months since the June 2026 low are left unscored, because it is too early to know what they were.
That is a definition, not a fact, so the rebuild was also scored against the plainer four-year-cycle view, in which a bull market runs from each cycle low to the next cycle top. The ranking came out the same: the current rule first, then the 200-day moving average, then the original rule.
What it got right
The verdict carried real information. On the days it said yes, Bitcoin rose at an annualised rate of 80%; on the days it said no, it fell at 8% a year. Each return is measured over the day after the answer was given, so this is not hindsight.
It also caught the October 2025 top. It first said no nine days after the high, and from 31 October held no for 296 days, through the fall to about $58,500.
What it got wrong
It flip-flopped. The original rule changed its answer about 20 times a year, and 64% of its answers lasted less than two weeks. In the ten weeks around the October 2025 top it changed its answer 11 times. The cause was arithmetic: on 28% of days the count sat within one vote of the line, so a single signal moving flipped the headline.
Two votes were stuck. Bitcoin dominance has been above 55% on every day since early 2025, and the 7-day average of funding cleared its +11% baseline on 3% of days in 2025 and on none in 2026, so both could only ever vote no.
And a change made earlier the same month did not hold up. The all-time-high signal had gained a second leg, bullish at 50% above the 52-week low. It scored 92% from 2023 on, the period it was designed around, but 53% for 2018 to 2022. After a sharp top the 52-week low is stale: through 2018 it still held 2017 prices, so the leg voted bullish on 218 of the 288 days from March to December 2018 while Bitcoin fell.
What changed
The verdict now changes only when one side leads by three or more: 8 of 13 bullish turns it yes, 5 or fewer turns it no, and the new lead has to last about a day. In between it keeps its last answer. Dominance and funding moved to context, where they are shown but never vote. And the all-time-high signal’s second leg now measures from the lowest close since the high.
Rebuilt the same way, the current rule matched the hindsight phases on 80% of days (73% for 2018 to 2022, 88% since 2023), and 82% on the four-year-cycle view. It changed its answer about four times a year, and its median answer lasted 46 days. That edges past the 200-day moving average on accuracy while changing its mind about half as often.
The cost is speed at some turns. Across the nine turns since 2018, the last not yet confirmed, it settled on its new answer sooner at five and later at four. The one that matters most recently: after the October 2025 top it would have said no after 39 days, not 25.
Three things the rebuild turned up
MVRV, the market price divided by the average price holders paid, stayed above 1 on every day of the 53% fall from about $126,000 to $58,500. Its textbook rule would have called that whole fall a bull market, which is why realized price now votes on its 60-day direction instead.
Bitcoin dominance has not been below 55% since early 2025. The altcoin season the old rule was waiting for never arrived in this cycle.
Funding has not cleared its neutral baseline on a single day in 2026. Leveraged traders have paid less than the default rate even through the recovery from the June low.
What this does not show
This is not a trading record. The current rule was chosen after seeing the data, the hindsight phases are a definition, and nine turns is a small sample. The verdict describes the market’s current state; it does not predict it. A yes has meant stronger days on average in the past, which is no promise about the next one.
Every rule and threshold behind these numbers is on the methodology page.
Frequently asked questions
How accurate is this bull market tracker?
Rebuilt back to March 2018, the original rule matched hindsight bull and bear phases on 72% of days and the current rule on 80%. The current rule was chosen after seeing that data, so treat 80% as a best case rather than a forecast.
Why not just use the 200-day moving average?
On its own it matched 78% of days, better than the original verdict, and it is a good single measure. But it changes its answer about eight times a year, often back and forth within days, and says nothing about why. The current rule edges past it on accuracy, changes about four times a year, and shows the flows, sentiment and on-chain activity behind each answer.
Was the rebuild tuned to fit the past?
Partly, yes. The September 2026 changes were chosen after seeing it, which flatters them. Splitting the years helps a little: the current rule beat the original on 2018 to 2022 (73% against 72%) as well as since 2023 (88% against 72%), but most of the gain is recent.
Does a yes verdict mean the price will rise?
No. In the rebuild, the days after a yes were far stronger on average than the days after a no, but that describes the past, not the next day or month. Nothing on this site is financial advice.
Why is Google Trends missing from the rebuild?
Replaying it would need daily search data for every day since 2018, and Google only provides daily data three months at a time. It is left out of every rebuilt day, which is also what the live site does whenever a source fails.