On 18 September Willy Woo posted a chart: the Fisher Transform on Bitcoin's monthly candles. The caption read "BTC bottoms: 3 for 3 without fake out. Latest cross is the 4th on record." Three bottoms, three hits, no false signal, and a fourth cross in progress.
That reads like a reason to buy. Before it becomes one, it needs an answer the post does not give: three out of how many?
A hit rate without a denominator is a list
"Three for three" is a fraction, and the post only gives the top half. Woo's own follow-up hints at the rest: Fisher turning points are not a reversal signal, price can consolidate and continue, and in some bull markets Fisher crosses bearish and then bullish again.
So the indicator has crossed more than three times. The other crossings are not counted as misses. They are filed as noise, because no bottom sat under them.
That sorting only works in hindsight. Anyone looking at a crossing today cannot tell which pile it will land on. A hit rate you can only compute after the fact describes the past. It is not a signal. "Without fake out" is the same problem in three words: what counted as a fake-out is decided once you know what came next.
For the Fisher Transform on the monthly chart, the denominator cannot be counted. For another famous cycle indicator it can, and at first the record looks spotless.
Pi Cycle: four crosses, four tops
Pi Cycle is the classic Bitcoin top indicator. It fires when the 111-day moving average rises above twice the 350-day moving average. Its reputation is "one signal per cycle, near-perfect", and until now that is also what the Arena strategy catalogue said. The Arena's event log reconstructs the rule on daily closes since December 2010 and finds exactly four upward crosses.
| Cross | Close that day | What it marked |
|---|---|---|
| 6 Apr 2013 | $142.63 | the all-time high of that day, days before the April 2013 peak |
| 3 Dec 2013 | $1,155 | the day before the cycle high ($1,237.96 on 4 Dec) |
| 16 Dec 2017 | $19,423.67 | the cycle-high day itself |
| 12 Apr 2021 | $59,839.58 | 2 % under the April 2021 high |
Four crosses, four tops, each within days. Anyone who sold on the cross of 16 December 2017 sold at the highest close of that cycle. Nothing was filed away as noise afterwards; there was no other upward cross.
If the story ended here, the reputation would be earned. It does not end here, because the same log also knows the tops where nothing crossed.
The highest top came without a signal
Above the line, the four tops where Pi Cycle crossed. Below it, the four where it stayed silent, the highest among them.
| Peak | Close | Signal |
|---|---|---|
| June 2019 | the 2019 high | none |
| November 2021 | about $67,500, 10 % above the April high | none |
| 13 March 2024 | $73,084.11, the highest close of that halving cycle | none |
| 6 October 2025 | $124,658.54, the all-time high | none |
On 6 October 2025 Bitcoin closed at $124,658.54, higher than ever. Pi Cycle did not cross. By mid-November price was 24 % lower, by the start of September 2026 37 % lower, per the Arena's Mayer-band log. Anyone waiting for Pi Cycle to sell is still waiting.
Four for four at the tops Pi Cycle called. Four for eight at the tops that happened. Both sentences are true. A list of hits shows where an indicator was right; where it stayed silent only shows up in the list of tops.
One of the four hits is also softer than it looks. Anyone who sold on the April 2021 cross avoided the summer correction and then watched Bitcoin close in November 10 % above the April high.
The Arena catalogue entry for Pi Cycle is being rewritten to that record: four dates, four misses, and the same verdict threshold that applies to every strategy on the platform, 30 completed trades. n = 4 reports a history, not an edge.
Back to Woo's chart. It carries a second problem, and this one concerns the fourth cross.
The fourth cross sat on an open candle
When Woo posted the chart, September's monthly candle had twelve days left to run. The Fisher Transform calculates on closes. The fourth cross therefore sat on a candle that did not exist yet, and it could still disappear before the month ended.
The first three crosses are measured on completed candles. Listing all four as a series puts three observations next to an interim reading. For any monthly signal, the first check is whether the candle has closed.
Even if the cross survives the monthly close, a deeper problem remains. It is not in the indicator.
Bitcoin has too little history for monthly track records
The Arena's Bitcoin series starts in July 2011. That is about 180 monthly candles, four halvings and three completed cycles. A bottom indicator on that chart gets three observations at most, a top indicator four, as Pi Cycle shows.
No backtest, method or amount of compute changes that. More Bitcoin history cannot be manufactured.
That is not a criticism of either tool. John Ehlers published the Fisher Transform in 2002, years before Bitcoin existed, so it is not fitted to this market. The limit is in the sample, not the tool.
So what is left for anyone who wants to use cycle indicators? The same rules, at a resolution that produces enough observations.
On daily data, a track record can be argued with
On daily data the Fisher Transform, Pi Cycle and any other crossing rule generate enough events for a real evaluation. Then there is a denominator, costs, drawdown depth and duration, a split between the period the rule was built on and the period it is tested on, and a check whether neighbouring settings deliver the same.
The result may be boring. It can be argued with, which a monthly track record cannot.
Tested on the monthly chart: you can save yourself the backtest. It cannot answer there, and knowing that protects you from trusting a position to a record of three cases.
Four questions before a track record moves money
- How many signals in total, not just hits?
- Who decided which signals counted, and when: before or after the outcome?
- Is the current signal on a completed candle?
- How many observations does this market offer at this resolution at all?
If the answer to the fourth is "three" or "four", the first three are already settled. And if someone says those were simply the three relevant signals, the next question is: who decided they were relevant?
The most dangerous number in a track record is rarely the wrong one. It is the one with no denominator.
FAQ
Do the Arena's Pi Cycle dates match the public charts? Closely, not always to the day. The log is reconstructed from the Arena's stored daily closes under the declared rule (111-day SMA at or above 2 × 350-day SMA), so a source using a different close or a different data feed can be a day off. The four hits and four misses do not depend on that.
Isn't a top that Pi Cycle "missed" just not a cycle top? That is the sorting problem again. October 2025 is the all-time high. If the highest price Bitcoin has ever printed does not count as a top for a top indicator, the definition has been fitted to the record.
Where do the $67,500 and the −24 % and −37 % come from? From the Arena's Mayer-band event log, which records the close and the drawdown from the all-time high at each flip: 28 December 2021, $41,837 and 38.1 % under the high, which puts that high at about $67,500; 3 November 2025, 24.4 % under the October high; 19 August 2026, 37.1 % under it.
The rules for the Pi Cycle log were declared before the events were read; the reconstruction runs from stored daily closes.
Not investment advice, not a recommendation, not a forecast. Historical patterns are not a promise.
Sources: Willy Woo on X, 18 September 2026, including the follow-up reply · John F. Ehlers, "Using the Fisher Transform", Technical Analysis of Stocks & Commodities, Vol. 20 No. 11, November 2002 · Backtesting Arena, signal event log (pi_cycle_cross, mayer_band), halving table and cycle snapshot, retrieved 24 September 2026
Study the Past — Improve your Future 🥋