[Crypto] Analysts Have a Stats Problem

Don’t Make It Yours Too

It’s almost impossible to dive into crypto analysis without being bombarded by statistical claims.

“This month closes bullish 70% of the time.”

“Bull flags often resolve to the upside.”

“Bitcoin follows a four-year price cycle: bear market, consolidation, bullish breakout, parabolic run. We’re entering the parabolic phase.”

“The Golden Cross has a 55% win rate.”

And then there are my favorites: “The more times a price level is touched, the weaker it gets.” Or, wait for it, “The more times a price level is touched, the stronger it gets.”

These statements sound authoritative, but here’s the thing: you can find statistics to prove or disprove just about anything. Numbers don’t lie, but they can be used to tell stories. And in crypto, those stories often belong in the fiction section.

Statistics, while powerful, don’t predict the future—they describe the past. When analysts present data as destiny, they build narratives that haven’t happened yet. For instance, saying, “This pattern played out 70% of the time in the past,” doesn’t guarantee it will this time. But we buy into these stories because our brains crave order and predictions. Numbers give us that. However, there’s another challenge: the way most statistics are presented doesn’t align with how our minds naturally work.

---

Statistical interpretations often fall into one of two camps: frequentist and Bayesian.

A frequentist example:

Imagine watching a basketball player shoot 100 free throws. They make 75 of them. Based solely on this data, you conclude their accuracy is 75%. So, the next time they step to the line, you think, “They have a 75% chance of making this shot.”

A Bayesian example:

As a Bayesian, you start with prior knowledge, “It’s usually sunny and hot this time of year where I live, and the chances of precipitation are very low.” But then, new information comes in— you look out your window and see it’s dark and cloudy. You bring your prior knowledge of historically regular sunny weather in the area and that dark, overcast skies lead to rain to update your beliefs and predictions for the day to conclude, “It’s more likely to rain, so I’ll take the umbrella, despite that being a rare occurrence in this area.”

The difference?

Frequentists base decisions solely on past observations. Bayesians incorporate context and update their beliefs as new information arrives. The latter mirrors how our brains naturally work: We adapt as we learn more. But most [crypto] analysis is presented in the frequentist style, which is devoid of context, static, and rigid.

---

Frequentist interpretations might work when you have a massive dataset—like analyzing thousands of trades. But crypto doesn’t give us that luxury in a higher timeframe (daily/weekly/monthly). Take Bitcoin’s annual cycles, for example. At best, we are looking at a 14-year history. That’s only three repetitions of the observed 4-year pattern. Yet since it is currently unbroken, people still hold it to play out as gospel.

Claims like “Bitcoin’s price follows a four-year cycle” or “bull flags resolve upward 60% of the time” are inherently weak. With so few data points, random chance plays a huge role. There isn’t enough historical data to rely on; the Law of Large Numbers, which establishes this reliability, can’t come into play.

---

My struggle with frequentist thinking led to an altered approach:

When I am wrong (the basketball player misses the 76th shot), do I continually adjust my stats to 75 out of 101? This may decrease the accuracy over time. It might even perpetually whittle away what once looked like an incredible stat. Do I hope my initial 75% holds out over time, given more shots?

Sure, I can keep rolling the dice, hoping the numbers will play out in the end, but it feels unrealistic to expect that there can’t be some external elements that can affect those outcomes. The more the player shoots, the more tired they will get, possibly decreasing accuracy. Maybe the player gets stressed out more because they missed a lot in a row, and that continually decreases accuracy because they go on tilt. Perhaps the player misses so many in a row that they slow down, concentrate on fundamentals and form then improve to an even better ratio in time.

So, how do we avoid falling into the trap of misleading statistics? Here are a few takeaways:

• When presented with data, ask: How many data points are in this analysis? What’s the context? Could external factors affect this outcome?

• Don’t rely solely on past patterns. Add layers of context, like macroeconomic conditions, social sentiment, or historical anomalies. Update your beliefs as new information emerges.

• Even if a setup “works” 70% of the time, there’s a 30% chance it doesn’t. Always weigh the risks.

• Markets evolve. Don’t cling to old statistics when new conditions make them less relevant. A flexible approach is far better than blind adherence to past trends.

Crypto markets are chaotic, and while stats can offer guidance, they’re no crystal ball. Analysts have a stats problem because they use them to tell stories of certainty where none exists. But by shifting to a Bayesian mindset that adapts, evolves, and contextualizes, you can craft a more realistic and actionable narrative.

Numbers don’t lie. But they also don’t predict the future. You must interpret them wisely, update your beliefs, and trade accordingly.

Here are a couple of accessible books that I enjoyed.

https://app.thestorygraph.com/books/aaf9bb6c-d05b-4d31-b959-249d0533e342

https://app.thestorygraph.com/books/8a889cb4-8d74-4e41-9f7b-ef1bdd218a41

@ThePrivacySmurf

Tomorrow, in the ANALYZE section of my Substack, I’ll showcase how I’ve integrated this thinking/math into my technical analysis to create expected trading ranges in the weekly updates.