Bitcoin trading strategy breakdown: celebrity predictions and classic models all fail, leaving only these four indicators in the end

Title: “Distilled My Accurate Bitcoin Trading Strategy Using AI for Everyone”
Author: JiaYi, Founder of GeekCartel

Author: Rhythm BlockBeats

Source:

Repost: Mars Finance

Since I am not a trader, one thing I must be clear about when developing BTC trading strategies is: which data can predict Bitcoin, and which data only adds confusion to the prediction.

Here’s the conclusion first: After completing it, I tested this system for a week, and at every key signal point, it provided me with the direction in advance.

Below is the complete logic.

1. Research Background: I reviewed all methods of “predicting BTC”

I am not a professional secondary trader. So I didn’t start by choosing indicators, but did something simpler—

Reviewed all Bitcoin prediction methods available from 2017 to 2025.

Divided into three categories:

First category: Celebrity opinions. VanEck predicts $180K by 2025. Not reached. Bitwise predicts $200K. Not reached. Tom Lee, Arthur Hayes, Novogratz, Cathie Wood—almost all major price forecasts in the past 8 years—systematically overestimated, with an average deviation over 50%.

Second category: Analytical methods. Stock-to-Flow model (PlanB’s set), logarithmic growth curves, cycle theories, Wyckoff school, Elliott Waves… Each has its own “historical accuracy,” but when run after 2024, almost all fail.

Third category: On-chain signals. MVRV Z-Score, SOPR, NUPL, Puell Multiple, Hash Ribbon, Reserve Risk… I’ve studied this category the longest because it’s not “prediction,” but “state description.”

After reviewing all three categories, I began filtering.

2. Filtering and Analysis: More data doesn’t mean more accuracy, it means more chaos

After filtering, I found an counterintuitive fact:

When massive data points point in different directions simultaneously, your judgment actually worsens.

After analysis, I divided them into two categories—

Unreliable (discard)

Celebrity predictions. Incentive structures force them to talk big. Saying “$500K” gets headlines, followers, repeated citations. Saying “$80K sideways” gets no attention. If they’re wrong, no one holds them accountable; if right, they’re always “gurus.” This structure won’t change, so predictions won’t be accurate either.

Models like Stock-to-Flow. Highly accurate before 2021, collapsed after 2022. Why? Because the model assumes “supply curve determines price,” but after ETFs entered, the price is driven by capital flows, not supply. The model itself isn’t wrong; the world it describes has changed.

Single emotional indicators (pure Fear & Greed). Historically, when Fear & Greed stays below 20 for a long time, it’s sometimes a bottom, sometimes a prelude to a -30 dip. Used alone, they generate too many false signals.

Reliable (keep)

MVRV Z-Score. Measures how much the current market cap deviates from the average cost of all holders. Historically, every time it entered the green zone, it precisely indicated a bottom within ±2 weeks—2018, March 2020, 2022—all three hits. But note: after 2024, its top-detection fails (e.g., it triggered overheating when BTC surged to $126K), because ETF trading happens off-chain, and it can’t see institutional holdings. So, only keep its bottom-detection ability.

SOPR 28-day moving average. Measures how many BTC sold in profit or loss during movement. Sustained below 1.0 = holders are taking losses = near bottom. This indicator has been very stable historically for bottom detection.

ETF net capital flow. A new core indicator after 2024. Institutional behavior must be observed here, as on-chain data can’t see it. Net inflow >180k for over 5 consecutive days = institutions are increasing holdings; net outflow >200k for over 5 days = institutions are reducing positions.

Macro liquidity. Fed policy direction + M2 growth rate. During easing cycles, go long; during tightening, reduce exposure. No short-term timing, only big-picture direction.

Fear & Greed as an auxiliary. Not used alone, only weighted when resonating with other signals.

After filtering, four dimensions remain. More than that feels excessive.

3. Strategy Formation: Four-dimensional resonance, act only when three or more signals point in the same direction

Once I clarified “which signals are reliable and why,” I turned this into a trading strategy.

Core logic: No chasing price targets, only judging direction and position.

Bottom detection: MVRV enters green zone + SOPR drops below 1.0 → on-chain holders are taking losses, historically a high-probability buy window.

Top detection: On-chain signals overheating + ETF continuous net outflow → institutions are withdrawing, reducing positions.

Macro background: Fed policy → easing for bullish, tightening for reducing exposure.

Sentiment aid: Fear & Greed < 20 → extreme panic, used as auxiliary weighting.

Any single signal is insufficient for action. Only when three or more signals align in the same direction is it a true entry point.

I then made it into an automated monitoring system:

· Automatically fetch BTC price, Fear & Greed, on-chain data, ETF capital flows daily
· No notifications if no signals trigger
· When triggered, notify me directly via Telegram
· Not a daily report, not noise. Only alert when truly worth attention.

Current signals (April 15, 2026)

This system currently gives me the reading:

BTC $71,631. Fear & Greed = 12, an extremely fearful level in history. MVRV Z-Score in the green buy zone. SOPR below 1.0, holders are selling at a loss.

All three on-chain signals are aligned.

The only counter-signal: recent ETF capital flows are weak, and institutions haven’t clearly started adding positions.

Historically, the triple on-chain resonance (extreme fear + MVRV green zone + SOPR < 1) has only appeared three times: late 2018 bottom, March 2020, late 2022 bottom. Each time, it produced over 100% returns within 12 months.

This isn’t predicting how high BTC will go. It’s an objective description of the current market state.

My biggest takeaway from this research is:

Predictions are others’ opinions; frameworks are your own judgment tools.

If you get it wrong, you lose everything. If your framework is wrong, at least you know where the problem lies and can iterate.

You can incorporate your preferences, such as contract multiples and cycle preferences, so that the signals AI gives you are tailored to your trading style.

Note: The above is based on historical patterns and is not financial advice.