Original Title: "I distilled my accurate Bitcoin trading strategy with AI for everyone"
Original Author: JiaYi, Founder of GeekCartel

Since I'm not a trader, I must be clear about one thing when making BTC trading strategies: what data can really predict Bitcoin, and what data will only make predictions more confusing.

Let me give you the conclusion first: After I finished, this system has been tested for a week, and it has provided me with direction at every key signal point.

Here is the complete logic.

1. Research Background: I reviewed all methods for "predicting BTC"

I am not a professional secondary trader. So instead of immediately choosing indicators, I did a foolish thing first—

I reviewed all methods available in the market for predicting Bitcoin from 2017 to 2025.

Divided into three categories:

First category: Celebrity opinions. VanEck said $180K by 2025. Did not happen. Bitwise said $200K. Did not happen. Tom Lee, Arthur Hayes, Novogratz, Cathie Wood—almost all recorded significant price predictions over the past 8 years, systematically biased high, with an average deviation of over 50%.

Second category: Analytical methods. Stock-to-Flow model (PlanB's set), logarithmic growth curve, cycle theory, Wyckoff school, Elliott wave... each has its own "historical accuracy," but if you run them after 2024, almost all of them fail.

Third category: On-chain signals. MVRV Z-Score, SOPR, NUPL, Puell Multiple, Hash Ribbon, Reserve Risk... This category is what I researched the longest because it is not "prediction," but "state description."

After reviewing all three categories, I started filtering.

2. Filtering and Analysis: More data does not mean more accuracy, more data means more confusion

After filtering, I found an intuitive counterpoint:

When vast amounts of data point in different directions simultaneously, your judgment will be worse.

After the analysis, I divided them into two categories—

Unreliable category (discard)

Celebrity predictions. Incentive structures dictate that they must make grand statements. Saying "$500K" makes headlines, gains followers, and is quoted repeatedly. Saying "$80K sideways" gets no shares. If wrong, no one is held accountable; if right, they are forever "great." This structure won’t change, hence the predictions won’t be accurate.

Pure models like Stock-to-Flow. Before 2021, its accuracy was high; after 2022, it collapsed. Why? Because the model's assumption is "supply curve determines price," but after the ETF enters the market, what determines price is capital flow, not supply. The model itself is not wrong; the world it describes has changed.

Single indicators of sentiment (pure Fear & Greed). Historically, when Fear & Greed is consistently below 20, sometimes it is a bottom, sometimes it is a precursor to "dropping to -30." When used alone, there are too many false signals.

Reliable category (retain)

MVRV Z-Score. Measures the deviation of current market value relative to the average cost of all holders. Historically, every time it entered the green zone, it accurately corresponded to a cycle bottom ±2 weeks — 2018, March 2020, and 2022, all three hit exactly. But note: after 2024, its judgment on the top becomes invalid ($73K triggered overheating, BTC rose to $126K), because ETF trading is off-chain; it cannot see the institutional portion of the chips. So only retain the bottom judgment ability.

SOPR 28-day moving average. Measures how much BTC is sold at a loss during movement. Sustained below 1.0 = holders are cutting losses = approaching the bottom. This indicator's historical judgment of the bottom has been very stable.

ETF net capital flow. A new core indicator after 2024. Institutional marginal behavior must be viewed from here, which on-chain data cannot see. Continuous net inflows for 5 or more days accumulating >$1 billion = institutions are increasing their holdings; continuous net outflows for 5 or more days = institutions are withdrawing.

Macroeconomic liquidity. Federal Reserve direction + M2 growth rate. Go long during an easing cycle, reduce exposure during a tightening cycle. No short-term timing, only determine the big direction.

Fear & Greed as auxiliary. Used not alone, only weighted when resonating with other signals.

After filtering, there are four remaining dimensions. One more and it feels excessive.

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

After clarifying "which ones are accurate, why they are accurate," I turned it into a trading strategy.

The core logic: do not chase price targets, only judge direction and position.

Bottom judgment: MVRV enters green zone + SOPR falls below 1.0 → on-chain holders are cutting losses, historical high win rate buy window

Top judgment: on-chain signals are overheated + continuous net outflows from ETFs → institutions are withdrawing, reducing holdings

Macroeconomic background: Federal Reserve direction → long when easing, reduce exposure when tightening

Sentiment auxiliary: Fear & Greed 20 → extreme panic, auxiliary weighting

No single signal is enough to act on. Three or more pointing in the same direction is the true basis for entry.

Then I turned it into an automatic monitoring system:

· Automatically pulls BTC prices, Fear & Greed, on-chain data, ETF capital flows daily

· No push notifications if no signal is triggered

· If triggered, directly notify me via Telegram

· Not a daily report, not noise. Only rings when worthy of attention

Current signal (April 15, 2026)

This system currently gives me the reading:

BTC $71,631. Fear & Greed = 12, historically extreme panic level. MVRV Z-Score is in the green buy zone. SOPR is below 1.0, holders are selling at a loss.

All three on-chain resonances are established.

The only counter signal: ETF capital flow is weak recently, institutions have not clearly begun to accumulate.

Historically, the on-chain triple resonance (extreme panic + MVRV green zone + SOPR 1) has only occurred three times: bottom in 2018, March 2020, and bottom in 2022. All produced 100%+ returns in the following 12 months.

This is not predicting how high BTC will go. This is an objective description of the current market conditions.

My biggest feeling after the research is:

Prediction is someone else's opinion, the framework is your own judgment tool.

If the prediction is wrong, you have nothing. If the framework is wrong, at least you know where the problem lies and can iterate.

You can input your own preferences, such as leverage and cycle preferences, so that the signals AI gives you are tailored to your operational characteristics.

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

Original link

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。