The preliminary quantitative model that was previously built has recently been updated.

Today's BTC quantitative score is 32.5, see the direction and conclusion below.

Mainly added a trading module to the previous function of trend judgment.

Includes daily suggestions for entry positions based on direction, take profit and stop loss suggestions, position and leverage suggestions.

Then today, it added tracking and management for open positions, providing suggestions for increasing or decreasing positions based on daily data scores.

It is considered a further improvement of the preliminary model, and the main logic can currently be divided into three layers.

1. Trend Layer



The initial version could only score the direction of long and short positions for the day based on long, medium, and short-term indicators with different weights; the higher the score, the more obvious the trend, which is used to guide the size of the position.

2. Execution Layer



There are two directions and strengths; in trading, it is also necessary to time positions based on short-term indicators.

The basic logic is to derive a suggested opening price and take-profit/stop-loss prices based on certain calculations involving buy/sell walls.

Since the original intention of this model was to serve my own trading, the style is low-frequency operation, with a clear direction and defined strength, allowing for heavy positions; just two or three trades a month are enough.

Thus, the execution price is more about micro-timing under the premise of a clear direction.

3. Tracking Layer



This is because once a position is opened, it may take two or three days to trigger my take-profit or stop-loss orders.

Moreover, the market environment changes in real-time; once a position is opened, the opened position is already exposed to market risks, so I designed a tracking system for the opened positions.

The logic is to compare the most recent day's data with the data from the day the position was opened to check if the direction has changed and whether the strength is increasing or decreasing.

If it is the former, I will close the position immediately; if it is the latter, I will consider appropriately adding or reducing the position.
However, I will only adjust the take-profit position up or down; the stop-loss position will only narrow, not widen.

1. Data Standards



When actually building the model, a very practical and tedious issue is the different data units; some are in billions, some in percentages, and others in basis points.

All I can do is write them all down in the Guide worksheet for quick reference and calibration when I forget the data standards.

2. Backtesting Validation and Trend Reversal Alerts


Once I decided to extend the prediction model to a trading model, the model gradually became more complex.

This is because it is necessary to consider not just the initial market judgment but also to control the position and price while tracking the opened positions.

I have now opened a Backtest worksheet to conduct 7D and 14D backtesting on the entry prices and directions provided daily to see how they perform within 7D and 14D.

This way, I can use the market to verify whether this model is effective for trading.

On the other hand, I opened a short position on the 16th for real-time tracking.

The main focus has been to establish an Inflection worksheet that provides early warning alerts when the direction and strength decrease, allowing for early position reduction.


3. Data Accumulation



At this point, the model's functionality for trading is quite complete, and all that remains is to accumulate some data to test its effectiveness.

But in reality, building the model is really painful, even though I relied on AI to help me write the formulas.

However, I can always find mistakes in the formulas, sometimes due to incorrect references and sometimes due to a misunderstanding of my logic, requiring repeated adjustments.

I am now disappointed with GPT, and after numerous conversations, it feels like GPT can no longer keep up.

Every conversation causes GPT to crash and lose responsiveness; I will try Gemini3 next time.

I hope that in a few months this model can run smoothly, and whether I can make money next year depends on it.