[Vibe Coding Part 3] Why Your Bot is Losing Money : Delete the '+3% Take-Profit' Code Now

 

Stop Hardcoding +3%: Architecting a Trend Following Quant Bot with Gemini

1. The Trap of Fixed Profit: Stop Automating Human Cowardice

In Part 2, we broke through the 32-bit barrier of broker APIs and completed the data_fetcher module to collect real-time prices. Once data starts flowing, novice quant developers get excited and immediately begin writing strategy code. In nine out of ten cases, their strategy.py file ends up with the worst possible code:

if current_price >= buy_price * 1.03: sell_all() (Sell everything at +3% profit)

They start system trading to eliminate emotion, but end up hardcoding the innate human loss-aversion psychology—the urge to quickly lock in small profits—directly into their system. This is not quant trading. It is merely automating human cowardice using Python.

This fixed take-profit approach completely destroys the Risk-Reward ratio. When a massive bull market arrives and the market skyrockets +50% or +100%, your bot bails out of the rocket with a mere 3% gain. Conversely, in a bear market, the stop-loss logic often tangles, leaving you to absorb massive -20% or -30% losses. You have built a system that caps your upside at +3% while leaving your downside infinitely open. This is the mathematical reason why amateur retail bots inevitably fail.

2. The Architect's Insight: Trend Following and Risk-Reward Optimization

There is only one true winning formula in the capital markets: "Cut your losses short and let your profits run." Implementing this architecture is precisely what the "Trend Following" sell logic does.

In this logic, the concept of a fixed target price does not exist. Instead, as the asset's price rises, the system dynamically trails the sell trigger line (trendline) upward from below. As long as the market pulls the price up, you hold the asset to maximize profits. The moment the trendline breaks and the upward momentum exhausts, you dump your position without hesitation to lock in the profit. What you must instruct the AI via Vibe Coding is not a simple target price calculator, but the quantification of a surfer's instinct to ride the massive wave to the very end.

3. System Implementation: Gemini Prompt for the Trend Following Module

Now, open the Gemini chat window and instruct it to write the core strategy module, strategy/trend_follower.py, based on our architectural principles (Modularity and Exception Handling).

[Gemini Strategy Module Request Prompt]

"Gemini, as a Senior System Architect, we will now develop the strategy/trend_follower.py module, which makes trading decisions based on the data fetched in our previous conversation.

Never use amateur logic like a fixed +3% take-profit. Our core philosophy is 'Trend Following: ride the profit to the end and sell when the trendline breaks.'

• Modularity: Design it as class TrendFollowingStrategy:. Separate the methods to receive inputs (current price and historical moving average, or trailing stop baseline) and return buy/sell/hold signals.

• Safety: To prepare for data omission or calculation errors, apply try-except blocks and traceback to all logic. If a calculation fails, unconditionally return a 'HOLD' signal to protect the system.

• Execution Control: Include an if __name__ == "__main__": block at the bottom of the file. Write test code to verify that the sell signal triggers accurately when a mock price array (mock data that rises and then drops) is inputted."

 

4. Analyzing the Production-Ready Architecture Code

The code generated by Gemini is no longer a beginner's toy. It is a professional quant engine that controls errors and chases macroeconomic profits.

If you run the code standalone (if __name__ == "__main__":), you can verify that the system never sells while the price climbs to 12,000 KRW. However, the moment the trend breaks and touches 11,400 KRW (a 5% drop from the peak), it mechanically fires a 'SELL' signal. This is the true essence of system trading: completely eliminating human emotion.

We have fetched the data (Part 2) and established a strategy to determine when to sell (Part 3). In Part 4, we will cover the process of building the trader/ (order execution module), which injects actual orders into the broker's server upon receiving this 'SELL' signal. We will also discuss the architecture of hedging macroeconomic risks by separating KRW/USD accounts.