Skip to main content

RaposaDivergenceV1 Strategy: The Divergence Hunter

Nickname: Reversal Hunter, Peak-Valley Detective
Job: Specializes in finding divergence signals, catching reversal points
Timeframe: 5 minutes


I. What is This Strategy?​

Simply put, RaposaDivergenceV1 is:

  • A strategy that specifically looks for "divergence" signals
  • Price and RSI indicator moving in opposite directions? It enters
  • Uses mathematical methods to precisely find peak and valley points

It's like a detective with a magnifying glass looking for turning points 🔍

⚠️ First, Let's Talk About a Big Pitfall​

The author wrote in the code:

"argrelextrema might have look-ahead bias so results in backtest will not be reliable"

Translation: This strategy's backtest results might be inaccurate because it uses a "future function." During backtesting, it might "see the future" to confirm signals, but live trading can't do that.


II. Core Configuration: Profit Targets Are Pretty Aggressive​

Take-Profit Rules (ROI Table)​

0 minutes: Make 15% and run!
60 minutes: 2% is fine too...
120 minutes: Even 1% is okay
After 180 minutes: Even 0.1% is good enough

Translation: Initially wants to make 15%, but after waiting an hour, only wants 2%. Expectations drop pretty fast.

Stop-Loss Rules​

Fixed stop-loss: -30% (30% again...)
Trailing stop: Activates after 2% profit, locks in 1%

Translation: Stop-loss is set pretty wide, but trailing stop provides some protection.


III. Buy Condition: Only Bullish Divergence​

Core Logic (Plain English Version)​

What is bullish divergence?

  • Price makes a new low
  • But RSI indicator doesn't make a new low
  • Meaning: Downward momentum is exhausting, might bounce!

In plain English:

"Price is still falling, but with less force — prepare to buy the dip!"

Buy Condition Breakdown​

Price forms a higher low (bullish divergence)
RSI also forms a higher low (confirmation)
RSI < 50 (not buying in overbought territory)
Has volume

Peak-Valley Detection Algorithm​

This strategy uses a mathematical function called argrelextrema to find peak and valley points.

Plain English:

"Like finding the highest and lowest points of a mountain range — automatically identified using mathematical methods."

Parameter explanation:

ParameterDefaultPlain English
order5Look at 5 candles on each side to determine if it's a peak/valley
K value2Need 2 consecutive extrema to confirm a real peak/valley

IV. Sell Condition: Bearish Divergence (But Author Says It Doesn't Work Well)​

Sell Logic​

What is bearish divergence?

  • Price makes a new high
  • But RSI indicator doesn't make a new high
  • Meaning: Upward momentum is exhausting, might drop!

Plain English:

"Price is still rising, but with less force — prepare to exit!"

But the Author Said...​

"sell signal seems like garbage"

Translation: The sell signal seems like trash...

My Advice: Don't count on the sell signal, rely on ROI and trailing stop to exit.


V. Protection Mechanisms: Just These Few​

Protection TypeParameterPlain English
Fixed stop-loss-30%"If losing big, run"
Trailing stop1% @ 2% profit"Lock in some profit after making some"
Tiered ROI15%→2%→1%→0.1%"The longer you wait, the less picky you are"

Simple and rough, no complex protection logic.


VI. This Strategy's "Personality Traits"​

✅ Advantages (Compliment Section)​

  1. Solid theory: Divergence is classic technical analysis, not some esoteric stuff
  2. Precise algorithm: Uses scipy scientific computing library, not self-made hacks
  3. Clear signals: Bullish divergence buy, bearish divergence sell, simple rules
  4. Few parameters: Just 4 parameters, unlike some strategies with dozens

⚠️ Disadvantages (Roast Section)​

  1. Future function: Backtest results are unreliable, this is the most fatal issue
  2. Poor sell signal: Author admits the sell signal is garbage
  3. Few signals: Divergence doesn't happen often, might not see a signal for days
  4. Stop-loss too big: 30% stop-loss, if you hit a trending market you're screwed

VII. Applicable Scenarios: When to Use It?​

Market EnvironmentRecommended ActionReason
Oscillating transitionCan useDivergence signals are more accurate
Single-direction trendDon't useWill get repeatedly stopped out
High volatilityIncrease orderFilter noise extrema
Low volatilityDecrease orderCapture more signals

VIII. Summary: How's This Strategy Really?​

One-Sentence Review​

"A divergence strategy with more academic value than live trading value. Backtest looks good but don't take it seriously."

Who Is It For?​

  • ✅ Programmers who want to learn divergence detection algorithms
  • ✅ Players in oscillating trend-transition markets
  • ✅ People who can accept differences between backtest and live trading
  • ✅ People willing to manually confirm signals

Who Is It NOT For?​

  • ❌ People who rely on backtest results to make decisions
  • ❌ Players in single-direction trending markets
  • ❌ People who need frequent trading (few signals)
  • ❌ People who can't accept 30% drawdown

My Advice​

  1. For learning only: This strategy's greatest value is learning peak-valley detection algorithms
  2. Use as auxiliary: Use divergence signals as auxiliary, not as the sole basis
  3. Reduce stop-loss: 30% is really too big, change to 10-15%
  4. Paper trading test: Run at least 2 weeks of paper trading before live, compare differences

IX. In What Markets Can This Strategy Make Money?​

9.1 Core Logic: Finding Reversal Points​

RaposaDivergenceV1 is a pure divergence strategy. About 200+ lines of code, mainly relies on scipy's argrelextrema to find peaks and valleys.

Its money-making philosophy: Enter when price and indicator "disagree."

  • Bullish divergence: Price still falling, but RSI says "can't fall anymore"
  • Bearish divergence: Price still rising, but RSI says "can't rise anymore"

9.2 Performance in Different Markets (Plain English Version)​

Market TypePerformance RatingPlain English Explanation
📈 Sustained uptrend⭐☆☆☆☆Bearish divergence keeps appearing, price keeps rising, stopped out all the way
📉 Sustained downtrend⭐☆☆☆☆Bullish divergence keeps appearing, price keeps falling, stopped out all the way
🔄 Oscillating transition⭐⭐⭐⭐⭐Home field! Divergence is designed for this
🔄 Oscillating no-transition⭐⭐⭐☆☆Many signals but might not be accurate

One-sentence summary: Divergence strategies only work during trend transitions; single-direction trends will slap you in the face.


X. Want to Run This Strategy? Check These Configurations First​

10.1 Trading Pair Configuration​

Configuration ItemRecommended ValueComment
Number of pairs10-20Few signals, need more pairs
VolatilityMediumToo stable = no divergence, too volatile = too many false signals
Timeframe5 minutesDefault setting

10.2 Configuration File Key Settings​

# config.json key configurations
"timeframe": "5m",
"startup_candle_count": 40, # Warmup needs 40 candles
"trailing_stop": true,
"trailing_stop_positive": 0.01,
"trailing_stop_positive_offset": 0.02

10.3 Hardware Requirements (Important!)​

This strategy has moderate computational load, scipy's peak-valley detection isn't too heavy:

Number of PairsMinimum MemoryRecommended MemoryExperience
1-10 pairs2GB4GBSmooth
10-50 pairs4GB8GBNormal
50+ pairs8GB16GBSlightly laggy

10.4 Backtest vs. Live Trading​

This is key!

Backtest might show: 70% win rate, 300% return
Live might show: 40% win rate, 50% return

Why? Because argrelextrema "sees the future."

Recommended Process:

  1. Take backtest results with a grain of salt, multiply by 0.5 for a more realistic estimate
  2. Run paper trading for at least 2 weeks, compare signal lag
  3. Test with small capital live, verify actual performance
  4. Manually confirm divergence signals before entering

XI. Bonus: The Strategy Author's "Little Secrets"​

Looking carefully at the code, you'll find some interesting things:

  1. Author roasts their own sell signal: sell signal seems like garbage

    "This sell signal is trash, don't count on it"

  2. Warning about future function: argrelextrema might have look-ahead bias

    "Backtest looks good but isn't accurate, I already told you"

  3. Parameter range design: order and K value ranges are both 1-32, but defaults are 5 and 2

    "Defaults are good enough, but you can adjust"


XII. Final Words​

One-Sentence Review​

"An honest strategy — the author themselves said the sell signal is garbage and backtest is unreliable. Good for learning divergence algorithms, use with caution in live trading."

Who Is It For?​

  • ✅ Programmers wanting to learn divergence detection algorithms
  • ✅ Players in oscillating transition markets
  • ✅ People who can accept backtest ≠ live trading
  • ✅ People willing to manually confirm signals

Who Is It NOT For?​

  • ❌ People who rely on backtests to make decisions
  • ❌ Players in single-direction trending markets
  • ❌ People who need high-frequency trading
  • ❌ People who can't accept 30% drawdown

Manual Trading Recommendations​

If you want to use this strategy's signals manually:

  1. Learn to identify divergence patterns (price and RSI moving in opposite directions)
  2. Wait for peak-valley confirmation before entering, don't rush to buy the dip
  3. Confirm with other indicators (like MACD, volume)
  4. Set smaller stop-loss (10-15%), don't copy the strategy's 30%

XIII. ⚠️ Risk Emphasis (Must Read)​

Backtest Is Beautiful, Live Trading Needs Caution​

RaposaDivergenceV1's historical backtest performance might be very good — but there's a big pitfall:

argrelextrema "sees the future" to confirm peak-valley points, but live trading can't do that.

Simply put: Backtest has x-ray vision, live trading doesn't.

Look-Ahead Bias​

What is a future function?

During backtest:
Candle #100 is confirmed as a low point because candles #101, #102, #103 are all higher

During live trading:
Candle #100 just closed, you don't know what #101, #102, #103 will be

Consequences:

  • Backtest signals "precisely" appear at the lowest point
  • Live signals need to wait several candles before confirmation
  • Missing the best entry timing

My Advice (Real Talk)​

1. Treat this strategy as learning material, don't count on it to make money
2. If you must use it, combine with other indicators for manual confirmation
3. Change stop-loss to 10-15%, 30% is too big
4. Paper trading test, compare differences between backtest and live
5. Use divergence signals as auxiliary, not as the sole basis

Remember: Divergence strategies are accurate during trend transitions, but will get slapped repeatedly in single-direction trends. Test with small positions, staying alive is most important! 🙏