Henry Lin

Posted on Oct 4

Lesson 7: Freqtrade Multi-Timeframe Backtesting

#cryptocurrency #tutorial #testing #algorithms

Lesson 7: Multi-Timeframe Backtesting

⏱ Duration: 2 hours
🎯 Learning Objectives: Understand the impact of timeframes on strategies
📚 Difficulty: ⭐⭐ Backtesting practical

📖 Course Overview

The same strategy can perform very differently on different timeframes. Through practical testing, this lesson will help you understand timeframe selection principles and find the most suitable time period for your strategy.

7.1 Timeframe Selection Principles

Three Core Principles

Principle 1: Trading Style Match

Your time investment → determines timeframe

Full-time monitoring (8h+/day)  → 1m-5m (ultra-short term)
Part-time monitoring (2-4h/day) → 15m-1h (short term)
Occasional checking (<1h/day) → 4h-1d (swing trading)
Fully automated → any timeframe

Principle 2: Risk Tolerance

Single trade volatility increases with timeframe

1m:  Single trade volatility 0.1-0.5%  → Low risk, low pressure
5m:  Single trade volatility 0.3-1%    → Moderate risk
1h:  Single trade volatility 1-3%      → Higher risk
1d:  Single trade volatility 3-10%     → High risk, high psychological pressure

Principle 3: Strategy Type Adaptation

Trend following strategies    → Long timeframes (1h-1d)
Mean reversion strategies    → Medium-short timeframes (5m-1h)
Breakout strategies          → Medium timeframes (15m-4h)
High-frequency arbitrage     → Ultra-short timeframes (1m-5m)

Timeframe Feature Comparison

Timeframe	Signal Frequency	Signal Quality	Fee Percentage	Data Requirements	Execution Requirements
1m	Extremely High	Low (lots of noise)	Very High (15-30%)	Massive	Millisecond-level
5m	High	Medium-Low	High (8-15%)	Large	Second-level
15m	Medium-High	Medium	Moderate (5-10%)	Medium	Minute-level
1h	Medium	Medium-High	Low (2-5%)	Moderate	Hour-level
4h	Low	High	Very Low (< 2%)	Small	Hour-level
1d	Very Low	Very High	Extremely Low (< 1%)	Very Small	Day-level

7.2 Characteristics of Different Timeframes

1. Ultra-Short Term (1m - 5m)

Detailed Features

Advantages:

✅ Extremely many trading opportunities (tens to hundreds of trades daily)
✅ Fast capital turnover, high utilization
✅ Small single trade risk (small stop loss)
✅ Quick strategy validation (see results in days)

Disadvantages:

❌ Many noise signals, frequent false breakouts
❌ High fee costs (may consume 30-50% of profits)
❌ High execution speed requirements (1-second delay can affect returns)
❌ Large data volume (1 year of 1m data > 500MB/pair)
❌ Slow backtesting speed
❌ Significant slippage impact

Fee Impact Case Study

5m timeframe, 30-day backtest:

Trade Count: 150 trades
Average Profit per Trade: +0.6%
Fees: 0.2% × 150 = 30%

Gross Profit: 0.6% × 150 = 90%
Net Profit: 90% - 30% = 60%

Fee erosion: 33.3% of profits!

Suitable For

Full-time traders
Those with exchange fee discounts (VIP users)
High-frequency algorithmic trading systems
Those well-prepared for latency and execution

Risk Warning

⚠️ Not recommended for beginners: Good backtesting performance ≠ Good live performance (slippage and latency will significantly reduce returns)

2. Short Term (15m - 1h)

Detailed Features

Advantages:

✅ Good signal quality (noise filtered)
✅ Moderate trading frequency (5-20 trades daily)
✅ Controllable fee impact (5-10%)
✅ Moderate data volume, fast backtesting
✅ Relaxed execution requirements (few minutes of delay acceptable)
✅ Suitable for beginners to practice

Disadvantages:

⚠️ Still requires regular monitoring (cannot be fully automated)
⚠️ Need to endure intraday volatility
⚠️ Overnight risk (1h may span overnight)

Best Practices

15m timeframe:

Each candle is 15 minutes
96 candles daily
Suitable for intraday trading (8:00 AM - 12:00 AM)
Typical holding time: 1-6 hours

1h timeframe:

Each candle is 1 hour
24 candles daily
Suitable for short swing trading
Typical holding time: 4-24 hours

Fee Impact Case Study

1h timeframe, 30-day backtest:

Trade Count: 40 trades
Average Profit per Trade: +1.5%
Fees: 0.2% × 40 = 8%

Gross Profit: 1.5% × 40 = 60%
Net Profit: 60% - 8% = 52%

Fee erosion: 13.3% of profits (acceptable)

Suitable For

Quantitative trading beginners (highly recommended)
Part-time traders (2-4 hours daily)
Those who want to see quick results
Balance returns and risks

3. Medium-Long Term (4h - 1d)

Detailed Features

Advantages:

✅ High signal quality (clear major trends)
✅ Extremely low fee percentage (< 2%)
✅ High single trade returns (average 3-10%)
✅ No need for frequent monitoring
✅ Small slippage impact
✅ Low psychological pressure (no need to watch screen)

Disadvantages:

❌ Few trading opportunities (few trades weekly)
❌ Low capital utilization (occupied for long periods)
❌ Large single trade volatility (need larger stop loss space)
❌ Overnight risk (weekend, holiday risks)
❌ Long validation period (need months of data)

Overnight Risk Analysis

Risk Sources:

US Market Close Impact: US market crashes → Crypto follows
Asian Opening Volatility: High volatility during China market opening hours
Sudden News: Regulatory policies, hacker attacks, exchange suspensions
Weekend Gaps: Price jumps from Friday close to Monday open

Countermeasures:

# Force close before weekend (advanced usage)
def custom_exit(self, pair, trade, current_time, **kwargs):
    # Force close after 22:00 on Friday
    if current_time.weekday() == 4 and current_time.hour >= 22:
        return 'weekend_exit'

Suitable For

Part-time traders (< 1 hour daily)
Those who dislike frequent operations
Those with large capital (can handle single trade volatility)
Long-term investment mindset

7.3 Timeframe and Trading Style Matching

Trading Style Classification

1. Day Traders

Features:

Open and close positions same day
No overnight holding
5-20 trades daily

Recommended Timeframes: 5m - 15m
Recommended Strategies: Mean reversion, breakout strategies
Key Settings:

# Ensure no overnight holding (set in strategy)
minimal_roi = {
    "0": 0.02,   # 2% immediate take profit
    "60": 0.01,  # 1% take profit after 1 hour
    "300": 0     # Close position after 5 hours
}

2. Swing Traders

Features:

Hold positions for days to weeks
Capture medium-term trends
2-10 trades weekly

Recommended Timeframes: 1h - 4h
Recommended Strategies: Trend following, momentum strategies
Key Settings:

minimal_roi = {
    "0": 0.10,      # 10% immediate take profit
    "1440": 0.05,   # 5% take profit after 1 day
    "4320": 0.02    # 2% take profit after 3 days
}

3. Trend Traders

Features:

Hold positions for weeks to months
Capture major trends
1-5 trades monthly

Recommended Timeframes: 1d
Recommended Strategies: Trend following, breakout strategies
Key Settings:

minimal_roi = {
    "0": 0.30,      # 30% immediate take profit
    "10080": 0.15,  # 15% take profit after 1 week
    "43200": 0.08   # 8% take profit after 1 month
}

Matching Test Table

Your Situation	Recommended Timeframe	Expected Trading Frequency
Can monitor 6h+ daily	5m - 15m	10-50 times/day
Can check 2-4h daily	15m - 1h	5-20 times/day
Can only check 1h daily	1h - 4h	2-10 times/day
Only check few times weekly	4h - 1d	5-20 times/month
Fully automated	Based on strategy type	Any

7.4 Multi-Timeframe Comparative Analysis

Practical Testing Process

Step 1: Prepare Data

# Download multiple timeframe data
freqtrade download-data \
  -c config.json \
  --pairs BTC/USDT \
  --days 90 \
  --timeframes 5m 15m 1h 4h 1d

Step 2: Batch Backtesting

# Test 5m
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 5m --timerange 20250701-20250930

# Test 15m
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 15m --timerange 20250701-20250930

# Test 1h
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 1h --timerange 20250701-20250930

# Test 1d
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 1d --timerange 20250701-20250930

Step 3: Use Shell Script for Batch Testing

Create test_timeframes.sh:

#!/bin/bash

STRATEGY="Strategy001"
CONFIG="config.json"
TIMERANGE="20250701-20250930"

for TIMEFRAME in 5m 15m 1h 4h 1d
do
  echo "========================================"
  echo "Testing $STRATEGY on $TIMEFRAME"
  echo "========================================"

  freqtrade backtesting \
    -c $CONFIG \
    --strategy $STRATEGY \
    --timeframe $TIMEFRAME \
    --timerange $TIMERANGE

  echo ""
  echo "Finished $TIMEFRAME, press Enter to continue..."
  read
done

echo "All tests completed!"

Run script:

chmod +x test_timeframes.sh
./test_timeframes.sh

Comparison Analysis Template

Create comparison table to record results:

Metric	5m	15m	1h	4h	1d
Trade Count
Win Rate %
Total Profit %
Avg Profit %
Max Drawdown %
Sharpe Ratio
ROI Exit %
Stop Loss Exit %
Avg Holding Time
Fee %

Real Case Analysis

Assume Strategy001 performance on different timeframes:

Timeframe	Trade Count	Win Rate	Total Profit	Avg Profit	Max Drawdown	Sharpe	Fee %	Rating
5m	247	68%	+12.5%	+0.05%	-8.2%	1.85	19.8%	⭐⭐
15m	89	75%	+18.3%	+0.21%	-6.5%	2.65	9.7%	⭐⭐⭐⭐
1h	31	81%	+15.7%	+0.51%	-4.8%	3.12	3.9%	⭐⭐⭐⭐⭐
4h	12	83%	+9.2%	+0.77%	-3.2%	2.88	2.6%	⭐⭐⭐⭐
1d	4	100%	+6.8%	+1.70%	-2.1%	3.45	1.2%	⭐⭐⭐

Analysis Conclusions

5m timeframe:

❌ Trading too frequent (247 trades)
❌ Serious fee erosion (19.8%)
❌ Extremely low average profit (0.05%)
❌ Low Sharpe Ratio (1.85)
Conclusion: Not suitable for this strategy

15m timeframe:

✅ Moderate trade count (89 trades)
✅ Good win rate (75%)
✅ High total profit (18.3%)
✅ Controllable fees (9.7%)
✅ Excellent Sharpe Ratio (2.65)
Conclusion: Very suitable, balances returns and frequency

1h timeframe:

✅ Highest win rate (81%)
✅ Highest Sharpe Ratio (3.12)
✅ Smallest drawdown (4.8%)
✅ Extremely low fees (3.9%)
⚠️ Total profit slightly lower than 15m
Conclusion: Best choice, optimal risk-return ratio

4h timeframe:

✅ High win rate (83%)
✅ High average profit (0.77%)
⚠️ Few trades (12 trades, small sample)
⚠️ Lower total profit (9.2%)
Conclusion: Optional, but needs longer validation

1d timeframe:

✅ 100% win rate (4 wins, 0 losses)
✅ Highest average profit (1.70%)
❌ Too few trades (only 4 trades)
❌ Insufficient sample, not representative
Conclusion: Needs longer validation (at least 1 year of data)

Optimal Timeframe Judgment Criteria

Comprehensive Scoring Formula:

Total Score = Total Profit(30%) + Sharpe(25%) + Win Rate(20%) + Trade Count Reasonableness(15%) + Fee Percentage(10%)

Applied to Case:

15m: (18.3×0.3) + (2.65×10×0.25) + (75×0.2) + (100×0.15) + ((100-9.7)×0.1) = 5.49 + 6.63 + 15 + 15 + 9.03 = 51.15
1h:  (15.7×0.3) + (3.12×10×0.25) + (81×0.2) + (90×0.15) + ((100-3.9)×0.1) = 4.71 + 7.80 + 16.2 + 13.5 + 9.61 = 51.82 ⭐

Conclusion: 1h is slightly better

💡 Practical Tasks

Task 1: Data Preparation

# Download BTC/USDT multiple timeframe data
conda activate freqtrade
freqtrade download-data \
  -c config.json \
  --pairs BTC/USDT \
  --days 90 \
  --timeframes 5m 15m 1h 1d

Task 2: Batch Backtesting

Choose a strategy (recommend Strategy001), test 4 timeframes:

# 5m
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 5m --timerange 20250701-20250930

# 15m
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 15m --timerange 20250701-20250930

# 1h
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 1h --timerange 20250701-20250930

# 1d
freqtrade backtesting -c config.json --strategy Strategy001 --timeframe 1d --timerange 20250701-20250930

Task 3: Create Comparison Table

Create Excel table, record the following information:

Metric	5m	15m	1h	1d
Trade Count	?	?	?	?
Win Rate	?	?	?	?
Total Profit %	?	?	?	?
Avg Profit %	?	?	?	?
Max Drawdown %	?	?	?	?
Sharpe Ratio	?	?	?	?
Avg Holding Time	?	?	?	?
Fee %	?	?	?	?

Task 4: Find Optimal Timeframe

Based on your comparison table, answer the following questions:

Which timeframe has the highest total profit?
Which timeframe has the highest Sharpe Ratio?
Which timeframe has the smallest drawdown?
Which timeframe has the lowest fee percentage?
Overall, which timeframe do you think is most suitable for this strategy?

Task 5: Write Analysis Report

Write a 200-300 word analysis report:

Report Template:

Strategy Name: ___________
Test Time Range: ___________
Test Trading Pair: ___________

Timeframe Comparison Analysis:
- 5m Performance: ___________ (Advantages/Disadvantages)
- 15m Performance: ___________ (Advantages/Disadvantages)
- 1h Performance: ___________ (Advantages/Disadvantages)
- 1d Performance: ___________ (Advantages/Disadvantages)

Optimal Choice: ___________
Reasons for Choice: ___________

Risk Warnings: ___________

📚 Knowledge Check

Basic Questions

Why is the fee percentage high for 5m timeframe?
What is the main risk of 1d timeframe?
If you can only check charts for 1 hour daily, what timeframe should you choose?

Answers

Many trades, each trade requires paying fees, cumulative percentage is high
Overnight risk (weekend gaps, sudden news) and small sample size (long validation period)
1h or 4h, so you don't need frequent monitoring but can still capture good trading opportunities

Advanced Questions

Why does the same strategy perform very differently on different timeframes?
How to determine if the sample size for timeframe testing is sufficient?
If short-term backtesting performs well, does it mean live trading will also be good?

Thought Questions

If a strategy performs poorly on all timeframes, what does it indicate?
Can you run the same strategy on multiple timeframes simultaneously?
What are the advantages of multi-timeframe combination strategies (e.g., use 1h to confirm trend, 5m for entry)?

🔧 Common Problems and Solutions

Problem 1: Insufficient data error during backtesting

Error Message:

No data found for timeframe 1d in the specified timerange

Solution:

# Confirm data downloaded for that timeframe
freqtrade list-data -c config.json

# Redownload
freqtrade download-data -c config.json --timeframes 1d --days 90

Problem 2: Only few trades on 1d timeframe

Causes:

Sample period too short (90 days = 90 1d candles)
Strategy itself has low trading frequency

Solution:

# Download longer period data (1-2 years)
freqtrade download-data -c config.json --timeframes 1d --days 730

Problem 3: Large variation in results across timeframes

Causes:

Market environment changes
Strategy overfitting
Insufficient sample size

Solutions:

Test longer time ranges
Test different market environments (bull, bear, sideways)
Conduct out-of-sample testing

📊 Timeframe Selection Decision Tree

Start
  │
  ├─ Can you monitor full-time?
  │   ├─ Yes → Do you have fee discounts?
  │   │        ├─ Yes → 5m (high-frequency trading)
  │   │        └─ No → 15m (intraday trading)
  │   │
  │   └─ No → How many times can you check daily?
  │            ├─ 2-4 times → 1h (short swing)
  │            ├─ 1-2 times → 4h (medium swing)
  │            └─ < 1 time → 1d (long swing)
  │
  └─ Your risk tolerance?
      ├─ Low → Choose long timeframes (1h-1d)
      ├─ Medium → Choose medium timeframes (15m-1h)
      └─ High → Choose short timeframes (5m-15m)

🔗 Reference Materials

Supporting Documentation

📄 TESTING_GUIDE.md - Different timeframe backtesting
📄 STRATEGY_SELECTION_GUIDE.md - Strategy selection guide

📌 Key Points Summary

No best timeframe, only most suitable one
Short timeframe = High frequency + High fees + High noise
Long timeframe = Low frequency + Low fees + High quality
Beginner recommended 15m-1h: Balance returns, risks, and operation frequency
Sample size is important: 1d needs at least 1 year of data
Good backtesting ≠ Good live performance: Short-term strategies heavily affected by slippage

➡️ Next Lesson Preview

Lesson 8: Strategy Batch Comparison

In the next lesson, we will:

Learn methods for batch backtesting multiple strategies
Quickly compare strategy performance
Master strategy selection decision tree
Create strategy recommendation list

Preparation:

✅ Ensure you have 5+ strategies available
✅ Select testing timeframe (recommend 15m or 1h)
✅ Prepare 90+ days of data

🎯 Learning Check Standards:

✅ Can independently test strategy performance on multiple timeframes
✅ Understand advantages and disadvantages of different timeframes
✅ Can choose suitable timeframe based on your situation
✅ Can create timeframe comparison analysis table

After completing these tasks, you have mastered the core skills of timeframe selection! Ready to move on to strategy comparison learning! 🚀

Lesson 7: Multi-Timeframe Backtesting

📖 Course Overview

7.1 Timeframe Selection Principles

Three Core Principles

Principle 1: Trading Style Match

Principle 2: Risk Tolerance

Principle 3: Strategy Type Adaptation

Timeframe Feature Comparison

7.2 Characteristics of Different Timeframes

1. Ultra-Short Term (1m - 5m)

Detailed Features

Fee Impact Case Study

Suitable For

Risk Warning

2. Short Term (15m - 1h)

Detailed Features

Best Practices

Fee Impact Case Study

Suitable For

3. Medium-Long Term (4h - 1d)

Detailed Features

Overnight Risk Analysis

Suitable For

7.3 Timeframe and Trading Style Matching

Trading Style Classification

1. Day Traders

2. Swing Traders

3. Trend Traders

Matching Test Table

7.4 Multi-Timeframe Comparative Analysis

Practical Testing Process

Step 1: Prepare Data

Step 2: Batch Backtesting

Step 3: Use Shell Script for Batch Testing

Comparison Analysis Template

Real Case Analysis

Analysis Conclusions

Optimal Timeframe Judgment Criteria

💡 Practical Tasks

Task 1: Data Preparation

Task 2: Batch Backtesting

Task 3: Create Comparison Table

Task 4: Find Optimal Timeframe

Task 5: Write Analysis Report

📚 Knowledge Check

Basic Questions

Answers

Advanced Questions

Thought Questions

🔧 Common Problems and Solutions

Problem 1: Insufficient data error during backtesting

Problem 2: Only few trades on 1d timeframe

Problem 3: Large variation in results across timeframes

📊 Timeframe Selection Decision Tree

🔗 Reference Materials

Supporting Documentation

Recommended Reading

📌 Key Points Summary

➡️ Next Lesson Preview