Visualization guide

Strategy Visualization Guide

How to use scripts/visualize_strategy.py + freqtrade built-in tools to actually understand a backtest result.

Using HonestTrend15mDry full history (2017-08 → 2026-04, BTC+ETH) as the running example.

Key numbers (look at these first)

Metric	Value	Interpretation
Total trades	570	8.6 years × 15m frequency → ~1 trade every 5.5 days on average (slow and steady)
Win rate	38.2%	Typical trend strategy — few wins, many losses, but wins are big
Total profit	+954.27%	10K → 104K (compound annual growth ~32%)
Avg per trade	+1.20%	Positive expected value
Avg win / Avg loss	$1,439 / −$620 = 2.32×	Winners are 2.3× larger than losers — that’s where the edge lives
Max drawdown	17,544 USDT (26.06%)	Occurred in the 2018 bear, lasted ~11.5 months
Sharpe (daily)	—	Use plotly-profit to gauge stability

One sentence: 62% of 570 trades lose money, but the 38% winners are on average 2.3× larger than the losers → systemic positive expectancy.

7 custom charts (by importance)

1. 01_equity_curve.html — Equity curve + drawdown (the single most important chart)

Upper half: green equity curve starting at 10K, ending around ~104K over 8 years. Gray dashed line marks the rolling peak. Lower half: red filled area shows current drawdown percentage.

How to read it:

• Gentle up-slope = healthy growth
• Sharp vertical drops + long time in negative territory = rough patch (e.g. 2018)
• Speed of drawdown recovery = strategy resilience
• New highs being hit continually = strategy is still earning

HonestTrend’s signature: crawled around −26% for all of 2018 (max DD segment), only reclaimed the peak in 2019 Q3. After that, 2020 and 2024 were two strong runs.

2. 02_drawdown.html — Rolling drawdown

Zoomed-in drawdown view. Automatically annotates the worst drawdown point.

Important thresholds:

• DD ≤ 15% — Normal volatility
• DD 15-20% — Warning (risk_manager PAUSE)
• DD > 20% — Strategy retirement trigger (risk_manager RETIRE)

HonestTrend hit −26% in 2018 — with the current kill-switch it would have been paused long before then. This is exactly why you run the kill-switch instead of trusting historical data blindly.

3. 03_per_pair.html — Per-pair performance

3 panels side by side: total profit / win rate / trade count.

Look for:

• Which pair contributes the most profit?
• Which pair has a low win rate?
• Big gap in trade counts (concentration vs spread)?

HonestTrend full history: BTC and ETH contribute about equally (strategy is symmetric).

4. 04_trade_distribution.html — 4-in-1 distribution plot

• Top-left: Profit distribution histogram. Is it normal? Long-tailed? Median near 0?
• Top-right: Holding duration distribution.
• Bottom-left: Profit vs holding duration scatter. If you see a “longer hold = higher profit” trend → the “let winners run” thesis holds empirically.
• Bottom-right: 20-trade rolling average profit. Staying above 0 = edge persists.

HonestTrend’s signature: 5–10 day holds are the big winners (scatter points beyond 60h cluster in positive returns); short-duration exits are mostly stops.

5. 05_monthly_heatmap.html — Monthly P&L heatmap

Rows are years, columns are months, color intensity is P&L. Green = win, red = loss.

At a glance:

• Which years earned the most (2020, 2024)
• Which months repeatedly turn red (historically May in crypto tends to drop — “sell in May”)
• Consecutive green/red months = clear regime

6. 06_exit_reasons.html — Exit reason analysis

Two bar charts:

• Left: trade count per exit reason
• Right: total profit per exit reason

For HonestTrend: almost every trade is trend_exit (EMA death cross). Because the strategy has no hard stoploss and no ROI take-profit, it only exits on signal reversal.

If one exit_reason is responsible for a lot of losses → that rule may need tuning.

7. 07_rolling_winrate.html — Rolling win rate

3 lines: 10 / 30 / 60-trade window win rate. 50% baseline.

How to read it:

• Consistently 40–50% = normal trend strategy
• Long-term < 40% = edge decay
• Sudden spike or drop = precursor to a regime shift

HonestTrend’s win rate oscillates between 35–45% long-term, consistent with theoretical expectations.

Freqtrade built-in charts (auto-generated)

8. 08_freqtrade_profit_plot.html — Official profit plot

Output of Freqtrade’s plot-profit command. Includes:

• Per-pair price curve
• Cumulative profit curve
• Per-pair average profit
• Open trades count over time

Use: verify the equity curve conclusions from chart 1.

9. 09_btc_candles_with_trades.html — Candles + trade markers

Output of Freqtrade’s plot-dataframe. Draws directly on the BTC candles:

• Green triangle = buy entry
• Red triangle = sell exit
• Overlaid EMA / ADX indicators

This is the most intuitive “what is the strategy actually doing” view. You can see first-hand:

• Under what price patterns does the strategy enter
• Where it exits
• Whether there are obvious “chase highs / panic sells” mistakes

Tabular analysis (CSV output)

freqtrade backtesting-analysis --analysis-groups 0 2 5 produces:

• group_0.csv — Enter tag summary
• group_2.csv — Enter × Exit tag matrix
• group_5.csv — Exit reason summary
• indicators.csv — Entry/exit indicator values for each trade

Open in pandas / Excel and filter, e.g.:

import pandas as pd
df = pd.read_csv('reports/full_history_btceth/indicators.csv')
# How did entries with FnG > 70 perform?
print(df[df['enter_FnG'] > 70][['pair', 'enter_date', 'profit_ratio']])

Cross-window comparison (regime analysis)

Want to compare 2018 vs 2024 strategy behavior? Run 2 backtests separately and store them in different reports subdirectories:

# 2018 crash
freqtrade backtesting --strategy HonestTrend15mDry ... --timerange 20180101-20181231 --export signals
python scripts/visualize_strategy.py --out reports/2018_crash

# 2024 bull
freqtrade backtesting --strategy HonestTrend15mDry ... --timerange 20240101-20241231 --export signals
python scripts/visualize_strategy.py --out reports/2024_bull

Compare 05_monthly_heatmap.html or 07_rolling_winrate.html to visually see regime differences.

References

• freqtrade · Strategy Analysis Example
• freqtrade · Advanced Backtesting
• WALK_FORWARD_FULL_HISTORY.md — full case study using these charts to diagnose the 2018 loss