Summary

CryptoBacktestEngine hardcodes periods_per_day=24 (assuming hourly bars), but the system's own download_crypto_data() function downloads daily data (timeframe="1d"). This internal inconsistency causes annualized returns to be inflated by orders of magnitude when using the standard data pipeline.

This is not a user-error scenario — it's a mismatch between two modules in the same system:

• auto_download.py:153 — download_crypto_data(timeframe="1d") — downloads daily candles
• crypto_backtest.py:45 — CryptoBacktestEngine(periods_per_day=24) — assumes hourly candles
• auto_evolve.py:141 — WalkForwardConfig(bars_per_year=8760) — assumes hourly candles

Steps to Reproduce

pip install stratevo ccxt  # v6.0.5
stratevo evolve --market crypto --generations 30 --population 20 --seed 42
# (auto-downloads daily data via download_crypto_data, then backtests assuming hourly)

If ccxt can connect to an exchange, download_crypto_data fetches daily candles (timeframe="1d") and feeds them to CryptoBacktestEngine which processes them as if they were hourly — producing walk-forward window returns like:

Window 0: +462.0%  (66 trades) ✅
Window 1: +1666.1%  (63 trades) ✅ ⚠️
Window 3: +149193.8%  (67 trades) ✅ ⚠️

Root Cause

The annualization formula in crypto_backtest.py:594-597:

periods_per_year = 365 * self.periods_per_day  # 8760 (but data is daily, should be 365)
years = periods_used / periods_per_year          # 194 daily bars → 0.022 years (should be 0.53)
annual_return = ((1 + total_return) ** (1/years) - 1) * 100  # exponential explosion

194 daily bars are treated as 194 hourly bars (~8 days). A 50% return over 194 days becomes: (1.5)^(1/0.022) - 1 = 8,939,994,666%.

Same mismatch affects hold_periods (inflated 12×), Sharpe ratio (~6× inflated), and walk-forward window CAGR chaining.

Controlled Experiment (seed=42, same daily data, same config)

"FIXED" = set periods_per_day=1 to match the daily data that download_crypto_data actually provides.

Suggested Fix

Either:

A. Make the download match the engine — change download_crypto_data default to timeframe="1h"

B. Make the engine match the download — auto-detect data frequency from date intervals after loading:

# Infer from date gaps: median interval >= 12h → daily
dates = data[first_symbol]["date"][:50]
intervals = [(dates[i+1] - dates[i]) for i in range(len(dates)-1)]
median_hours = sorted(intervals)[len(intervals)//2].total_seconds() / 3600
periods_per_day = 1 if median_hours >= 12 else 24

Then propagate to CryptoBacktestEngine.periods_per_day, WalkForwardConfig.bars_per_year, and CAGR chaining in auto_evolve.py:1313.

Option B is more robust since users may provide data at any frequency via --data-dir.

Environment

• stratevo 6.0.5 (PyPI)
• Python 3.12.13, macOS Darwin 25.5.0