# Cross-Venue Arbitrage Backtesting

Backtests of BTC and ETH cross-venue spreads from 2023–2025 show persistent but decaying edges. A latency-tolerant model targeting holding periods of 1–15 seconds captures the majority of exploitable divergence while minimising adverse selection.

## Dataset

Tick and 1-second order-book data from Binance, OKX, Bybit, Coinbase and Kraken for BTC/USDT and ETH/USDT. Period: January 2023 – March 2026. Combined sample >2.5 billion observations.

## Findings

- **Average fee-adjusted spread:** 0.18% on BTC majors, 0.35% on ETH.
- **Mean reversion half-life:** ~4.2 seconds.
- Entry signals using 3-sigma z-score generated 6–10 trades per day per pair after filtering for liquidity events.

## Risk Profile

Drawdowns concentrated during exchange outages or sudden volatility spikes (e.g. ETF approval, exchange-specific risk events). Stop logic based on quote-rate collapse reduced max drawdown from 14% to 8%.

## SoVael Design

Build an event-driven backtester whose latency, fees and fill assumptions are configurable per exchange. Use it to gate live deployment and to set maximum position limits per venue pair.

---

*Source: SoVael Trading internal backtest framework; augmented with CoinMetrics, Kaiko and CryptoQuant data.*