Backtesting

The sim engine replays historical L2 book snapshots and trade ticks through your WASM algo, producing fills, PnL, and performance metrics. All replay is deterministic - same seed + same data = identical output.

Quick Start

bash

# Build + backtest in one command
sequence backtest BTC-USD --start 2026-03-01 --end 2026-03-08
 
# Multi-symbol portfolio backtest
sequence backtest --symbols BTC-USD,ETH-USD --start 2026-03-01 --end 2026-03-08
 
# Direct sim-engine invocation
sim-engine run --wasm algo.wasm --symbol BTC-USD --start 2026-03-01 --end 2026-03-08 --capital 100000 --seed 42

Multi-Symbol Mode

When using --symbols, events from all symbols are merged into a single timeline sorted by exchange timestamp. Your algo receives L2Book updates with symbol_id set:

rust

fn on_book(&mut self, book: &L2Book, state: &AlgoState, _features: &OnlineFeatures, actions: &mut Actions) {
    match book.symbol_id {
        1 => { /* First symbol in --symbols list */ }
        2 => { /* Second symbol */ }
        _ => {}
    }
}

Each (symbol, venue) pair gets its own matching engine with independent queue state, latency sampling, and microstructure EMAs.

Realism Features

The sim engine includes configurable market microstructure models. All are off by default so existing backtests are unaffected.

Maker Adverse Selection

When enabled, passive fills are scored for toxicity based on trade size, order-flow imbalance, volatility regime, and spread regime. High-toxicity fills may be probabilistically skipped or receive a price penalty.

bash

MAKER_ADVERSE_ENABLED=true sim-engine run --wasm algo.wasm --symbol BTC-USD ...

Key parameters (all configurable via env vars):

MAKER_ADVERSE_SKIP_RATE - max probability of fill skip (default: 0.3)
MAKER_ADVERSE_PENALTY_K - price penalty scale (default: 0.5)
MAKER_ADVERSE_BIAS - sigmoid center; higher = less toxic by default (default: 2.0)

Self-Impact (Shadow Book)

When enabled, your aggressive orders consume visible depth and your passive orders add to it. The algo sees an adjusted book reflecting its own footprint:

code

effective_depth = historical_depth - consumed_shadow + own_displayed_depth

bash

SELF_IMPACT_ENABLED=true sim-engine run --wasm algo.wasm --symbol BTC-USD ...

Consumed depth decays exponentially (SELF_IMPACT_DECAY_RATE, default 0.5/sec) and partially clears when new raw depth replenishes at consumed price levels.

Volatility-Conditioned Latency

Submit and report latency parameters scale with recent volatility and trade intensity:

bash

LATENCY_VOL_MU_SCALE=0.3 LATENCY_VOL_SIGMA_SCALE=0.2 sim-engine run ...

When scale factors are 0.0 (default), latency sampling is identical to the constant-parameter model.

Queue Cancel Attrition

Exponential decay on queue_ahead models background cancellations thinning the queue over time:

bash

QUEUE_CANCEL_DECAY_RATE=0.5 sim-engine run ...

The decay is capped at QUEUE_CANCEL_MAX_DT_S (default 5s) to prevent unrealistic jumps during sparse event gaps.

Calibration

Tune queue, impact, and adverse selection parameters against historical data:

bash

# Queue + latency model
sim-engine calibrate --symbol BTC-USD --venue kraken --target queue
 
# Impact model (k_temp, k_perm, adverse_k)
sim-engine calibrate --symbol BTC-USD --venue kraken --target impact
 
# All parameters jointly
sim-engine calibrate --symbol BTC-USD --venue kraken --target full
 
# Maker adverse selection - Step 1: fill rate
sim-engine calibrate --symbol BTC-USD --venue kraken --target maker-adverse-fill-rate
 
# Maker adverse selection - Step 2: penalty (freezes step 1 params)
sim-engine calibrate --symbol BTC-USD --venue kraken --target maker-adverse-penalty

Calibrated parameters are saved to ./data/calibration/{venue}_{symbol}_latest.json.

Stress Testing

Test your algo under synthetic adversity:

bash

sim-engine stress --wasm algo.wasm --symbol BTC-USD --start 2026-03-01 --end 2026-03-08 \
  --scenario liquidity-shock --factor 0.2

Available scenarios: liquidity-shock, spread-widen, depth-truncation, feed-outage, gap-up, gap-down.

Output

The sim engine produces:

Tearsheet with Sharpe, Sortino, Calmar, max drawdown, win rate, profit factor
MM metrics with 1s/5s/30s markouts, maker/taker ratio, inventory time-at-risk, fill rate
Equity curve sampled every 1000 events
Trade list with per-fill P&L attribution