ammbt 0.2.0

High-performance vectorized backtesting engine for AMM LP positions

ammBT - AMM Backtesting Engine

High-performance vectorized backtesting engine for AMM liquidity provider positions.

Inspired by vectorbt, ammBT enables testing thousands of LP strategies simultaneously using vectorized operations and Numba compilation.

Features

• Multi-AMM Support: Uniswap v2 (CPAMM), Uniswap v3 (CLMM), Meteora (DLMM)
• Vectorized Backtesting: Test thousands of strategy variants in seconds
• Path-Dependent Simulation: Accurate swap-by-swap pool state simulation
• Comprehensive Metrics: IL, fees, PnL, Sharpe ratio, capital efficiency
• Interactive Visualization: Plotly-based charts and heatmaps

Architecture

ammBT uses a hybrid approach:

• Sequential (Numba): Swap processing through time (path-dependent)
• Vectorized (NumPy): Parallel strategy evaluation across parameter space

This allows testing N strategies with time complexity of O(swaps) instead of O(swaps × N).

Supported AMM Types

Uniswap v2 (CPAMM)

• Constant product formula: x × y = k
• Full-range liquidity
• Fixed 0.3% fee

Uniswap v3 (CLMM)

• Concentrated liquidity with tick ranges
• Multiple fee tiers (0.05%, 0.3%, 1%)
• Capital efficiency through range orders

Meteora DLMM

• Bin-based liquidity distribution
• Dynamic fees based on volatility
• Solana-native implementation

Status

⚠️ Work in Progress - This project is under active development.

Installation (Development)

git clone <repo>
cd ammbt
pip install -r requirements.txt
pip install -e .
python test_install.py

Quick Start

import ammbt as amm

# Generate synthetic swap data
swaps = amm.generate_swaps(
    n_swaps=10000,
    volatility=0.02,
    drift=0.0
)

# Define strategy space
strategies = {
    'initial_capital': [10000, 50000, 100000],
    'rebalance_threshold': [0.0, 0.05, 0.10],
    'rebalance_frequency': [0, 100, 500],
}

# Run backtest
backtester = amm.LPBacktester(amm_type='v2')
results = backtester.run(swaps, strategies)

# Analyze
print(results.summary())
amm.plot_performance(results, strategy_idx=0).show()

See examples/uniswap_v2_demo.ipynb for full walkthrough.

Using Real Swap Data

Historical swap data loaders are available for Uniswap V2/V3 (The Graph), and Solana venues like Meteora and Raydium (Birdeye APIs). Install the optional data dependencies first:

pip install -e ".[data]"

import time
import ammbt as amm
from ammbt.data import UniswapV3Loader

loader = UniswapV3Loader(network="ethereum")
data = loader.load(
    "0x8ad599c3a0ff1de082011efddc58f1908eb6e6d8",
    start_time=1700000000,
    end_time=1700500000,
    limit=20000,
)
swaps = data.to_backtest_format()

bt = amm.LPBacktester(amm_type="v3", fee_rate=0.003)
results = bt.run(swaps, strategies)

Live Swap Streaming (Polling)

Loaders can also poll their data sources and stream new swaps. This is useful for near-real-time simulation or building a rolling dataset.

import time
import ammbt as amm
from ammbt.data import UniswapV2Loader

loader = UniswapV2Loader(network="ethereum")
stream = loader.stream_swaps(
    "0xb4e16d0168e52d35cacd2c6185b44281ec28c9dc",
    start_time=int(time.time()) - 3600,
    poll_interval=5.0,
    batch_limit=500,
)

for batch in stream:
    # batch is normalized swaps with timestamp/amount0/amount1/tx_hash
    swaps = batch  # accumulate and backtest as desired
    break

Project Status

• Architecture design
• Uniswap v2 implementation (COMPLETE)
• Uniswap v3 implementation (COMPLETE)
• Meteora DLMM implementation
• Data loaders for real swap data (The Graph/Birdeye)
• Live swap polling via loader streaming
• Record system for event tracking

License

MIT