<div align=center>

WindTunnel : Market Simulation by Generation

</div>

WindTunnel is a modular, multi-layer generative market simulation framework designed to evaluate trading strategies and market impact under synthetic yet realistic conditions.

It combines macro regime generation, high-frequency order flow simulation, LOB microstructure modeling, and plug-in execution strategies (e.g. VWAP, TWAP, Dynamic Stop Loss), supporting both controlled experimentation and generative scenario construction.

---

Installation

You can install WindTunnel via pip or conda:

# Install via conda
conda install -c jialuechen::windtunnel

---

Example Config (configs/market_crash_scenario.json)

{
  "regime": "volatile",
  "volatility": 0.35,
  "liquidity": "low",
  "lob": {
    "levels": 5,
    "latency_ms": 10
  },
  "strategy": {
    "type": "vwap",
    "params": {
      "target_volume": 100,
      "time_horizon": 10
    }
  }
}

---

Run a Simulation

python examples/run_simulation.py

from windtunnel import WindTunnel
from windtunnel.plotting import plot_executions
import json

with open("configs/market_crash_scenario.json") as f:
    config = json.load(f)

gm = WindTunnel(config)
metrics = gm.run()

print("Simulation Metrics:")
for k, v in metrics.items():
    print(f"{k}: {v}")

# Visualize execution result
plot_executions(metrics.get("executions", []))

---

Strategy Plugins

• VWAPStrategy – Volume-weighted execution across time horizon
• TWAPStrategy – Time-weighted slices across fixed intervals
• DynamicStopLossStrategy – Trailing or fixed stop loss based on market conditions

You can create your own strategies and add them under windtunnel/strategy_plugins/.

---

New Features

Natural Language Scenario Configuration

Generate simulation configurations from plain English descriptions using OpenAI's GPT API. Example:

from windtunnel.nlp_config_generator import NLPConfigGenerator

api_key = "your_openai_api_key"
generator = NLPConfigGenerator(api_key=api_key)

description = "Simulate a high-volatility market with low liquidity. Use a VWAP strategy with a target volume of 1000 shares over a 1-hour period."
config = generator.generate_config(description)

print("Generated Configuration:")
print(config)

# Save the configuration to a file
with open("generated_config.json", "w") as f:
    f.write(config)

---

Interactive Visualization

Visualize execution prices, order flow, and order book depth interactively using Plotly.

from windtunnel.interactive_plotting import InteractivePlotting

# Example data for execution prices
timestamps = ['2025-04-13 10:00', '2025-04-13 10:01', '2025-04-13 10:02']
prices = [100, 102, 101]

InteractivePlotting.plot_execution_prices(prices=prices, timestamps=timestamps)

# Example data for order book depth
price_levels = [99, 100, 101, 102, 103]
bid_depth = [50, 70, 60, 40, 30]
ask_depth = [30, 40, 50, 60, 70]

InteractivePlotting.plot_order_book_depth(bid_depth=bid_depth, ask_depth=ask_depth, levels=price_levels)

---

Parallelized Simulation

Accelerate order flow generation and LOB simulation using multi-threading.

from windtunnel.order_flow_generator import OrderFlowGenerator
from windtunnel.lob_simulator import LOBSimulator

# Example parameters for order flow generation
order_flow_params = [
    {"base_price": 100, "steps": 10},
    {"base_price": 200, "steps": 10},
    {"base_price": 300, "steps": 10},
]

# Example parameters for LOB simulation
lob_params = [
    {"base_bid": 100, "base_ask": 101},
    {"base_bid": 200, "base_ask": 201},
    {"base_bid": 300, "base_ask": 301},
]

# Initialize generators
order_flow_generator = OrderFlowGenerator(flow_params={})
lob_simulator = LOBSimulator(lob_params={})

# Generate order flows in parallel
order_flows = order_flow_generator.generate_order_flow_parallel(order_flow_params, num_threads=3)
print("Order Flows:", order_flows)

# Simulate LOBs in parallel
lobs = lob_simulator.simulate_lob_parallel(lob_params, num_threads=3)
print("LOBs:", lobs)

---

Applications

• Strategy benchmarking under stress scenarios
• Execution cost modeling and market impact analysis
• Synthetic data generation for pre-trade analytics
• Reinforcement learning environment (future support)

---

Roadmap

• [x] Integrate pretrained diffusion/VAE market generators
• [x] Add LLM-driven natural language scenario parser
• [x] Extend to multi-asset & cross-venue simulation
• [ ] Add dashboard interface for real-time simulation control

---

License

MIT License. See LICENSE.

---

Citation

@misc{windtunnel2025,
  title   = {WindTunnel: A Multi-Layer Generative Market Simulation Framework},
  author  = {Jialue Chen},
  year    = {2025},
  note    = {https://github.com/jialuechen/windtunnel}
}