Skip to content

openbb_widget_description.md

Latest commit

 

History

History
133 lines (95 loc) · 4.97 KB

openbb_widget_description.md

File metadata and controls

133 lines (95 loc) · 4.97 KB
Raw

Pulse — Synthetic Market Data by Simudyne

Pulse is Simudyne's agent-based market simulator. It generates synthetic tick-level order book data for financial instruments. These are calibrated to real historical microstructure and delivered via a Python SDK and REST API. Use it to test execution algorithms, stress-test strategies, and build training data for market models.

Get started free: Sign Up · API Docs · API Console

What the API Does

Browse and download pre-run simulations

The free tier gives immediate access to a library of cached simulations across HKEX symbols. No job submission needed, just browse, pick a symbol, and download full L2 LOB data and tick data.

from simudyne import PulseABM

client = PulseABM(api_key="pk_live_...")

sims = client.simulation.list_cached()
df   = client.simulation.get_sim_data(sim_id="...", filename="sim_data.parquet")

Each simulation returns full millisecond-resolution L2 order book data (10 levels each side) plus an order-by-order event stream as a Polars DataFrame.

Run custom simulation jobs

Submit a job against any symbol, calibration date, and market scenario. Pulse runs up to 100 Monte Carlo realisations in parallel, giving you an empirical distribution of outcomes rather than a single path.

job = client.simulation.run(
    symbol="700.HK",
    cal_date="2025-09-01",
    n_runs=100,
    scenario="flash_crash",
    exec_algos=["TWAP", "VWAP"]
)
status  = client.simulation.get_job_status(job.job_id)
results = client.simulation.get_job_results(job.job_id)

Monte Carlo mid-price paths for 700.HK — normal vs flash crash scenarios

Available scenarios:

Scenario What it injects
normal No directional shock — background agents only
flash_crash Large rapid SELL (22× order size, 500 ms intervals)
buy_panic Large rapid BUY (22× order size, 500 ms intervals)
gradual_selloff Sustained SELL pressure (10×, 5 s intervals)
trending_up Slow steady BUY drift (5×, 30 s intervals)
trending_down Slow steady SELL drift (5×, 30 s intervals)

Test execution algorithms under stress

Include TWAP, VWAP, or a Custom Static Schedule (CSS) in any simulation job. Pulse runs the algo inside the simulated order book and returns implementation shortfall, fill rate, and market impact metrics alongside the tick data.

job = client.simulation.run(
    symbol="9988.HK",
    cal_date="2025-09-01",
    scenario="gradual_selloff",
    exec_algos=["TWAP", "VWAP", "CSS"]
)

Output files per simulation:

File Contents
sim_data.parquet Full L2 book + order events at tick resolution
mid_price_by_min.parquet Mid-price in 1-minute bars
l2_by_second.parquet L2 snapshot (10 levels) sampled per second
exec_results.parquet Slippage, fill rate, and market impact per algo
exec_schedule.parquet Order schedule with times and quantities
results.json Job-level summary metrics

Bulk download across multiple simulations

dfs = client.simulation.get_bulk_data(
    sim_ids=[...],
    include_sim_data=True,
    include_mid_price=True
)

Returns a ZIP of parquet files. This is useful for building datasets across many symbols, dates, or scenarios in one call.

How We Validate the Data

Synthetic data is only useful if it is realistic. Pulse is validated against historical microstructure using a suite of established academic benchmarks, the same methods used to evaluate leading market simulators in recent literature.

Benchmark What It Checks
Cont (2001) Stylised Facts 11 statistical properties that all real markets exhibit
LOB-Bench Distributional accuracy across 16 order book metrics
Bouchaud Impact Response Price impact of individual order events across lag horizons

What You Can Build With It

  • Algo benchmarking — compare TWAP vs VWAP slippage across symbols and stress regimes without live market exposure
  • Stress testing — run the same strategy across all six scenarios and 100 Monte Carlo paths to get tail-risk estimates
  • Training data — generate large volumes of realistic synthetic order flow for reinforcement learning agents or market impact models
  • Pre-trade TCA — estimate expected execution costs for a given order size before placing it in the real market
  • Liquidity research — study how spread, depth, and order imbalance evolve under different shock types across the HKEX universe

Tiers

Free Pro
Cached simulations ✓ ✓
Tick-level L2 download ✓ ✓
Custom calibration dates — ✓
All 6 pre-built scenarios — ✓
Up to 100 Monte Carlo runs — ✓
Execution algo testing — ✓

Start free — no credit card required: pulse.simudyne.com