← All case work
08 · EXPERIMENT & SIMULATION

Data modelling and research simulation

Reproducible systems for prediction, classification, factor research, strategy testing, and model comparison.

Keep data versions, experiment parameters, cost assumptions, and risk exposure together so results can be reproduced and challenged.

Custom engineeringCase workData path validation
Discuss a similar workflow ↗Inspect the interface ↓
WEILIAN ANXIN / quant-simulationPRODUCT INTERFACE
Prediction, Quant Research, and Simulation Systems primary interfacePrediction, Quant Research, and Simulation Systems secondary interface
Data cleaning and versioningBusiness system interface
THE BUSINESS PROBLEM

Where daily work and management decisions become harder.

Experiments are scattered across scripts and sheets

Backtests omit cost, liquidity, or data bias

Several model and factor versions must be compared

A research process needs to become a team product

Where it fits

Built for work that
standard software misses.

We start with one high-value workflow, make its scope and operating risks explicit, and expand only after the first version is proven in use.

01

Experiments are scattered across scripts and sheets

02

Backtests omit cost, liquidity, or data bias

03

Several model and factor versions must be compared

04

A research process needs to become a team product

Operating flow

How work moves
through the system.

Roles, permissions, review points, and exception rules are adapted to the customer’s current operating model.

01
Fix the data version

Record the sample, fields, cleaning steps, and information available at the time.

02
Build features and factors

Create candidate signals and test missingness, bias, and stability.

03
Train and test independently

Use time-based splits and out-of-sample data for evaluation.

04
Add real-world constraints

Include fees, slippage, capacity, trading halts, and risk limits.

05
Compare risk and attribution

Inspect returns, drawdown, exposure, and the sources of variation.

06
Register the experiment

Save data, code, parameters, outcomes, and the next falsifiable check.

EXPERIMENT RECORD

Data, parameters, constraints, and outcomes stay reproducible

Select a stage to inspect the operating rhythm that distinguishes this class of system.

Product surfaces

Inspect the controls,
states, and outputs.

These interfaces come from real business workflows. Public screens remove customer-sensitive data while preserving the operating structure.

Prediction, Quant Research, and Simulation Systems product interface 1
Data cleaning and versioningCreate candidate signals and test missingness, bias, and stability.
Prediction, Quant Research, and Simulation Systems product interface 2
Training and evaluationUse time-based splits and out-of-sample data for evaluation.
Scope

Choose the modules
the workflow requires.

Modules can be delivered independently or connected to existing software after discovery confirms the data and permission boundaries.

01Data cleaning and versioning
02Feature and factor research
03Training and evaluation
04Backtesting and simulation
05Risk and attribution
06Experiment registry
Engineering decisions

Technology follows
the operating requirement.

Data volume, latency, security, and current systems determine the final architecture.

ENGINEERING 01

Reproducible experiment tracking

Record data, code, parameters, environment, and outcomes for team review.

ENGINEERING 02

Out-of-sample robustness

Use time splits, stress scenarios, and sensitivity tests to reduce overfitting.

ENGINEERING 03

Realistic operating constraints

Represent fees, slippage, capacity, halts, and risk exposure in simulation.

Project boundaries

Scope and operating
conditions stay explicit.

  • Public screens remove strategy, performance, and commercially sensitive information.
  • Historical testing does not predict future performance and no return is promised.
  • Data licensing, model use, and production execution require separate authorisation.
IMPLEMENTATION & HANDOVER

Agree the scope, integration needs and ongoing support.

Agree development, deployment, training, maintenance and customer involvement, with cost and acceptance documented for the project.

Delivery scope

  • Data and experiment pipeline
  • Research and backtest workspace
  • Metrics, risk, and attribution
  • Testing and deployment documentation

Existing systems and information

  • Market and alternative data
  • Python / notebooks / warehouse
  • Model training platforms
  • Schedulers and compute
  • BI and reporting
  • Permissions, audit, and version control
START A CONVERSATION

Start with the work that most needs to improve.

Tell us how the work happens today, where time is lost, which systems you use and what needs to improve. We can then discuss scope, implementation and collaboration.

Discuss your business needs ↗