Using Synthetic Data to Test Trading Rules and Portfolio Methods
Summary
This document explains how synthetic data can help investigate trading systems when historical observations are too limited to support strong conclusions. It distinguishes simulated price paths for testing individual rules, correlated asset-return series for portfolio optimisation, and strategy-level return series for studying account curves. Each simulation requires assumptions about relevant properties, such as a process plus noise, or returns with specified means, volatility, correlation, and skew.
The proposed uses include checking rule behaviour in chosen market conditions, estimating holding periods and trading costs, exploring parameter sensitivity, and studying drawdowns or capital scaling. The suggested workflow uses real data for out-of-sample forecasts and allocations, while synthetic data helps design rules and calibrate portfolio and risk methods. The author cautions that simple Gaussian assumptions cannot capture every feature: linked markets, asymmetric returns, jumps, and unusual kurtosis require more complex models. Synthetic results depend on chosen assumptions and are a tool for controlled analysis, not a substitute for real out-of-sample evaluation.
Key ideas
- Synthetic price paths can test how individual trading rules respond to specified conditions.
- Simulated correlated returns can help explore portfolio allocation methods.
- Strategy-level return simulations can illustrate distributions of returns and drawdowns.
- Use real data for out-of-sample forecasts and allocations within the proposed workflow.
- Simple Gaussian simulations omit features such as jumps, skew, and cross-market links.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.