Strategy Validation Lab
A controlled strategy-validation demonstration using holdout testing, transaction costs, drawdown and parameter sensitivity.
The Strategy Validation Lab demonstrates how MFXG tests a trading rule beyond its development sample. The workflow includes signal lagging, transaction costs, a fixed holdout period, drawdown measurement, risk-adjusted statistics, and nearby-parameter sensitivity.
MFXG Internal R&D Demonstration — synthetic or controlled inputs; not client work or live investment performance.
Research question
Can a simple rule retain useful evidence outside the sample used to develop it, after transaction costs, and do nearby parameter choices tell a reasonably consistent story?
The demonstration uses a long-only moving-average crossover because the objective is to test the validation process rather than showcase a proprietary strategy.
Validation design
| Control | Implementation | Why it matters |
|---|---|---|
| Look-ahead control | Trading position is lagged one bar before returns are applied | Prevents a same-bar signal from benefiting from information that was not yet available |
| Sample separation | 65% development sample and 35% holdout sample | Keeps part of the synthetic series outside the initial evaluation |
| Costs | 3 basis points applied to turnover | Prevents trading activity from being treated as frictionless |
| Risk | Annualized volatility and maximum drawdown | Shows risk alongside return |
| Risk-adjusted evidence | Sharpe ratio | Relates return to observed variability |
| Sensitivity | Nearby fast/slow moving-average combinations | Tests whether the conclusion depends on one exact parameter pair |
What the recorded evidence shows
The base configuration uses a 20-period fast average and an 80-period slow average.
| Sample | Observations | Total return | Sharpe | Maximum drawdown |
|---|---|---|---|---|
| Development sample | 812 | -15.04% | -0.482 | -18.72% |
| Holdout sample | 438 | 1.91% | 0.148 | -21.68% |
These are synthetic-data results. The weak development result and modest holdout result are retained because the purpose of a validation workflow is to reveal uncomfortable evidence as well as favorable evidence.
Parameter sensitivity
The holdout sensitivity grid evaluates 20 nearby parameter combinations. Seventeen produced a positive holdout total return, while the median holdout Sharpe across the grid was approximately 0.20.
That does not establish market alpha. Some combinations were negative, and the base configuration's holdout Sharpe remained low. The evidence is therefore classified as mixed demonstration evidence rather than proof of a profitable strategy.
What this demonstrates
The useful output is the research discipline: separate development and holdout evidence, account for costs, inspect drawdown, challenge parameter dependence, and preserve results that weaken the original idea.
It does not demonstrate future profitability, live trading performance, or suitability for a particular investor.
Explore the working demonstration
Open the Strategy Validation Lab to inspect the human-facing demonstration built from the verified synthetic evidence.
For related work, see Quantitative Research & Strategy Validation.