PT-BRContact us

Methodology

The backtest approves. The market rejects.

Strategy-generation platforms produce thousands of candidates, and the vast majority, even after extensive robustness testing, fail when running in real MT4/MT5 environments.

Why simulated validation fails

Illustrative animation, not actual results.
Transcript
  1. Backtest: a strategy's equity curve rises smoothly, and the backtest approves it.
  2. Live at the broker: the same strategy runs under real conditions. Spread and commissions flatten the slope, slippage adds small losses, latency adds noise, and the live feed drifts from history.
  3. The market rejects: the live curve ends well below the backtest. That is the gap the backtest never sees.
  4. xCube measures that gap with real money: the haircut.
  5. Validated by real execution, not by backtest.

Four places where a simulation flatters a strategy, and what xCube measures instead.

  • Overfitting

    What the backtest assumesThe strategy is optimized on the same history it is judged on, so a fine curve can be fit to noise and leaked data.

    What xCube measures at the brokerOut-of-sample data reserved at generation, then months of walk-forward on live demo accounts the strategy has never seen.

  • Execution costs

    What the backtest assumesFixed or average spreads and commissions.

    What xCube measures at the brokerThe spreads, commissions and pricing logic the broker actually applied to every trade.

  • Slippage and latency

    What the backtest assumesOrders fill instantly at the requested price.

    What xCube measures at the brokerReal fills, with live slippage and latency. A real-money sentinel account measures the gap that remains between demo and real.

  • Data feed

    What the backtest assumesA historical price database.

    What xCube measures at the brokerThe live broker feed the strategy will actually trade on.

Each row is a source of the gap between backtest and live results. xCube validates on the right-hand column, so a strategy earns its place with trades the broker really executed.
The central question: how to identify, validate and combine robust strategies under real execution conditions, instead of relying on simulated environments?

Hypotheses

Three hypotheses, continuously tested

01

Real execution > backtest

Strategies with acceptable performance on live demo accounts are statistically more likely to perform on funded real accounts than strategies validated by backtest alone.

02

Population + decorrelation

Large populations of mutually low-correlated strategies form portfolios with lower drawdown, less stagnation and better risk-adjusted returns.

03

Directional decomposition

Treating the long and short sides of each strategy as independent entities increases composition flexibility and reduces exposure to asymmetric regimes.

All three hypotheses are continuously tested against the execution data the operation itself generates.

Hypothesis 03 in practice

Every strategy is split into a Buy leg and a Sell leg, and each leg goes through the pipeline on its own.

One strategy

Trades both directions

Judged as a whole, a weak direction can hide a strong one.

Buy leg

Scored on its own

Its own trades and its own Individual Score.

Qualifies

Joins portfolios

Combined with legs of other strategies, balancing long and short exposure.

Sell leg

Scored on its own

Its own trades and its own Individual Score.

Held back

Keeps measuring in demo

Its trades keep accumulating on the demo account; it can qualify later.

Illustrative: which leg qualifies varies from strategy to strategy.

Splitting each strategy into Buy and Sell legs: a weak direction no longer drags down a strong one, and portfolios balance long and short exposure leg by leg.

The pipeline

From generator to real capital

The full pipeline is in production today. Every strategy travels eight stages, grouped in four phases, before touching real capital. Each phase admits only what passed the one before.

  1. Simulation

    Generate & filter

    EntersThousands of generated candidates

    1. Generation

      Strategies generated on M15 over 10 years of history, with out-of-sample data reserved.

    2. Filters & robustness

      Sensitivity, perturbation and Monte Carlo tests eliminate weak candidates.

  2. Live demo

    Measure at the broker

    EntersOnly the survivors of the robustness filters

    1. Live demo

      Months on demo accounts at the broker, each strategy uniquely identified.

    2. Individual Score

      Quality × evidence maturity, with the sentinel haircut applied.

  3. Scores

    Compose

    EntersOnly strategies with a sufficient Individual Score

    1. Portfolio construction

      The best strategies are combined into low-correlated portfolios.

    2. Portfolio Score

      Every portfolio is scored by proprietary models, alongside correlation and redundancy filters that keep each portfolio unique.

  4. Real capital

    Operate

    EntersOnly the best-scored portfolios

    1. Real account

      The best portfolios are copied to the real-money account.

    2. Monitoring

      Real vs. expected behavior, tracked continuously.

Diverges? Rotate. When live behavior drifts from what was expected, the portfolio leaves the real account and a new composition takes its place.

UnderneathContinuous trade collection, ingestion and analysis in the xCube platform, and long/short decomposition at every evaluation stage.

The eight stages in four phases. Each phase filters what reaches the next; monitoring feeds back into composition, so the process keeps replacing strategies instead of trusting any single one.

In depth

What each stage actually does

Generation and cheap pre-filtering

Multiple filtering frameworks (profit factor, stagnation, return/drawdown, minimum activity) and perturbation frameworks: UTC to UTC+4 time zones, M5/M30/H1 timeframes, Monte Carlo for outliers and spread/slippage. The goal is not to find perfect strategies: it is to eliminate weak candidates cheaply, before the expensive validation.

Real walk-forward

The demo accounts are continuous walk-forward environments at the broker, generating a growing base of trades under real spread, slippage and latency conditions.

Long/short decomposition

Each strategy is split into independent Buy and Sell legs for the whole pipeline. The regime bias this introduces is acknowledged and mitigated by extended demo periods before eligibility.

Score → composition → rotation

Only strategies with a sufficient Individual Score are combined into portfolios; portfolios receive a Portfolio Score, correlation verification and a redundancy filter before competing for real capital, and are rotated when behavior diverges from expectations.

Talk to the partners

We speak with qualified investors, institutional partners and brokers interested in our research and execution process.

contact@xcube.financeContact details