Quant Lab Tools
Home / Methodology
How this works

Methodology

Every tool and guide here comes out of an actual research workflow, not a content brief. This page is that workflow, end to end — plus who runs the site and how the material is made and checked.

Who runs this

A
ridingyo

A systematic-trading developer with a software background. I build and validate expert advisors on MetaTrader (MQL4/MQL5) and research agentic trading architectures, and I apply Marcos López de Prado's validation methodology — Deflated Sharpe Ratio, Probability of Backtest Overfitting, purged and embargoed walk-forward, and combinatorially purged cross-validation — to everything I keep. The tools on this site are the ones I reach for in my own work; the guides are what I learned the hard way.

I am not a financial adviser, and nothing here is a recommendation. I write about how to test whether a strategy is real — the engineering and statistics of validation — not about what to trade.

What this site is, and isn't

Quant Lab Tools is a set of diagnostics and explanations for people who build their own systematic strategies. It exists because the methods that actually separate a real edge from a lucky backtest live in dense papers and code libraries, and almost nowhere in a form you can just use.

It is calculators that implement published methods faithfully, and guides that explain them in plain language with real examples. It is not a signal service, a course funnel, or a place that tells you a number will go up. There are no trade ideas here and there never will be.

The validation workflow

This is the order I move through, and the order the tools are meant to be used in. The sequence matters: each step exists to stop a specific way a backtest lies to you.

  1. Structural hypothesis first

    Before any search, I write down why an edge should exist — what inefficiency, what behaviour, on what instrument and timeframe. A reason invented after seeing a winner is a story, not a hypothesis, and it is the root of most overfitting.

  2. Realistic costs from the start

    Spread, commission and slippage are modelled before I read any performance number. I treat transaction-cost modelling as the single most critical element of a backtest, because I have watched a clearly profitable gross result flip to a net loss with nothing else changed.

  3. Log every trial

    Every variant, threshold and discarded idea is counted. That count feeds the Deflated Sharpe Ratio, because the honesty of a result depends entirely on how many tries stand behind it.

  4. Out-of-sample without leakage

    Validation uses purged, embargoed walk-forward — training samples whose outcomes overlap the test window are removed, with a buffer after. A holdout reused many times is no longer out-of-sample, so I treat it as a spending budget.

  5. Combinatorial cross-validation

    Where it matters, CPCV builds many backtest paths from combinations of purged folds, giving a distribution of out-of-sample outcomes rather than one. A genuine edge looks similar across paths; an overfit one falls apart on most.

  6. Deflated Sharpe and PBO gates

    A candidate has to clear the luck line — the Sharpe an unskilled strategy would reach given the trial count — with a Deflated Sharpe Ratio at or above 0.95. Below that, the result is labelled provisional, not promoted. PBO checks whether the selection process itself is picking noise.

  7. Paper trading before live

    Nothing touches live capital on the strength of a backtest. Strategies run forward on paper first, and automation is never given decision authority over live money until it has been validated forward — and even then, conservatively. The goal of the discipline is survival, not a leaderboard.

How the tools and guides are made

I think the honest version of this matters, so here it is:

Principles I won't break

Start here

The tools are meant to be used together. If you have a backtest in hand, the natural entry point is the overfitting check.

Deflated Sharpe calculator → Overfitting guide → Cost calculator → Position sizing →
Educational, not investment advice. Quant Lab Tools provides statistical diagnostics and educational material on strategy validation. It does not recommend any security, strategy, or trade, and does not predict performance. Simulated and historical results do not guarantee future outcomes. Verify any figure independently before relying on it.