Backtesting and scorecard methodology

Why this page exists

AKQX attaches labels to backtests — Paper Trading Ready, Live Trading Ready, Rejected. A label is only worth anything if you can see the rule that produced it. This page states the data, the assumptions, the scoring and the thresholds, so that you can decide whether they are strict enough for your own purposes, and disagree with them on the evidence.

It is also the page that will date fastest. Thresholds and models change, and when they do this page and its last-updated date change with them.

Data sources and coverage

The research database holds about 41.5 million (4.15 crore) historical candles, collected from public exchange and broker feeds covering crypto and Indian markets, at every timeframe from 1 minute to 1 day.

Three constraints on that data are worth stating plainly:

• It is third-party data and may be delayed or inaccurate. AKQX does not originate prices.
• It is for use inside AKQX and may not be copied or redistributed. AKQX does not sell market data.
• Coverage is not uniform. Crypto has tick-level depth that supports full footprint order flow; Indian equity order-flow data is partial, which is why those views are limited and in beta.

Cost assumptions: fees and slippage

Every simulated trade has estimated trading fees and slippage deducted. A cost-free backtest is not a neutral starting point, it is a systematically optimistic one, and the error grows with trade frequency: the more a strategy trades, the more a cost-free simulation flatters it.

Position sizing is calculated with those costs included rather than against a cost-free ideal, and a fee-to-risk breaker refuses setups where estimated cost is large relative to the risk taken. A minimum stop distance exists for the same reason.

The 350-point scorecard

A result is scored across seven areas:

• Profitability — 50 points
• Risk — 50 points
• Consistency — 40 points
• Quantitative checks — 60 points
• Hedge-fund-style checks — 50 points
• Intelligence — 50 points
• Execution quality — 50 points

Profitability is 50 of 350. That weighting is the whole design: risk, consistency and the quantitative and hedge-fund checks together account for 200 points, so a strategy that made money fragilely cannot score well. If you think profitability should count for more, that is a reasonable disagreement to have — and now it is one you can have with a number in front of you.

Verdict thresholds

The total maps onto six bands:

• Rejected — 0 to 112
• Review Required — 113 to 168
• Paper Trading Ready — 169 to 224
• Live Trading Ready — 225 to 280
• Professional Grade — 281 to 315
• Institutional Grade — 316 to 350

These names describe score ranges and nothing more. “Live Trading Ready” means a simulation cleared 225 points and the hard conditions below; it is not a recommendation to trade, not a prediction, and not a statement that the strategy is safe. Whether anything ever reaches real money remains entirely your decision, and for Indian equities AKQX does not offer real execution at all.

Auto-reject rules

Some failures are not worth scoring around. A result is rejected outright when any of these is true:

• Expectancy is zero or negative — the average trade does not make money, so there is nothing to scale.
• Profit factor is below 1.0 — gross losses matched or exceeded gross wins.
• Maximum drawdown exceeds 30 percent — rejected regardless of profitability, because an unsurvivable equity curve is not a strategy.
• Walk-forward score below 5 — the edge did not hold on data the parameters were not fitted to.
• Monte Carlo score below 5 — the result did not survive resampling.
• Overfitting score below 5 — the parameters describe this history specifically rather than any general behaviour.
• Too few trades — fewer than 100 for crypto and forex, or fewer than 30 for Indian markets, where higher timeframes genuinely produce fewer signals. Below that, an edge cannot be told apart from luck.

Robustness checks in more detail

Walk-forward. Parameters are fitted on one slice of history and evaluated on the next. This is the closest a backtest gets to asking an honest out-of-sample question.

Monte Carlo. Trade outcomes are resampled across 1,000 simulations, and a drawdown of 30 percent from peak is treated as ruin. The useful output is not the average run but the proportion of runs that ended in ruin, because the average run is not the one that stops you trading.

Overfitting. Scores how much the result depends on precise parameter values. A strategy whose performance collapses when a parameter moves slightly was fitted to noise.

What this model does not account for

A transparency page that only lists strengths is marketing. These are the known gaps:

• Your broker's exact fees. Modelled costs are estimates and may differ from what you are charged.
• Order-queue position. A simulated limit fill does not know how many orders were ahead of it.
• Liquidity at your size. Slippage is estimated, not computed from the book your order would actually have hit.
• Exchange and broker outages, rejections and API failures. Live trading meets these; history does not.
• Funding-rate and fee-schedule changes over the tested period.
• Regime change. No backtest, however robust, can price in a market that starts behaving differently after the test window.

Every one of these pushes in the same direction: a backtest is more optimistic than reality. That is precisely why paper trading on live prices is a required second step, and why the drift between a backtest and its paper run is a more informative number than either result alone.

Versioning

The scoring configuration is versioned, and changing a threshold changes the version. The visible last-updated date on this page moves whenever the method behind it does, so a verdict can always be read against the rules that produced it rather than against today's rules.

Questions about methodology

Related pages