Elite Features

Quant Lab

11 min read

What it is

5 quantitative analysis tools that run against REAL OHLCV from 11 backtestable exchanges, no synthetic data: Parameter Sweep (try N values of one parameter), Walk-Forward (optimize on past, validate on unseen, detect overfitting), Monte Carlo (permute trade order for confidence intervals), Multi-Symbol (run one strategy across many assets to expose asset-specific overfit), and Compare (2-4 strategies side-by-side on the same window). Save promising experiments to the experiments list for follow-up. Backtest data caches per-exchange-pair-timeframe so reruns are instant.

How to think about it

Single backtests lie. A strategy that returns +85% in one run could be lucky on trade ordering, over-fit to one asset, or memorizing the in-sample period. Quant Lab gives you the 4 industry-standard ways to detect each of those failures: parameter sweep finds the true peak (vs the random spike), walk-forward detects in-sample overfitting, Monte Carlo gives a confidence interval not a point estimate, multi-symbol exposes asset bias. Use them in sequence before trusting any strategy with real capital.

Step-by-step

  1. Pick a tool from the 5 tabs

    Tool tabs across the top: Parameter Sweep / Walk-Forward / Monte Carlo / Multi-Symbol / Compare. Each has a one-line description. Start with Parameter Sweep — simplest and most useful as a first pass on any strategy.

  2. Parameter Sweep — find the true optimum

    Pick a strategy template + exchange + pair + timeframe + date range. Pick which parameter to sweep (e.g., RSI threshold 20-40 in steps of 5) and the result metric (Sharpe / Return / Win Rate). Engine runs N backtests, plots metric per parameter value. Reveals whether your strategy has a true peak (real edge) or a flat noisy curve (overfit).

  3. Walk-Forward — detect in-sample overfitting

    Splits the date range into in-sample (training) and out-of-sample (validation) windows. Engine optimizes parameters on in-sample, then runs them UNCHANGED on out-of-sample. If in-sample Sharpe is +2.5 and out-of-sample drops to +0.3, the strategy is over-fit. Honest strategies hold ≥60% of their in-sample performance.

  4. Monte Carlo — confidence interval on outcomes

    Takes your strategy's actual trade list, then RESHUFFLES trade order 1000 times. Plots the distribution of final equity curves. The 5th-percentile curve is your "realistic worst case". If 5th-percentile final equity is still positive, you have margin. If 5th-percentile bankrupts, your strategy is fragile to the order of wins/losses.

  5. Multi-Symbol — detect asset-specific overfit

    Runs the SAME strategy across BTC + ETH + SOL + AVAX + LINK simultaneously over the same window. Plots returns per symbol. If returns are +50% / -20% / +5% / -30% / +10%, the strategy works on BTC ONLY — it is asset-specific. A genuinely good strategy is positive (even if smaller) across most majors.

  6. Compare — side-by-side 2-4 strategies

    Pick 2-4 strategies. Same exchange / pair / timeframe / window. Engine runs all of them, plots equity curves on the same chart, shows side-by-side metrics table (Sharpe, return, max DD, win rate, profit factor) AND the correlation matrix between their returns. If two strategies have correlation > 0.85, they are duplicating each other — pick one, drop the other.

  7. Save promising experiments

    After each run, the Save Experiment row appears. Give it a name + notes. Saved experiments appear in the experiments list at the bottom of the page. Useful for tracking what you have already tried — "RSI sweep on BTC/USDT 1h — peak at threshold 30, Sharpe 1.4" — so you do not re-run identical experiments.

Tips & pitfalls

  • Use the tools IN SEQUENCE on each new strategy. 1) Parameter Sweep to find the optimal config. 2) Walk-Forward at the optimal config to validate out-of-sample. 3) Monte Carlo to size confidence interval. 4) Multi-Symbol to confirm cross-asset robustness. A strategy that survives all 4 has earned the right to be paper-traded.
  • Walk-Forward is the single highest-value tool. Most retail "backtested strategies" fail walk-forward because they were optimized on the same data they were tested on. If you only use one Quant Lab tool, use this one.
  • Monte Carlo with 1000 shuffles is the standard. The 5th-percentile equity curve is your realistic worst case — plan for this, not the median. If the median says +20% but the 5th percentile says -40%, you live in the world where -40% is plausible.
  • Multi-Symbol with all green = robust strategy. Some green + some red = asset-specific. All red = bad strategy. The pattern matters more than the average return — averaging hides the variance.
  • Compare with correlation matrix exposes "diversification theater". If you run 4 trend-following strategies and they all have 0.95 correlation, you have 1 strategy with 4 names. Diversification requires the underlying returns to be uncorrelated, not just the strategy names to be different.
  • Real OHLCV from 11 backtestable exchanges (Coinbase, Binance, Binance US, Kraken, Bybit, OKX, KuCoin, Gate.io, Bitget, HTX, Crypto.com). No synthetic data — the strategies are tested against actual market conditions that actually happened.
  • First run on a new (exchange, pair, timeframe, window) tuple takes 5-30 seconds to fetch candles from the exchange. Subsequent runs hit the cache and complete in 1-3 seconds. Run your full Sweep + Walk-Forward + Monte Carlo + Multi-Symbol on the same window — they all hit the same cache after the first fill.
  • Quant Lab requires a strategy template to run. If the templates list is empty (red banner), click Retry. Strategies you build in Strategy Builder are NOT auto-imported — use the Templates tab in Quant Lab specifically.