I Traded StrategyQuant X Strategies Live for 2 Years. Here’s How I Did It.

Who / what this is about

• Brandon (Trivium System Trading) discusses running a fully systematic, intraday breakout strategy portfolio using StrategyQuant X (SQX), with live trading running for just over two years.
• He emphasizes that backtest performance often degrades live if the research process doesn’t incorporate robustness and realistic trading costs/conditions.

Markets, instruments, and tickers mentioned

Primary instruments (mostly CFDs):

• Bitcoin
• US indices (mentions NASDAQ, S&P 500, Dow Jones)
• German index
• Japanese index
• Gold
• Oil
• Forex markets (mentions “Forex pairs”)

No specific stock/ETF tickers were explicitly named in the subtitles.

Key methodology / step-by-step framework (SQX workflow + robustness + portfolio construction)

1) Data + execution realism setup (before generating strategies)

• Ensure CFD contract specifications match the actual broker (ticks/tick size, contract specs, etc.).
• Use accurate spreads and commissions for that broker.
• Download/use tick data where possible; use correct time zone alignment.
• Clone data to the correct time zone.
• For time/cost realism:
  • Historical prices differ dramatically over long histories (e.g., gold’s “current spread” may not apply to older regimes).
  • Add pessimism via +5–10% spreads and some slippage when stress-testing.

2) Strategy generation choices in SQX

• Use the Builder to generate large numbers of candidate strategies quickly (he mentions “thousands per second”).
• Prefer signals with low/zero parameters for robustness:
  • Zero-parameter / binary logic example: “Yesterday’s close > yesterday’s open”
  • One fixed threshold example: “Daily close above a 200-period moving average” (where the 200 is fixed, not optimized)
• Turn off many default indicator blocks (e.g., RSI, MACD) because they introduce parameter richness (increased overfitting risk).
• Trading direction:
  • For index-heavy markets that drift upward over time, start with long-only rather than optimizing short-only strategies.
• Timeframe preference:
  • Builds commonly on 30-minute charts (also mentions hourly).

3) Trade construction (entry/exit design)

• Build simple breakout / trend-following logic.
• Use stop orders for every trade (mandatory stop-loss).
• Profit target:
  • Optional, and many strategies end up without profit targets.
  • Breakouts can “run” (especially via end-of-day closure) to improve risk/reward.
• Position lifecycle:
  • Trades close at end of day, increasing average trade frequency.
  • Typical trade frequency cited: ~100–200 trades/year, sometimes up to ~400 trades/year.
• Stop sizing:
  • Uses ATR only (not fixed pips).

4) Robustness testing stack

• Out-of-sample (OOS) testing
  • He mentions two out-of-sample steps in his workflow.
  • OOS is framed as a high-impact robustness filter (using an 80/20 heuristic).
• Multimarket testing
  • Test the same strategy logic across many markets (correlated and uncorrelated).
  • Evaluate median/average performance across markets, not perfect results everywhere.
  • This can act as extra OOS.
  • Strategy counts cited: up to 50,000 trades sometimes, and as high as 100,000 trades in cases (many are out-of-sample).
• Monte Carlo tests
  • Uses Monte Carlo trade resampling, data permutation, and parameter permutation.
  • He claims these improve average OOS performance in his testing series (details not fully quantified beyond later Sharp ratio examples).

5) Portfolio construction (SQX Portfolio Maker + correlation + risk)

• After candidates pass robustness filters:
  • Use SQX portfolio maker and correlation analysis.
• He targets a monthly P&L correlation threshold:
  • Typically < 0.2 to 0.25 (described as a “sweet spot”).
• He chooses unrelated returns (even if markets are correlated), focusing on return correlation, not just asset correlation.
• He initially reduces a pool:
  • From hundreds to as low as 2–3 strategies (depending on the market).
• Then:
  • Check cross-market portfolio-level correlation and “kick out” strategies with heavy correlation.
• Risk management implementation:
  • Does portfolio-level Monte Carlo maximum drawdown analysis in Excel after exporting trades from SQX.
  • Then defines risk and deploys.

Risk management details, stops, drawdowns, and explicit controls

Stop-loss / take-profit rules

• Stop-loss is mandatory on every trade.
• Risk per trade:
  • If the stop triggers, the strategy takes only about 0.1%–2% risk hit (position sizing rule).
• Profit targets:
  • Often none; breakouts typically run until:
    • stop-loss, optional trailing stop, or
    • end-of-day closure.

Stop sizing method

• Uses ATR (explicitly rejects fixed pip stops because price levels evolve).

Drawdown management and maximum acceptable loss

• Portfolio max drawdown (master account):
  • Max drawdown set to 20%.
• Returns vs drawdown relationship:
  • He doesn’t focus on maximizing returns, but cites:
    • historical average performance around ~3.6% per month
    • roughly 40–60% annual in that ballpark (approximate wording)
  • Implicit risk/reward cited:
    • “conservative estimate” of about 40% return vs 20% max drawdown → roughly 2:1 risk-to-reward.

Why drawdown breaches may not mean “strategy edge is dead”

He outlines three drivers of live performance:

1. True edge decay over time (statistical)
2. Randomness / variance (Monte Carlo helps estimate worst paths)
3. Live conditions differ from backtests:
   • broker spread widening
   • commissions
   • slippage
   • platform/VPS failures (e.g., missed trades)
   • data resolution differences

Biggest gaps between backtest and live (explicit cautions)

• He says a major source of gap is not “edge decay” but mismatch between backtest modeling and live execution.
• Specifically:
  • Strategies that look good on 1-minute OHLC / high-low-close can degrade significantly on tick data.
• He recommends validating in MT5 using his broker data, then matching:
  • entry/exit timestamps
  • costs
  • P&L
  • overall equity curve shape
• If strategies still don’t match after tick testing/live monitoring:
  • discard the strategy and restart.

Performance metrics and numbers cited

Robustness / overfitting mitigation metrics

• Overfitting lesson: manual parameter tuning created “perfect equity curves” that collapsed live.
• In an experiment described as “random strategies”:
  • No robustness testing
    • average Sharp ratio ~0.65 in-sample
    • average out-of-sample Sharp ratio around ~0.4–0.44
    • implies about ~50% reduction in Sharp ratio
  • After robustness testing
    • multimarket testing described as boosting OOS Sharp to roughly ~0.5–0.6, depending on strictness
  • (These are reported illustrative averages from his series.)

Live track record / attribution

• Public track record runs just over two years.
• Within the portfolio:
  • A large portion of profitability comes from:
    • Bitcoin (roughly a quarter to a third of profitability)
    • NASDAQ strategies (mentions running ~9 or 10 NASDAQ strategies, the most from any single market)
• Monthly variability:
  • There are losing months—not a steady upward line—and risk controls are what allow survival.

Explicit recommendations / cautions (actionable takeaways)

• Don’t trust backtests blindly: prioritize robustness and realism.
• Prioritize data accuracy:
  • correct CFD specs, spreads, commissions, tick data, time zone.
• Prefer simple / low-parameter / zero-parameter signals to reduce overfitting.
• Use OOS + multimarket + Monte Carlo to increase probability of real edge.
• Use ATR for stops (not fixed pips).
• Validate strategy output in your live-compatible platform (he uses MT5) with broker data, and verify:
  • trade timestamps
  • costs
  • P&L
  • equity curve shape
• If performance doesn’t match backtests under realistic conditions: throw it out.
• Risk:
  • Set a portfolio-level max drawdown (he uses 20% on the master account).
  • Use Monte Carlo worst-case framing to set risk, not a single historical drawdown.

Disclosures / disclaimers

• The subtitles include no clear “not financial advice” disclaimer in the provided text.

Presenters / sources mentioned (at end)

• Brandon (Trivium System Trading)
• Cornell (interviewer; referenced as “Cornell and Thomas” at the start)
• Thomas (interviewer/colleague)
• Clonex (source of indicator/blog ideas; credited for multimarket testing inspiration)
• Larry Williams (credited for patterns such as Kangaroo Tail and Smash Day)
• Mentioned by name: René Bala (systematic trader interview referenced)
• Mentions using AI tools: ChatGPT / GPT, Claude (and “Bad GPT” as a subtitle error)