Algorithmic Futures Portfolio — 12 Strategies on One Prop Account

The problem

Prop-firm accounts are closed by drawdown, not by a weak Sharpe ratio. One profitable strategy usually can’t be sized up without breaching the trailing-drawdown limit, so the real target is the highest monthly income at the contract count the drawdown budget can fund. That means combining strategies whose losing streaks don’t overlap, and measuring drawdown the way the firm does: intraday and marked to market.

What I built

Sole engineer: research, backtesting, portfolio sizing and live execution.

• Data pipeline: a custom parser for 4.65 GB of Databento 1-second data, front-month roll selection, and an 82M-bar Parquet catalog across 6 contracts (NQ, ES, YM, RTY, GC, CL).
• Research: 260+ candidate strategies, each put through random-period sampling, split-half replication, concentration checks and an adversarial review that tries to refute the result.
• Backtesting: NautilusTrader at 1-second resolution with bracket/OCO orders, partial-fill protection, and slippage and commission models per contract.
• Portfolio sizing: the whole account is marked to market every second, and contract counts come from bootstrap resampling rather than the one lucky path.
• Live: a NinjaTrader 8 C# add-on with a Python bridge on a Windows VPS, plus a Next.js dashboard behind token auth.
• Safety net: a golden regression suite replays every strategy trade for trade before any live restart.

Hypothetical backtest at deployed size (2 micro contracts per strategy), Jan 2020 – Sep 2026, net of commissions and modeled slippage. Not a promise of future results.