Intraday AI Signals on ES & NQ: Q4 2025 Benchmark + First Look at 2023–2024 Backtests

This benchmark summarizes the out‑of‑sample performance of our pre-trained intraday AI signals for US equity index futures E-mini S&P 500 (ES) and E-mini Nasdaq-100 (NQ). The signals are designed to predict short-horizon mid-price trends (Up / Stable / Down) using Level II limit order book (LOB) and microstructure features.

Detailed results here

What’s new in this release

1) Q4 2025 results (this year’s Q4) — out-of-sample vs training through Sep 2025
The headline update is that we tested on Q4 2025 with the latest model trained through September 2025 (Pythia‑v0.1.1‑Sep25). That means the Q4 2025 benchmark is designed to be a clean out‑of‑sample check as we roll the horizon forward.

2) First publication of 2023 and 2024 backtests
For the first time, we are also publishing backtested results for 2023 and 2024, alongside 2025, so readers can compare behavior across multiple market regimes and volatility conditions.

3) Small model/training pipeline improvements
The current benchmark uses Pythia‑v0.1.1‑Sep25, which includes incremental improvements versus the version used in our prior benchmark cycle (Q4 2024 to Q3 2025). These changes are focused on training and retraining mechanics—how retraining candidates are generated, which candidates are promoted, and which parts of the network are held fixed vs. updated when the horizon is rolled forward.

Benchmark setup (high level)

To keep the benchmark comparable over time, we evaluate the predictions inside a simple intraday trading harness:

• Prediction cadence: 1-minute signals during 09:30–16:00 ET.
• Decision cadence for the benchmark strategy: Every 5 minutes (09:45–15:30 ET), the strategy selects long/short/flat using 1/70 of daily starting capital per decision; positions are entered over the next 5 minutes and exited at the close (15:55–16:00 ET).
• Costs: 1 bp round-turn assumption (plus roll rules on the front-month continuous contract).

Results snapshot

Q4 2025 (this year’s Q4)

• ES (25Q4): Sharpe 2.650, Ann. Return 10.401%, Ann. Vol 3.926%, Max DD -1.412%
• NQ (25Q4): Sharpe 2.504, Ann. Return 12.360%, Ann. Vol 4.937%, Max DD -1.002%

Full-year context (2025, 2024, 2023)

• ES
  • 2025: Sharpe 1.382, Ann. Return 12.714%, Max DD -4.374%
  • 2024: Sharpe 2.248, Ann. Return 13.630%, Max DD -3.483%
  • 2023: Sharpe 1.074, Ann. Return 6.230%, Max DD -3.340%
• NQ
  • 2025: Sharpe 2.337, Ann. Return 23.481%, Max DD -3.834%
  • 2024: Sharpe 2.205, Ann. Return 17.551%, Max DD -4.762%
  • 2023: Sharpe 0.913, Ann. Return 6.465%, Max DD -3.229%

Performance commentary by symbol

NQ (Nasdaq-100) — 2025

In 2025, NQ performance is solid, with the strategy curve providing an “outside envelope” to the index price—i.e., the strategy’s net curve remains comparatively stable while the underlying price experiences larger swings (see the 2025 NQ chart).

ES (S&P 500) — 2025

For ES in 2025, the strategy tends to track the underlying’s movements closely (directionally), with a smoother net curve relative to the index path (see the 2025 ES chart).

2024 (ES and NQ)

• ES 2024: Performance is consistently positive across quarters, highlighted by a very strong Q3 2024 (Sharpe 4.939, low drawdown), which helped drive the strong full‑year risk-adjusted profile.
• NQ 2024: Results are also strong overall, with notably good performance in Q1 and Q3, and a weaker patch in Q2—consistent with how sensitive NQ can be to regime shifts.
• Regimes: higher-for-longer rates plateau; disinflation but above-target inflation; election-year volatility premium; continued AI/mega-cap concentration.

2023 (ES and NQ)

• ES 2023: A more mixed year: strong performance early (especially Q2), followed by a choppier second half where Q3 is negative and Q4 is closer to flat.
• NQ 2023: Modestly positive overall, again with the strongest contribution in Q2, and more muted (but still positive) results in the second half.
• Regimes: disinflation + final rate hikes; March banking stress; late‑year rate/term-premium shock; AI/tech-led concentration.

Important notes: These results are hypothetical/simulated, use stated cost assumptions, and past performance is not necessarily indicative of future results. Futures trading involves substantial risk.