An open-source project (github.com/sstklen/trump-code) that analyzes all of Trump's Truth Social and X posts since taking office, runs brute-force statistical analysis across 31.5 million model combinations, and identifies patterns with statistically significant predictive power over S&P 500 daily moves. The surviving rules are published openly and updated daily.

Is the 61.3% hit rate statistically significant?

Yes — the reported z-score is 5.39, which corresponds to p < 0.00001. That's not a fluke. However, hit rate alone doesn't make a strategy profitable — position sizing, transaction costs, and whether signals arrive in time to trade on all matter. The raw statistical finding is solid; the trading implementation requires more work.

What is the most actionable finding?

Truth Social publishes posts 6.2 hours before they appear on X (38/39 posts verified). If a market-moving signal appears on Truth Social at 7 AM, it won't hit X until 1 PM. That's a 6-hour window. Monitoring Truth Social directly — not X — is the structural edge.

What are the strongest individual signals?

Pre-market posts containing RELIEF-type language average +1.12% same-day S&P. Pure tariff days average -1.057% (Apr 3: -4.84%, Apr 4: -5.97%). Silence days (zero posts) are 80% bullish, averaging +0.409%. Late-night tariff tweets are 62% wrong — the model auto-inverts them to LONG.

How does this relate to StockScout or algorithmic trading?

StockScout scores individual stocks based on fundamental signals. Trump Code is a macro regime filter — it signals whether the day is likely up or down for the whole market. Used together: Trump Code sets the macro regime, StockScout selects the stocks to trade within it.

Is the project fully open source?

The core repo, data, and surviving rules are open source at github.com/sstklen/trump-code with a live dashboard at trumpcode.washinmura.jp. The repo does contain a Stripe payment link suggesting some premium tier, but the analysis logic and data files are publicly available.

Trump Code: Someone Ran 31 Million Models on 7,400 Presidential Posts to Find Market Signals

Q: What is the most actionable finding?

Truth Social publishes posts 6.2 hours before they appear on X (38/39 posts verified). If a market-moving signal appears on Truth Social at 7 AM, it won't hit X until 1 PM. That's a 6-hour window. Monitoring Truth Social directly — not X — is the structural edge.

Q: What are the strongest individual signals?

Pre-market posts containing RELIEF-type language average +1.12% same-day S&P. Pure tariff days average -1.057% (Apr 3: -4.84%, Apr 4: -5.97%). Silence days (zero posts) are 80% bullish, averaging +0.409%. Late-night tariff tweets are 62% wrong — the model auto-inverts them to LONG.

Q: How does this relate to StockScout or algorithmic trading?

StockScout scores individual stocks based on fundamental signals. Trump Code is a macro regime filter — it signals whether the day is likely up or down for the whole market. Used together: Trump Code sets the macro regime, StockScout selects the stocks to trade within it.

Q: Is the project fully open source?

The core repo, data, and surviving rules are open source at github.com/sstklen/trump-code with a live dashboard at trumpcode.washinmura.jp. The repo does contain a Stripe payment link suggesting some premium tier, but the analysis logic and data files are publicly available.

By Prahlad Menon Published 2026-03-16 1 min read

One developer. 7,400 posts. 31.5 million model combinations.

The premise is simple: Trump is the only person alive who can move global markets with a single social media post. So someone decided to treat his posting behavior as a signal — not a sentiment, not a vibe — and run actual statistics on it.

The result is Trump Code: an open-source project that has found, verified, and published 551 trading rules that survived two-stage train/test validation. The hit rate across 566 verified predictions is 61.3% with a z-score of 5.39 (p < 0.00001).

That’s not noise.

What the Data Actually Found

The project’s findings table is worth reading carefully:

Finding	Evidence	Market Impact
Pre-market RELIEF language	Apr 9, 2025: S&P +9.52%	Avg +1.12% same-day
Pure tariff day	Apr 3: -4.84%, Apr 4: -5.97%	Avg -1.057%
Silence (zero posts)	566-day analysis	80% bullish, avg +0.409%
TARIFF→SHORT signal	Circuit breaker analysis	70% wrong → auto-inverted to LONG
Late-night tariff tweets	62% wrong	Model inverts to LONG
4-signal combo	12 occurrences, 66.7% up	Avg +2.792%

The TARIFF→SHORT inversion is the most interesting. The intuition is that when tariff posts hit after hours or late night, the market has already priced in the fear — by open, the move has exhausted itself and the trade is the other way.

This is the finding that matters most for anyone who wants to act on it.

Of 39 posts analyzed, 38 appeared on Truth Social an average of 6.2 hours before the same content appeared on X (formerly Twitter). If a signal fires at 7 AM on Truth Social, most traders relying on X won’t see it until 1 PM.

That’s not a statistical artifact — it’s a structural information advantage. The developer found 203 China-related signals that appeared on Truth Social and never appeared on X at all. Those get a 1.5x weight boost in the model.

The practical implication: if you’re monitoring presidential posts for market signals, monitoring X is insufficient. Truth Social is the primary source.

How It Works Under the Hood

The methodology is brute-force statistical search:

Data collection — 7,400+ Truth Social posts scraped from 3 independent sources and cross-verified
Feature extraction — 316 signal features extracted per post (keywords, timing, platform, tone, length, etc.)
Model search — 31.5 million combinations of features tested against S&P 500 daily returns
Survival filtering — two-stage train/test validation; only rules that hold up on out-of-sample data survive
551 surviving rules — published openly in surviving_rules.json
Daily updates — the system scrapes new posts every 5 minutes and runs predictions against the surviving ruleset

The architecture:

Trump posts on Truth Social
    ↓ (detected every 5 min)
Detect → Classify signals → Dual-platform boost
    → Event pattern check → Snapshot PM + S&P 500
    → Predict → Track at 1h/3h/6h → Verify

Results log in predictions_log.json — publicly auditable.

How to Use It With StockScout

Trump Code is a macro regime filter — it predicts whether today is broadly up or down for the market. StockScout scores individual stocks within the universe.

Used together:

Check Trump Code signal at 8 AM — is today bullish or bearish regime?
If bullish: run StockScout picks normally, full position sizing
If bearish or high-uncertainty: reduce position size or skip entirely
If pure tariff day signal fires: stay flat

This is exactly the kind of macro overlay the current StockScout v2 intel layer is trying to approximate with geopolitical signals. Trump Code does it empirically, from the actual data.

Caveats Worth Knowing

The 61.3% hit rate is on past data. The model was built on the same administration’s posts — it may have overfit to Trump’s 2025 communication patterns, which he reportedly changed in August 2025 (the project claims to have detected this shift and updated the model).

Hit rate ≠ profitability. 61.3% correct directional calls sounds good. But if the wrong 38.7% of calls are on days with large moves and the right 61.3% are on quiet days, the strategy loses money. The +2.792% average on 4-signal combos is more useful than the headline hit rate.

The repo has a Stripe payment link. The core project is open source, but there may be a premium tier. Everything needed to replicate the analysis is publicly available.

This is a one-person project. Impressive, but not institutionally validated. Treat it as a research artifact, not a production system.

Getting Started

git clone https://github.com/sstklen/trump-code

Key files:

data/surviving_rules.json — all 551 validated rules
data/predictions_log.json — full prediction history with outcomes
Live dashboard: trumpcode.washinmura.jp

The data updates daily. The rules update as new posts are processed. It’s the kind of project that gets more interesting the longer it runs.

GitHub: sstklen/trump-code · Related: StockScout V1 Backtest Results · MiroFish: Swarm Intelligence Prediction Engine