1. What We Predict
When a NASDAQ stock hits a Limit Up-Limit Down (LULD) trading halt, HaltPredict's AI model predicts the direction of the reopen: will the stock reopen above or below its halt price?
The model outputs a score from 0 to 100. Scores above 50 indicate a predicted UP reopen; below 50 indicates DOWN. The further from 50, the higher the model's confidence.
2. Dataset
54,000+
Total LULD halts collected
37,533
Halts with reopen prices (scored)
7 years
Date range (2019–2026)
Data is collected from the NASDAQ trading halt RSS feed (live) and historical LULD halt records. Price data comes from the Polygon.io API. Each halt record includes: symbol, halt time, halt price, halt number (for cascade sequences), pre-halt momentum, reopen price, and reopen gap percentage.
3. Model Architecture (V6.2)
The model is a rule-based scoring system with three weighted components, calibrated on historical halt outcomes:
Stock Fundamentals
25 points
Market cap, sector, historical halt frequency, and price level. Small-cap stocks with frequent halts behave differently than large-cap first-time halts.
Recent Activity
25 points
Pre-halt momentum direction and magnitude, recent price trend, and warm-start recency (how recently the stock was last halted).
Halt Session Data
50 points
Halt number in the session (1st, 2nd, 5th+), cascade probability, time of day, and halt-specific momentum signals. This is the most predictive component.
V6.2 Enhancements
- Cascade threshold raised to 50% (from 42%) based on validated data
- Warm-start modifier: stocks halted in the last 1–3 days get a small accuracy penalty (−2.5pp)
- Halt-number confidence: H1 predictions slightly discounted (−2pp), H3+ boosted (+2pp)
- Momentum clamping: H1 pre-halt momentum capped at ±20% to handle outlier data
- DNA stability signal removed (not validated on out-of-sample data)
4. Validated Performance
Historical Accuracy
55.9%
37,533 halts · 2019–2026
Live Accuracy
54.4%
612 live halts · Jan–Feb 2026
AVOID Precision
63.4%
When model says “avoid”, it's right 63.4%
Random Baseline
50.0%
Coin flip comparison
By Halt Number
Accuracy varies significantly by halt number in the sequence:
| Halt # | Accuracy | Sample Size | Notes |
|---|
| H1 (first halt) | 57.6% | 10,285 | Best single-halt accuracy |
| H2 (second halt) | 53.8% | 8,412 | Cascade signal weakens |
| H3+ | 55.2% | 18,836 | Deep cascades recover accuracy |
| H5+ (exhaustion) | 58.1% | 6,203 | Exhaustion reversal signal is strong |
5. Limitations
The model is not a crystal ball. A 55.9% accuracy means it gets the direction wrong 44.1% of the time. Here's what the model cannot do:
- Cannot predict magnitude — it predicts direction (UP/DOWN), not how far the stock will move
- Cannot predict timing — halts may reopen in 5 minutes or 60+ minutes; the model does not know when
- Cannot account for news — FDA decisions, earnings, SEC actions, and breaking news are not inputs to the model
- Cannot guarantee future accuracy — market conditions change; past performance does not guarantee future results
- Accuracy varies by market regime — the model performs differently in high-volatility vs low-volatility environments
6. Cascade Detection
Beyond direction prediction, the platform monitors for cascade patterns — when a first halt leads to subsequent halts on the same stock. Key findings:
- 42% of LULD halts are followed by at least one more halt on the same stock the same day
- H1 gaps above +20% or below −10% predict a second halt 60–66% of the time
- H5+ halts show strong exhaustion-reversal signals
- Average cascade sequence length is 3.2 halts when cascading occurs
7. Data Sources
- NASDAQ Trading Halts RSS Feed — real-time halt notifications, polled every 15 seconds
- Polygon.io — minute-level stock price data for halt prices, reopen prices, and pre-halt momentum
- Historical LULD Records — 7 years of halt data from public NASDAQ records
8. Disclaimer
HaltPredict is an informational and entertainment platform. Nothing on this site constitutes financial advice.
AI predictions, community predictions, accuracy statistics, cascade alerts, and all other data are provided for informational purposes only. Past performance does not guarantee future results. Trading halted securities carries significant risk including total loss of capital.
Always do your own research and consult a qualified financial advisor before making investment decisions.