1. Start with the date
The page is a published snapshot, not a live quote feed. The source date tells you which completed close the latest forecasts use. Filters explore the published results; they do not download new prices or retrain the model.
Daily direction research
Stock direction
Will the next observed session close higher? Explore the model’s UP/DOWN forecasts, compare price trends, and see how often historical forecasts were correct.
Published snapshot · Yahoo Finance adjusted closes through Sep 11, 2026 · Imported Sep 12, 2026
Stocks covered
3 / 3
UP forecasts
3
Walk-forward accuracy
49.5%
Evaluated forecasts
753
UP probability
Above 50% becomes UP; 50% or below becomes DOWN. Probability does not predict the size of a price move.
Price performance
Each stock starts at 100. A reading of 120 means a 20% increase over this chart window.
- AAPL
- MSFT
- NVDA
Latest forecasts
Based on the latest completed close in this snapshot. No orders are placed.
| Stock | Signal date | Adjusted close (USD) | UP probability | Direction |
|---|---|---|---|---|
| AAPL | Sep 11, 2026 | $332.27 | 55.2% | UP |
| MSFT | Sep 11, 2026 | $495.63 | 51.4% | UP |
| NVDA | Sep 11, 2026 | $218.29 | 50.3% | UP |
Historical accuracy by month
Correct directions divided by evaluated forecasts. Always predicting UP scored 51.7% on the same 753 observations. Compare the model with this baseline, not just 50%.
- Accuracy
- Always UP
How to read this page
2. Read UP and DOWN correctly
UP means the model estimates more than a 50% chance that the next observed session closes above the signal close. DOWN covers a lower or unchanged close. A 55% UP probability is not a forecast of a 55% price increase.
3. Compare price paths
The price chart uses Yahoo adjusted closes, rebased to 100 so different share prices can be compared. It describes past price behavior. It is not a portfolio equity curve or the model’s trading return.
4. Check historical evidence
Walk-forward evaluation refits at the start of each target month using outcomes already known at the training cutoff. The model and always-UP baseline use the same observations. One stock on one target session counts as one evaluated forecast.
5. Understand the model
Pooled logistic regression combines ten features: 1-, 5- and 20-session returns; volatility; three moving-average ratios; 60-session momentum; weekday; and month. Each symbol contributes up to 504 closes before warmup and label filtering. Latest forecasts use a separate fresh fit.
6. Know the scope
This selected Yahoo watchlist does not reproduce QuantConnect’s historical top-20 universe. Adjusted history may be revised. Accuracy is not trading profitability: opening gaps, execution, fees, slippage and portfolio returns are not modeled here.
Source: Yahoo Finance via yfinance · Research and educational information · Back to Aizevin Stocks