USER GUIDE
How to read a PMRC report.
Start with direction, then regime magnitude, confidence, supporting market context and finally historical model-quality metrics.
Up / Neutral / Down
Up is Small Up + Large Up. Down is Small Down + Large Down. These are model probability masses, not guarantees.
Five-State Distribution
Shows how the model divides 100% probability across Large Down, Small Down, Neutral, Small Up and Large Up.
Confidence & Entropy
Confidence describes how concentrated the distribution is. Entropy approaches 2.32 bits when all five outcomes are similarly likely; lower entropy means greater concentration.
20D Momentum
The stock's percentage price change over roughly 20 trading days. Positive means recent appreciation; negative means recent decline.
Annualized Volatility
Recent return variability scaled to an annual rate. Higher values indicate a more volatile market environment.
RSI 14
A 0–100 momentum measure. Values above 70 are conventionally called overbought and below 30 oversold, but PMRC does not treat these as automatic trading signals.
Beta vs SPY
Measures sensitivity to broad-market returns over the rolling window. Beta above 1 indicates greater historical sensitivity than SPY.
Correlation vs SPY
Ranges from -1 to +1 and measures recent linear co-movement with SPY. It does not imply causation.
Excess vs SPY
The stock's 20-day performance minus SPY's performance. Negative values indicate recent underperformance.
Skewness & Kurtosis
Skewness describes asymmetry in returns. Kurtosis describes tail/extreme-event behavior relative to a normal-like distribution.
5th / 95th Percentiles
Empirical tail markers from recent returns. They summarize historically weak and strong observations in the rolling window.
Accuracy / Macro F1 / Log Loss / Brier
These describe held-out historical model performance. Higher accuracy/F1 is better; lower log loss and Brier are better. They are not today's forecast probabilities.
Crispen Chachengwa
Machine Learning & FinTech Developer
MSc in FinTech · University of Central Florida
crischachengwa@gmail.com
Dr. Mesut Ozdag
Ph.D.: Computer Science, University of Central Florida
Assist. Prof.: Karabuk University
For Overleaf: mst.ozdag@gmail.com
PMRC is an experimental educational tool, not investment advice, a recommendation, solicitation, or guarantee of future performance. Predictions can be wrong.