Model Performance

Residual Stacked GRU · Evaluation metrics & visualizations

Model Performance Analysis

Using cached fallback metrics

Metric	Open Returns	Close Returns	Open Price	Close Price
R²	-0.5826	-3.7990	0.9231	0.7927
MSE	0.0001	0.0004	41621.4514	110373.6803
RMSE	0.0120	0.0194	204.0134	332.2253
MAE	0.0097	0.0162	164.0673	276.3165
Direction Accuracy	0.568	0.517	—	—
Direction Accuracy (Large)	0.559	0.407	—	—

R² (Coefficient of Determination)

Proportion of variance explained. Higher is better (max 1.0).

MSE (Mean Squared Error)

Average squared error. Lower is better.

RMSE (Root Mean Squared Error)

Square root of MSE. Same units as output. Lower is better.

MAE (Mean Absolute Error)

Average absolute error. Lower is better.

Type: Residual Stacked GRU

Input: 40-day windows of log-returns

Output: Open & Close price log-returns

Loss Function: Huber Directional Loss (δ=0.05, direction_weight=0.25)

Data: 84 months of historical data with 85% train / 15% test split