Description: The article explores a closed-form forecasting method that combines Exponentially Weighted Moving Average (EWMA), weighted linear regression, and ridge regression (L2 regularization). The approach addresses two key challenges in time-series forecasting: adapting to nonstationarity and mitigating overfitting due to multicollinearity. EWMA assigns more weight to recent observations, while ridge regression penalizes large coefficients to stabilize estimates. A cross-validation strategy across rolling time windows selects optimal hyperparameters: the EWMA decay factor (α) and the ridge penalty (λ). The method is tested using repeated regressions of stock returns, showing that most hyperparameter combinations outperform ordinary least squares (OLS) under the (λ=1, α=0) setting.