This paper presents a simple yet effective approach for volatility prediction using robust least squares regression, applied to the CBOE Volatility Index (VIX) from 2011 to 2022. The authors first decompose the logarithmic return time series into long-term trend, seasonal, and remainder components using the Kolmogorov-Zurbenko filter, which allows separate analysis of different scales of motion. Then, they apply robust least squares curve fitting with and without the least absolute residuals (LAR) method to derive generalized prediction equations for volatility and each component. The LAR method, which minimizes absolute residuals, is found to be more robust to outliers and yields significantly lower root mean square errors (RMSE) for training data: volatility (0.01366 vs. 0.07559), trend (0.01615 vs. 0.0745), seasonal (0.001123 vs. 7.054), and remainder (0.0001294 vs. 0.4691). For testing data, LAR also provides better predictions for all components, with RMSE values of 0.10087 (trend), 0.010343 (seasonal), and 0.014783 (remainder), while volatility testing RMSE is similar between methods. The study contributes generalized prediction equations that can be directly used by researchers for market volatility analysis, and highlights the importance of decomposing volatility into its constituent components for more accurate forecasting.