Member-only story
Beyond RMSE: Using STL Decomposition to Quantify Directional Predictability in Time Series
Season-Trend with Loess is a great tool for Time Series Analysis.
Introduction
It can be frustrating seeing an impressive Root Mean Squared Error (RMSE) on your time series forecast after spending precious time modeling it, and still get bad forecast. Standard error metrics like RMSE or Mean Absolute Error (MAE) are good for measuring magnitude, but they can also miss the mark on real-world utility.
In the world of finance, inventory management, or macroeconomic analysis, knowing the sign (the direction of movement up or down) is often worth more than an exact magnitude.
Think about it: a forecast of $101.00 when the actual price is $100.90 is technically a “miss” by RMSE. However, you got the direction right! You knew it was going up. That directional win is a massive edge.
So, how do we look past the size of the error and get to the heart of directional confidence?
The answer lies in a classic technique: STL Decomposition.
STL Decomposition
STL stands for Seasonal-Trend decomposition using Loess. It’s an indispensable tool that breaks any time series…
