I'm a programmer and a trader, and a passionate algo / ML geek.
I'm planning on making a toolset for analysis of markets. The thing is that there's a lot of tools that math introduces, statisticians use and environments like R and python get packages for that. Things that are interesting, resistant to outliers or more stable.
I'm personally a fan of such things, and I hesitated too long before showing the world what I really specialise in, and before I make my brand as a quant, algo/indicator developer.
If you don't mind, I'll be reminding the store and giving fresh updates from time to time on this r/ .
Today I'm happy to - and hopefully you will be happy too - to share a demo of VDM.
VMD Stock Analyzer · Streamlit
Choose you stock and and check if this gives you any additional insight.
Don't forget to share your feedback, please. Be kind, I'm not a seasoned creator. But don't sugar coat it, thanks.
** What is Variational Mode Decomposition? **
Variational Mode Decomposition - an overview | ScienceDirect Topics
Most traders know SMA, EMA, maybe even WMA or TMA. They’re just different flavors of moving averages — all smoothing price with varying lag. Useful, but still primitive.
VMD (Variational Mode Decomposition) is something else entirely. Instead of just averaging, it decomposes a signal (price series) into multiple “modes” with different frequencies. Think of it as separating noise from real market structure:
- short-term noise gets isolated
- mid-term cycles get revealed
- long-term trend emerges cleaner
That makes it interesting for trading because you’re no longer blindly smoothing - you’re adaptively filtering. You can choose which component of price action you want to follow, and ignore the rest.
It’s heavier math (variational calculus, optimization), but conceptually it’s a smarter filter than any MA. Potential use: trend detection, noise reduction, feature extraction for algos/ML.
Most people who stumble into signal decomposition for trading first meet EMD (Empirical Mode Decomposition). It works, but it’s messy:
- EMD is data-driven but unstable → tiny changes in input can give totally different outputs.
- It suffers from mode mixing (different frequencies bleed into the same component).
- It’s not well-suited for real-time use because it’s inconsistent.
VMD (Variational Mode Decomposition) was built to fix this. Instead of iteratively guessing modes like EMD, VMD formulates the problem as a variational optimization: it finds modes simultaneously by minimizing bandwidth in the frequency domain.
- No mode mixing, much cleaner separation.
- More robust to noise.
- Much more stable for online / real-time analysis.
More to come... frequently.