Earlier market-model experiments
My early notebooks trying to predict market prices with machine learning.
- Status
- Research notebook collection
- Year
- 2024
- Role
- Solo
- Area
- Markets
In plain English
Like many people learning machine learning, I tried to predict prices. These notebooks trace that progression, from technical indicators to transformer and LSTM models. Their lasting value is what they taught me about testing fairly.
How it works
- Compute technical indicators such as RSI (a momentum measure) and check how they relate to prices.
- Build sequences of past values for the models to learn from.
- Train a transformer model in PyTorch on macro and price data.
- Train an LSTM (a model built for sequences) in Keras to predict Bitcoin prices.
What I did
I wrote these notebooks solo, using public market data and standard model libraries.
What I learned
Time-series models can leak future information in sneaky ways, and fixing that leak is the first thing I now check in any forecast.