Index / 05 Published
Statistics Thesis
Can classical econometrics beat deep learning at forecasting Swedish small-caps?
Academic
At a glance
- Role
- Academic
- Models
- ARMA-GARCH vs LSTM
- Index
- OMXSSC small-cap
- Stack
- Python · R · Jupyter
Overview
A study comparing an econometric model, ARMA-GARCH, against a deep-learning model, LSTM, for forecasting the Swedish small-cap index OMXSSC, then testing each as an algorithmic trading strategy.
The headline result: ARMA-GARCH beat LSTM on both accuracy and returns, and only ARMA-GARCH produced a strategy that outperformed the index over the study period. A win for classical statistics over the hyped model.
I designed and ran the full study, implementing both models and the backtests. The result surprised me: the classical model beat the deep-learning one I had expected to win.