Quant · Dec 2023
LSTM Based Financial Time Series Forecasting
A stacked LSTM trained on over ten years of daily equity price and volume data, evaluated out-of-sample on the 2022 to 2023 period against ARIMA, GARCH, and random walk baselines using walk-forward error decomposition.
10+ yrs daily
Training data
2022 to 2023
Held out
ARIMA, GARCH, RW
Baselines
Problem
Deep models for financial time series are frequently reported against weak baselines. I aimed to conduct a rigorous walk-forward comparison.
Approach
I trained a stacked LSTM on over ten years of daily price and volume data, tuned sequence length and dropout, and evaluated the model out-of-sample on the 2022 to 2023 period against ARIMA, GARCH, and random walk baselines.
Results
Walk-forward error decomposition identified the specific conditions under which the LSTM outperforms its baselines and where it does not, yielding insight into the underlying drivers of model risk rather than a single headline accuracy figure.
Stack