Machine Learning: Better than Traditional Techniques?
summary
Compares ARIMA and machine learning on daily S&P500 (1950s–2018) and FTSE100 (2001–2018) data, fitting an ARIMA(2,1,0) baseline and several ML models including a LSTM recurrent neural network and an SVM that shows a negative R² for SVM. Uses RMSE/MAE/MAPE and Diebold–Mariano tests. Finds LSTM vastly outperforms ARIMA on predictive accuracy, ARIMA is fast and transparent (~1s) while RNN tuning can take up to an hour, and ML gives actionable gains without prohibitive run time.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:56Background
- 6:43Machine Learning in Finance
- 8:18Research Questions
- 9:16Implementation - The Data
- 10:13Implementation - Python
- 12:10Implementation - The Models
- 14:14Comparing Predictions
- 16:35ARIMA - Pros and cons
- 18:26Machine Learning - Pros and cons
- 20:43Empirical Results
- 21:58Conclusions and Further Research
related talks
39mForecasting With Classical and Machine Learning Methods
25mTimeSeries Forecasting with ML Algorithms and there comparisons
24mMachine learning and financial time series analysis
31mAdvanced machine learning time series....
43mFeature Engineering for Time Series Forecasting
01h 26mApplying Statistical Modeling & Machine Learning to Perform Time-Series Forecasting
31mML time series forecasting methods
21m2D ARIMA, Capturing New Trends for Distant Time Horizons