Machine learning and financial time series analysis
summary
Compares market vs transactional series, claims stationarity, seasonality, asynchronous timestamps and inter-channel correlation cause most failures, and shows concrete fixes: exponential moving average and tanh mapping for spikes, RNNs and CNNs for dynamics, clustering plus learned embeddings for grouped assets, and adding time-interval features for irregular events. Covers preprocessing choices (normalization vs moving averages), when long- vs short-term history matters, and tradeoffs rather than lofty promises — practical recipes and failure modes.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:44Financial time series
- 2:26Time series characteristics
- 6:30Problem input
- 8:32Feature engineering
- 11:33Traditional ML models
- 12:37Challenges with financial data
- 15:13Nonlinear mapping
- 21:14Interchannel cross correlation
- 23:31Summary
related talks
01h 26mApplying Statistical Modeling & Machine Learning to Perform Time-Series Forecasting
43mFeature Engineering for Time Series Forecasting
31mAdvanced machine learning time series....
29mMachine Learning: Better than Traditional Techniques?
25mTimeSeries Forecasting with ML Algorithms and there comparisons
29mA dive into time series for the energy sector
45mForecasting at Scale
30mMachine Learning: Beyond the Tutorial