Synthetic Data in Financial Time Series
summary
Generates a TimeGAN-style pipeline using RNNs to produce synthetic financial series and an autoencoder-based anomaly detector, trained on WTI and Brent crude from 1987–2020 sourced via frad and Alpha Vantage API (plus Yahoo finance and Bloomberg). Uses rolling windows with sequence length 24, minmax scaling, Adam optimizer and adversarial + reconstruction losses. Synthetic series are smoother, miss some downward volatility, anomaly threshold is 0.15, PCA and UMAP show decent overlap but reduced variability versus real data.
AI-written summary · about
chapters (7)
- 0:00Synthetic data in finance
- 3:14Why synthetic data matters
- 7:35Generative adversarial networks basics
- 11:28TimeGAN for time series
- 18:29Model training process
- 23:24Evaluating results
- 25:54Future work and limits
related talks
01h 01mModel validation using deep generation of stress data
45mSynthetic Data for Machine Learning Applications
20mSynthetic Data: The new Super Power for AI
24mMachine learning and financial time series analysis
30mRealtime Time Series Anomaly Detection in Production
25mThe Importance of Synthetic Data in Data-Centric AI
01h 00mThe Synthetic Data Advantage: How It Helps You to Get Both AI Privacy & Explainability
59mAI in Finance: Scope and Examples