Advanced Topics in Autoencoders and Autoencoder Market Models (AEMM): 13.30 – 17.30

Session One: Machine Learning Architecture (VAE, VEGD) – 13:30 to 15:00

During this session we will train autoencoders to optimally represent the yield curve using one, two, or three model state variables, and compare our results to implicit and explicit factor representations used by popular classical models.

• Introduction to Variational Autoencoders (VAE)
  • The roles of encoder and decoder, latent space
  • Deliberately introducing uncertainty in reconstruction
  • Loss function and optimization loop
  • Conditional VAE (CVAE) vs. unconditional VAE
  • Reconstruction with VAE
  • Generation with VAE
• VAE for the yield curve
  • Training to single currency dataset
  • Training to multi-currency dataset
  • Few-shot learning with multi-currency dataset
• Hands-on examples with Python:
  • Swap curve VAE trained to single currency dataset
  • Swap curve VAE trained multi-currency dataset
  • Few-shot VAE for currencies with short time series

Coffee Break – 15:00 to 15:30

Session Two: Application to Interest Rate Models – 15:30 to 17:00

For five popular model families (three in Q-measure and two in P-measure), we will review a representative classical model in each family and then build its machine learning counterpart.

• Introduction to Autoencoder Market Models (AEMM)
  • Dimension Reduction as Compression
  • Combining VAE with stochastic process in latent space
  • Deterministic and stochastic volatility AEMM
• Models in Q-Measure
  • One Factor Short Rate Models
    • One Factor Hull-White Model (classical)
    • One Factor Short Rate AEMM (ML)
  • Two Factor Short Rate Models
    • Two Factor Hull-White Model (classical)
    • Two Factor Short Rate AEMM (ML)
  • Forward Rate Models
    • HJM, LMM and SABR-LMM Models (classical)
    • AFNS and FHJM Models (classical)
    • Forward Rate AEMM (ML)
• Models in P-Measure
  • Forecasting Models
    • Dynamic Nelson-Siegel Model (classical)
    • Forecasting AEMM (ML)
  • Stochastic Models
    • CKLS Model (classical)
    • Stochastic AEMM (ML)
• Hands-on examples with Python
  • Compare curve shapes generated by classical Q-measure models and AEMM
  • Compare interest rate forecasting by classical P-measure models and AEMM

Q&A – 17:00 to 17:30