TCNGAN: Generative Adversarial Network for Financial Time-Series

Goal:

TCNGAN are developed by adding TCN block to vanilla GAN for finanical time-series data generation to support strategy backtesting and modification.

Discussion:

First, the model can be improved to incorporate prior information to better estimate the tail. Second, the baseline models should include traditional financial statistical model for thorough evalualtion.

Result:

TCNGAN Generated Cumulative Log Return

TCNGAN Generated Cumulative Log Return V.S. Historical Cumulative Log Return