Two generative-model pipelines for producing synthetic stock-market data when real data is scarce or has a low signal-to-noise ratio. A sector-based method classifies stocks by sector and cleans their signal using non-local total-variation smoothing, Fourier bandpass filtering, and denoising diffusion implicit models; a recursive method synthesizes variable-length sequences for newly listed stocks using pattern recognition, Markov models, and sub-time-level augmentation. Students learn how deep generative modelling and classical signal processing can be combined to enrich thin financial datasets and improve downstream predictive models.
The financial industry is increasingly seeking robust methods to address the challenges posed by data scarcity and low signal-to-noise ratios, which limit the application of deep learning techniques …