Chain-structured neural architecture search for financial time series forecasting
This work addresses the lack of NAS studies for time series data, specifically for financial forecasting, but it is incremental as it applies existing NAS methods to a new domain without major methodological innovations.
The paper tackled the problem of applying neural architecture search (NAS) to financial time series forecasting, comparing three NAS strategies on chain-structured search spaces, and found that Bayesian optimization and hyperband performed best, with RNN and 1D CNN as top architectures, though results showed high variance due to dataset challenges.
Neural architecture search (NAS) emerged as a way to automatically optimize neural networks for a specific task and dataset. Despite an abundance of research on NAS for images and natural language applications, similar studies for time series data are lacking. Among NAS search spaces, chain-structured are the simplest and most applicable to small datasets like time series. We compare three popular NAS strategies on chain-structured search spaces: Bayesian optimization (specifically Tree-structured Parzen Estimator), the hyperband method, and reinforcement learning in the context of financial time series forecasting. These strategies were employed to optimize simple well-understood neural architectures like the MLP, 1D CNN, and RNN, with more complex temporal fusion transformers (TFT) and their own optimizers included for comparison. We find Bayesian optimization and the hyperband method performing best among the strategies, and RNN and 1D CNN best among the architectures, but all methods were very close to each other with a high variance due to the difficulty of working with financial datasets. We discuss our approach to overcome the variance and provide implementation recommendations for future users and researchers.