12 Abstract
12.1 Predicting Stock Market with Bayesian Neural Network
The randomness of the stock market challenge investments to be reliable. Many approaches have been introduced to find the hidden pattern behind the transitions. However, error estimation with the non-parametric method is in the early stage. In this project, we used a Bayesian neural network to predict discrete time-series data with Taken’s embedding theorem. In this project, The Unadjusted Langevin Monte Carlo, Metropolis-adjusted Langevin algorithm (MALA) and Hamiltonian Monte Carlo (HMC) are applied to measure the posterior distribution. We found that Langevin method performed best for the log likelihood optimization. The purpose of this project is to provide a model-free approach with uncertainty quantification that is essential to the investment strategy