Fig 1.
The time series (shown in red circles) is used as input for the neural network which predicts 5 steps-ahead in time (shown by black circles).
A sliding window approach is used to reconstruct the dataset in this way using Taken’s theorem.
Fig 2.
Bayesian neural network and MCMC sampling.
Note that the posterior distribution is shown that represents weights in Panel (a).
Fig 3.
An overview of the different replicas that are executed on a parallel computing architecture.
Table 1.
Time span of data considered for each stock-price pre-COVID-19.
Fig 4.
Prediction performance (mean of normalized RMSE) for the three models (Bayes-FNN, FNN-Adam, and FNN-SGD).
Note that the predictions were normalized in range of [0, 1].
Table 2.
Multi-step-ahead prediction (RMSE).
Fig 5.
Prediction and uncertainty (shaded region) for test data of stock MMM.
Fig 6.
Prediction and uncertainty (shaded region) for test data of stock 600118.SS.
Fig 7.
Prediction and uncertainty (shaded region) for test data of stock CBA.AX.
Fig 8.
Prediction and uncertainty (shaded region) for test data of stock DAI.DE.
Table 3.
Timespan considered for respective stocks during COVID-19.
Fig 9.
Prediction performance (mean of normalized posterior RMSE) of respective stocks during COVID-19 using Bayes-FNN (Setup 1 vs Setup 2).
Note that the predictions were normalized in range of [0, 1].
Table 4.
Multi-step-ahead prediction (RMSE) during COVID-19.
Fig 10.
Prediction and uncertainty over test data of MMM (data setup 1 vs data setup 2).
Fig 11.
Prediction and uncertainty over test data of 600118.SS (data setup 1 vs data setup 2).
Fig 12.
Prediction and uncertainty over test data of CBA.AX (data setup 1 vs data setup 2).
Fig 13.
Prediction and uncertainty over test data of DAI.DE (data setup 1 vs data setup 2).
Fig 14.
Stock price time series and monthly volatility for stocks MMM and 600118.SS.
Fig 15.
Stock price time series and monthly volatility for stock CBA.AX and DAI.DE.