Blockchain-Federated Learning using advanced Bayesian LSTM-FAN for Real Option Valuation prediction

Real Option Valuation (ROV) forecasting is an estimate of the value of flexible investment opportunities in the presence of uncertainty treating them as financial options. Unlike conventional valuation techniques, ROV considers the acknowledgment of strategic choices that can be decided by managers, like waiting, growing, or shelving a project depending on evolving market conditions. Forecasting methodologies tend to employ simulations and complex models, such as Deep Learning (DL), to predict uncertain variables affecting option value. This strategy delivers a more realistic and flexible evaluation, enabling organizations to make more informed decisions in the context of uncertainty. This work aims to design a hybrid DL model for ROV prediction. The key objective is to design a blockchain-Federated Learning (FL) model integrated with DL for ROV prediction. FL constructs a global model by aggregating locally trained models on the blockchain so that the collaboratively trained global model can give improved and more accurate predictions. The local model learns with the proposed Bayesian Long Short-Term Network-Fourier Analysis Networks (BayesianLSTM-FAN), a fusion of Bayesian Long Short-Term Network (BayesianLSTM) and Fourier Analysis Networks (FAN). The data is preprocessed first by data cleaning and scaling (StandardScaler) and then by statistical feature extraction (Rolling window), and prediction is done using the proposed BayesianLSTM-FAN. Lastly, the model also enables simultaneous training of a global model with the use of blockchain technology as well as FL. Experimental results show that the proposed approach obtained an MSE value of 0.001, an MAE of 0.024, and an R-squared value of 0.965.