cryptocurrency return probabilistic forecasting using blended deep learning model

cryptocurrency return probabilistic forecasting using blended deep learning model

after the global financial crisis in 2008 and with the genesis of bitcoin, which is a blockchain-supported peer-to-peer payment system, the cryptocurrency market has quickly direct vulgarity. consequently, the volatility of cryptocurrencies' value influences real consideration from both investors and researchers. it is a insubordinate study to forecast the return of cryptocurrencies due to the non-stationary value and the stochastic consequences in the market. furthermore, calculating the uncertainty in the predictions of a cryptocurrency is crucial for financial investment.

not only do we want our models to make precise predictions, but we also tend to have an accurate estimation of uncertainty along with each forecast. probabilistic forecasting, which is the approach where our model outputs a full probability distribution over the long-term and short-term timeline, is calculated via three deep learning models and finally, blended and optimized with a novel linear programming approach to obtain the optimized mean probabilistic forecasted return. therefore, our approach can be utilized in both portfolio construction and cryptocurrency trading. finally, we are using arima as a benchmark and comparing the forecast results with the benchmark. our measurement shows that the developed blended model outperforms the benchmark in most of the applied criteria.