چکیده:
Finding the best way to optimize the portfolio is one of the concerns of activists in the investment management industry. In recent years, the introduction of economic and mathematical models in the prediction of Bitcoin has helped many investors to optimize portfolios. Therefore, in this study, we introduce models of GARCH family composition and recurrent and convolutional neural network to predict the daily yield of Bitcoin will be paid during the period of 1398-1392. In this study, the Bitcoin is examined using GARCH and EGARCH short-term memory models. Of the two variables, the price of crude oil and the Gold as factors that their shocks and fluctuations have a major impact on Bitcoin are used as control variables. In addition to using long-term memory models, considering the better performance of combined models (compared to individual models) In anticipation In this study, all models of the GARCH family (both short and long run) with the recurrent and convolutional neural network were combined and using the combined models, the efficiency of the Bitcoin for the next 10 days were predicted step by step and its accuracy Based on the evaluation criteria.
خلاصه ماشینی:
Therefore, the innovations of the aforementioned article are as follows: - Developing the research literature in the field of new financial markets (cryptocurrencies such as Bitcoin) in domestic studies - Developing long-memory models FIGARCH and FIEGARCH - Presenting a hybrid model of the GARCH family - convolutional and recurrent hybrid neural network - Including gold price and oil price as control variables in the hybrid model Research Hypotheses Since the questions of this article are not survey-based, the answers to the questions have been obtained during the course of the research.
Kristjan Puller and Hernandez 18 (2017), using daily data of three metals: gold, copper, and aluminum during the period of September 7, 2009, to May 20, 2014, and using control variables such as the dollar, yuan, and euro exchange rates, oil prices, and stock indices of China, India, and America, addressed the prediction of the index volatility of these three metals with the GARCH model and then applied the model as an input to a neural network.
GARCH family models To provide a suitable depiction of conditional mean volatility and conditional variance models, the returns of time series are shown as r, conditional on the information t-1, as follows: (Refer to the page image) The above equation represents the ARIMA conditional moving average model.
The obtained results indicate that hybrid models of the GARCH family with long-term memory and neural networks, specifically the RNN-CNN-FIEGARCH-t, are more efficient and have lower forecasting errors in predicting Bitcoin return volatility.
(
, 
, 
)
