Risk Determination for Digital Currency Portfolio Optimization Based on Gaussian Mixture Clustering and Barra Multi-factor Stock Selection Model

In this paper, we propose a method to determine the portfolio risk of digital currencies on blockchain based on Gaussian mixture clustering and Barra multi-factor model. 36 digital currencies are randomly selected from 8 aspects, and the prediction data of currency prices are obtained by long and short-term memory recurrent neural network for a uniform time period of 84 days, and the goodness of fit is above 99%, which not only preserves the nature of each currency itself but also facilitates data analysis, The Garch model is used to calculate the volatility of currency prices and volume for the same time period, and the systematic risk factor is calculated by combining with the capital asset pricing model, and the Barra multi-factor model is used to construct the influencing factors on the basis of which the validity and linear correlation tests are performed to ensure that the most influential factors are selected. Gaussian mixture clustering is used to assemble the portfolio of currencies with the closest price and volume volatility and expected return. A multi-factor model was used to predict the expected return of each currency in a single portfolio over the same time period, and the portfolio performance analysis and value-at-risk theory were combined to further analyze the strengths and weaknesses of the portfolio. The results show that the model prediction accuracy is 90.0% under the influence of the assumptions of style factors (profitability Rit, price daily volatility VL+σn2, systematic risk factor β and statistical factors such as size ait, momentum RSTRit, liquidity Lit).