Research On Multivariate LSTM Price Prediction Based On Herding Effect In Cryptocurrency Marke

During the development of cryptocurrency market,asset prices fluctuate sharply,and there are many observable extreme fluctuations and abnormal trends.Under the influence of herding,investors are constantly pouring money into the market during the boom period,and the positive feedback function of the market has been strengthened to form a huge speculative bubble.During the market downturn,the rising trend of prices lost support,and the whole cryptocurrency market collapsed.Herding is an important branch of behavioral finance.It mainly studies the convergent investment anomalies that investors show in the process of investment and cannot be properly explained by traditional financial theory,so as to explain the abnormal trading behavior that cannot be explained by traditional financial theory and methods.Starting from the existence test of herding in cryptocurrency market,this paper uses CSAD model to study the overall market,cryptocurrency segments with different sizes and liquidity.Based on the results of existence test,according to the causes of herding and its impact on market efficiency,combined with Carhart pricing method,identify the authenticity of herding in each market segment.On this basis,this paper constructs the herding index of cryptocurrency market and applies it to cryptocurrency price prediction combined with LSTM network model to deal with the impact of extreme fluctuations and improve the accuracy of price prediction.The results show that herding has obvious asymmetry.In the overall market,there is only pseudo herding in the upward market during the market downturn,and there are both true and pseudo herding in the downward market during the market boom.There is no herd effect in the small market value cryptocurrency market,the pseudo herd effect is not obvious in the large market value cryptocurrency market,and there is real herd effect in the up and down markets in non-downturn periods.There is no herd effect in the cryptocurrency market with small trading volume.In the cryptocurrency market with large trading volume,there is pseudo herd effect in the up and down markets in the non-downturn period,and there is true herd effect in the up and down markets in the boom period.LSTM model can better fit the price trend of different cryptocurrencies.After the overall herding effect is included in the prediction index,it can most effectively improve the accuracy of cryptocurrency price prediction in the face of extreme price fluctuations,especially for extreme price fluctuations.