Proyección del precio de criptomonedas basado en tweets empleando LSTM

The modeling and prediction of time series is an arduous and essential task for financial optimization procedures. Numerous studies have been carried out to reduce investor uncertainty, by forecasting the price of currencies and shares. However, the emergence of a new type of coins with their own characteristics, known as cryptocurrencies, present additional challenges. In this sense, the paper seeks to understand up to what extent comments in social networks can capture the collective expectations of investors, and affect the future value of the currency. The objective is to predict the daily performance of a market based on two components: those that define the behavior of the cryptocurrency itself (volume, opening value, closing value, maximum value and minimum value) and the expectations and interactions of the environment, through the collected tweets. For this, the use of a type of recurrent neural network known as “Long Short Term Memory” (LSTM) is proposed. The methodology used for the preprocessing of the data and the application of this time series forecasting technique allows obtaining a prediction with a Mean Absolute Percent of 34.92%; This indicates that the representation of the perception variable in social networks has not been that relevant and, therefore, motivates new works for better representation using other NLP techniques.