Project description
Forecasting cryptocurrency value by measuring sentiment
It is difficult to forecast the cryptocurrency market. Unlike stocks and foreign exchange rates, which are influenced by economic indicators, cryptocurrency prices are driven by sentiments. The EU-funded CryptoVolatility project will explore new ways to forecast cryptocurrency volatility. Specifically, it will build a machine that produces discrete sentiment phases each day using news articles and internet search data. Using artificial neural networks to measure sentiments, this tool is expected to assist regulators and investors learn more about cryptocurrency volatility. In the long-term, this new tool will also make it possible to design policies to help overcome financial crises.
Objective
Forecasting cryptocurrency volatility is a topic of interest in quantitative finance. A growing number of studies argue that compared to equty price cryptocurrency prices are to a large and perhaps abnormal degree driven by sentiments. However, econometric studies focus on forcing conditional volatility models developed for equity return volatility to fit on cryptocurrency data despite being aware that estimation techniques developed for analyzing equity price or commodity price volatility lack robustness and do not work as intended. Is it possible to propose solutions to deal with the mentioned shortcomings? Is it possible to suggest a new family of models? If so, how? The purpose of New, realistic and robust models for cryptocurrency volatility is to answer these questions by suggesting new and more realistic conditional volatility models accompanied with reliable cross-disciplinary estimation techniques to forecast cryptocurrency price volatility. What is novel and innovative about the suggested framework is that contrary to the current literature our point of departure is the empirical features observed in cryptocurrency prices combined with a useful tool, namely, artificial neural networks used to measure sentiments. Our aim is to build a machine that produces discrete sentiment phases each day using news articles and internet search data. Once we have identified the number of phases and determined, which phase an observation at a given time-period belongs to following neural network estimation, we can estimate the model parameters, jumps and filter out the continuous conditional volatility process contemporaneously using particle filtering techniques. Besides academics, this proposal is also relevant for regulators and investors as they can learn a great deal by understanding how cryptocurrency volatility actually behaves. Regulators can use sentiment labels from the neural network to design policies to contrast and overcome financial crises in the future.
Fields of science
Programme(s)
Funding Scheme
MSCA-IF - Marie Skłodowska-Curie Individual Fellowships (IF)Coordinator
33100 Tampere
Finland