Bridging the Gap Between Bitcoin Velocity and Speculation in the Price Discovery Phase of Crypto-Ass

Abstract

In this study, we come up with a novel algorithm for calculating speculation, and we use this value as a feature in our LSTM network to generate an approximator that takes advantage of this advanced domain knowledge. The goal of this paper is to use machine learning techniques to better understand how speculation may affect the price of the bitcoin cryptocurrency by testing predictors for average price. Speculation is defined as a belief in the future of the currency. Speculation can be thought of as a reasonably thoughtful human guess into the future of a cryptocurrency.This is a valuable contribution to the field, as it will bring knowledge to space on how to evaluate a metric for speculation, and how speculation may affect the pricing of a cryptocurrency in its price discovery phase. We implement a baseline ARIMA model that has an unnormalized test RMSE of $1031 while the optimal LSTM model has an unnormalized test RMSE of $151. The normalized train and test set error of the optimal LSTM model, 2% and 3% respectively, exceeding the oracle’s RMSE by 3.87% to 5%. By manipulating the feature set on which the model was trained we gained insight into the relationship between speculation and velocity, specifically that 4velocity ! 4speculation, but not necessarily the converse. This insight mainly helps us to understand the new digital asset behavior throughout the current price discovery phase and thus, pre-existing crypto-asset valuation models still remain valid for when the asset has matured out of this phase. Ultimately, we propose a model for crypto-asset valuation during the price discovery stage.

Read more here: