Bitcoin Price Predictive Modeling Using Expert Correction

TL;DR

This paper proposes a hybrid Bitcoin price prediction model combining linear regression features with expert correction, enhanced by Bayesian methods to handle outliers and improve accuracy.

Contribution

It introduces a novel approach integrating regression models with expert correction and Bayesian techniques for better Bitcoin price forecasting.

Findings

The combined model outperforms standalone regression or expert methods.

Bayesian approach effectively manages outliers in Bitcoin price data.

Pattern deviations are simpler and more predictable than raw price series.

Abstract

The paper studies the linear model for Bitcoin price which includes regression features based on Bitcoin currency statistics, mining processes, Google search trends, Wikipedia pages visits. The pattern of deviation of regression model prediction from real prices is simpler comparing to price time series. It is assumed that this pattern can be predicted by an experienced expert. In such a way, using the combination of the regression model and expert correction, one can receive better results than with either regression model or expert opinion only. It is shown that Bayesian approach makes it possible to utilize the probabilistic approach using distributions with fat tails and take into account the outliers in Bitcoin price time series.