Presentation Title

Modelling and Forecasting Bitcoin’s Valuation Using Common Risk Factors

Faculty Mentor

Patrick Convery, Randall R. Rojas

Start Date

17-11-2018 3:00 PM

End Date

17-11-2018 5:00 PM

Location

CREVELING 95

Session

POSTER 3

Type of Presentation

Poster

Subject Area

business_economics_public_administration

Abstract

Bitcoin is an emerging alternative currency that has gained tremendous attention from the finance, business and technology industry. Recent applications of this new currency, such as merchant adaptations and initial coin offerings, have fortified Bitcoin’s integration in the economy. Due to Bitcoin’s extreme volatility in price in the past six months, it is increasingly important to understand the driving factors behind its valuation. Though previous papers disagreed on the fundamental value of Bitcoin and cryptocurrencies alike, most of them were published before Bitcoin’s incredible surge in price in September 2017. Using cross-sectional empirical data, we regress Bitcoin price against three key components of value: the difficulty of mining, the rate of performance of the network, and the cost per transaction. We also introduce a novel sentiment factor derived from Google Trend data to capture the speculative aspects of Bitcoin trading and determine the causality relationship between sentiment and price. Machine learning methods such as random forests are also used to determine the most important factors in predicting Bitcoin’s price. We then combine conventional time series analysis techniques with sentiment components to successfully forecast Bitcoin prices and returns. Finally, we compare and contrast the behavior found by our model to data prior to September 2017, and the behavior estimated on data after September 2017. Analysis shows that technological, economic and sentiment factors account for 84.53 percent of variations in price. Our study reveals potential economic factors, in addition to a speculative component, that drive Bitcoin’s valuation formation.

Summary of research results to be presented

Regression analysis shows that Hash rate, mining difficulty, and cost per transaction accounts for 78.69% of the variations in Bitcoin price. Both hash rate and cost per transaction are significant (p<0.01), but not difficulty of mining. With the addition of Google trend variable (p<0.01), the R-squared is improved to 84.53%. However, there may exist a collinear relationship between hash rate and mining difficulty. Hash rate has negative effect on price, while cost per transaction and Google Trend has positive effect on price.

Random forest model shows that difficulty is the most important factor in predicting price, followed by popularity, cost per transaction and hash rate. Our model accuracy is 87.7%. Random Forests yields a better accuracy rate compared to our linear regression model (RMSE for linear regression is 1369.504 and for Random Forest is 898.0063).

Using the Granger Causality test we find that before the September 2017 price surge, price granger causes changes in google trend searches but this causality is flipped after the price surge, where google trends granger caused change in price. This further suggests that sentiment drove valuation during the price surge.

Our time series analysis helps examine the differences before and after the price surge. Returns of Bitcoin post the price surge follow a ARIMA(4,1,0) process and we evaluate the model performance using the rolling scheme. Returns prior to the price surge is white noise.

This document is currently not available here.

Share

COinS
 
Nov 17th, 3:00 PM Nov 17th, 5:00 PM

Modelling and Forecasting Bitcoin’s Valuation Using Common Risk Factors

CREVELING 95

Bitcoin is an emerging alternative currency that has gained tremendous attention from the finance, business and technology industry. Recent applications of this new currency, such as merchant adaptations and initial coin offerings, have fortified Bitcoin’s integration in the economy. Due to Bitcoin’s extreme volatility in price in the past six months, it is increasingly important to understand the driving factors behind its valuation. Though previous papers disagreed on the fundamental value of Bitcoin and cryptocurrencies alike, most of them were published before Bitcoin’s incredible surge in price in September 2017. Using cross-sectional empirical data, we regress Bitcoin price against three key components of value: the difficulty of mining, the rate of performance of the network, and the cost per transaction. We also introduce a novel sentiment factor derived from Google Trend data to capture the speculative aspects of Bitcoin trading and determine the causality relationship between sentiment and price. Machine learning methods such as random forests are also used to determine the most important factors in predicting Bitcoin’s price. We then combine conventional time series analysis techniques with sentiment components to successfully forecast Bitcoin prices and returns. Finally, we compare and contrast the behavior found by our model to data prior to September 2017, and the behavior estimated on data after September 2017. Analysis shows that technological, economic and sentiment factors account for 84.53 percent of variations in price. Our study reveals potential economic factors, in addition to a speculative component, that drive Bitcoin’s valuation formation.