*5.3. Comparative Analysis*

In this section, we compare the performance of the VAR and BVAR models with some of the well-known autoregression and Bayesian regression algorithms, including the autoregression integrated moving average (ARIMA) (Chu et al. 2017; Hencic and Gouriéroux 2017) and Bayesian regression (BR) (Shah and Zhang 2014). ARIMA is a commonly used model to predict the price, and the model is a combination of three basic time-series models: autoregressive, moving average, and autoregressive moving average. Bayesian regression uses statistical analysis within the context of Bayesian inference rules. The comparison was made based on the values of the root mean squared Error (RMSE), the mean absolute error (MAE), and the mean absolute percentage error (MAPE) (Tan and Kashef 2019; Tobin and Kashef 2020). In this section, we focus on the data timeframe from Experiment B [January 2011–August 2020] and the variable of interest MKPRU (the equilibrium closing price of the BTC market as denominated by the US dollar). As shown in Tables 3–5, for the Full timeframe, the VAR model had the best performance. For the Post-boom timeframe, both the VAR and the BVAR models had the lowest RMSE, MAPE, and MAE values. Finally, for the Year of 2020, the VAR and the BVAR models had better performance than the ARIMA and BR models.


**Table 3.** Accuracy of forecasting models: Full Timeframe.

**Table 4.** Accuracy of forecasting models: Post-boom timeframe.


**Table 5.** Accuracy of forecasting models: Year of 2020 timeframe.


### **6. Conclusions and Future Directions**

In this paper, two VAR models were developed to analyze and understand the mechanics of the BTC market. The developed models were tested in predicting the endogenous variables using selected features of exogenous variables. The two models were compared with the state-of-the-art forecasting models in order to show their efficiency. This research presents a powerful way to predict Bitcoin market price and an interesting look at what factors of this BTC network can shape new innovations in blockchain and the future of digital currency. As a new currency not administered by the government, there are many interesting behaviors that can be studied. From the perspective of miners, investors, or users of BTC, these findings may be useful for understanding the movements of the price of the BTC, and could help to understand what influence each of the exogenous factors has on the price of BTC. Future experiments for BTC prices will use a non-linear or dynamic VAR, which is suitable for BTC simulation. Dynamic VAR accounts for the change in a relationship by allowing the coefficients to change over time, which makes it much more challenging to analyze. The technical indicator could be extended as an exponential moving average or volume-weighted average price. Different priors can be suggested for future directions, such as the independent normal-Wishart. Additionally, analyzing the daily market returns in order to understand the distribution of daily behavior could provide insight into the classification of upward and downward trends. Incorporating the classification would enable research to understand price action in more depth with increasingly sophisticated machine-learning or nonlinear models. Finally, further investigation combing machine-learning prediction models is recommended.

**Author Contributions:** Software, A.I., E.V.; Supervision, R.K.; Visualization, A.I., E.V., M.L., and E.H.; Writing—original draft, A.I.; Writing—review & editing, A.I. and R.K.; Validation, A.I. and M.L.; Source—A.I., M.L., E.V., and E.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.
