The Relationship between bitcoin and traditional financial assets in the context of trading decisions: a DCC-GARCH approach with machine learning

Abstract

Today's economy requires fast, inexpensive, and reliable transactions. Bitcoin (BTC), the cryptocurrency constructed by Satoshi Nakamoto in 2008, is one of the innovative tools used for transactions and payments. BTC, one of the earliest cryptocurrencies discovered, is based on a decentralized network that allows private, anonymous user-to-user transactions anywhere. On the contrary, price prediction problems and investments are significant for both financial analysts and traders. Moreover, in the past few years, machine learning methods have found many applications in time series forecasting. Therefore, in this study, for the first time, a Dynamic Conditional Correlation - Generalized Autoregressive Conditional Heteroscedasticity artificial neural network method is proposed and applied to the investment decision of BTC. On the other hand, policymakers, central bankers, portfolio managers, investors, and traders need to find the correlations with precious metal, petroleum, and the American currency in BTC investment forecasts. However, much-existing literature examines the correlation between BTC and conventional practices. Based on economic theory, this study examines the asymmetric cointegration, asymmetric causality, and vibrant connection among gold, petroleum, and the American dollar. Considering these factors, the following questions are proposed in this article. (1) Is there an asymmetric cointegration relationship between BTC and traditional investment assets, i.e., Gold, crude Oil, and the U.S. dollar? (2) Is there an asymmetric causality between BTC and traditional investment assets, i.e., Gold, crude Oil, and the U.S. dollar? (3) Is there a dynamic correlation between BTC and traditional investment assets, i.e., Gold, crude Oil? (4) What economic theory underlies the BTC investment forecasts? (5) Is the ANN-DCC-GARCH model good enough for BTC investment decisions? To answer the research questions, this paper separately examines the correlation between BTC and the traditional investment asset ie. Gold, oil, US dollars. At the same time, we propose a new model for BTC investment decisions to analyze and answer the posed questions empirically. The empirical analysis involves three parts: in the first part, we selected the data of BTC, gold, crude oil, and dollar index from Yahoo Finance. Weekly data from January 1, 2015, to June 15, 2023, are analyzed, and the methodology uses asymmetric cointegration and asymmetric causality tests. The empirical conclusions show no cointegration in the traditional sense between BTC and traditional financial assets, but there is an asymmetric cointegration instead. There is a cointegrating correlation between the increase of BTC and the decrease in the U.S. dollar index and between the decline of BTC and the growth and reduction of all three financial assets. Crude oil is a Granger causality for BTC; while gold and US dollar are not. Before the epidemic, the fall of gold was the Granger causality for the rise of BTC. Following the COVID-19 pandemic, the drop in crude oil price was the Granger causality for the decline in BTC price. The COVID-19 outbreak altered the causal connection between BTC and traditional financial assets. However, the US dollar did not cause a shift in BTC price.

In the latter section, we opt for data on a weekly basis, covering the period from January 2014 to April 2022, and assess the fluctuating correlation between BTC and crude oil, or BTC and gold, using the DCC-GARCH model. The empirical results show that (1) BTC is riskier than gold and crude oil, while gold has the lowest risk. However, crude oil is more dangerous in the early stages of the COVID-19 epidemic. (2) Returns on BTC are negatively correlated with risk, while returns on gold and crude oil are not significantly correlated with risk. (3) The relationship between BTC and crude oil, as well as BTC and gold, demonstrates noteworthy volatility. We observe a notable rise in the positive correlation between BTC and crude oil during the initial stages of the COVID-19 pandemic. Conversely, the opposite relationship between BTC and gold became more noticeable at the beginning of the COVID-19 pandemic. In the third part of this paper, a DCC-GARCH artificial neural network method is proposed and applied to the investment decision of BTC, which provides historical information on the correlation and covariance of BTC with traditional financial assets. The information is sourced from the Wind database. The dataset comprises of daily observations, spanning from September 17, 2014, to December 23, 2022. The input variables encompass the highest daily values, low, and opening prices of BTC and binary variables for gold, the US dollar, and crude oil. 0 represents a price decline, and 1 represents a price increase. All input variables are lagged in one period. All variables are normalized using entropy weights except for the binary variables. The 2019 data are considered out-of-sample before the COVID-19 outbreak, and the 2022 data are out-of-sample after the COVID-19 outbreak. Each timeframe is segmented into a training subset and a forecasting subset correspondingly. The training subset is employed to establish the ANN-DCC-GARCH model that yields the most accurate prediction, and the prediction set is used to test the performance of BTC investment decisions. The practical findings demonstrate that the ANN-DCC-GARCH model has a cumulative return of 318% in 2019 and can reduce the loss by 50% in 2022. Therefore, historical information such as correlation, volatility, and covariance between BTC and traditional financial assets is indeed instructive for improving investment transactions in BTC. In addition, the overall findings suggest that the ANN-DCC-GARCH model works well for BTC investment decisions, but we need to figure out how well the model predicts other financial assets. Our future research can explore the application of the ANN-DCC-GARCH model in diversified financial asset portfolios and further analyze the model's predictive effect on investment transactions in other financial assets. As the market economy has advanced, the rise of digital currencies, notably BTC, has drawn considerable interest from investors, leading to increased scrutiny of its investment worth. For investors how to develop an investment strategy for BTC has become a concern for investors. For investment strategies, an important research goal of macroeconomics and econometrics is to test hypotheses and estimate the relationship between economic variables based on economic theory. However, for non-stationary time series, since the traditional tests are no longer valid, one either cannot analyze them at all or can draw completely wrong conclusions. In this paper, under the concept of combining econometric and machine learning models, we propose the ANN-DCC-GARCH model and apply it for the first time to the investment transaction decision of BTC. It is found that a small number of scholars only focus on the causal relationship between Bitcoin and US dollar or crude oil, and do not comprehensively determine the asymmetric causal relationship between BTC and traditional financial assets. Hence, it is imperative to examine the non-symmetric covariance and non-symmetric causal relations between BTC and conventional financial assets, specifically gold, crude oil, and the US dollar in this paper, as well as predicting BTC investment and trading strategies before and after the outbreak of COVID-19 respectively. The prediction results of BTC show that the ANN-DCC-GARCH model has good practicality and operability, and also verify that the ANN-DCC-GARCH model is completely superior to the ANN model. Many scholars use the DCC-GARCH model to analyze the dynamic correlation and volatility of traditional financial assets and give investment recommendations based on the conclusions of dynamic correlation and volatility. Therefore, our future research can explore the application of the ANN-DCC-GARCH model in diversified financial asset portfolios and further analyze the model's predictive effect on investment transactions in other financial assets.