Abstract
This study proposed an optimal model to examine the relationship between the Bitcoin price and six macroeconomic variables – the Bitcoin price, Standard and Poor's 500 volatility index, US treasury 10-year yield, US consumer price index, gold price and dollar index. It also examined the effectiveness of the vector error correction model (VECM) in analyzing the interrelationship among these variables. The authors employed the following approach: first, the authors sampled the period August 2010–February 2022. This is because Bitcoin achieved a market capitalization of more than US$1 tn over this period, gaining market attention and acceptance from retail, corporate and institutional investors. Second, the authors employed a VECM with the six macroeconomic variables. Finally, the authors expanded the long-run equilibrium relationship (time-invariant cointegration)-based VECM to develop a time-varying cointegration (TVC) VECM. The authors estimated the TVC VECM using the Chebyshev polynomial specification based on various information criteria. The results showed that the Bitcoin price can be modeled with the VECM (p = 1, r = 1). The TVC approach generated more explanatory power for Bitcoin pricing, indicating the effectiveness of the approach for modeling the long-run relationship between Bitcoin price and macroeconomic variables.
Keywords
Citation
Lee, Y. and Rhee, J.H. (2022), "A VECM analysis of Bitcoin price using time-varying cointegration approach", Journal of Derivatives and Quantitative Studies: 선물연구, Vol. 30 No. 3, pp. 197-218. https://doi.org/10.1108/JDQS-01-2022-0001
Publisher
:Emerald Publishing Limited
Copyright © 2022, Yong Lee and Joon Hee Rhee
License
Published in Journal of Derivatives and Quantitative Studies: 선물연구. Published by Emerald Publishing Limited. This article is published under the Creative Commons Attribution (CC BY 4.0) licence. Anyone may reproduce, distribute, translate and create derivative works of this article (for both commercial and non-commercial purposes), subject to full attribution to the original publication and authors. The full terms of this licence maybe seen at http://creativecommons.org/licences/by/4.0/legalcode.
1. Introduction
Bitcoin exceeded a market capitalization of US$1 tn in February 2021, thereby attracting the interests of institutional investors and listed corporations in the cryptocurrency market (Coinmarketcap https://coinmarketcap.com/). Its market capitalization can be compared to the market capitalization rate of gold, silver, Apple and Samsung Electronics at 6.7, 58.6, 29.7, and 208%, respectively, as of March 2022 (8marketcap https://www.8marketcap.com/). In addition to this appreciation in the Bitcoin's market capitalization since 2010, there has been growing debate on Bitcoin's price, which is fundamental to the analysis of Bitcoin's price model.
In this context, while some studies use the security market analysis theory, such as the capital asset pricing model (CAPM) and the Fama–French multi-factor model, to predict the Bitcoin Price, some others use historical time series-based models, such as the univariate autoregressive integrated moving average (ARIMA) model and the multivariate vector autoregressive (VAR) models to explain Bitcoin's price dynamics.
From the perspective of the Bitcoin price metric, there has been a rise in the Bitcoin price since the mining of the genesis block in early 2009. Specifically, the price increased from $1 to $10 to $100 to below $1000 by April 2011, August 2012, April 2013, and the end of 2016, respectively, and it has been hovering at $10,000 since July 2020. From the market capitalization perspective, Bitcoin has staged only 10% of the time horizon from our study period. This study captures the period when Bitcoin exceeded a market capitalization of US$100 bn computed based on the total number of mined coins, which holds significance for the financial market.
Despite Bitcoin's price and market capitalization history and its higher return/volatility characteristics as an emerging asset class (Figure 1), previous studies have failed to analyze Bitcoin price in relation to the macro, financial and economic variables. Most of the Bitcoin price analysis, in mid-2010, have reported the significant relationship between the Bitcoin price and the factors (e.g. search volume and media exposure) extracted from the social network service-based statistics. However, they have paid little attention to the causality between the macroeconomic variables. Despite functioning as an alternative investment asset class facilitating portfolio diversification and yield enhancement, studies have produced inconsistent outcomes on the relationship between the Bitcoin price and the macroeconomic variables explaining the rapid price appreciation and market adoption (see Figure 2).
Based on the previous Bitcoin research, we adopt the following approach to analyze the Bitcoin price. First, we extend the data observation period to early 2022 when Bitcoin was more exposed to financial market interactions. Second, we interpret data using the VAR and vector error correction model (VECM) and the impulse response function (IRF) to analyze the explanatory power among the macroeconomic variables. Given that most of the explanatory variables are integrated of order one (I(1)), we conduct a cointegration test to determine the applicability of VECM. The results lead to the selection of the VECM over the VAR. Third, under the general VECM structure, the cointegration equation shows a long-run equilibrium with a constant coefficient (termed “time-invariant cointegration” [TIC]). However, this study presents a VECM model based on a time-varying coefficient (termed “time-varying cointegration” [TVC]). To check the significance of the TVC VECM model against the general TIC VECM, we test the null hypothesis using the likelihood ratio (LR) test across multiple TVC parameters. Based on the results, we propose an optimal TVC parameter based on information criteria.
The remaining paper is organized as follows. Chapter 2 reviews the literature review, and Chapter 3 presents the data and methodology. Chapter 4 estimates the TIC VECM and TVC VECM and explains the hypothesis used to report the optimal model parameter. Chapter 5 concludes the study.
2. Literature review
Based on the correlation between the Bitcoin price and the macroeconomic variables, Son and Kim (2019) presented Bitcoin as a safe-haven asset during high-volatility periods. The study considered the period from January 2019 to June 2019. It sourced data on the Bitcoin price (Korean Won) from Bithumb and on the macroeconomic variables from the Federal Reserve Economic Data of the Federal Reserve Bank of St. Louis. The results showed a low correlation between gold and the Bitcoin price. They did not show Granger causality between the price and the six macroeconomic variables – the Bitcoin price, Standard and Poor's 500 volatility index, US treasury 10-year yield, US consumer price index, gold price and dollar index.
Lee et al. (2019a, b) empirically analyzed the determinants of the Bitcoin price in the Korean market for the period July 2015–May 2018. They classified the pricing factors into three groups composed of 12 variables – supply/demand (7), real economic (4), and psychological factors (1) – and examined the effect of these factors on the Bitcoin price. They employed five regression analysis models. The results showed a significant relationship of the price only with one psychological variable extracted from the Naver Trend Index. They also showed a significant negative relationship between the net stock purchase from the retail investors and the price. This finding implied the role of Bitcoin as an alternative investment from retail investors' standpoint.
Kim et al. (2019) analyzed the mutual relationship between Bitcoin and real, speculative and currency assets using VECM and daily data from July 2010 to April 2018. They found that Bitcoin is more closely associated with the speculative than with real or currency assets.
Bouoiyour and Selmi (2015) analyzed the Bitcoin pricing factors using the autoregressive distributed lag (ARDL) bounds test with data spanning from May 2010 to July 2014. They showed Bitcoin's strong speculative characteristics and its weak role as a safe-haven asset.
Zhu et al. (2017) performed Bitcoin analysis using VECM. From 2011 to 2016, they used monthly data on macroeconomic variables such as the price index, dollar index, Dow Jones industrial average, index, Fed fund rate, gold price and bitcoin price. They showed that the economic variables have explanatory power to Bitcoin price on a long-term basis. They also that the dollar index most significantly influenced the Bitcoin price movement, while gold had the least influence. They concluded that the Bitcoin as a speculative asset may not be a stable currency.
Thaker and Mand (2021) examined the relationship between the Bitcoin price and the Asian stock market index using the VECM, generalized autoregressive conditional heteroskedasticity in the mean-GARCH and wavelet analysis. They found a long-term significant relationship between the Bitcoin price and the stock market index.
Syafiqah and Mohamad (2021) included various cryptocurrencies (e.g. Bitcoin, Monero and Stella) to model the long-term relationship between cryptocurrencies and economic variables, using VECM and monthly data from January 2016 to December 2020. They found that only Monero had a significant long-run relationship with macroeconomic variables.
Yang et al. (2020) analyzed the price discovery relationship between the Bitcoin spot and futures price employing the time-varying Granger causality, cointegration and information sharing approaches. They found that the Bitcoin Futures price granger causes the Bitcoin spot price and that the former contributes more to the price discovery function than the latter.
This study is different from the previous research in the following aspects. First, it takes a longer data observation period of 10 years covering the US$1 trillion market capitalization period. Second, it expands the general VECM used in the initial stage of the research to include the TVC VECM in order to capture the dynamic nature of the long-term equilibrium relationship embedded in the VECM.
In summary, until 2015, the research used the GARCH and ARDL to determine the relationship between the Bitcoin price and the macroeconomic variables. Since 2017, studies have been using the VECM and TVC for exploring the long-term relationship between the variables of interest. However, this study expands the Bitcoin price model based on the TVC VECM to explain the dynamic long-run equilibrium between the Bitcoin price and the macroeconomic variables.
3. Methodology
3.1 Analysis flow
Despite the short history and idiosyncratic nature of its functioning, Bitcoin has been significantly influencing the financial markets. This background motivates our Bitcoin price analysis based on the VECM approach. In this context, we consider the dynamic nature of the macroeconomic variables and examine the long-run relationship by expanding the VECM to include the TVC approach. To test the significance of the TVC approach, we set a null hypothesis. This is because of the same cointegration coefficient relationship between the TIC VECM and the TVC VECM. We estimate the coefficient of the TVC VECM using the Chebyshev polynomials. We also propose an optimal TVC VECM under any Chebyshev polynomial specification, based on various information criteria outcomes (Bierens and Martins, 2010).
We employ the following steps to apply the VECM. First, we check the stationarity of the VAR variables using the augmented Dickey–Fuller (ADF) tests. Based on the test, we configure the variables as an I(0) or I(1) variable. For an appropriate lag selection for the ADF test, we use the Akaike information criterion (AIC) test. Second, we apply the VAR estimation with numerous information criteria outcomes to select a VAR lag model including the intercept adoption. Third, we estimate the VAR (p = lag) model. Fourth, given that most of the macroeconomic variables are I(1), we conduct Johansen tests. This helps us to set up the model correctly in the presence of cointegration among I(1) variables. Fifth, after confirming the presence of cointegration, we set the VECM model as the analysis model. We validate the model from various standpoints by employing additional tests such as the test for the significance of the model coefficient and the test for the characteristics of the residuals generated from the model. The test for residuals included the tests for heteroscedasticity, autocorrelation of residuals and normality of residuals. Sixth, based on the estimated VECM model IRF and the forecast error variance decomposition (FEVD), we explain the relationship among the selected variables. Finally, we adjust for the time-varying long-run relationship of the cointegration equation by using the Chebyshev polynomials. This included the following steps. First, we estimate the TVC VECM model (Bierens and Martins, 2010). Second, we run a hypothesis of the TIC VECM over the TVC VECM using an LR test. Third, we choose optimal the TVC VECM model with some Chebyshev polynomials of (m) to illustrate the trend of each TVC VECM long-run cointegration parameter alongside the polynomial (m).
3.2 Data
The dataset comprised six macroeconomic variables (https://data.nasdaq.com) sampled monthly from August 2010 to February 2022. Table 1 presents the summary statistics of these variables.
The six economic variables are the Bitcoin price (in US dollars; BTCUSD), Standard and Poor's (S&P) 500 volatility index (VIX), US treasury 10-year yield (US 10Y), US consumer price index (US CPI), gold price and dollar index. VIX derived from option price of the S&P500 index is a volatility index representing broad market uncertainties moving in the opposite direction of the stock market or risky asset. US10Y standing for the US treasury's 10-year yield is regarded as the constant maturity treasury (CMT). Rate operates as a safe asset during high volatility periods and absorbs the inflationary pressure with rising yields. BTCUSD is the log of the Bitcoin price, USCPI stands for the US CPI, GOLD stands for the gold price in US$, and DXY represents the dollar index.
3.3 VECM and cointegration
As the starting point for the multivariate analysis, we use VAR, where this model assumes the stationarity of the times series dataset. If there are k variables for the analysis with time lag = p (VAR (lags = p)), the VAR model can be defined as below (Sims, 1980).
From Equation (1),
Given that VAR(p) model assumes the stationarity of the time series variables, if these variables fail to achieve stationarity at the raw level during the analysis, we differentiate the time series until the time series achieves stationarity and facilitates the application of VAR(p) model. However, the stationarity of the underlying variables can be obtained by differentiating the nonstationary time series at the expense of the cointegration information among variables. Hence, we use the Johansen test to check for the presence of cointegration among variables. This helps in validating the eligibility of the VECM.
Using the test results in Table 2, we conduct the ADF to test the time series stationarity (ADF lags were chosen based on the AIC).
According to ADF test result, all variables, except for USCPI, were reported as nonstationary, under no intercept condition (test regression: none). When we included the Trend term, the ADF test reported USCPI and DXY as nonstationary time series. The first differencing of all variables led to stationarity in the time series, except for USCPI with the Trend condition.
From Equation (1), if the solution of
The long-run relationship between the Bitcoin and the macroeconomic variables under VECM can be explained by the linear cointegration vector. In this case (transitory), the VECM can be represented as follows (Engle and Granger, 1987).
From Equation (2),
3.4 Time varying cointegration
Utilizing the VECM from Equation (2), the TVC VECM can be modeled as follows.
Our study uses the Chebyshev polynomials with different m (dimension) to estimate smoothly changing
We use the LR test to test for the effectiveness of the TVC VECM over the TIC VECM. This test is expressed as below in Equation (7).
We use the following values for the hypothesis test:
4. Empirical analysis
4.1 VECM and cointegration
To study the Bitcoin pricing model using macro-economic variables, we test the VAR lag selection procedure without and with the constant + Trend term. We also use the following information criteria: the AIC, final prediction error (FPE), Schwarz information criterion (SIC) and Hannan–Quinn (HQ) information criterion. The AIC and FPE produce Lag = 2, while the HQ and SC produce Lag = 1 (see Table 3).
Our analysis adopted the AIC of Lag = 2 and VAR (p = 2) with Trend model. According to the ADF test, most of variables were I(1). Additionally, we use the Johansen test to check for the presence of cointegration. This helps in the selection of a suitable model between VAR and VECM. The Johansen test utilized the trace and Eigen values. The test confirmed the existence of a single cointegration equation; this presence is regardless of Trend term being under the 5% significance level. Table 4 summaries the outcomes of the Johansen test.
Based on the results of the Johansen test, we select the VECM (p = 1, r = 1) [4]. In the following sections, we present the weights (or loading matrix) and coefficient of cointegration from the VECM (see Table 5).
Concerning the significance of the loading matrix (or weights), VIX, BTCUSD and GOLD were significant at the 5% level with the Trend term. In the absence of the constant term, only VIX and GOLD were significant at the 5% level (see Table 6).
With the loading matrix and the cointegrating coefficient, we present
F-statistics of
Table 8 summarizes the residual analysis for the VECM (p = 1, r = 1). The heteroscedasticity test failed to reject the null hypothesis, implying the absence of an ARCH effect. The Portmanteau test (PT) for serial correlation also failed to reject the null hypothesis, implying the absence of a serial correlation. The Jarque–Bera (JB) test also rejected the null hypothesis, rejecting the normality of residuals.
4.2 Granger causality, impulse response function and forecast error variance decomposition analyses
To understand the economic relationship between variables, we employed the Granger causality test. According to the analysis, the USCPI and DXY granger causes the Bitcoin price (10 and 5% significance levels, respectively). When analyzing the opposite direction, we find that only DXY granger causes Bitcoin at a 5% significance level. This outcome contradicts the previous results in Kim et al. (2019). Specifically, they reported a reciprocal Granger causality between VIX and the Bitcoin, while we found a mutual Granger causality between DXY and Bitcoin [5] (Table 9).
The IRF results with lag (n = 12) are summarized by n = 1, 3, 5 and 12 (left axis = impulse variable). In case of the impulse from the VIX, the Bitcoin price first moved negatively and gradually mitigated negative shocks over time. Concerning the impulse from the USCPI and US10Y, they provided a (+) positive impulse to the Bitcoin price. Concerning GOLD and DXY, these safe assets exerted a persistent (−) negative impact until n = 5 (roughly 5 months). In summary, Bitcoin has moved positively toward the inflation linked variables, like USCPI and USCPI, while it moved against the GOLD and DXY impulse shock (Table 10).
Concerning the outcomes of FEVD with lag (n = 12), they are summarized as follows. In case of VIX, GOLD and Bitcoin accounted for 19.3 and 2.2%, respectively, each at n = 12 (at the 12th month). For Bitcoin at n = 12, Bitcoin, DXY and GOLD accounted for 89.6, 3.7 and 2.1%, respectively. For DXY, when n increases from 1 to 12, the explanatory weights of GOLD decreased from 21 to 12.8%. However, the explanatory power of DXY, USCPI and GOLD increased from 52 to 55.9, 2.0–2.9, and 0.2–1.7%, respectively. The explanatory power of GOLD declined from 85.6 to 57.5%, when n moved from 1 to 12. However, US10Y to GOLD increased from 8.3 to 16.6%, USCPI to GOLD increased from 4.6 to 11.0% and Bitcoin to GOLD also increased from 1.0 to 2.7% (Table 11).
4.3 Time-varying cointegration test
Given the dynamic relationship between the Bitcoin price and the macroeconomic time series, it would be appropriate to consider the varying degree of their long-run relationship. This means that a time-varying approach should replace a constant coefficient equation from the general VECM. Hence, we develop the TVC VECM and test the feasibility of this expansion using the LR test. The cointegration rank for both the TIC and TVC VECM was set at r = 1 (see Table 12).
Null hypothesis: The TIC VECM and TVC VECM share the same cointegration coefficient relationship across the Chebyshev polynomials m = 1, 2, 3 and 4
The tests yield the following results. First, when Chebyshev polynomials (m) = 1, the null hypothesis was not rejected at a 10% significance level (P value = 0.18). Second, when Chebyshev polynomial of m = 2 or m > 3, the null hypothesis was rejected (P value = 0.00152 for m = 2, P value = 0.00233 for m = 3, P value = 0.00019 for m = 4), with TVC revealed visually. When we applied the HQ information criteria, the optimal Chebyshev polynomials m were 2 (Bierens and Martins, 2010).
When we used the Chebyshev polynomial of m = 1, the null hypothesis was not rejected. As shown in Figure 3, the TVC parameters have shown like constant numbers across the time horizon. At this point, we believe that they are similar to the TIC VECM cointegration parameters (Beta chart with m = 1). Hence, the null hypothesis holds.
When we use the Chebyshev polynomial of m = 2, the null hypothesis was rejected, implying the significance of the TVC parameters. Figure 4 shows the smooth and gradual evolution of the long-run relationship coefficient of three economic variables, including US10Y.
While we have shown the coefficient of the long-run relationship using the Chebyshev polynomials of m (see Appendix 1 for m = 3 and 4), the following Figures 5–7 exhibit the
5. Conclusion
This study analyzes the Bitcoin price in relation to six macroeconomic variables studied over a 10-year time horizon, using the VECM. Most studies have reported the speculative characteristics of Bitcoin owing to its short history and association with the financial market. Hence, we use a period when Bitcoin was more exposed to the financial market interactions, that is, when it achieved and crossed a market capitalization of US$1 tn. We also construct a VAR/VECM model suitable for capturing the interrelationship among the sampled macroeconomic variables. We validate these relationships using the time series stationary test and an optimal lag selection. We also test for the long-run cointegration both under the TIC VECM (p = 1, r = 1) and TVC VECM. Given the dynamic nature of the macroeconomic variables, based on the results of the LR test, we use a TVC VECM for Bitcoin modeling.
This study has implications in demonstrating that the emerging digital assets, including Bitcoin, would prevail and consolidate. Given this, it is important to analyze the dynamic long-run relationship between digital assets and macroeconomic variables from a modeling perspective. This study has used the time-varying parameter approach for long-run cointegration from the VECM. However, the future studies may apply the time-varying approach to loading factor.
Figures
Summary statistics
VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | |
---|---|---|---|---|---|---|
# of Observation | 139 | 139 | 139 | 139 | 139 | 139 |
Max | 9.5100 | 0.5500 | −2.7806 | 218.3120 | 1060.0000 | 85.5999 |
Min | 53.5400 | 3.4700 | 11.0305 | 283.7160 | 1964.9000 | 122.8165 |
Mean | 18.3071 | 2.1023 | 6.3133 | 243.5492 | 1438.6662 | 105.3285 |
Median | 16.4800 | 2.1600 | 6.4841 | 240.2360 | 1326.5000 | 110.0848 |
Stdev | 6.9079 | 0.6448 | 3.2873 | 14.7723 | 238.3072 | 10.9168 |
Skewness | 1.9509 | −0.2706 | −0.7036 | 0.5648 | 0.5271 | −0.3695 |
Kurtosis | 5.2782 | −0.2170 | −0.2192 | −0.2483 | −1.0416 | −1.4860 |
Results of augmented Dickey–Fuller and test regression with none condition
VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | |
---|---|---|---|---|---|---|
A. Level | ||||||
Test regression, None | ||||||
ADF(t-value) | −0.09 | −0.49 | 0.74 | 2.53 | 0.39 | 1.00 |
P-value | 0.928740 | 0.625000 | 0.463180 | 0.012767** | 0.698000 | 0.322400 |
Lag (from AIC) | 7 | 2 | 10 | 4 | 2 | 11 |
Test regression, Trend | ||||||
ADF(t-value) | −4.46 | 2.08 | −4.44 | 0.53 | −2.82 | −1.64 |
P-value | 0.0000212*** | 0.0399** | 0.000024*** | 0.597698 | 0.005723*** | 0.1046 |
Lag (from AIC) | 2 | 2 | 9 | 4 | 2 | 8 |
VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | |
---|---|---|---|---|---|---|
B. 1st Difference | ||||||
Test regression, None | ||||||
ADF(t-value) | −6.69 | −6.53 | −2.83 | 0.60 | −7.64 | −2.83 |
P-value | 0.0000000013*** | 0.0000000024*** | 0.005598*** | 0.553300 | 0.00000000001*** | 0.00571*** |
Lag (from AIC) | 6 | 2 | 8 | 11 | 2 | 10 |
Test regression, Trend | ||||||
ADF(t-value) | −6.65 | −6.47 | −3.10 | −4.44 | −8.03 | −3.34 |
P-value | 0.0000000016*** | 0.0000000033*** | 0.00253*** | 0.000022*** | 0.00000000000*** | 0.00121*** |
Lag (from AIC) | 6 | 2 | 9 | 3 | 2 | 10 |
Note(s): *** significant at 1% level, ** significant at 5% level and * significant at 10% level
Results of the VAR order selection criteria
None | Both (=Constant + Trend) | ||||||||
---|---|---|---|---|---|---|---|---|---|
Lag | AlC(n) | HQ(n) | SC(n) | FPE(n) | Lag | AIC(n) | HQ(n) | SC(n) | FPE(n) |
1 | 5.8999 | 6.2210 | 6.6901 | 365.1590 | 1 | 5.7723 | 6.2004 | 6.8258 | 321.5622 |
2 | 5.7691 | 6.4112 | 7.3494 | 321.2424 | 2 | 5.6384 | 6.3875 | 7.4820 | 282.3844 |
3 | 5.9645 | 6.9277 | 8.3349 | 393.4763 | 3 | 5.7789 | 6.8491 | 8.4126 | 328.0954 |
4 | 6.2795 | 7.5638 | 9.4400 | 547.0851 | 4 | 6.0630 | 7.4543 | 9.4869 | 443.6370 |
5 | 6.6174 | 8.2227 | 10.5680 | 786.0142 | 5 | 6.4024 | 8.1147 | 10.6164 | 640.8920 |
6 | 6.7399 | 8.6663 | 11.4807 | 922.3374 | 6 | 6.6185 | 8.6520 | 11.6227 | 829.9974 |
7 | 6.7850 | 9.0324 | 12.3159 | 1,017.9781 | 7 | 6.5402 | 8.8946 | 12.3345 | 814.7621 |
8 | 6.8309 | 9.3995 | 13.1520 | 1,147.4201 | 8 | 6.6051 | 9.2807 | 13.1896 | 943.1980 |
Results of the Johansen test
None (lag = 2) | Critical value | Trend (lag = 2) | Critical value | ||||||
---|---|---|---|---|---|---|---|---|---|
Null hypothesis | Test statistics | 1% | 5% | Null hypothesis | Test statistics | 1% | 5% | ||
Trace | r ≤ 2 | 31.77 | 55.43 | 48.28 | Trace | r ≤ 2 | 37.62 | 70.05 | 62.99 |
r ≤ 1 | 52.07 | 78.87 | 70.6 | r ≤ 1 | 71.18 | 96.58 | 87.31 | ||
r = 0 | 96.37 | 104.2 | 90.39 | r = 0 | 115.77 | 124.75 | 114.9 | ||
Eigen | r ≤ 2 | 15.87 | 32.14 | 32.14 | Eigen | r ≤ 2 | 16.67 | 36.65 | 31.46 |
r ≤ 1 | 20.31 | 38.78 | 33.32 | r ≤ 1 | 33.56 | 42.36 | 37.52 | ||
r = 0 | 44.3 | 44.59 | 39.43 | r = 0 | 44.59 | 49.51 | 43.97 |
VECM loading matrix (or weights) and statistical significance test
Trend* | None | ||||||
---|---|---|---|---|---|---|---|
Coefficient | t-value | Pr(>|t|) | Coefficient | t-value | Pr(>|t|) | ||
VIX | −0.31694 | −4.13220 | 0.00010 | −0.36320 | −4.42370 | 0.00003 | |
US10Y | 0.00101 | 0.32600 | 0.75200 | 0.00076 | 0.22920 | 0.82400 | |
ECT1** | BTCU SD | 0.01108 | 2.29080 | 0.02796 | 0.00894 | 1.69930 | 0.10160 |
USCPI | −0.00804 | −0.84160 | 0.41600 | −0.00961 | −0.93280 | 0.36700 | |
GOLD | 2.88630 | 2.80290 | 0.00743 | 3.22430 | 2.90770 | 0.00554 | |
DXY | −0.00893 | −0.34750 | 0.73600 | −0.01685 | −0.60810 | 0.55600 |
Note(s): * VECM with Trend adopted from the analysis ** ECT1: error correction term (α)
VECM cointegration relationship: the coefficient (β) of cointegrating equation estimated under Trend and none assumption
Trend* | None | ||||||
---|---|---|---|---|---|---|---|
Coefficient | t-value | Pr(>|t|) | Coefficient | t-value | Pr(>|t|) | ||
VIX | 1.0000 | 1.0000 | |||||
US10Y | −4.5776 | −2.0368 | 0.0420 | −3.6978 | −1.8067 | 0.0710 | |
BTCUSD | −2.7270 | −2.4217 | 0.0150 | −1.4801 | −1.9976 | 0.0460 | |
ßeta | USCPI | 0.3617 | 0.9594 | 0.3370 | 0.5122 | 2.3408 | 0.2190 |
GOLD | −0.0443 | −5.0238 | 0.0000 | −0.0376 | −5.1166 | 0.0070 | |
DXY | −0.6762 | −1.7510 | 0.0800 | −0.3418 | −1.9713 | 0.1730 | |
Trend | 0.2281 | 0.9121 | 0.3620 |
Note(s): * VECM with Trend adopted from the analysis
VECM (p = 1, r = 1) representation
D.BTCUSD | Trend | ||||
---|---|---|---|---|---|
Coefficient | t-value | Pr(>|t|) | |||
ECT1 | 0.0111 | 2.2230 | 0.0280 | ||
Const | 0.5023 | 2.5960 | 0.0l05 | ||
D.VIX(l1) | −0.0081 | −1.3130 | 0.1915 | ||
D.US10Y(l1) | 0.0571 | 0.3750 | 0.7085 | ||
D.BTCUSD(l1) | 0.2279 | 2.6890 | 0.0081 | Adj R square | 0.1625 |
D.USCPI(l1) | 0.0469 | 1.2530 | 0.2126 | F-statistics | 4.3220 |
D.GOLD(l1) | −0.0007 | −1.4620 | 0.1463 | Pr(>|t|) | 0.0001 |
D.DXY(l1) | −0.0422 | −1.9740 | 0.0505 |
Note(s): ECT1: error correction term (α); coefficient: short-term relationship;
adjusted R square: 0.1625; F-statistics: 4.322; p-value: 0.001 (significance of ΔBTCUSD equation)
Model diagnostic statistics using ARCH, PT and JB
VECM(r = 1) | Trend (Y/N) | Multivariate ARCH (lags = 5) | P value | Portmanteau test (asymptotic, lags = 16) | P value | JB normality test | P value |
---|---|---|---|---|---|---|---|
p = 1 | Y | 2251.6 | 0.23990 | 527.33 | 0.28860 | 370.2 | 2.20E-16 |
Granger causality test based on F-statistics with lag = 2 (Wald test comparing the unrestricted with the restricted model)
Caused | BITCUSD | Causing | BITCUSD | ||||
---|---|---|---|---|---|---|---|
F-value | Pr(>|t|) | F-value | Pr(>|t|) | ||||
VIX | 1.38310 | 0.25440 | VIX | 0.34470 | 0.70910 | ||
US10Y | 1.21360 | 0.30040 | US10Y | 1.11040 | 0.33250 | ||
Causing | USCPI | 2.38990 | 0.09559 | Caused | USCPI | 0.25840 | 0.77260 |
GOLD | 0.79880 | 0.45200 | GOLD | 0.07260 | 0.93000 | ||
DXY | 3.39360 | 0.03654 | DXY | 2.52590 | 0.08384 |
Impulse response function
Response Variable | Response Variable | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
n = 1 | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | n = 3 | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY |
VIX | 4.945 | −0.052 | −0.059 | −0.064 | 4.801 | 0.818 | VIX | 2.586 | −0.071 | −0.087 | −0.408 | 16.131 | 0.883 |
US10Y | – | 0.192 | −0.020 | 0.101 | −19.110 | −0.097 | US10Y | −0.486 | 0.220 | 0.029 | 0.310 | −25.838 | −0.077 |
BTCUSD | – | – | 0.306 | −0.066 | −6.621 | −0.081 | BTCUSD | 0.202 | 0.021 | 0.385 | −0.174 | −8.916 | −0.237 |
USCPI | – | – | – | 0.600 | 14.244 | −0.234 | USCPI | 0.113 | −0.013 | 0.042 | 1.148 | 20.474 | −0.275 |
GOLD | – | – | – | – | 61.415 | −0.759 | GOLD | 1.039 | −0.023 | −0.048 | 0.118 | 48.074 | −0.612 |
DXY | – | – | – | – | – | 1.195 | DXY | 0.133 | 0.004 | −0.072 | −0.111 | −11.656 | 1.276 |
IRF outcome (vertical: impulse variables), n = number of time lapse | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Response Variable | Response Variable | ||||||||||||
n = 5 | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | n = 12 | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY |
VIX | 1.809 | −0.058 | −0.054 | −0.505 | 19.529 | 0.891 | VIX | 1.433 | −0.050 | −0.030 | −0.537 | 21.359 | 0.882 |
US10Y | −0.383 | 0.220 | 0.037 | 0.384 | −26.297 | −0.088 | US10Y | −0.293 | 0.218 | 0.036 | 0.416 | −26.499 | −0.091 |
BTCUSD | 0.397 | 0.021 | 0.378 | −0.201 | −10.93 | −0.239 | BTCUSD | 0.493 | 0.019 | 0.370 | −0.214 | −11.74 | −0.234 |
USCPI | 0.226 | −0.017 | 0.055 | 1.313 | 21.969 | −0.298 | USCPI | 0.337 | −0.019 | 0.057 | 1.382 | 22.184 | −0.306 |
GOLD | 1.240 | −0.027 | −0.060 | 0.140 | 47.007 | −0.601 | GOLD | 1.304 | −0.029 | −0.065 | 0.146 | 46.743 | −0.599 |
DXY | 0.173 | 0.004 | −0.080 | −0.143 | −12.647 | 1.291 | DXY | 0.161 | 0.004 | −0.082 | −0.157 | −12.748 | 1.293 |
Forecast error variance decomposition by dependent variables
Dependent variable | VIX | Dependent variable | US10Y | Dependent variable | BTCUSD | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
n = | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY |
1 | 100.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 6.8% | 93.2% | 0.0% | 0.0% | 0.0% | 0.0% | 3.5% | 0.4% | 96.0% | 0.0% | 0.0% | 0.0% |
2 | 96.6% | 0.5% | 0.0% | 0.0% | 2.8% | 0.1% | 9.7% | 89.6% | 0.3% | 0.1% | 0.3% | 0.0% | 5.7% | 0.2% | 92.0% | 0.3% | 0.5% | 1.4% |
3 | 94.1% | 1.0% | 0.1% | 0.0% | 4.7% | 0.1% | 9.5% | 89.3% | 0.5% | 0.2% | 0.6% | 0.0% | 5.3% | 0.3% | 90.9% | 0.6% | 0.8% | 2.1% |
4 | 91.4% | 1.2% | 0.3% | 0.1% | 6.9% | 0.2% | 9.0% | 89.5% | 0.6% | 0.2% | 0.7% | 0.0% | 4.6% | 0.4% | 90.4% | 0.9% | 1.1% | 2.5% |
5 | 88.8% | 1.3% | 0.5% | 0.2% | 8.9% | 0.2% | 8.4% | 89.7% | 0.6% | 0.3% | 0.9% | 0.0% | 4.0% | 0.5% | 90.1% | 1.1% | 1.4% | 2.8% |
6 | 86.4% | 1.4% | 0.8% | 0.3% | 10.9% | 0.2% | 8.0% | 90.0% | 0.7% | 0.3% | 1.0% | 0.0% | 3.5% | 0.6% | 90.0% | 1.3% | 1.6% | 3.1% |
7 | 84.1% | 1.5% | 1.1% | 0.4% | 12.7% | 0.3% | 7.6% | 90.3% | 0.7% | 0.4% | 1.1% | 0.0% | 3.1% | 0.6% | 89.9% | 1.4% | 1.7% | 3.3% |
8 | 82.0% | 1.5% | 1.4% | 0.5% | 14.3% | 0.3% | 7.3% | 90.5% | 0.7% | 0.4% | 1.1% | 0.0% | 2.8% | 0.7% | 89.8% | 1.5% | 1.8% | 3.4% |
9 | 80.2% | 1.6% | 1.6% | 0.6% | 15.7% | 0.3% | 7.0% | 90.6% | 0.7% | 0.5% | 1.2% | 0.0% | 2.6% | 0.7% | 89.7% | 1.5% | 1.9% | 3.5% |
10 | 78.5% | 1.6% | 1.8% | 0.7% | 17.1% | 0.3% | 6.8% | 90.8% | 0.7% | 0.5% | 1.2% | 0.0% | 2.4% | 0.7% | 89.7% | 1.6% | 2.0% | 3.6% |
11 | 77.0% | 1.6% | 2.0% | 0.8% | 18.2% | 0.3% | 6.6% | 90.9% | 0.7% | 0.5% | 1.3% | 0.0% | 2.2% | 0.7% | 89.7% | 1.6% | 2.1% | 3.7% |
12 | 75.6% | 1.6% | 2.2% | 0.9% | 19.3% | 0.3% | 6.5% | 91.0% | 0.7% | 0.5% | 1.3% | 0.0% | 2.1% | 0.7% | 89.6% | 1.7% | 2.1% | 3.7% |
Dependent variable | USCPI | Dependent variable | GOLD | Dependent variable | DXY | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
n = | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY | VIX | US10Y | BTCUSD | USCPI | GOLD | DXY |
1 | 1.1% | 2.7% | 1.2% | 95.1% | 0.0% | 0.0% | 0.5% | 8.3% | 1.0% | 4.6% | 85.6% | 0.0% | 24.4% | 0.3% | 0.2% | 2.0% | 21.0% | 52.0% |
2 | 5.8% | 4.4% | 1.7% | 87.1% | 0.7% | 0.3% | 1.0% | 14.3% | 1.0% | 6.7% | 74.9% | 2.2% | 26.1% | 0.2% | 1.0% | 2.0% | 16.2% | 54.4% |
3 | 8.1% | 5.2% | 1.8% | 83.6% | 0.8% | 0.5% | 2.9% | 15.3% | 1.4% | 8.0% | 69.8% | 2.6% | 26.3% | 0.2% | 1.3% | 2.2% | 15.1% | 54.9% |
4 | 9.3% | 5.7% | 1.8% | 81.8% | 0.8% | 0.7% | 4.1% | 15.8% | 1.7% | 9.0% | 66.4% | 3.0% | 26.5% | 0.2% | 1.5% | 2.4% | 14.3% | 55.1% |
5 | 10.0% | 6.0% | 1.8% | 80.7% | 0.8% | 0.7% | 5.2% | 16.1% | 1.9% | 9.6% | 64.0% | 3.2% | 26.5% | 0.2% | 1.6% | 2.5% | 13.9% | 55.3% |
6 | 10.4% | 6.2% | 1.8% | 80.0% | 0.8% | 0.8% | 6.1% | 16.3% | 2.1% | 10.0% | 62.2% | 3.3% | 26.5% | 0.2% | 1.6% | 2.6% | 13.6% | 55.4% |
7 | 10.7% | 6.3% | 1.8% | 79.5% | 0.9% | 0.8% | 6.7% | 16.4% | 2.3% | 10.3% | 60.9% | 3.4% | 26.5% | 0.2% | 1.7% | 2.7% | 13.4% | 55.6% |
8 | 10.8% | 6.4% | 1.8% | 79.2% | 0.9% | 0.9% | 7.3% | 16.4% | 2.4% | 10.5% | 59.9% | 3.5% | 26.5% | 0.2% | 1.7% | 2.7% | 13.2% | 55.7% |
9 | 11.0% | 6.5% | 1.8% | 78.9% | 0.9% | 0.9% | 7.7% | 16.5% | 2.5% | 10.7% | 59.1% | 3.5% | 26.5% | 0.2% | 1.7% | 2.8% | 13.1% | 55.7% |
10 | 11.1% | 6.6% | 1.8% | 78.7% | 0.9% | 0.9% | 8.0% | 16.5% | 2.6% | 10.8% | 58.5% | 3.6% | 26.4% | 0.2% | 1.7% | 2.8% | 13.0% | 55.8% |
11 | 11.1% | 6.6% | 1.8% | 78.6% | 0.9% | 0.9% | 8.3% | 16.6% | 2.7% | 10.9% | 57.9% | 3.6% | 26.4% | 0.2% | 1.7% | 2.8% | 12.9% | 55.9% |
12 | 11.2% | 6.7% | 1.8% | 78.5% | 0.9% | 0.9% | 8.5% | 16.6% | 2.7% | 11.0% | 57.5% | 3.6% | 26.4% | 0.3% | 1.7% | 2.9% | 12.8% | 55.9% |
Summary results on time-varying cointegration
Model | VECM(p = 1, r = 1) | |||
---|---|---|---|---|
m = 1 | m = 2 | m = 3 | m = 4 | |
LR TVC statistics | 8.87395 | 31.73887 | 39.65449 | 56.63646 |
P-value | 0.18079 | 0.00152*** | 0.00233*** | 0.00019*** |
Log likelihood | −1511.23162 | −1499.79916 | −1495.84136 | −1487.35037 |
AIC | 22.70411 | 22.59561 | 22.82980 | 22.76424 |
BIC | 23.64192 | 23.61867 | 24.27913 | 24.29883 |
HQC | 23.08521 | 23.01135 | 23.41877 | 23.38786 |
Note(s): *** significant at 1% level, ** significant at 5% level and * significant at 10% level
Notes
See Appendix 2
This is the case when Chebyshev polynomials m = 0
For the maximum likelihood estimates and LR TVC, refer to Bierens and Martins (2010)
As the VAR model was estimated with lag = 2 (VAR (p = 2)), the VECM model is presented as lag = 1 (VECM (p = 1)).
Kim et al. (2019) used the daily data from July 2010 to April 2018, while used monthly data from August 2010 to Feb 2022.
Cointegration Relation with m = 3,4 were reported at Appendix 3 and normalization in charts was based on the coefficient of BTCUSD.
Orthonormal Chebyshev Polynomials
Using and substituting
Let
Coefficient β of TIC (time invariant cointegration) for null hypothesis is derived from
References
Bierens, H.J. and Martins, L.F. (2010), “Time-varying cointegration. econometric theory 143–1490”, Econometric Theory, Vol. 26 No. 5, pp. 1453-1490, doi: 10.1017/S0266466609990648.
Bouoiyour, J. and Selmi, R. (2015), “What does bitcoin look like?”, Analysis of Economics and Finance, Vol. 16 No. 2, pp. 449-492.
Engle, R.F. and Granger, C.W. (1987), “Co-integration and error correction: representation, estimation, and testing”, Econometrica: Journal of the Econometric Society, Vol. 55 No. 2m, pp. 21-276, doi: 10.2307/1913236.
Kim, J.M. Kang, N.Y. and Park, Y.J. (2019), “A study on cross – effects of prices of bitcoin, traditional assets, and traditional currencies”, Korean Business Education Review, Vol. 34 No. 5, pp. 151-169, doi: 10.23839/kabe.2019.34.5.151.
Lee, G.S., Joe, Y.M. and Jeong, J.H. (2019a), “An investigation of dynamic price movements of the cryptocurrency coin in Korea”, Asian Review of Financial Research, Vol. 32 No. 3, pp. 383-400.
Lee, K.K., Cho, S.J., Min, G.S. and Yang, W.C. (2019b), “The determinant of bitcoin prices in Korea”, Korean Journal of Financial Studies, Vol. 48 No. 4, pp. 393-415, doi: 10.26845/KJFS.2019.08.48.4.393.
Pfaff, B. (2008), “VAR, SVAR and SVEC models: implementation within r package vars”, Journal of Statistical Software, Vol. 27 No. 4, doi: 10.18637/jss.v027.i04.
Sims, C.A. (1980), “Macroeconomics and reality”, Econometrica, Vol. 48 No. 1, pp. 1-48, doi: 10.2307/1912017.
Son, J.H. and Kim, J.Y. (2019), “A study in bitcoin volatility through economic factors”, The Journal of Society for E-Business Studies, Vol. 24 No. 4, pp. 109-118, doi: 10.7838/jsebs.2019.24.4.109.
Syafiqah, I. and Mohamad, Y. (2021), “An analysis on cryptocurrencies and macroeconomic variables using VECM, ASEAN”, Journal of Management and Business Studies, Vol. 3 No. 1, pp. 08-15, doi: 10.26666/rmp.ajmbs.2021.1.2.
Thaker, H.M.T. and Mand, A.A. (2021), “Bitcoin and stock markets: a revisit of relationship”, Journal of Derivatives and Quantitative Studies, Vol. 29 No. 3.
Yang, H., Yang, G.H. and Oxley, L. (2020), “What role do futures markets play in bitcoin pricing? Causality, cointegration and price discovery from a time-varying perspective?”, International Review of Financial Analysis, Vol. 72, doi: 10.1016/j.irfa.2020.101569.
Zhu, Y., Dickinson, D. and Li, J. (2017), “Analysis on the influence factors of bitcoin's price based on VEC model”, Financial Innovation, Vol. 3 No. 3, pp. 1-13, doi: 10.1186/s40854-017-0054-0.
Further reading
Badev, A. and Chen, M. (2014), “Bitcoin: technical background and data analysis”, Working Papers, No. 2014-104, Finance and Economics Discussion Series (FEDS), Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, DC, pp. 1-39.
Barai, M. and Kundu, S. (2019), “The role of the federal reserve in the U.S housing crisis: a VAR analysis with endogenous structural breaks”, Journal of Risk Financial Management, Vol. 12 No. 3, doi: 10.3390/jrfm12030125.
Berentsen, A. and Schar, F. (2018), “A short introduction to the world of cryptocurrencies. Federal Reserve Bank of St. Louis Review”, First Quarter, Vol. 100 No. 1, doi: 10.20955/r.2018.1-16.
Casas, I. and Fernandez-Casal, R. (2019), “tvReg: time-varying coefficient linear regression for single and multi-equations in R”, available at: https://ssrn.com/abstract=3363526.
Dyhrberg, A.H. (2015), “Bitcoin, gold and the dollar – a GARCH volatility analysis”, Finance Research Letters, Vol. 16, pp. 85-92, doi: 10.1016/j.frl.2015.10.008.
Choi, S.Y. and Shin, J.S. (2019), “Analysis of cryptocurrency volatility”, The Korean Journal of Financial Management, Vol. 36 No. 2, pp. 65-82.
D'Agostino, A., Gambeti, L. and Giannone, D. (2011), “Macroeconomic forecasting and structural change”, Journal of Applied Econometrics, Vol. 28 No. 1, pp. 82-101, doi: 10.1002/jae.1257.
Glaser, F., Zimmermann, K., Haferkorn, M., Weber, M. and Siering, M. (2014), “Bitcoin – asset or currency? revealing user's hidden intentions”, Twenty Second European Conference on Information Systems, Tel Aviv 2014, Electronic copy, available at: http://ssrn.com/abstract=2425247.
Gottschalk, J. (2001), “An introduction into the SVAR methodology: identification, interpretation and limitations of SVAR models”, Kiel Working Paper No. 1072, Kiel Institute for the World Economy (IfW Kiel), available at: http://hdl.handle.net/10419/17887.
Haslbeck, J., Bringmann, L. and Waldorp, L. (2020), “A tutorial on estimating time-varying vector autoregressive models”, Multivariate Behavioral Research, Vol. 56 No. 1, pp. 120-149, doi: 10.1080/00273171.2020.1743630.
Jang, S.I. and Kim, J.Y. (2017), “A study on the asset characterization of bitcoin”, The Journal of Society for E-Business Studies, Vol. 22 No. 4, pp. 117-128, doi: 10.7838/jsebs.2017.22.4.117.
Johansen, S. (1988), “Statistical analysis of cointegration vectors”, Journal of Economic Dynamics and Control, Vol. 12 Nos 2-3, pp. 231-254, doi: 10.1016/0165-1889(88)90041-3.
Kjaerland, F., Khazal, A., Krogstad, E.A., Nordstrom, F.B.G. and Oust, A. (2018), “An analysis of bitcoin's price dynamics”, Journal of Risk and Financial Management, Vol. 11, doi: 10.3390/jrfm11040063.
Kuschnig, N. and Vashold, L. (2020), “BVAR: Bayesian vector autoregressions with hierarchical prior selection in R”, Working paper, 296, WU Vienna University of Economics and Business, available at: https://epub.wu.ac.at/id/eprint/7216.
Lange, A., Dalheimer, B., Herwartz, H. and Maxand, S. (2021), “An r package for data-driven identification in multivariate time series analysis”, Journal of Statistical Software, Vol. 97 No. 5, doi: 10.18637/jss.v097.i05.
Lee, J.S., Kim, K.W. and Park, D.H. (2018), “Empirical analysis on bitcoin price change by consumer, industry and macro-economy variables”, Journal of Intelligence and Information Systems, Vol. 24 No. 2, pp.195-220, doi: 10.13088/jiis.2018.24.2.195.
Martins, L.F. (2018), “Bootstrap tests for time varying cointegration”, Econometric Reviews, Vol. 37 No. 5, pp. 466-483, doi: 10.1080/07474938.2015.1092830.
Meynkhard, A. (2019), “Fair market value of bitcoin: halving effect”, Investment Management and Financial Innovations, Vol. 16 No. 4, doi: 10.21511/imfi.16(4).2019.07.
Nam, Y.S. and Kang, S.H. (2022), “Effect of quantitative growth of the cryptocurrency market on herding phenomenon”, The Korean Journal of Financial Management, Vol. 39 No. 1, pp. 63-96, doi: 10.22510/kjofm.2022.39.1.003.
Thaker, H.M.T. and Mand, A.A. (2021), “Bitcoin and stock markets: a revisit of relationship”, Journal of Derivatives and Quantitative Studies, Vol. 29 No. 3, pp. 234-256, doi: 10.1108/JDQS-07-2020-0016.
Yoo, J.H., Kang, J.Y. and Park, S.U. (2018), “Measuring return and volatility spillovers across major virtual currency market”, The Journal of Information Systems, Vol. 27 No. 3, pp. 43-62, doi: 10.5859/KAIS.2018.27.3.43.