How do uncertainties affect the connectedness of global financial markets? Changes during the Russia-Ukraine conflict

ABSTRACTWe utilize the spectral decomposition of TVP-VAR connectedness to examine the dynamics of connectedness among six global financial markets. Additionally, we employ dynamic model averaging with retrospective analysis to ascertain the impact of uncertainties on the connectedness network. The findings reveal a shift from a declining trend in both total and directional connectedness to an ascending trend during the Russia-Ukraine conflict. The US dollar has the largest outward and inward spillovers in the short-term, but G7 MSCI and EFM MSCI are the largest outward and inward spillovers in the medium- and long-term, respectively. Among the five uncertainties under study, the financial stress index and VIX consistently hold significant influence throughout the sample period. Meanwhile, geopolitical risk and Twitter-based economic uncertainty demonstrate significance during the conflict period. Nonetheless, the impacts of these uncertainties diverge. The financial stress index and Twitter-based economic uncertainty exhibit positive effects, whereas VIX and geopolitical risk tend to weaken connectedness. Our findings underscore the need for investors to remain cautious of shifts in market connectedness patterns as they manage their assets.KEYWORDS: Russia-Ukraine conflictuncertaintydynamic connectednessspectral decompositiondynamic model averagingJEL CLASSIFICATION: G01G15C32C11 Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1. See report by ELPAIS: https://english.elpais.com/international/2022-03-03/ukrainian-exodus-could-be-europes-biggest-refugee-crisis-since-world-conflict-ii.html.2. We use ‘risk’ and ‘uncertainty’ interchangeably in this paper.3. For example, the J.P. Morgan Global Research: https://www.jpmorgan.com/insights/research/russia-ukraine-crisis-market-impact.4. This was reported, for example, in a report in CNBC https://www.cnbc.com/2022/02/24/forex-markets-russia-ukraine-invasion-euro-dollar.html.5. There is an alternative approach to estimating a TVP-VAR model, which is based on the Kalman Filter (Antonakakis et al. Citation2018). However, the MCMC procedure for the TVP-VAR approach could help us identify the confidence interval of the estimated parameters. Appendix A1 shows the details about the identification procedure of confidence intervals. Additionally, to reduce the autocorrelations of the draws in the MCMC process, we take every tenth draws in the MCMC process as the effective draws.6. The borders a and b define the frequency domains of short-term and ML-term. In this study, the borders of the short-term are [1, 5], and the borders of the ML-term are [6, 126]. The H chosen is 126, which is the number of trading days in a half-year.7. There is a term called ‘pairwise connectedness’ that measures the spillovers from one market to another market. In this paper, we do not discuss this measure because it does not help understand the effects of the potential determinants on the role of one specific market in the spillover network.8. We choose February 23, 2022 as the end of the sample period by considering the data available.9. The WTI crude oil spot price is from the US Energy Information Administration; the nominal broad US dollar index is collected from the Federal Reserve Bank of St. Louis; daily indexes of G7 MSCI and EFM MSCI are collected from the MSCI website; the global-aggregate bond total return index is collected from Bloomberg; and the global gold price is collected from the World Gold Council.10. Here and thereafter, the level rates are the logarithm of the initial price series.11. The VIX is collected from the CBOE website. The daily index of the US EPU is collected from https://www.policyuncertainty.com/us_monthly.html. The daily GPR index is collected from https://www.matteoiacoviello.com/gpr.htm. The TEU is collected from https://www.policyuncertainty.com/twitter_uncert.html. The FSI is the OFR financial stress index collected from office of financial research https://www.financialresearch.gov/financial-stress-index/.12. The ID-EMV is collected from https://www.policyuncertainty.com/EMV_monthly.html.13. The EFFR is collected from the Federal Reserve Bank of St. Louis.14. We do not impose any transformations on the FSI and EFFR.15. In our practice, we estimated the connectedness based on a TVP-VAR(2) model.16. Recall that we estimated the TVP-VAR for the period from January 1, 2021 to February 21, 2022, which contains 426 trading days.17. EquationEquation (5)(5) Fromconnectedness:Ci←⋅,t(a,b)=Σj=1,j≠inθ˜ij,t(a,b),(5) claims the from connectedness has a positive value. To compare the to and from connectedness within the same plot, we take the opposite of the from connectedness when drawing the figure.18. Considering that the short-term connectedness measures the market linkage by forecasting the shock responses for five days, we choose 0.871 as the decay factor in the DMA when estimating the short-term total connectedness model, which assumes the half-live of the data memory is five days. Similarly, we choose 0.966 as the decay factor when estimating the ML-term total connectedness model, which assumes the half-live of the data memory is 20 days. When modeling the aggregated total connectedness, we assume the half-live of the data memory is 10 days, and choose 0.933 as the decay factor.19. For brevity purposes, the expected coefficients are presented in Appendixes A2 and A3.Additional informationFundingThis research is supported by the Humanity and Social Science Youth Foundation of Ministry of Education of China (Grant No. 21YJC790043), the Major Project of Philosophical and Social Science Research in Hubei Universities (Grant No. 22ZD016), the Fundamental Research Funds for the Central Universities (Zhongnan University of Economics and Law: 2722023DK064) and Shandong Provincial Natural Science Foundation (Grant No. ZR2022QG069).