Introduction Methodological Discussion Data Results Conclusion Bonus! Exploring EME Asset Returns Conditional Comovement Nico Katzke Equity Quantitative Analyst, Fairtree Capital [email protected] July 2015 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Table of contents 1 Introduction Overview Plots Main Idea 2 Methodological Discussion Approach DCC ADCC GoGARCH Which model should be preferred? 3 Data 4 Results Statistics Fitting Procedures Graphs Variables Used Regression Outputs 5 6 Conclusion Bonus! Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Introduction In this paper, I explore the dynamics of comovement between emerging market asset and currency returns series since 2000. The main idea is to explore which events and / or factors contributed to the increased comovement of returns. For emerging market securities and currencies, there has been a steady increase in comovement since 2000, characterized by several events that drove return comovements higher. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Introduction In this paper, I explore the dynamics of comovement between emerging market asset and currency returns series since 2000. The main idea is to explore which events and / or factors contributed to the increased comovement of returns. For emerging market securities and currencies, there has been a steady increase in comovement since 2000, characterized by several events that drove return comovements higher. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Introduction In this paper, I explore the dynamics of comovement between emerging market asset and currency returns series since 2000. The main idea is to explore which events and / or factors contributed to the increased comovement of returns. For emerging market securities and currencies, there has been a steady increase in comovement since 2000, characterized by several events that drove return comovements higher. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Logarithmicplot of EME indexes 9 8 7 6 5 4 3 0 100 200 300 400 Nico Katzke 500 600 700 Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Main Idea The main idea is to isolate the bivariate return comovements, aggregate the bivariate pairs of return comovement, and then use a panel regression to assess which factors significantly influence the comovements between the return series collectively. This will also allow us to gain insight into which factors had been contemporaneous to return comovements as a whole. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Main Idea The main idea is to isolate the bivariate return comovements, aggregate the bivariate pairs of return comovement, and then use a panel regression to assess which factors significantly influence the comovements between the return series collectively. This will also allow us to gain insight into which factors had been contemporaneous to return comovements as a whole. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Main Idea The main idea is to isolate the bivariate return comovements, aggregate the bivariate pairs of return comovement, and then use a panel regression to assess which factors significantly influence the comovements between the return series collectively. This will also allow us to gain insight into which factors had been contemporaneous to return comovements as a whole. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Overview Plots Main Idea Main Idea The main idea is to isolate the bivariate return comovements, aggregate the bivariate pairs of return comovement, and then use a panel regression to assess which factors significantly influence the comovements between the return series collectively. This will also allow us to gain insight into which factors had been contemporaneous to return comovements as a whole. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling The ability to dynamically and jointly model the full multivariate density dynamics of multiple returns series has very important implications for risk and portfolio management. In particular, the growing literature on MV volatility models allow modelers to efficiently isolate the correlation components from the second order moments. The most widely used methods of studying second order persistence: GARCH models. Generalizing the univariate GARCH processes to the multivariate framework has proven to be challenging due to curse of dimensionality. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling The ability to dynamically and jointly model the full multivariate density dynamics of multiple returns series has very important implications for risk and portfolio management. In particular, the growing literature on MV volatility models allow modelers to efficiently isolate the correlation components from the second order moments. The most widely used methods of studying second order persistence: GARCH models. Generalizing the univariate GARCH processes to the multivariate framework has proven to be challenging due to curse of dimensionality. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling The ability to dynamically and jointly model the full multivariate density dynamics of multiple returns series has very important implications for risk and portfolio management. In particular, the growing literature on MV volatility models allow modelers to efficiently isolate the correlation components from the second order moments. The most widely used methods of studying second order persistence: GARCH models. Generalizing the univariate GARCH processes to the multivariate framework has proven to be challenging due to curse of dimensionality. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling The ability to dynamically and jointly model the full multivariate density dynamics of multiple returns series has very important implications for risk and portfolio management. In particular, the growing literature on MV volatility models allow modelers to efficiently isolate the correlation components from the second order moments. The most widely used methods of studying second order persistence: GARCH models. Generalizing the univariate GARCH processes to the multivariate framework has proven to be challenging due to curse of dimensionality. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling Generalization of univariate GARCH models to the multivariate domain is a conceptually simple exercise: Given the stochastic process, xt {t = 1, 2, ...T } of financial returns with dimension N × 1 and mean vector µt , given the information set It−1 : xt |It−1 = µt + εt (1) where the residuals of the process are modelled as: εt = Ht 1/2 zt , Nico Katzke (2) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling Generalization of univariate GARCH models to the multivariate domain is a conceptually simple exercise: Given the stochastic process, xt {t = 1, 2, ...T } of financial returns with dimension N × 1 and mean vector µt , given the information set It−1 : xt |It−1 = µt + εt (1) where the residuals of the process are modelled as: εt = Ht 1/2 zt , Nico Katzke (2) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling Generalization of univariate GARCH models to the multivariate domain is a conceptually simple exercise: Given the stochastic process, xt {t = 1, 2, ...T } of financial returns with dimension N × 1 and mean vector µt , given the information set It−1 : xt |It−1 = µt + εt (1) where the residuals of the process are modelled as: εt = Ht 1/2 zt , Nico Katzke (2) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling 1/2 Ht is a N × N positive definite matrix such that Ht is the conditional covariance matrix of xt . zt is a N × 1 i.i.d. N(0,1) series. Several techniques have been proposed that map the Ht matrix into the multivariate plain: VECH, BEKK, CCC-GARCH, DCC-GARCH, GOGARCH, etc. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling A large literature has developed that attempts to map Ht into the multivariate space, But MV Volatility models typically suffer from curse of high dimensionality. Particularly in this study, as I am looking at 21 series, it leads to a variance-covariance matrix that becomes very large (21 × 21 matrix). Studying correlation between these series, implies a total of 210 unique bivariate combinations. This requires a parsimonious and effective means of isolating the conditional comovements. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling A large literature has developed that attempts to map Ht into the multivariate space, But MV Volatility models typically suffer from curse of high dimensionality. Particularly in this study, as I am looking at 21 series, it leads to a variance-covariance matrix that becomes very large (21 × 21 matrix). Studying correlation between these series, implies a total of 210 unique bivariate combinations. This requires a parsimonious and effective means of isolating the conditional comovements. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling A large literature has developed that attempts to map Ht into the multivariate space, But MV Volatility models typically suffer from curse of high dimensionality. Particularly in this study, as I am looking at 21 series, it leads to a variance-covariance matrix that becomes very large (21 × 21 matrix). Studying correlation between these series, implies a total of 210 unique bivariate combinations. This requires a parsimonious and effective means of isolating the conditional comovements. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling A large literature has developed that attempts to map Ht into the multivariate space, But MV Volatility models typically suffer from curse of high dimensionality. Particularly in this study, as I am looking at 21 series, it leads to a variance-covariance matrix that becomes very large (21 × 21 matrix). Studying correlation between these series, implies a total of 210 unique bivariate combinations. This requires a parsimonious and effective means of isolating the conditional comovements. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling A large literature has developed that attempts to map Ht into the multivariate space, But MV Volatility models typically suffer from curse of high dimensionality. Particularly in this study, as I am looking at 21 series, it leads to a variance-covariance matrix that becomes very large (21 × 21 matrix). Studying correlation between these series, implies a total of 210 unique bivariate combinations. This requires a parsimonious and effective means of isolating the conditional comovements. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Multivariate Volatility Modelling For this reason, two methods will be used that provide convincing and effective means of isolating conditional correlations from the Ht matrix. DCC-GARCH model Generalized Orthogonal (GO)-GARCH model. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? DCC GARCH Engle (2002) and Tse & Tsuy (2002) introduced a class of models that allow the direct estimation of time-varying conditional correlation estimates. The DCC model can be defined as: Ht = Dt .Rt .Dt . (3) Equation 3 splits the varcovar matrix into identical diagonal matrices and an estimate of the time-varying correlation. Estimating RT requires it to be inverted at each estimated period, and thus a proxy equation is used (see Engle, 2002): Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH DCC equation: 0 Qij,t = Q̄ + a zt−1 zt−1 − Q̄ + b Qij,t−1 − Q̄ (4) 0 = (1 − a − b)Q̄ + azt−1 zt−1 + b.Qij,t−1 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH DCC equation: 0 Qij,t = Q̄ + a zt−1 zt−1 − Q̄ + b Qij,t−1 − Q̄ (4) 0 = (1 − a − b)Q̄ + azt−1 zt−1 + b.Qij,t−1 Note that equation 4 is similar in form to a GARCH(1,1) process, with non-negative scalars a and b, and with: Qij,t the unconditional (sample) variance estimate between series i and j, Q̄ the unconditional matrix of standardized residuals from each univariate pair estimate. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH DCC equation: 0 − Q̄ + b Qij,t−1 − Q̄ Qij,t = Q̄ + a zt−1 zt−1 (4) 0 = (1 − a − b)Q̄ + azt−1 zt−1 + b.Qij,t−1 We then use eq. 4 to estimate the Rt = diag (Qt )−1/2 Qt .diag (Qt )−1/2 . (5) Which has bivariate elements: Rt = ρij,t = √ Nico Katzke qi,j,t qii,t .qjj,t (6) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH With the resulting DCC model formulated as: εt ∼ N(0, Dt .Rt .Dt ) Dt2 ∼ Univariate GARCH(1,1) processes ∀ (i,j), i 6= j zt = Dt−1 .εt Qt = Q̄(1 − a − b) + a(zt0 zt ) + b(Qt−1 ) Rt = Diag (Qt−1 ).Qt .Diag (Qt −1 ) (7) Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH The time-varying conditional correlations are then estimated using the following log-likelihood function,which can be convincingly and parsimoniously decomposed into volatility and correlation components, with the latter then estimated directly: LL = T 1 X N log (2π) + 2 log |Dt | + log |Rt | + zt0 .Rt−1 .zt0 2 i=1 = T 1 X N log (2π) + 2 log |Dt | + ε0t .Dt−1 .Dt−1 .εt 2 i=1 − T 1 X 0 z zt + log |Rt | + zt0 .Rt−1 zt0 2 i=1 t = LLV (θ1 ) + LLR (θ1 , θ2 ) Nico Katzke (8) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH The time-varying conditional correlations are then estimated using the following log-likelihood function,which can be convincingly and parsimoniously decomposed into volatility and correlation components, with the latter then estimated directly: LL = T 1 X N log (2π) + 2 log |Dt | + log |Rt | + zt0 .Rt−1 .zt0 2 i=1 = T 1 X N log (2π) + 2 log |Dt | + ε0t .Dt−1 .Dt−1 .εt 2 i=1 − T 1 X 0 z zt + log |Rt | + zt0 .Rt−1 zt0 2 i=1 t = LLV (θ1 ) + LLR (θ1 , θ2 ) Nico Katzke (8) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH The time-varying conditional correlations are then estimated using the following log-likelihood function,which can be convincingly and parsimoniously decomposed into volatility and correlation components, with the latter then estimated directly: LL = T 1 X N log (2π) + 2 log |Dt | + log |Rt | + zt0 .Rt−1 .zt0 2 i=1 = T 1 X N log (2π) + 2 log |Dt | + ε0t .Dt−1 .Dt−1 .εt 2 i=1 − T 1 X 0 z zt + log |Rt | + zt0 .Rt−1 zt0 2 i=1 t = LLV (θ1 ) + LLR (θ1 , θ2 ) Nico Katzke (8) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 1: DCC GARCH The time-varying conditional correlations are then estimated using the following log-likelihood function,which can be convincingly and parsimoniously decomposed into volatility and correlation components, with the latter then estimated directly: LL = T 1 X N log (2π) + 2 log |Dt | + log |Rt | + zt0 .Rt−1 .zt0 2 i=1 = T 1 X N log (2π) + 2 log |Dt | + ε0t .Dt−1 .Dt−1 .εt 2 i=1 − T 1 X 0 z zt + log |Rt | + zt0 .Rt−1 zt0 2 i=1 t = LLV (θ1 ) + LLR (θ1 , θ2 ) Nico Katzke (8) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 2: ADCC GARCH The second model used to extract time-varying conditional correlation estimates is the ADCC-GARCH model. The specification generalizes the DCC model to control for leverage effects of the standardized residuals: zt0− . ADCC model: εt ∼ N(0, Dt .Rt .Dt ) Dt2 ∼ Univariate GARCH(1,1) processes ∀ (i,j), i 6= j zt = Dt−1 .εt Qt = Q̄(1 − a − b − G ) + a(zt0 zt ) + b(Qt−1 ) + G 0 zt− zt0− G Rt = Diag (Qt−1 ).Qt .Diag (Qt −1 ) Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 2: ADCC GARCH The second model used to extract time-varying conditional correlation estimates is the ADCC-GARCH model. The specification generalizes the DCC model to control for leverage effects of the standardized residuals: zt0− . ADCC model: εt ∼ N(0, Dt .Rt .Dt ) Dt2 ∼ Univariate GARCH(1,1) processes ∀ (i,j), i 6= j zt = Dt−1 .εt Qt = Q̄(1 − a − b − G ) + a(zt0 zt ) + b(Qt−1 ) + G 0 zt− zt0− G Rt = Diag (Qt−1 ).Qt .Diag (Qt −1 ) Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 2: ADCC GARCH The second model used to extract time-varying conditional correlation estimates is the ADCC-GARCH model. The specification generalizes the DCC model to control for leverage effects of the standardized residuals: zt0− . ADCC model: εt ∼ N(0, Dt .Rt .Dt ) Dt2 ∼ Univariate GARCH(1,1) processes ∀ (i,j), i 6= j zt = Dt−1 .εt Qt = Q̄(1 − a − b − G ) + a(zt0 zt ) + b(Qt−1 ) + G 0 zt− zt0− G Rt = Diag (Qt−1 ).Qt .Diag (Qt −1 ) Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model Another different approach to estimating second order persistence models is based on the assumption that returns are generated by a set of unobserved orthogonal conditionally heteroskedastic factors (c.f. van der Weide (2002), Boswijk & van der Weide (2006, 2011)). These are measured by identifying independent and uncorrelated factors that make up the var-covar matrix Ht . The statistical transformations are done as follows: rt = µt + εt (10) εt = A.ft (11) with A linking the unobserved uncorrelated components with the observed residual process. A: constant and invertible ; ft : normalized, unit variance. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model Another different approach to estimating second order persistence models is based on the assumption that returns are generated by a set of unobserved orthogonal conditionally heteroskedastic factors (c.f. van der Weide (2002), Boswijk & van der Weide (2006, 2011)). These are measured by identifying independent and uncorrelated factors that make up the var-covar matrix Ht . The statistical transformations are done as follows: rt = µt + εt (10) εt = A.ft (11) with A linking the unobserved uncorrelated components with the observed residual process. A: constant and invertible ; ft : normalized, unit variance. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model Another different approach to estimating second order persistence models is based on the assumption that returns are generated by a set of unobserved orthogonal conditionally heteroskedastic factors (c.f. van der Weide (2002), Boswijk & van der Weide (2006, 2011)). These are measured by identifying independent and uncorrelated factors that make up the var-covar matrix Ht . The statistical transformations are done as follows: rt = µt + εt (10) εt = A.ft (11) with A linking the unobserved uncorrelated components with the observed residual process. A: constant and invertible ; ft : normalized, unit variance. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model Another different approach to estimating second order persistence models is based on the assumption that returns are generated by a set of unobserved orthogonal conditionally heteroskedastic factors (c.f. van der Weide (2002), Boswijk & van der Weide (2006, 2011)). These are measured by identifying independent and uncorrelated factors that make up the var-covar matrix Ht . The statistical transformations are done as follows: rt = µt + εt (10) εt = A.ft (11) with A linking the unobserved uncorrelated components with the observed residual process. A: constant and invertible ; ft : normalized, unit variance. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model Also: ft represents the unobserved independent factors assigned to each series (factor weights), such that: ft = H 1/2 .zt (12) with HT and zt as before and: E [ft ] = 0; E [ft ft0 ] = IN E [εt ] = 0; E [εt ε0t ] = AA0 (13) So that the conditional covariance matrix is given by Σt = AHt A0 Nico Katzke (14) Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model What is happening is that we are using the residual series of a simple mean return series to decompose the var-covar matrix into a factor loading matrix, A, which is made up of: a sample (unconditional) covariance estimate Σ1/2 ; an orthogonal matrix, U, mapping the uncorrelated factors onto the covar matrix: estimated using Independent Component Analysis (ICA). Thus it assumes the variance process is driven by independent orthogonal latent components. The aim is then to model the independent factors as independent univariate GARCH processes. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model What is happening is that we are using the residual series of a simple mean return series to decompose the var-covar matrix into a factor loading matrix, A, which is made up of: a sample (unconditional) covariance estimate Σ1/2 ; an orthogonal matrix, U, mapping the uncorrelated factors onto the covar matrix: estimated using Independent Component Analysis (ICA). Thus it assumes the variance process is driven by independent orthogonal latent components. The aim is then to model the independent factors as independent univariate GARCH processes. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model What is happening is that we are using the residual series of a simple mean return series to decompose the var-covar matrix into a factor loading matrix, A, which is made up of: a sample (unconditional) covariance estimate Σ1/2 ; an orthogonal matrix, U, mapping the uncorrelated factors onto the covar matrix: estimated using Independent Component Analysis (ICA). Thus it assumes the variance process is driven by independent orthogonal latent components. The aim is then to model the independent factors as independent univariate GARCH processes. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model What is happening is that we are using the residual series of a simple mean return series to decompose the var-covar matrix into a factor loading matrix, A, which is made up of: a sample (unconditional) covariance estimate Σ1/2 ; an orthogonal matrix, U, mapping the uncorrelated factors onto the covar matrix: estimated using Independent Component Analysis (ICA). Thus it assumes the variance process is driven by independent orthogonal latent components. The aim is then to model the independent factors as independent univariate GARCH processes. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model What is happening is that we are using the residual series of a simple mean return series to decompose the var-covar matrix into a factor loading matrix, A, which is made up of: a sample (unconditional) covariance estimate Σ1/2 ; an orthogonal matrix, U, mapping the uncorrelated factors onto the covar matrix: estimated using Independent Component Analysis (ICA). Thus it assumes the variance process is driven by independent orthogonal latent components. The aim is then to model the independent factors as independent univariate GARCH processes. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model An example of a Normally distributed GoGARCH(1,1) model is: εt = A.ft = Σ1/2 .U (15) With each uncorrelated component driven by a GARCH(1,1) process: Ht = diag (h1,t , ..., hN,t ), (16) (with hi : conditional variances of the factors, ft ). 2 hi,t = γi + αi .fi,t−1 + βi .hi,t−1 , ∀i (17) The conditional covariances of εt are then driven by: Vt = ZHt Z Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model An example of a Normally distributed GoGARCH(1,1) model is: εt = A.ft = Σ1/2 .U (15) With each uncorrelated component driven by a GARCH(1,1) process: Ht = diag (h1,t , ..., hN,t ), (16) (with hi : conditional variances of the factors, ft ). 2 hi,t = γi + αi .fi,t−1 + βi .hi,t−1 , ∀i (17) The conditional covariances of εt are then driven by: Vt = ZHt Z Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model An example of a Normally distributed GoGARCH(1,1) model is: εt = A.ft = Σ1/2 .U (15) With each uncorrelated component driven by a GARCH(1,1) process: Ht = diag (h1,t , ..., hN,t ), (16) (with hi : conditional variances of the factors, ft ). 2 hi,t = γi + αi .fi,t−1 + βi .hi,t−1 , ∀i (17) The conditional covariances of εt are then driven by: Vt = ZHt Z Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The most widely used statistical technique for orthogonal transformation (required here to estimate U): PCA. For non-Gaussian data (which return series normally correspond to), we can use Independent Component Analysis (ICA). ICA separates a multivariate signal (x1 , ..., xN ) into maximally independent (non-Gaussian) additive subcomponents,(s1 , ...sN ), so that: x = B.s si are thus independent factors estimated using the FastICA technique proposed by Hyvarinen and Oja (2000), which does not require Gaussianity or distributional assumption of the series. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The most widely used statistical technique for orthogonal transformation (required here to estimate U): PCA. For non-Gaussian data (which return series normally correspond to), we can use Independent Component Analysis (ICA). ICA separates a multivariate signal (x1 , ..., xN ) into maximally independent (non-Gaussian) additive subcomponents,(s1 , ...sN ), so that: x = B.s si are thus independent factors estimated using the FastICA technique proposed by Hyvarinen and Oja (2000), which does not require Gaussianity or distributional assumption of the series. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The most widely used statistical technique for orthogonal transformation (required here to estimate U): PCA. For non-Gaussian data (which return series normally correspond to), we can use Independent Component Analysis (ICA). ICA separates a multivariate signal (x1 , ..., xN ) into maximally independent (non-Gaussian) additive subcomponents,(s1 , ...sN ), so that: x = B.s si are thus independent factors estimated using the FastICA technique proposed by Hyvarinen and Oja (2000), which does not require Gaussianity or distributional assumption of the series. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The two-step FastICA estimation procedure basically works as follows: ˆ .εˆ , with the standardized residual process is decomposed into: zt = Σ−1/2 t ˆ Σ−1/2 obtained from PCA type eigenvalue decomposition. Step two uses the estimate of the independent factor components, ft , so obtained to estimate (using the relatively strong assumption of independence of the factors) the likelihood function of the GoGARCH model, where the conditional log-likelihood is expressed as the sum of the individual conditional marginal density log-likelihoods of the individual factor components in step 1. see Hyvarinen and Oja (2002) for details. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The two-step FastICA estimation procedure basically works as follows: ˆ .εˆ , with the standardized residual process is decomposed into: zt = Σ−1/2 t ˆ Σ−1/2 obtained from PCA type eigenvalue decomposition. Step two uses the estimate of the independent factor components, ft , so obtained to estimate (using the relatively strong assumption of independence of the factors) the likelihood function of the GoGARCH model, where the conditional log-likelihood is expressed as the sum of the individual conditional marginal density log-likelihoods of the individual factor components in step 1. see Hyvarinen and Oja (2002) for details. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The two-step FastICA estimation procedure basically works as follows: ˆ .εˆ , with the standardized residual process is decomposed into: zt = Σ−1/2 t ˆ Σ−1/2 obtained from PCA type eigenvalue decomposition. Step two uses the estimate of the independent factor components, ft , so obtained to estimate (using the relatively strong assumption of independence of the factors) the likelihood function of the GoGARCH model, where the conditional log-likelihood is expressed as the sum of the individual conditional marginal density log-likelihoods of the individual factor components in step 1. see Hyvarinen and Oja (2002) for details. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The two-step FastICA estimation procedure basically works as follows: ˆ .εˆ , with the standardized residual process is decomposed into: zt = Σ−1/2 t ˆ Σ−1/2 obtained from PCA type eigenvalue decomposition. Step two uses the estimate of the independent factor components, ft , so obtained to estimate (using the relatively strong assumption of independence of the factors) the likelihood function of the GoGARCH model, where the conditional log-likelihood is expressed as the sum of the individual conditional marginal density log-likelihoods of the individual factor components in step 1. see Hyvarinen and Oja (2002) for details. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Model 3: GoGARCH model The two-step FastICA estimation procedure basically works as follows: ˆ .εˆ , with the standardized residual process is decomposed into: zt = Σ−1/2 t ˆ Σ−1/2 obtained from PCA type eigenvalue decomposition. Step two uses the estimate of the independent factor components, ft , so obtained to estimate (using the relatively strong assumption of independence of the factors) the likelihood function of the GoGARCH model, where the conditional log-likelihood is expressed as the sum of the individual conditional marginal density log-likelihoods of the individual factor components in step 1. see Hyvarinen and Oja (2002) for details. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Main drawbacks of the models The DCC model estimations allow for the direct modelling of the conditional correlation. Although efficient, this method has some drawbacks, as e.g. it assumes a constant structure to the correlation dynamics. The GoGARCH model uses a highly efficient and less parameter intensive estimation technique to decompose the var-covar matrix into orthogonal sources of volatility, which could then be calculated using an additive log-likelihood approach because of the independence of volatility factors. Although computationally highly efficient, the assumption of linearly indendent volatility factors is strong. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Main drawbacks of the models The DCC model estimations allow for the direct modelling of the conditional correlation. Although efficient, this method has some drawbacks, as e.g. it assumes a constant structure to the correlation dynamics. The GoGARCH model uses a highly efficient and less parameter intensive estimation technique to decompose the var-covar matrix into orthogonal sources of volatility, which could then be calculated using an additive log-likelihood approach because of the independence of volatility factors. Although computationally highly efficient, the assumption of linearly indendent volatility factors is strong. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Main drawbacks of the models The DCC model estimations allow for the direct modelling of the conditional correlation. Although efficient, this method has some drawbacks, as e.g. it assumes a constant structure to the correlation dynamics. The GoGARCH model uses a highly efficient and less parameter intensive estimation technique to decompose the var-covar matrix into orthogonal sources of volatility, which could then be calculated using an additive log-likelihood approach because of the independence of volatility factors. Although computationally highly efficient, the assumption of linearly indendent volatility factors is strong. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Main drawbacks of the models The DCC model estimations allow for the direct modelling of the conditional correlation. Although efficient, this method has some drawbacks, as e.g. it assumes a constant structure to the correlation dynamics. The GoGARCH model uses a highly efficient and less parameter intensive estimation technique to decompose the var-covar matrix into orthogonal sources of volatility, which could then be calculated using an additive log-likelihood approach because of the independence of volatility factors. Although computationally highly efficient, the assumption of linearly indendent volatility factors is strong. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Main drawbacks of the models As all the approaches have some benefits and drawbacks not discussed here in detail, I will use all three the DCC, ADCC and GoGARCH (FastICA) models to calculate the dynamic conditional moment estimates. This should provide insight into which periods led to increased comovements (and to what degree) of EME returns over the last decade. These estimates will then be used in a fixed-effects panel regression to identify the influence of common global factors on aggregate bivariate return comovement between emerging market series. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Main drawbacks of the models As all the approaches have some benefits and drawbacks not discussed here in detail, I will use all three the DCC, ADCC and GoGARCH (FastICA) models to calculate the dynamic conditional moment estimates. This should provide insight into which periods led to increased comovements (and to what degree) of EME returns over the last decade. These estimates will then be used in a fixed-effects panel regression to identify the influence of common global factors on aggregate bivariate return comovement between emerging market series. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Approach DCC ADCC GoGARCH Which model should be preferred? Main drawbacks of the models As all the approaches have some benefits and drawbacks not discussed here in detail, I will use all three the DCC, ADCC and GoGARCH (FastICA) models to calculate the dynamic conditional moment estimates. This should provide insight into which periods led to increased comovements (and to what degree) of EME returns over the last decade. These estimates will then be used in a fixed-effects panel regression to identify the influence of common global factors on aggregate bivariate return comovement between emerging market series. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Statistics Data and Frequency The data consists of the weekly aggregate equity market indexes (TRI, and denominated in Dollars), of 21 EME markets as defined by the MSCI EME index. It also consists of the currencies of these countries relative to the US Dollar. The period is from 2000/01/21 to 2015/02/27 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Statistics Statistics Descriptive Statistics Min Max Stdev Skewness Kurtosis LB(1)-pval LB(5)-pval ARCH-LM(1) Brazil 0.215 0.526 −33.056 25.617 5.255 −0.623 7.426 0.016 0.043 0.000 0.000 Chile 0.180 0.360 −34.607 19.134 3.356 −1.543 18.528 0.118 0.095 0.000 0.000 China 0.152 0.406 −22.110 17.937 4.231 −0.405 5.277 0.331 0.387 0.000 0.000 Colombia 0.316 0.368 −28.530 15.111 3.896 −0.895 8.703 0.160 0.000 0.054 0.000 Czech Egypt Greece Mean 0.262 Median 0.417 −29.812 21.635 3.888 −0.717 9.652 0.360 0.024 0.000 JB p-value 0.000 0.285 0.457 −21.922 15.513 4.274 −0.554 6.016 0.031 0.043 0.000 0.000 −0.209 −0.008 −29.108 18.765 5.172 −0.656 5.947 0.806 0.114 0.000 0.000 Hungary 0.061 0.289 −43.448 22.731 5.174 −0.977 10.781 0.610 0.072 0.000 0.000 India 0.258 0.466 −21.879 18.366 4.010 −0.464 5.616 0.063 0.005 0.000 0.000 Indonesia 0.296 0.387 −26.842 21.542 4.909 −0.227 5.918 0.907 0.000 0.000 0.000 Korea 0.218 0.447 −27.901 28.635 4.741 −0.279 7.151 0.052 0.309 0.000 0.000 0.276 −12.513 Malaysia 0.240 0.283 7.300 Mexico 0.249 0.511 −30.677 22.572 4.087 −0.622 10.348 0.012 0.001 0.000 0.000 Peru 0.312 0.440 −29.362 22.054 14.370 3.980 2.698 −0.535 8.802 0.424 0.103 0.414 0.009 0.000 0.003 0.000 0.000 Philippines 0.152 0.328 −20.803 15.242 3.482 −0.311 5.834 0.466 0.108 0.006 0.000 Poland 0.129 0.355 −26.459 24.067 4.622 −0.525 6.682 0.901 0.431 0.000 0.000 Russia 0.283 0.485 −28.062 44.899 5.971 −0.044 9.033 0.977 0.147 0.000 0.000 South.Africa 0.239 0.531 −18.425 27.601 3.937 −0.121 7.723 0.022 0.020 0.000 0.000 Taiwan 0.094 0.361 −14.386 19.364 3.653 −0.145 5.400 0.597 0.213 0.049 0.000 Thailand 0.237 0.348 −29.026 17.264 4.214 −0.470 7.239 0.527 0.000 0.004 0.000 Turkey 0.197 0.573 −73.767 38.610 6.971 −1.253 20.532 0.159 0.290 0.000 0.000 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Statistics Data The returns show typical returns series behaviour: excess kurtosis and skewness, as well as evidence of significant serial autopersistence as well as conditional heteroskedasticity. This motivates the use of the MV GARCH models employed to extract the second order persistence series. The data is also scaled by subtracting each series’ mean prior to estimating the models. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Statistics Data The returns show typical returns series behaviour: excess kurtosis and skewness, as well as evidence of significant serial autopersistence as well as conditional heteroskedasticity. This motivates the use of the MV GARCH models employed to extract the second order persistence series. The data is also scaled by subtracting each series’ mean prior to estimating the models. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Statistics Data The returns show typical returns series behaviour: excess kurtosis and skewness, as well as evidence of significant serial autopersistence as well as conditional heteroskedasticity. This motivates the use of the MV GARCH models employed to extract the second order persistence series. The data is also scaled by subtracting each series’ mean prior to estimating the models. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Statistics GoGARCH estimates: (Equities) F_1 omega F_2 F_3 F_4 F_5 F_6 F_7 F_8 F_9 F_10 0.07422551 0.007901069 0.02013517 0.05510071 8.672800e-04 0.09001604 0.02278301 0.02556973 0.03573081 0.003315177 0.0011 alpha1 0.11796657 0.054758822 0.05854870 0.10030617 2.578081e-09 0.07654145 0.08330919 0.08237253 0.03719492 0.058134152 0.0105 beta1 0.81151789 0.936294065 0.91978245 0.84566417 9.989999e-01 0.83264833 0.89327224 0.89545198 0.92673715 0.940865831 0.9873 omega 0.01935749 0.05709150 0.01980954 0.01706659 0.004233564 0.008182664 0.007875477 0.009033343 0.002555149 0.09481719 F_12 F_13 F_14 F_15 F_16 F_17 F_18 F_19 F_20 F_21 alpha1 0.04500304 0.04209236 0.05453154 0.08944808 0.056328861 0.069675464 0.049737695 0.029490802 0.039384955 0.19876847 beta1 0.93417703 0.90035262 0.92353795 0.89456896 0.939139202 0.924631787 0.939655777 0.960569894 0.959615035 0.71825986 ============================= GoGARCH Parameter estimates Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Statistics GoGARCH estimates: (Currencies) F_1 omega F_2 F_3 F_4 F_5 F_6 F_7 F_8 F_9 F_10 F_1 0.01623448 0.03099957 0.0924535 0.008274239 0.03796785 0.04458345 0.003321745 0.02055873 0.01543423 0.01628214 0.0245642 alpha1 0.03782264 0.12455670 0.2255800 0.082767297 0.13514496 0.10511909 0.042638709 0.18679680 0.09627177 0.14218538 0.1857924 beta1 0.94464983 0.85017091 0.7032481 0.914802500 0.83175104 0.85499662 0.970799878 0.81220318 0.88943244 0.81054104 0.8132074 omega 0.0920156 0.1972719 0.03257469 0.003670243 0.04927513 0.1935921 0.009981032 0.05914329 F_12 F_13 F_14 F_15 F_16 F_17 F_18 F_19 alpha1 0.3549753 0.2147538 0.06641944 0.140942882 0.07413239 0.2891759 0.077231573 0.24591772 beta1 0.6027820 0.6041746 0.90242859 0.858057117 0.87921705 0.5129753 0.914310758 0.71207333 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Fitting Procedures DCC Fitting: Assume all the series follow AR(1)-GJR GARCH processes, which controls for leverage in univariate GARCH specifications in the first step. Use a DCC(1,1) order, assuming a MV Normal distribution (for simplicity, as the Std-t adds a layer of parameter complexity which undermines accuracy for large N). GoGARCH Fitting: Assume the factor series all follow AR(1) GARCH(1,1) processes, with factor estimation carried out using the FastICA method used. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs DCC series: Dynamic Conditional Correlation Aggregates (Equities): DCC: 0.45 0.40 DCC mean estimates 0.50 Mean DCC estimates across countries ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ●● ●● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 05/2006 GFC BNP Greece QE2 Taper QE3 2000−01−21 2003−02−14 2006−03−10 2009−04−03 2012−04−27 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs DCC series: Dynamic Conditional Correlation Aggregates (Equities): ADCC: 0.45 0.40 DCC mean estimates 0.50 Mean DCC estimates across countries ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ●● ●● ●● ● ● ● ● ● ● ● ● ●● ●●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ●● ●● ●● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 05/2006 GFC BNP Greece QE2 Taper QE3 2000−01−21 2003−02−14 2006−03−10 2009−04−03 2012−04−27 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs DCC series: Dynamic Conditional Correlation Aggregates (Equities): GoGARCH: 0.9 Mean DCC estimates across countries ● ● ● ● 0.7 ● ● ● ● 0.6 0.5 0.4 ● ● ● 0.3 DCC mean estimates 0.8 ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ● ●● ● ●● ● ● ● ●●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ●● ●● ● ● ● ● ●● ● ● ● ●● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ●● ● ●● ● ●●● ●● ●● ●●● ● ● ● ●● ●● ●● ● ●● ● ● ●●● ● ● ● ● ● ●●●● ● ●● ● ●● ● ● ●●● ● ●● ●● ● ● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ●●● ● ●● ● ● ●● ● ● ●● ● ● ● ●● ●● ●● ● ●● ● ● ● ●● ● ● ● ● ●● ● ●● ●● ●●● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ●● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●●● ●●●● ● ●● ●●● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●●● ●● ●● ●● ● ●● ● ●●● ● ●● ●● ● ● ● ●● ● ● ● ●● ●●● ●● ● ●●● ● ●● ● ● ● ● ● ●● ● ●● ●● ● ● Taper ●●● ● ● QE3 BNP Greece QE2 ●●● ● ●GFC ●● ● ● ● ● 05/2006 ●● ●● ●● ● ● ● ●● ●● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ●●● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 2000−01−21 2003−02−14 2006−03−10 2009−04−03 2012−04−27 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs DCC series: Dynamic Conditional Correlation Aggregates (Currencies): DCC: 0.30 0.25 DCC mean estimates 0.35 Mean DCC estimates across countries ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●●● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ●● ●● ●● ●●● ● ● ●● ● ● ● ●●● ● ● ● ● ●●● ● ● ●● ●● ● ● ● ●●●● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ●● ● ● ● ● ●● ●● ● ● ● ●● ● ●● ● ●● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ●● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● 05/2006 GFC BNP Greece QE2 Taper QE3 2000−01−21 2003−02−21 2006−03−24 2009−04−24 2012−05−25 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs DCC series: Dynamic Conditional Correlation Aggregates (Currencies): ADCC: 0.30 0.25 DCC mean estimates 0.35 Mean DCC estimates across countries ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ●● ●● ● ●● ●● ● ● ● ●● ●● ●●● ● ● ● ●● ● ● ●● ●● ●● ● ●●● ●● ●● ●● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●● ●●● ● ●●● ● ●● ● ● ● ● ● ●● ●● ●● ●● ● ● ● ●● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ●● ● ●●● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● 05/2006 GFC BNP Greece QE2 Taper QE3 2000−01−21 2003−02−21 2006−03−24 2009−04−24 2012−05−25 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs DCC series: Dynamic Conditional Correlation Aggregates (Currencies): GoGARCH: 0.6 Mean DCC estimates across countries ● 0.5 ● ● ● ● ● 0.4 0.3 0.2 DCC mean estimates ● ● ● ●●● ●● ● ● ●● ● ● ●● ● ●● ●● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●●● ● ●● ●● ● ● ●●● ● ●● ● ● ●●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●●● ● ●●●● ● ●● ● ● ●● ● ● ●●● ● ●● ● ● ● ●● ● ● ● ● ●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●●● ● ●● ● ● ●●● ● ● ● ● ● ●● ●● ●● ● ● ● ●●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ●●● ●●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ●● ●● ● ● ● ●● ●●●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ●●● ●● ● ● ●● ●● ● ● ● ● ● ● ●● ●● ●● ● ● ● ●● ●● ●● ● ● ● ● ●●● ● ●● ● ●● ●●● ● ●●●● ●● ● ●● ●● ●● ●● ●● ●●● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ●●●●● ●● ●● ●● ● ● ● ●● ●● ● ●● ● ●● ●●● ● ● ● ● ● ● ●● ● ● ● ● ●● ●●● ● ● ●●●● ●● ● ● ●●● ● ● ●●● ●●●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ● ● ●●● ● ● ●● ● ● ● ● ●●● ● ●●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ●●● ●● ● ●● ● ●● ● ●● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ●● ●● ● ● ● ●●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●●● ●● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● Taper QE3 05/2006 BNP Greece QE2 GFC ● 2000−01−21 2003−02−21 2006−03−24 2009−04−24 2012−05−25 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs VIX Equity comovement with VIX index series: Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs VIX Equity comovement with VIX index series: DCC: Mean DCC estimates across countries 0.45 0.40 DCC mean estimates 0.50 80 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ●● ● ●● ● ●● ● ● ● ● ●● ●● ● ●● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 70 60 50 40 30 20 10 2000−01−21 2003−08−01 2007−02−09 2010−08−20 2014−02−28 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs VIX Equity comovement with VIX index series: ADCC: Mean DCC estimates across countries 0.50 0.45 0.40 DCC mean estimates 0.55 80 ● ● ● ● ● ● ● ● ● ● ● 70 ● ● ● ● 60 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ●● ● ● ●● ●●● ● ● ● ● ● ●●● ● ●● ● ●● ●● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ●● ●● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●● ● 50 40 30 20 10 2000−01−21 2003−08−01 2007−02−09 2010−08−20 2014−02−28 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs VIX Equity comovement with VIX index series: GoGARCH: 0.9 Mean DCC estimates across countries 80 ● ● ● ● 0.7 ● ● ● ● 60 ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●● ● ● ● ● ● ● ● ●●● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●●●● ● ● ●●●●● ● ●●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ●●● ● ● ●●● ● ● ●● ● ● ● ● ● ● ●●● ●● ●● ●●● ●● ● ● ●●● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ●● ● ● ●●● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●●● ● ●● ● ●● ●● ● ●● ●● ● ● ● ●● ●● ● ● ●● ● ●● ● ● ● ● ● ● ●●● ● ●●●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ●● ●● ● ● ●●● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ●● ● ● ● ●●● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ●●● ● ●● ●● ● ●●● ● ●● ●● ● ● ● ●● ●● ● ● ● ●● ●●● ●● ●● ● ●● ● ● ● ● ● ●●● ●● ●● ● ●● ● ●● ● ●●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ●●● ● ● 50 0.4 0.5 0.6 ● 0.3 DCC mean estimates 0.8 70 ● 40 30 20 10 2000−01−21 2003−08−01 2007−02−09 2010−08−20 2014−02−28 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs GSCI index Currency comovement with log(GSCI) series: Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs GSCI index Currency comovement with log(GSCI) series: DCC: 0.30 0.25 DCC mean estimates 0.35 Mean DCC estimates across countries ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ●● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ●● ●●● ● ● ●● ●● ●● ●●● ● ●●● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●●● ● ● ●● ● ● ●● ● ● ● ● ● ●● ●● ●● ● ●● ● ●●● ●● ●● ●● ● ● ● ● 6.5 6.0 5.5 2000−01−21 2003−08−15 2007−03−09 2010−10−01 2014−04−25 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs GSCI index Currency comovement with log(GSCI) series: ADCC: 0.30 0.25 DCC mean estimates 0.35 Mean DCC estimates across countries ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ●● ● ● ●● ● ● ● ●●● ●● ● ● ● ● ●● ●● ● ● ● ●● ●●● ● ● ●● ●● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● 6.5 6.0 5.5 2000−01−21 2003−08−15 2007−03−09 2010−10−01 2014−04−25 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs GSCI index Currency comovement with log(GSCI) series: GoGARCH: 0.6 Mean DCC estimates across countries ● 6.5 0.5 ● ● ● ● ● ● ● ●●● ● ● ●● ●● ● ● ●● ● ●● ●● ●●●●● ● ● ● ● ● ● ● ●● ● ● ● ●●● ● ● ● ●● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●●● ● ● ● ●● ● ● ●● ● ●●●● ● ● ● ●● ● ● ● ● ●●● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ●● ●● ● ●● ● ●● ● ● ● ● ●● ● ●●● ● ● ● ● ●● ● ● ●● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ●●● ● ● ●● ●● ●● ● ●● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ●● ●●●● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ●● ● ●● ●● ●●●● ●● ● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ●● ● ● ● ●● ● ●● ● ● ● ●● ● ●● ● ●●● ●● ● ●● ●● ● ●● ● ● ● ● ●● ● ●● ●●● ● ● ● ●● ● ●● ●● ●● ●● ●● ● ●● ● ● ● ● ● ● ● ●● ●●● ● ● ● ● ●● ● ● ● ● ●●● ● ● ●● ● ● ● ●● ●●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ●● ● ● ●● ●● ● ● ● ●● ● ● ● ● ●●● ●● ● ●● ●●●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ●●● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●●● ● ● ●● ● ● ● ●●● ● ●● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ●● ●● ● ●●● ●● ● ●● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ●● ●● ● ● ●● ● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● 0.3 0.4 ● 6.0 5.5 0.2 DCC mean estimates ● 2000−01−21 2003−08−15 2007−03−09 2010−10−01 2014−04−25 Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Variables used After fitting the three different models (which all adhere to their constraints and all converged), the next step is to create a Panel data set. Included in the Panel set are the following variables: QE1, QE2, QE3: all indicator variables identifying periods of US quantitative easing measures. GFC: Global Financial Crisis period (Lehman : 31/12/2009). AQE1, AQE2, AQE3: IVs corresponding to the two month period after each round of QE - so as to measure the unwinding effect of QE policies. VIX30, which corresponds to periods where VIX is above 30, indicating periods of high uncertainty in global markets, as proxied for by the CBOE VIX series. GSCI: the global commodity price index, used as a proxy for global price pressures. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Variables used After fitting the three different models (which all adhere to their constraints and all converged), the next step is to create a Panel data set. Included in the Panel set are the following variables: QE1, QE2, QE3: all indicator variables identifying periods of US quantitative easing measures. GFC: Global Financial Crisis period (Lehman : 31/12/2009). AQE1, AQE2, AQE3: IVs corresponding to the two month period after each round of QE - so as to measure the unwinding effect of QE policies. VIX30, which corresponds to periods where VIX is above 30, indicating periods of high uncertainty in global markets, as proxied for by the CBOE VIX series. GSCI: the global commodity price index, used as a proxy for global price pressures. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Variables used After fitting the three different models (which all adhere to their constraints and all converged), the next step is to create a Panel data set. Included in the Panel set are the following variables: QE1, QE2, QE3: all indicator variables identifying periods of US quantitative easing measures. GFC: Global Financial Crisis period (Lehman : 31/12/2009). AQE1, AQE2, AQE3: IVs corresponding to the two month period after each round of QE - so as to measure the unwinding effect of QE policies. VIX30, which corresponds to periods where VIX is above 30, indicating periods of high uncertainty in global markets, as proxied for by the CBOE VIX series. GSCI: the global commodity price index, used as a proxy for global price pressures. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Panel Fixed effect Panel approach used, as the means between the country estimates differ significantly: Equities (GoGARCH): 0.7 Heterogeineity across DCC estimates ● ● ● ● 0.6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.5 ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.3 0.4 ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● 0.2 DCC mean estimates ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● dcc1 dcc12 dcc142 dcc167 dcc19 dcc21 dcc41 dcc62 dcc83 Nico Katzke Date Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Panel Fixed effect Panel approach used, as the means between the country estimates differ significantly: Currencies (GoGARCH): Heterogeineity across DCC estimates 0.8 ● ● ● ● ● 0.6 ● ● ● ●● ● ● 0.4 ● ● ● 0.2 ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ●●● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● dcc1 dcc116 dcc136 dcc156 Nico Katzke ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● 0.0 DCC mean estimates ● ● ● ● ● ●● ●● ● ● ● ● ●●● ● dcc2 dcc36 dcc53 dcc7 dcc86 Date Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Panel: Equities Dependent variable: Aggregate Dynamic Conditional Comovement Estimates (DCC) (ADCC) (GoGARCH) (DCC) (ADCC) (GoGARCH) 0.038∗∗∗ 0.036∗∗∗ 0.060∗∗∗ 0.042∗∗∗ 0.041∗∗∗ 0.071∗∗∗ (0.0003) (0.0003) (0.001) (0.0003) (0.0003) (0.001) qe2 −0.009∗∗∗ −0.013∗∗∗ −0.029∗∗∗ (0.0004) (0.0004) (0.001) qe3 −0.029∗∗∗ −0.026∗∗∗ −0.033∗∗∗ (0.0003) (0.0003) (0.001) 0.011∗∗∗ 0.017∗∗∗ 0.048∗∗∗ (0.0005) (0.001) (0.001) −0.009∗∗∗ −0.005∗∗∗ 0.015∗∗∗ gfc aqe2 aqe3 0.049∗∗∗ (0.0005) (0.001) (0.001) 0.024∗∗∗ 0.040∗∗∗ 0.010∗∗∗ (0.0003) (0.0003) (0.001) (0.0003) (0.0003) (0.001) lgsci 0.040∗∗∗ 0.034∗∗∗ 0.092∗∗∗ 0.033∗∗∗ 0.027∗∗∗ 0.078∗∗∗ (0.0002) (0.0002) (0.0005) (0.0002) (0.0002) (0.0004) 165,690 Observations 0.027∗∗∗ 0.009∗∗∗ vix30 165,690 165,690 165,690 165,690 165,690 R2 0.351 0.294 0.270 0.315 0.270 Adjusted R2 0.350 0.294 0.269 0.315 0.270 0.266 17,864.670∗∗∗ 13,794.590∗∗∗ 12,220.070∗∗∗ 15,236.990∗∗∗ 12,264.970∗∗∗ 12,031.160∗∗∗ F Statistic (df = 5; 165475) Note: Nico Katzke 0.267 p<0.1; p<0.05; Returns p<0.01 Exploring EME Asset Conditional Comovement ∗ ∗∗ ∗∗∗ Introduction Methodological Discussion Data Results Conclusion Bonus! Fitting Procedures Graphs Variables Used Regression Outputs Panel: Currencies Dependent variable: Aggregate Dynamic Conditional Comovement Estimates (DCC) (ADCC) (GoGARCH) (DCC) (ADCC) (GoGARCH) 0.025∗∗∗ 0.022∗∗∗ −0.001 0.026∗∗∗ 0.023∗∗∗ 0.004∗∗∗ (0.0003) (0.0003) (0.001) (0.0003) (0.0003) (0.001) qe2 0.014∗∗∗ 0.014∗∗∗ −0.012∗∗∗ (0.0005) (0.0005) (0.001) qe3 −0.008∗∗∗ −0.012∗∗∗ −0.023∗∗∗ (0.0004) (0.0004) (0.001) 0.025∗∗∗ 0.024∗∗∗ 0.012∗∗∗ (0.001) (0.001) (0.001) −0.005∗∗∗ −0.011∗∗∗ 0.074∗∗∗ gfc aqe2 aqe3 −0.005∗∗∗ (0.001) (0.001) (0.001) 0.009∗∗∗ −0.001 0.005∗∗∗ (0.0003) (0.0003) (0.001) (0.0003) (0.0003) (0.001) lgsci 0.042∗∗∗ 0.045∗∗∗ 0.059∗∗∗ 0.040∗∗∗ 0.043∗∗∗ 0.052∗∗∗ (0.0002) (0.0002) (0.001) (0.0002) (0.0002) (0.0005) 136,287 Observations 0.007∗∗∗ 0.007∗∗∗ vix30 136,287 136,287 136,287 136,287 136,287 R2 0.251 0.290 0.089 0.256 0.291 Adjusted R2 0.251 0.289 0.089 0.256 0.291 0.103 9,143.967∗∗∗ 11,098.710∗∗∗ 2,668.757∗∗∗ 9,363.376∗∗∗ 11,167.790∗∗∗ 3,124.307∗∗∗ F Statistic (df = 5; 136111) Note: Nico Katzke 0.103 p<0.1; Asset p<0.05; p<0.01 Exploring EME Returns Conditional Comovement ∗ ∗∗ ∗∗∗ Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Models In conclusion, we see the following: The dynamics of the GoGARCH model are significantly more responsive to new information. This follows from the comparatively lower persistence parameters in the univariate GARCH estimates. In contrast, the aggregate joint correlation dynamics of the DCC model has a = 0.009 and b = 0.95 (a = 0.006, b = 0.932 and g = 0.012). This implies that the DCC model reacts slower to new information. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Models In conclusion, we see the following: The dynamics of the GoGARCH model are significantly more responsive to new information. This follows from the comparatively lower persistence parameters in the univariate GARCH estimates. In contrast, the aggregate joint correlation dynamics of the DCC model has a = 0.009 and b = 0.95 (a = 0.006, b = 0.932 and g = 0.012). This implies that the DCC model reacts slower to new information. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Panel Findings From the panel regressions, we note the following: Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Panel Findings From the panel regressions, we note the following: The aggregate bivariate return comovement of equities and currencies are roughly 40% and 30%, respectively. Currency returns (relative to the Dollar) are also less likely to spike significantly compared to equity returns. Significant spikes in equity comovement following: Greek bailout in 2010; Bernanke’s Taper talk in 2013 ; following the unwinding of QE3 GoGARCH picks up sustained higher comovement following QE2’s unwinding, with a spike in comovement of currencies again following unwinding of QE3. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Panel Findings From the panel regressions, we note the following: The GFC saw a significant rise in equity comovement (equities: between 4-7% rise). In contrast, the GFC had a comparatively moderate effect on currency comovement (currencies: between 1 - 2%, insignificant for GoGARCH estimates). Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Panel Findings From the panel regressions, we note the following: QE2 produced relatively little change in comovements, with it being negative for equities and slightly positive for currencies (between 1% and -1% for DCC and GoGarch estimates). During QE3, equity and currency comovements again experienced a relative dampening of comovements (roughly 3% for equities and 1-2%for currencies). Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Panel Findings From the panel regressions, we note the following: The two months following QE2 saw a modest rise in comovements of both eq’s and curr’s of roughly 2-4% for eq’s and 2% for currencies. The two month period following QE3 also saw a modest downward adjustment for eq’s and curr’s (DCC) and upward adjustment using the GoGARCH estimates. Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Panel Findings From the panel regressions, we note the following: Periods of large equity market uncertainty saw equity comovements rise significantly (1%, 2.7% and 4.9% for the DCC, ADCC & GoGARCH respectively) Periods of large equity market uncertainty saw no great influence on currency comovement (0.1%, 0.7% and -0.5% for the DCC, ADCC & GoGARCH respectively) Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Conclusion: Panel Findings From the panel regressions, we note the following: Commodity price movements saw equity comovements rise significantly (4%, 3% and 9% price elasticities for the DCC, ADCC & GoGARCH respectively) Commodity price movements saw currency comovements rise significantly(4%, 4.5% and 6% for the DCC, ADCC & GoGARCH respectively) Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! BONUS: BRICS equity return comovements Equity comovement: BRICS returns Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! BONUS: BRICS equity return comovements Equity comovement: BRICS returns DCC: 0.7 Mean DCC estimates across BRICS 0.6 0.5 0.4 DCC mean estimates ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●05/2006 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● GFC BNP Greece QE2 Taper QE3 0.2 0.3 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ●● ● ● ● 2000−01−21 2003−02−14 2006−03−10 2009−04−03 2012−04−27 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! BONUS: BRICS equity return comovements Equity comovement: BRICS returns ADCC: 0.7 Mean DCC estimates across BRICS 0.6 0.5 DCC mean estimates ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● 05/2006 ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● GFC BNP Greece QE2 Taper QE3 0.3 0.4 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● 2000−01−21 2003−02−14 2006−03−10 2009−04−03 2012−04−27 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! BONUS: BRICS equity return comovements Equity comovement: BRICS returns GoGARCH: Mean DCC estimates across BRICS 0.6 0.5 0.4 0.3 0.2 DCC mean estimates 0.7 0.8 ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ●● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ●● ● ● ●●●●● ●●● ● ● ● ●●● ●● ● ● ●● ● ● ●● ● ●● ● ● ● ●●●●● ●● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ●● ●● ●●● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ●●●● ● ● ●● ● ● ● ● ●● ●●●●● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ●● ● ●● ● ●●●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ●●● ● ●● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ● ●● ● ● ●● ●● ●● ● ●● ●● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ●● ● ● ●● ●● ●● ● ●● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●●●● ● Taper QE3 BNP Greece QE2 ● ●●● ● ●GFC ● ● ● ●● ● ● ● ● ● ● 05/2006 ● ● ● ● ●● ● ● ●●●● ● ●●● ● ●● ●●● ● ● ●● ● ●● ● ●● ●● ● ● ● ● ●●● ● ●●● ● ● ●● ●● ● ● ●● ● ● ●●● ● ● ● ●● ● ● ● ● ● ●●● ● ● 2000−01−21 2003−02−14 2006−03−10 2009−04−03 2012−04−27 Date Nico Katzke Exploring EME Asset Returns Conditional Comovement Introduction Methodological Discussion Data Results Conclusion Bonus! Comments? Nico Katzke Exploring EME Asset Returns Conditional Comovement
© Copyright 2026 Paperzz