Why variance-based connectedness measures cannot tell risk transmission from hedging, with a minimal R demonstration.
Published
January 29, 2026
Modified
September 29, 2026
The connectedness approach (Diebold, F. X., K. Yilmaz, 2012; Diebold, F. X., K. Yilmaz, 2014) is a popular framework for understanding the transmission of shocks across assets or economic variables. It is based on the generalized forecast error variance decomposition(Pesaran, H. H., Y. Shin, 1998) of a vector autoregressive (VAR) model, which measures how much of the forecast error variance of one variable is explained by shocks to others.
While widely used, interpretation of connectedness measures requires care—especially in financial contexts where correlations (signed relationships) often guide investment and risk management decisions.
Variance vs. correlation: why the sign matters
Variance and covariance are inherently unsigned measures of association. They capture the magnitude of dependence, but not its direction. In contrast, correlation provides a signed indicator of whether assets tend to move together (positive correlation) or in opposite directions (negative correlation).
As a result, connectedness metrics—being variance-based—do not distinguish between risk comovement and hedging (safe-haven) behavior. For example:
A strong stock–gold connectedness does not necessarily imply risk transmission. Gold may act as a safe haven, moving inversely with stocks during crises.
Similarly, stock–bond or stock–uncertainty pairs often exhibit negative correlations. Without considering the expected sign, high connectedness could be misinterpreted as contagion, when it may actually reflect diversification.
When connectedness makes sense
The connectedness approach is most interpretable when applied to systems of similar assets, where the underlying economic drivers are expected to move them in the same direction in response to shocks. Typical examples include:
multiple cryptocurrencies;
equity indices across markets;
interest rate or credit spread series within the same region.
In such systems, a high degree of connectedness genuinely indicates risk propagation and systemic linkages.
Minimal demonstration
This example uses the ConnectednessApproach and zoo packages (install.packages(c("ConnectednessApproach", "zoo"))). We simulate two bivariate systems that differ only in the sign of the relationship between x and y.
Despite opposite correlations, both systems show similar connectedness magnitudes. The connectedness table alone cannot tell the co-moving system from the hedging one; always consider the theoretical sign of relationships when interpreting results.
Recommendations
Use the connectedness approach only when the system variables have theoretically consistent relationships.
Avoid mixing assets with fundamentally opposite expected signs (e.g., stock–gold) unless you explicitly address sign effects.
Consider supplementing variance-based connectedness with correlation or coherence analysis to capture directionality.
Always interpret connectedness in the context of economic theory and empirical sign expectations.
This limitation applies to all variants of the approach, including asymmetric, quantile, time-varying parameter (TVP), R² and R² quantile connectedness measures.
References
Diebold, F. X., K. Yilmaz. (2014). On the network topology of variance decompositions: Measuring the connectedness of financial firms. Journal of Econometrics, 182, (1), 119–134.
Diebold, F. X., K. Yilmaz. (2012). Better to give than to receive: Predictive directional measurement of volatility spillovers. International Journal of Forecasting, 28, (1), 57–66.
Pesaran, H. H., Y. Shin. (1998). Generalized impulse response analysis in linear multivariate models. Economics Letters, 58, (1), 17–29.
Source Code
---title: "Connectedness Approach in Financial Systems"date: 2026-01-29date-modified: 2026-09-29categories: [Connectedness, Econometrics]description: "Why variance-based connectedness measures cannot tell risk transmission from hedging, with a minimal R demonstration."bibliography: references.bibformat: html: toc: true code-tools: trueexecute: echo: true message: false warning: false freeze: auto---The connectedness approach [@diebold2012; @diebold2014] is a popular framework for understanding the transmission of shocks across assets or economic variables. It is based on the **generalized forecast error variance decomposition** [@pesaran1998] of a vector autoregressive (VAR) model, which measures how much of the forecast error variance of one variable is explained by shocks to others.While widely used, interpretation of connectedness measures requires care—especially in financial contexts where correlations (signed relationships) often guide investment and risk management decisions.## Variance vs. correlation: why the sign mattersVariance and covariance are inherently **unsigned** measures of association. They capture the *magnitude* of dependence, but not its direction. In contrast, **correlation** provides a signed indicator of whether assets tend to move together (positive correlation) or in opposite directions (negative correlation).As a result, connectedness metrics—being variance-based—do not distinguish between risk comovement and hedging (safe-haven) behavior. For example:- A strong **stock–gold** connectedness does not necessarily imply risk transmission. Gold may act as a safe haven, moving inversely with stocks during crises.- Similarly, **stock–bond** or **stock–uncertainty** pairs often exhibit negative correlations. Without considering the expected sign, high connectedness could be misinterpreted as contagion, when it may actually reflect diversification.## When connectedness makes senseThe connectedness approach is most interpretable when applied to systems of similar assets, where the underlying economic drivers are expected to move them in the same direction in response to shocks. Typical examples include:- multiple **cryptocurrencies**;- **equity indices** across markets;- **interest rate** or **credit spread** series within the same region.In such systems, a high degree of connectedness genuinely indicates **risk propagation** and **systemic linkages**.## Minimal demonstrationThis example uses the **ConnectednessApproach** and **zoo** packages (`install.packages(c("ConnectednessApproach", "zoo"))`). We simulate two bivariate systems that differ only in the sign of the relationship between `x` and `y`.### 1. Positive correlation```{r}set.seed(123)library(zoo)library(ConnectednessApproach)n <-1000idx <-seq(as.Date("2010-01-01"), by ="day", length.out = n)x <-rnorm(n)y <-0.5* x +rnorm(n, 0, 0.2)df_p <-zoo(cbind(x, y), order.by = idx)fit_p <-VAR(df_p, configuration =list(nlag =4, method ="pearson"))dca_p <-TimeConnectedness(Phi = fit_p$B,Sigma = fit_p$Q,nfore =10,generalized =TRUE)# Correlationcor(df_p[, 1], df_p[, 2])# Connectedness tableknitr::kable(dca_p$TABLE)```### 2. Negative correlation```{r}y <--0.5* x +rnorm(n, 0, 0.2)df_n <-zoo(cbind(x, y), order.by = idx)fit_n <-VAR(df_n, configuration =list(nlag =4, method ="pearson"))dca_n <-TimeConnectedness(Phi = fit_n$B,Sigma = fit_n$Q,nfore =10,generalized =TRUE)# Correlationcor(df_n[, 1], df_n[, 2])# Connectedness tableknitr::kable(dca_n$TABLE)```Despite opposite correlations, both systems show similar connectedness magnitudes. The connectedness table alone cannot tell the co-moving system from the hedging one; always consider the theoretical **sign of relationships** when interpreting results.## Recommendations- Use the connectedness approach only when the system variables have theoretically consistent relationships.- Avoid mixing assets with fundamentally opposite expected signs (e.g., stock–gold) unless you explicitly address sign effects.- Consider supplementing variance-based connectedness with correlation or coherence analysis to capture directionality.- Always interpret connectedness in the context of economic theory and empirical sign expectations.This limitation applies to **all** variants of the approach, including **asymmetric**, **quantile**, **time-varying parameter (TVP)**, **R²** and **R² quantile** connectedness measures.