An International Perspective on Risk Management Quality
| Author | Nicholas Taylor,Svetlana Mira |
| DOI | http://doi.org/10.1111/j.1468-036X.2011.00611.x |
| Published date | 01 November 2013 |
| Date | 01 November 2013 |
Financial crises highlight the interconnected nature of financial markets. Perhaps most notably, it is apparent from such episodes that risk (asset/portfolio return volatility) evolves across international financial markets; that is, ‘volatility spillovers’ occur (see, e.g., Engle et al. (), Fleming and Lopez (), Christiansen () and Asgharian and Nossman ()). From a risk management perspective, such financial contagion is not problematic providing that the level of risk in a particular market (given previous levels in other markets) can be accurately estimated. In this case, risk management policies can be accordingly adjusted in anticipation of future risk levels. However, if risk levels cannot be anticipated then such policies will fail and financial instability will be further exaggerated. It is this issue that is investigated in the current paper. In particular, using a recently proposed testing framework we investigate risk management quality by examining the transmission of unanticipated downside risk (referred to as extreme risk) across international equity markets.
In the current paper, extreme risk is said to occur when asset (or portfolio) returns fall below an expected downside limit. Consequently, we require some notion of what downside risk is expected to be at any particular time: that is, a dynamic model of downside risk is needed. To this end, we make use of various augmentations of the conditional Value at Risk (VaR) model recently proposed by Xiao and Koenker (). Born out of regulatory demand following a series of high profile financial failures (including the October 1987 stock market crash), VaR models are the most commonly‐used risk management measurement device, with VaR measures defined as the lower quantile levels on the conditional return distribution. Given a set of such measures, we proceed by examining the extent to which extreme risk is transmitted across international equity markets. This is achieved via multiple applications of the kernel‐based extreme downside risk test of Hong et al. (). In conducting such an exercise, we contribute to the literature by providing a novel simultaneous examination of the interdependence of financial markets and the quality of state‐of‐the‐art VaR models from an international perspective.
Conventional tests of VaR model efficiency are conducted by considering (either separately or jointly) whether VaR measures correctly cover the data, and are independent. The latter quality is often examined by considering whether VaR violations (extreme risk measures) are orthogonal to a pertinent information set (cf. Christoffersen () and Engle and Manganelli ()). This can be achieved by regressing an extreme risk indicator variable upon each element of the pertinent information set. A potential problem with this approach, particularly in the current international context, is that there may be a large number of elements in this set. This would indeed be the case if one wishes to examine whether extreme risk in a particular market is independent of past values of extreme risk in all other markets. Moreover, one has to make an explicit assumption regarding the exogeneity of the variables considered. To avoid these issues we carry out multiple applications of the (non‐)causality test of Hong et al. () using all possible market pairings, and consider the significance of the rejection rates associated with meaningful clusters of markets via use of a simple bootstrap procedure.
Previous studies in the extant literature show that while the statistical significance of volatility spillovers is strong (see, e.g., Engle et al. ()), their economic significance appears less so. For instance, using data from four developed equity markets, McAleer and da Veiga () find that VaR measures are generally insensitive to the inclusion of such spillovers. We add to this debate by using data from 45 developed and emerging markets to show that VaR models based only on domestic information deliver significant bivariate causality in extreme risk across national borders. Thus, it would appear that such VaR models are inefficient (from a forecasting perspective), and that risk managers are vulnerable to external international downside shocks. However, we also demonstrate that simple parsimonious augmentations of these VaR models (to include such shocks) deliver efficient extreme risk measures that are invariant (orthogonal) to extreme risk in other international markets. Consequently, extreme risk (if defined correctly) does not appear to evolve across financial markets, and VaR measures are available to risk managers that adequately capture the interdependence of financial markets. In short, models that fail to acknowledge the international dimension to the risk evolution process are seriously misspecified.
The rest of the paper is organised as follows. The next section provides a description of the testing framework, while Section 3 contains the empirical results. The final section concludes.
In this section we first provide an outline of four downside risk models, followed by a formal definition of extreme risk and a description of the testing framework used to test for (non‐)causality in such risk.
We construct downside risk measures using four different versions of the conditional quantile generalised autoregressive conditional heteroscedasticity (CQ‐GARCH) model introduced by Xiao and Koenker (). While a multitude of other risk models are available, this particular class of models is used because of its applicability to a wide range of conditional return distributions, and its documented superior performance over competing models, viz. the RiskMetrics, Gaussian GARCH, and conditional autoregressive value‐at‐risk (CAViaR) models, (ibid.).
The CQ‐GARCH(p, q) model considered in this paper is a quantile version of the popular linear (standard deviation) GARCH(p, q) model (Taylor, ). The latter model can be specified as follows:
The first step involves construction of an estimate of σt (denoted ) based on repeated quantile estimation of the sieve approximation to the above linear GARCH(p, q) model. The second step uses this estimate of σt to deliver the conditional τs‐quantile of Rt via the following quantile regression:
In addition to the above model, we also consider three augmented versions of this model. Each of these augmentations is defined with respect to the content of the vector of explanatory variables, , in (2). The first of these allows for asymmetric responses to previous good and bad news. Specifically, in the spirit of the threshold GARCH‐type models of Glosten et al. (), we replace with
Given the abundant volatility spillover evidence (see, e.g., Engle et al. ()), we wish to allow for risk transmission across national boundaries. This could be achieved including past returns from all other markets in the above specification. However, in the current application this would invoke the ‘curse of dimensionality’ as a large number of individual markets are considered in the subsequent analysis. To avoid this problem we assume that returns in these markets are driven by a small number of observed factors, and use these factors are representations of all individual markets. Specifically, we assume an international CAPM in which returns are driven by two world factors, viz. returns to an emerging market portfolio and a developed market portfolio. Continuing the assumption of asymmetric responses in the presence of good and bad news, we thus have
The final augmentation permits changes in the transmission of risk from international markets during periods of financial crisis (referred to as ‘shift contagion’, Beirne et al. ). Specifically, we assume that
where turbt equals unity during episodes of market turbulence in mature markets (and zero otherwise). We follow Beirne et al. () and define market turbulence as periods in which the implied volatility on S&P 500 index options (VIX) is very high (above 30), or increasing rapidly (the five‐day moving average is greater then the 52‐week moving average by more than 30%). This model is henceforth referred to as the international shift contagion CQ‐GARCH(p, q) model.
The parsimony of the above international models is achieved via use of an economically‐motivated factor model (i.e., the international CAPM) in which it is assumed that returns in each market are driven by two specific world factors (i.e., developed and emerging market returns). To examine the sensitivity of this assumption we also consider an alternative factor model, viz. a statistical factor model in which we conduct a principal component analysis (PCA) to determine the factors that best explain the covariance of individual...
Get this document and AI-powered insights with a free trial of vLex and Vincent AI
Get Started for FreeUnlock full access with a free 7-day trial
Transform your legal research with vLex
-
Complete access to the largest collection of common law case law on one platform
-
Generate AI case summaries that instantly highlight key legal issues
-
Advanced search capabilities with precise filtering and sorting options
-
Comprehensive legal content with documents across 100+ jurisdictions
-
Trusted by 2 million professionals including top global firms
-
Access AI-Powered Research with Vincent AI: Natural language queries with verified citations
Unlock full access with a free 7-day trial
Transform your legal research with vLex
-
Complete access to the largest collection of common law case law on one platform
-
Generate AI case summaries that instantly highlight key legal issues
-
Advanced search capabilities with precise filtering and sorting options
-
Comprehensive legal content with documents across 100+ jurisdictions
-
Trusted by 2 million professionals including top global firms
-
Access AI-Powered Research with Vincent AI: Natural language queries with verified citations
Unlock full access with a free 7-day trial
Transform your legal research with vLex
-
Complete access to the largest collection of common law case law on one platform
-
Generate AI case summaries that instantly highlight key legal issues
-
Advanced search capabilities with precise filtering and sorting options
-
Comprehensive legal content with documents across 100+ jurisdictions
-
Trusted by 2 million professionals including top global firms
-
Access AI-Powered Research with Vincent AI: Natural language queries with verified citations
Unlock full access with a free 7-day trial
Transform your legal research with vLex
-
Complete access to the largest collection of common law case law on one platform
-
Generate AI case summaries that instantly highlight key legal issues
-
Advanced search capabilities with precise filtering and sorting options
-
Comprehensive legal content with documents across 100+ jurisdictions
-
Trusted by 2 million professionals including top global firms
-
Access AI-Powered Research with Vincent AI: Natural language queries with verified citations
Unlock full access with a free 7-day trial
Transform your legal research with vLex
-
Complete access to the largest collection of common law case law on one platform
-
Generate AI case summaries that instantly highlight key legal issues
-
Advanced search capabilities with precise filtering and sorting options
-
Comprehensive legal content with documents across 100+ jurisdictions
-
Trusted by 2 million professionals including top global firms
-
Access AI-Powered Research with Vincent AI: Natural language queries with verified citations