AXA IM | Recent years have shown the importance of financial linkages across markets and assets. In particular, spillovers were evident during the Global Financial Crisis (GFC) when financial turmoil originating in the US mortgage markets spread around the globe and caused an unprecedented crisis. More recently, sovereign debt concerns in the euro area led to volatility in global financial markets. Identifying and measuring these spillover mechanisms is key to both understanding past market stress episodes and anticipating the propagation of a shock that might occur in a specific market.
Return vs volatility spillovers
Price (or return) spillovers are observed when markets are integrated (i.e. when returns are interdependent), whereas volatility spillovers are observed when unexpected shocks are transmitted across asset prices. In statistical terms, we identify return spillovers when means move together and volatility spillovers when variances move together. These two types of spillovers do not necessarily come together. For instance, in equity markets we usually observe high return spillovers among individual stocks but not necessarily high volatility spillovers. Looking at stocks from the DJIA1 Index, prices of individual stocks tend to move together; on average, stock and index performance are similar. However the impact from bad news for a specific company or sector is likely to be contained to that company/sector affecting only the variance of a limited number of stocks. When observing G4 sovereign markets we notice that yields tend to move together, suggesting high linkages and spillovers across series.
Constructing a measure of spillovers
As the magnitude of spillovers varies across time, we follow a time-varying approach to capture changes of intensity and the origin of such effects. We proceed following Diebold and Yilmaz (2009), basing our analysis on the commonly-used vector auto-regression (VAR) technique which allows to model multiple variables according to their past values and their relationships. From our VAR model, we apply an error variance decomposition on forecasted returns in order to assess the part of unexpected changes in one asset price due to shocks coming from other assets.
We then aggregate these cross-innovations into one single measure called the spillover index. By repeating this approach on a six-month rolling window and decomposing two-week ahead forecast errors, we are able to compute a time-varying index ranging from one (no spillover) to 100 (perfect transmission of shocks), that indicates the level of spillover across assets. Using a six-month rolling window allows us to capture both longer term structural / macro-economic trends and short-term dynamics stemming from market shocks.
Our analysis is performed on 10-year government bond prices across the US, Germany, the UK and Japan using daily data from August 1995 to July 2016. Daily total returns are used to compute the return spillover index while the volatility spillover index is obtained using weekly standard deviations of each market’s yield. Volatility spillovers in the most recent period. Since 2015, the volatility spillover index has been relatively high and above the historical average. This might stem from higher uncertainties in the current market and prolonged QE schemes from major central banks. Among other salient features, we notice that the spillovers originating from Japan have been much more important over the recent period than historically (12% in 2015-2016 vs68% previously). As a consequence of the ECB’s PSPP , we also note that the influence of the German Bund is greater since mid-2015 than in previous periods. Overall, the index increased from about 25 to nearly 40 between the stock market sell-off of in August 2015 and October 2015. Furthermore, we observed a sharp rise in the index around the stock market correction of early February 2016 and, unsurprisingly, another UK-driven peak around the Brexit vote.
Net and directional spillovers
Moreover, we are able to derive directional spillovers (spillovers from one market to another) and net spillovers (spillovers “supplied” minus spillovers “received” by a specific market). Net spillovers provide useful information about the magnitude of volatility in a market caused by events outside that market. This is hardly ever the case for US Treasuries, which are permanent “net suppliers” of spillovers, meaning that investors can safely focus on US economic/market events. It is very different for UK Gilts which have historically been mostly “net receivers” of spillovers from other markets. The case of JGBs and Bunds is more complex as these markets have been alternatively “suppliers” and “receivers” of spillovers.
Directional spillovers help us quantify how much a news shock is transmitted from one market to another. For instance, if directional spillovers from USTs to Bunds is around 40%, it implies that, all things being equal, an increase of 1% volatility in USTs leads to an increase of 0.4% of volatility in Bunds. For a static view, directional spillovers can be summarised at a specific date in a spillover table. We can also draw four different charts to have a closer look at their dynamics across time.
Looking at directional spillovers, first we note that in the recent period, only about 20%-30% of unexpected changes in UST yields are explained by foreign yields, whereas about 70% of unexpected changes in Gilt yields is caused by foreign markets. This is consistent with our observations from net spillovers – the US is a strong domestically driven market while the UK is largely influenced by global conditions. One striking feature is that Japan has been “receiving” very limited spillovers while accounting for a large share of spillovers to other markets in the recent period, especially to the US. We observe that since 2015, spillovers from Japan to other markets were higher and in particular it accounted for most of the spillover “received” by the US.
This a recent phenomenon that requires consideration from investors, especially for those holding Bunds and Gilts as they are currently net “receivers” of spillovers. We also note that spillovers from the UK to the US and Germany are higher since the Brexit vote, which is consistent with the economic intuition.
The large spillovers from the UK to Japan are a bit more puzzling but may be explained by the huge swings on Japanese holdings of Gilts. Japanese bank holdings of Gilts went from £700mn in November 2014 to £1,043mn in February 2015, and back to £653mn in November 2015. We must also mention that the Gilt is influenced by the US and to a lesser extent by Germany, and finally an investor in German Bunds needs to pay attention to developments in the US.
