Informations and abstract
Keywords: Contingent Capital; Financial Regulation; Outlier Detection.
This paper develops and applies sophisticated data mining techniques to detect in an early stage potential risks regarding the stability of institutions by making use of the market information of their issued contingent capital instruments. Data mining is an important new topic within the financial world. The detection of observations which are different from the majority, called outliers, can be of interest for market analysts, risk managers, regulators and traders. These exceptions might be caused by extraordinary circumstances that may potentially require extra hedging or can be seen as trading opportunities. They could also give regulators an early warning and signal for potential trouble ahead. In this paper we first explain and apply the new risk measure, called the Value-at-Risk Equivalent Volatility (VEV). The concept was introduced by the European authorities in the new PRIIPs1 regulation and needs to be implemented for all structured products by January 1st 2018. This risk-measure is an extension of the classical volatility measure by taking into account skewness and kurtosis. This measure however works in a one-dimensional setting. In this paper we apply outlier detection and the VEV concept to CoCo bonds. CoCos are hybrid high yield securities that convert into equity or write down if the issuing financial institution is in a distressed situation. Further we want to detect outliers in the CoCo market taking into account multiple variables such as the CoCo market returns and the underlying equity return. Based on a multiple-dimension distance we can detect CoCos that are outlying compared to previous time periods but also taking into account extreme moves of the market situation. To some extent, CoCos can be seen as derivative instruments with some capital ratio (CET1) as underlying driver. In this perspective, a CoCo market price is just the price of a derivative and hence contains forward looking information or at least the market's anticipated view on the financial health of the institution and the level of the relevant trigger. This paper develops data mining techniques that incorporate such forward looking view by comparing historical data with current CoCo market prices.