Tail risk parity

Tail risk parity is an extension of the risk parity concept that takes into account the behavior of the portfolio components during tail risk events. The goal of the tail risk parity approach is to protect investment portfolios at the times of economic crises and reduce the cost of such protection during normal market conditions. In the tail risk parity framework risk is defined as expected tail loss. The tail risk parity concept is similar to drawdown parity

Traditional portfolio diversification relies on the correlations among assets and among asset classes, but these correlations are not constant. Because correlations among assets and asset classes increase during tail risk events and can go to 100%, TRP divides asset classes into buckets that behave differently under market stress conditions, while assets in each bucket behave similarly. During tail risk events asset prices can fall significantly creating deep portfolio drawdowns. Asset classes in each tail risk bucket fall simultaneously during tail risk events and diversification of capital within buckets does not work because periods of negative performance of portfolio components are overlapped. Diversification across tail risk buckets can provide benefits in the form of smaller portfolio drawdowns and reduce the need for tail risk protection.

Baitinger, Dragosch, and Topalova in their article "Extending the Risk Parity Approach to Higher Moments: Is There Any Value Added?" propose an extension of the classical risk parity portfolio optimization approach from Maillard et al. (2010) to incorporate higher moments such as skewness and kurtosis. They present a methodology for consistently incorporating higher moments like skewness and kurtosis into the risk parity optimization framework developed by Maillard et al. (2010). This allows tail risks to be considered in the optimization. Empirical analysis on four real-world datasets by Baitinger, Dragosch, and Topalova finds mixed results. Their higher moment risk parity methods tend to outperform classical risk parity significantly when the underlying data exhibits high non-normality and co-dependencies. But they provide little value-add in other datasets. Simulation studies confirm the value of higher moment methods increases with degree of non-normality and correlation in the data. The inferred optimization approach also does better when provided enough data.