[Paper Review] Copula-based Hierarchical Aggregation of Correlated Risks. The behaviour of the diversification benefit in Gaussian and Lognormal Trees

Jean-Philippe Bruneton|arXiv (Cornell University)|Nov 4, 2011

Risk and Portfolio Optimization3 citations

TL;DR

This paper proposes a copula-based hierarchical aggregation framework for modeling correlated risks in portfolios using Gaussian and Lognormal trees. It analytically derives the diversification benefit in Gaussian trees, showing that 'thin' trees (with higher depth) consistently outperform 'fat' trees (shallow, wide structures) due to a natural reduction in effective dependence, a phenomenon that also holds in more realistic Lognormal models.

ABSTRACT

The benefits of diversifying risks are difficult to estimate quantitatively because of the uncertainties in the dependence structure between the risks. Also, the modelling of multidimensional dependencies is a non-trivial task. This paper focuses on one such technique for portfolio aggregation, namely the aggregation of risks within trees, where dependencies are set at each step of the aggregation with the help of some copulas. We define rigorously this procedure and then study extensively the Gaussian Tree of quite arbitrary size and shape, where individual risks are normal, and where the Gaussian copula is used. We derive exact analytical results for the diversification benefit of the Gaussian tree as a function of its shape and of the dependency parameters. Such a "toy-model" of an aggregation tree enables one to understand the basic phenomena's at play while aggregating risks in this way. In particular, it is shown that, for a fixed number of individual risks, "thin" trees diversify better than "fat" trees. Related to this, it is shown that hierarchical trees have the natural tendency to lower the overall dependency with respect to the dependency parameter chosen at each step of the aggregation. We also show that these results hold in more general cases outside the gaussian world, and apply notably to more realistic portfolios (LogNormal trees). We believe that any insurer or reinsurer using such a tool should be aware of these systematic effects, and that this awareness should strongly call for designing trees that adequately fit the business. We finally address the issue of specifying the full joint distribution between the risks. We show that the hierarchical mechanism does not require nor specify the joint distribution, but that the latter can be determined exactly (in the Gaussian case) by adding conditional independence hypotheses between the risks and their sums.

Motivation & Objective

• To rigorously formalize copula-based hierarchical aggregation of correlated risks in tree-structured portfolios.
• To quantify the diversification benefit in Gaussian trees as a function of tree topology and dependence parameters.
• To investigate whether structural effects in hierarchical aggregation systematically reduce overall portfolio dependence.
• To extend findings from Gaussian to more realistic Lognormal risk models.
• To clarify whether the full joint distribution can be recovered from hierarchical copula structures with conditional independence assumptions.

Proposed method

• Defines hierarchical aggregation via top-down tree decomposition, where risks are aggregated at each level using copulas to model dependence.
• Uses the Gaussian copula to model dependence between sub-portfolios at each aggregation step, with marginal distributions assumed normal or lognormal.
• Applies conditional independence assumptions to derive exact expressions for the full joint distribution in Gaussian trees.
• Employs recursive covariance decomposition to compute effective correlations between leaves based on their path of coalescence in the tree.
• Derives analytical formulas for the total variance and diversification benefit using recursive variance propagation through tree levels.
• Validates results in Lognormal trees by applying the same hierarchical copula mechanism with non-normal marginals, confirming robustness of structural effects.

Experimental results

Research questions

• RQ1 How does the topology of a hierarchical aggregation tree (e.g., depth vs. width) affect the diversification benefit in the presence of correlated risks?
• RQ2 To what extent does hierarchical aggregation via copulas reduce the effective dependence between individual risks compared to direct aggregation?
• RQ3 Does the diversification benefit in Gaussian trees depend on the choice of copula parameters, and how is this modulated by tree structure?
• RQ4 Can the full joint distribution of risks be reconstructed from hierarchical copula aggregation under conditional independence assumptions?
• RQ5 Do the structural effects observed in Gaussian trees—such as reduced effective dependence in thin trees—persist in more realistic Lognormal risk models?

Key findings

• For a fixed number of individual risks, 'thin' trees (with higher depth and lower branching factor) consistently achieve greater diversification benefits than 'fat' trees (shallow, high-branching structures).
• Hierarchical aggregation using Gaussian copulas naturally reduces the effective dependence between risks, even when the same copula parameter is used at each level, due to the recursive structure of the tree.
• The diversification benefit in Gaussian trees is analytically tractable and depends explicitly on the tree's shape and the copula correlation parameter, with closed-form expressions derived for total variance and risk measures.
• The same structural tendency—thin trees outperforming fat trees—holds in Lognormal trees, indicating that the effect is not an artifact of normality but a general feature of hierarchical copula aggregation.
• Under conditional independence assumptions, the full joint distribution of risks in a Gaussian tree can be exactly recovered from the hierarchical copula structure and the marginal distributions.
• The effective correlation between two leaves in the tree depends only on the level at which their paths first coalesce, with a recursive formula derived for this dependence.