Using Asymmetric Correlations to Build More Resilient Portfolios
Summary
This review explains why diversification may help more during declines than during gains, and why average correlations can hide that difference. It recommends estimating an asset pair’s downside relationship by conditioning on the portfolio’s main growth asset alone. Conditioning on both assets being below a threshold misses cases where the second asset rises and offsets the first asset’s loss. Because conditional correlations can vary mechanically even for normally distributed returns, the review says observed correlations should be compared with a normal-distribution benchmark before treating the difference as genuine asymmetry.
The article examines six broad asset classes and reports that relationships vary by pairing; government bonds showed broadly favorable asymmetry in the historical sample, though the authors caution that very low interest rates could weaken this behavior. For portfolio construction, it contrasts dynamic reallocations with static portfolios that account for downside behavior, and describes global optimization using a utility function that penalizes losses past a threshold. The analysis is historical and model-dependent: results depend on the sample, chosen threshold, and investor preferences, and do not ensure protection in future selloffs.
Key ideas
- Asset correlations can differ between rising and falling markets, so a single full-sample estimate may miss useful portfolio behavior.
- Estimate downside correlation by conditioning on the main growth asset rather than requiring both assets to decline.
- Compare empirical conditional correlations with a distribution-based expectation to separate genuine asymmetry from mechanical effects.
- The review reports that government bonds were broadly favorable diversifiers in its historical asset-class analysis, with caveats about low rates.
- Global optimization with loss-sensitive utility can incorporate asymmetric relationships into portfolio weights.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.