Robust Risk Parity Portfolio Optimization under Cardinality and Transaction-Lot Constraints

Kusumawati, Rosita, Abdurakhman, Gunardi
Operations Research Forum · Springer Nature · 2026-10-05 · DOI: 10.1007/s43069-026-00699-y

Abstract

Traditional risk parity (RP) portfolio optimization relies on covariance estimates that are highly sensitive to outliers and departures from normality. Although robust covariance estimators have been widely studied in portfolio optimization, their behavior within discretized risk parity portfolios remains insufficiently understood. This issue is particularly relevant in practice, where portfolio allocations must satisfy cardinality and transaction-lot constraints, transforming the continuous RP problem into a mixed-integer optimization problem and altering the set of attainable risk allocations. This study develops a risk parity portfolio with Cardinality and Transaction-Lot Constraints (RP-CTLC) and investigates the interaction between covariance robustness and discrete implementation constraints. Five high-breakdown covariance estimators, namely the Minimum Covariance Determinant (MCD), S-estimator, MM-estimator, Stahel-Donoho estimator, and Orthogonalized Gnanadesikan-Kettenring (OGK) estimator, are incorporated into the RP-CTLC framework and compared with the classical maximum likelihood estimator (MLE). The resulting non-convex mixed-integer optimization problem is solved using an Integer Particle Swarm Optimization algorithm. Performance is evaluated through controlled contamination experiments and empirical backtesting on IDX80 stocks, with a constrained Global Minimum Variance portfolio (GMV-CTLC) serving as a benchmark. The results demonstrate that robust covariance estimation affects RP and GMV portfolios differently under discrete trading constraints. In simulation experiments, the S and OGK estimators achieve significantly lower volatility than the MLE within the RP-CTLC framework. Conversely, robust estimators lead to significantly deeper drawdowns in GMV-CTLC under severe contamination, revealing a nontrivial risk-return trade-off. Empirical results further indicate that the MCD estimator provides the most favorable balance between risk-adjusted performance and drawdown control for RP-CTLC, particularly at shorter rebalancing horizons. These findings provide new evidence on the interaction between covariance robustness and implementable risk parity portfolios, highlighting the importance of considering discrete trading constraints when evaluating robust portfolio construction methods. Sensitivity analyses indicate that the effects of cardinality and transaction-lot constraints depend on portfolio size, initial capital, stock prices, and the portfolio construction method, with the 100-share lot constraint potentially causing material changes in portfolio composition and realized performance.

Read at publisher →

Similar articles