Federal Reserve Board research proposes PCA-based portfolio margining to better handle correlation shifts and reduce model computation

The Federal Reserve Board published a research paper proposing an alternative portfolio margining method for central counterparties that applies filtered historical simulation (FHS) to latent risk factors derived from principal component analysis (PCA), rather than filtering each risk factor separately. The paper argues that traditional FHS reacts quickly to changes in market volatility but does not address time-varying correlations, while also creating a computational burden that grows linearly with the number of risk factors. In tests on simulated and constructed portfolios, the PCA latent-factor approach performs materially better when correlations change sharply, and performs well when correlation is more stable, although it can require care for certain concentrated portfolios. The authors backtest the methods using data from 2020 during COVID-19 market stress.