Från SPAN till filtrerad historisk simulering VaR: En utvärdering av marginalkravsmodeller för centrala motparter

In Central Counterparty (CCP) clearing, there is an ongoing shift from the historically dominant SPAN margin model to Filtered Historical Simulation (FHS) Value at Risk (VaR) based margin models. While the academic literature on FHS VaR estimation often favors GARCH-type models, CCPs in practice typically implement EWMA-based approaches. This thesis evaluates GARCH, GJR-GARCH, FIGARCH, EGARCH, and EWMA volatility filters with varying lookback windows in the context of FHS VaR margin models. Using data from 2013 to early 2025, the analysis is based on portfolios consisting of futures contracts, with all models designed to comply with regulatory requirements. The margin models is built in eight steps. The core FHS VaR component, representing the model before regulatory adjustments, is assessed using the Kupiec and Christoffersen statistical tests. The primary evaluation is based on a loss function designed to capture key properties such as margin size, exceedance behavior, and margin variability. The findings show that FHS VaR-based margin models generally outperform SPAN, with the EWMA-based model using a 250-day lookback window achieving the best overall results. Compared to SPAN, EWMA-based models offer more responsive margin setting and reduced exceedance frequency, with more gradual daily adjustments in contrast to SPAN’s infrequent but often abrupt updates. These results support the industry’s transition to FHS VaR based margin models and point to EWMA as a particularly practical and effective choice for CCPs.