We evaluate predictive models for Credit Default Swap (CDS) spreads, investigating the predictive power of past idiosyncratic equity returns on future CDS spread changes. We aim to identify which modeling approach best captures the structural nature of credit spreads while minimizing model error during periods of market stress.
Two distinct weighting methodologies are constructed and backtested:
- The Boxcar Model (Stability): A rolling, equal-weighted moving average over a fixed window.
- The Exponential Model (Reactivity): A decay-weighted model that heavily prioritizes recent data. This is designed to capture rapid structural shifts and momentum, risking potential overreaction to temporary market noise.
The analysis confirms that past idiosyncratic equity returns possess significant predictive power for future CDS spread changes. However, the efficacy of the signal is heavily regime-dependent.
- Performance: Delivered the lowest Root Mean Square Error (RMSE), particularly during the severe market shocks of the 2020 crisis period.
- Mechanism: Its equal-weighting mechanism effectively smoothed out acute volatility spikes that misled more reactive models, proving that filtering out noise is often more valuable than capturing immediate shifts during credit panics.
- Performance: Superior for capturing "momentum" names (e.g., Netflix) during low-volatility, trending regimes.
- Vulnerability: By heavily weighting recent observations, the model systematically extrapolated short-term noise spikes as genuine trend shifts. This sensitivity significantly increased error rates and capital destruction during periods of high market stress.