Exponential decay controls how information about the underlying risk factors is weighted. Given the time-varying properties of these factors, it is often desirable to weight recent data more heavily. Our calculation engine allows for 2 separate decays for correlation and volatility.
The half life in months is controlled by the parameters: volatility_half_life and correlation_half_life. For example:
volatility_half_life = 2
correlation_half_life = 6
The above parameters reflect an exponentially weighted scheme to the risk factors such that 50% will apply to the sample corresponding to 2 months prior to the portfolio date when volatility is calculated and 50% will apply to 6 months when correlation is calculated. Such a model would be quite sensitive to changes in correlation and specially volatility and might be appropriate for portfolios that have shorter holding periods.
Conversely, if you would like to avoid any decay, i.e. apply the same weight for every sample when making calculations, you need to use a zero (0) decay for either correlation or volatility. For example:
volatility_half_life = 0
correlation_half_life = 6
The above parameters indicate a model that does not decay the information at all when volatility is calculated. Such a model is slower to react to sudden changes of volatility and might be more appropriate for portfolios that have longer holding periods.
More details of the formulas behind the double decay can be found in our white paper.