The "filtered" aspect in filtered historical simulation refers to the fact that the historical returns are adjusted, or "filtered," by the conditional volatility (B) from the GARCH model.

Here's why:

1. GARCH Model: The GARCH model captures the time-varying volatility (conditional volatility) of the returns. It recognizes that volatility tends to clustering (periods of high volatility followed by more high volatility, and vice-versa).
2. Standardization: By dividing each historical return by its corresponding GARCH-estimated volatility, the analyst creates a series of standardized returns. These standardized returns have a mean of approximately zero and a standard deviation of approximately one, removing the heteroskedasticity (varying volatility) present in the original returns.
3. Resampling: The resampling is then performed on these standardized returns. This is the key "filtering" step. It ensures that when these resampled standardized returns are used to simulate future scenarios, they are scaled by future volatility forecasts from the GARCH model. This incorporates the expected changes in volatility into the risk estimation.

A is incorrect. Zero-sum weighting is not related to the filtering process in this context. C is incorrect. Filtering does not involve discarding any returns, positive or negative. D is incorrect. While some risk management techniques use age-based weighting or discarding of old data, this is not the defining characteristic of "filtered" historical simulation.