Why is it important to model time varying stock return volatility when estimating Value-at-Risk (VaR) based on the normal distribution? Outline one approach that can be used for this purpose, with details of how to choose the amount of weight given to older data. What trade-offs are faced in making this choice?