Since markets generally do not trade during overnight period, volatility cannot be estimated on a high-frequency basis. We adopt a new forecasting approach by using squared overnight return, pre-open volatility of the same asset and realized volatilities of related assets from other markets, where intraday data is still available while the Australian Stock Exchange (ASX) is closed, to predict stock volatility. We use a number of different specifications of the Heterogeneous Autoregressive (HAR) model to identify an optimal way to incorporate this additional information. We evaluate the forecasting performance of 45 ASX 200 stocks, categorized in three groups based on their annual total trading volumes, three Global Industry Classification Standard (GICS) indices and the S&P/ASX 200 index using a rolling estimation method. Our empirical analysis of the ASX constituents confirms the usefulness of using pre-open volatility of the same asset and realized volatilities of related assets from other markets when the ASX is closed for forecasting future volatility. Furthermore, we find that the predictive power of overnight information for all stocks and indices is higher during the market opening period and declines gradually over the trading day. However, the decrement is steeper for active stocks, suggesting that the predictive power is higher for inactively traded stocks. Finally, we evaluate the economic significance of the augmented HAR model that includes realized volatilities of related assets from other markets, and we find that it provides significant utility gains to a typical mean-variance investor.