This paper revisits the study of Cochrane (2005), to estimate the risk and returns of venture capital investments, while correcting for the selection bias. We use an up-to-date dataset and enhance it to account for missing firm valuations using machine learning. The model is able to infer, with a median error of less than 4%, the true log-value of the firm, for a total of nearly 120,000 observations, or six times more than the original paper, from 2010 to 2022.