Cambodia

This Selected Issues paper focuses on Cambodia’s satellite data for nowcasting. Cambodia faces limited institutional capacity in the production and timely release of quality official statistics, limiting policymakers’ ability to make agile and effective policy decisions. The machine learning method applies quarterly satellite indicators, along with the traditional variables, for training the nowcasting model. The nowcasting model uses the random forest machine-learning algorithm to predict year-on-year quarterly gross domestic product growth rate. The random forest machine-learning technique demonstrates a strong fit, pointing to a nowcasting result of 5.7 and 6.7 percent gross domestic product growth year-on-year in 2025q1 and 2025q2, respectively with the underlying stories. The addition of satellite variables to the list of indicators used to train the model improved the model accuracy by over 20 percent. This percentage might not seem substantial because of the weight of the traditional indicators, which is significantly higher. However, in situations where these traditional indicators are scarce or not collected on time, satellite indicators can fill the gap and contribute more to the accuracy of models.