Forecasting Commodity Prices (WB Pink Sheet)
The World Bank generates a "Pink Sheet" each month with global commodity prices. This information is incredibly useful for a number of puposes - we previously used it to track food commodity prices as well as oil and energy prices as indicators of food insecurity.
So I wondered if the information could be used to forcast future prices as well as identify clusters of commodities that fluctuate together. This project built several different approaches to forecast commodities including XGboost, ARIMA, PCA, and Random Forrest. Suprisingly, the ARIMA model performed the best while the ML approaches did not work that well. I guess its not too surprising since there was no additional information use to predict.
Clustering of commodities was not really that interesting either, but I didnt really do any exploration or refinement of the models.