EPU_GeneticAlgo

• This project presents a new approach to building category-specific Economic Policy Uncertainty (EPU) indices by extracting keywords from texts via the genetic algorithm.
• Knowing that the EPU indices vary with different sets of EPU terms, we let the EPU terms vary to fit a specific target variable to obtain the final optimized EPU index.
• We proceed in three steps:
  • First, with initial EPU terms, we use a pre-trained word embedding space to extend the set of candidate keywords.
    • A word embedding space maps words into high dimensional vectors. Words with similar semantic meanings are close to each other in the space. We use the pre-trained embedding space from Pennington et al. with GloVe Technique.
  • Second, using the genetic algorithm, we search for the best subset of candidate keywords that matches the dynamics of a pre-determined target variable.
  • Last, following the same steps proposed by Baker et al. (2016), we build the new category-specific EPU index.