Uncertainty Quantification and Latent Group Analysis in Financial Statement Forecasting

The problem considered in this thesis is, in the first instance, the prediction of financial statements. Given what a company has filed over the three most recent fiscal years, the task is to forecast how its twenty reclassified accounting variables will change over the following year, and to do so probabilistically: the model returns a predictive distribution for each variable rather than a single number, so that the uncertainty of the forecast is part of the output. On top of this, the thesis asks a second question: whether companies can be separated into groups that are described by different predictive functions, and whether such groups correspond to the categories that an expert rule-based taxonomy would assign.