The aim of this paper is to assess inflation forecasting acurracy over the short-term horizon using Consumer Price Index (CPI) disaggregated data. That is, aggregating forecasts is compared with aggregate forecasting. In particular, three questions are addressed: i) one should bottom-up or not, ii) how bottom one should go and iii) how one should model at the bottom. In contrast with the literature, di erent levels of data dis-aggregation are allowed, namely a higher disaggregation level than the one considered up to now. Moreover, both univariate and multivariate models are considered, such as SARIMA and SARIMAX models with dynamic common factors. An out-of-sample forecast comparison (up to twelve months ahead) is done using Portuguese CPI dataset. Aggregating the forecasts seems to be better than aggregate forecasting up to a five-months ahead horizon. Moreover, this improvement increases with the disaggregation level and the multivariate modelling outperforms the univariate one in the very short-run.