This paper studies the construction of an inflation model and its forecast using a large set of predictors. The decision about the optimal number of common factors is made using Bai and Ng (2002) information criteria; and, a One Covariate at a Time - Multiple Testing approach (OCMT) (Chudik et al., 2018) is implemented to choose the optimal predictors and lags of the dependent variable; and, afterwards, 1 to 12 month ahead forecasts are constructed to predict the inflation in Colombia using 60 macroeconomic variables from 2006 to 2021. During this period, a rolling-window approach is used. The OCMT model consistently shows significant better performance than a Phillips curve based Vector Autoregressive (VAR) model and an Autoregressive Integrated Moving Average (ARIMA) model when using the entire dataset and, also, performs closely to these models when estimated with a pre-COVID dataset.
