nowcastDFM

Run dynamic factor models (DFM) in R. Adapted from Bok et al. 2017, MATLAB code. The package provides the ability to estimate a DFM model using the expectation–maximization method, obtain predictions from estimated models, and obtain the impact of new data releases on model predictions.

Installation

devtools::install_github("dhopp1/nowcastDFM")

Functionality

Example

Given data is a dataframe (not a tibble) with a date column and 4 columns for various seasonally adjusted growth rates of economic series with missing values of NA:

library(nowcastDFM)

# estimate a DFM with one block for all variables
output_dfm <- dfm(data) 

# estimate a DFM with two different blocks
blocks <- data.frame(block_1 = c(1,1,1,0), block_2 = c(0,0,1,1)) # defining two blocks
output_dfm <- dfm(data, blocks = blocks)

# get predictions from estimated DFM for the following 3 months
# new data is dataframe with same columns as data the model was trained on, but newer data
predictions <- predict_dfm(new_data, output_dfm, months_ahead = 3)

# get impact of new data on predictions for a particular variable and time period
# old_data and new_data are dataframes with same columns as the data the model was trained on, but with older and newer data
news <- gen_news(old_data, new_data, output_dfm, target_variable = "target_name", target_period = "2020-01-01")