R/macro_helpers_general.R

In ooelrich/oscbvar: Stuff for Bayesian vector autoregressions

#' @title Generate list of atomic models
#' 
#' @description Generates a list of all the available atomic models. Passed as
#'   an argument to the atomic prediction generator (and the gen_all_data fun).
#' 
#' @export

gen_atomic_list <- function() {
    atomic_models <- c("bvar_3", "bvar_7", "bvar_3_o2", "bvar_7_o2", 
                     "svbvar_3", "svbvar_7", "svbvar_3_o2", "svbvar_7_o2",
                     "bart_7", "tvpsvbvar_3", "bvar_3_basic", "bvar_7_basic")
    return(atomic_models)
}

#' @title Generate atomic data frame for predictions
#' 
#' @description
#' Generates an empty data frame to store atomic predictions.
#' 
#' @details 
#' Separate from the actual notebook functions to ensure that each function
#' stores its predictions in identical data frames, since otherwise the merging
#' won't work.
#' 
#' @keywords internal
 
gen_atomic_df <- function() {
    df <- data.frame(matrix(ncol = 5, nrow = 0))
    x <- c("pmean", "lpdens", "method", "t", "ytrue")
    colnames(df) <- x
    return(df)
}

#' @title Generate Z matrix
#' 
#' @description Generates a Z (design) matrix for a VAR model.
#' 
#' @details 
#' Given a data frame, a starting point, a number of lags, and whether to
#' include an intercept or not, creates a design matrix with correctly names
#' columns.
#' 
#' @param data Data frame (Y).
#' @param start_t Which observation is considered as t = 1. Needs to be greater 
#'   than the number of lags.
#' @param lags Order of the VAR.
#' @param include_intercept Boolean. Determines is an intercept should be used
#'   or not.
#' @export

gen_Z <- function(data, start_t, lags, include_intercept) {

    intercept <- rep(1, nrow(data) - start_t + 1)
    Z <- data.frame(intercept)
    for (i in seq_len(lags)) {
        tmp <- data[(start_t - i):(nrow(data) - i), ]
        colnames(tmp) <- paste0(colnames(tmp), sprintf("_lag_%i", i))
        Z <- cbind(Z, tmp)
    }
    if (include_intercept == FALSE) {
        Z <- Z[, -1]
    }
    return(as.matrix(Z))
}


#' @title Omega_0 for the Minnesota prior
#' 
#' @description Generates an Omega_0 according to a (the) Minnesota prior.
#' 
#' @details 
#' Prior variance for the intercept (if included) is set to 10 000. Cross series
#' tightness is set to 1.
#' 
#' @param overall_tightness Overall tightness.
#' @param lag_decay Lag decay.
#' @param m Dimensionality of each y-vector (or columns in Y).
#' @param include_intercept Whether or not the model includes an intercept.
#' 
#' @keywords internal

omega_minnesota <- function(lags, overall_tightness, lag_decay, m, include_intercept) {
    om_vec <- c()
    if (include_intercept == TRUE){
        om_vec <- c(100^2)
    }
    for (i in seq_len(lags)) {
        om_vec <- c(om_vec, rep((overall_tightness / i)^2, m))
    }
    return(diag(om_vec))
}


#' @title Generate state_of_the_world
#' 
#' @description 
#' Generates a data frame with the state of the world, including the t-notation
#' used by the rest of the package.
#' 
#' @param start_t Which observation to consider as t = 1.
#' @keywords internal

gen_sotw <- function(start_t) {
    max_t <- nrow(oscbvar::macrodata)
    sub <- oscbvar::macrodata[start_t:max_t, ]
    t <- c(1:nrow(sub))
    sotw <- cbind(t, sub)
    return(sotw)
}