R/create_nl_data.R

In lpirfs: Local Projections Impulse Response Functions

#' @name create_nl_data
#' @title Compute data for nonlinear model with instrument variable approach
#' @description Function to create data for nonlinear model with instrument variable approach.
#' @param specs A \link{list}(). Inputs are created in \link{lp_nl_iv}.
#' @param endog_data A \link{data.frame} with dependent variables.
#' @return List with filled matrices of lagged left (y_nl) and right hand side (x_nl) variables.
#' @import dplyr
#' @keywords internal
#' @author Philipp Adämmer


create_nl_data <- function(specs, endog_data){

  # Check whether lag lengths have to be determined
  if (is.nan(specs$lags_criterion)) {

    # Get transition probabilities by logistic function?
    if(isTRUE(specs$use_logistic)){

    fz      <- get_vals_switching(specs$switching, specs)

                   } else {

    fz      <- specs$switching

    # Use first lag of value from switching function?
    if(isTRUE(specs$lag_switching)){

    fz      <-  dplyr::lag(fz, 1)


    }

   }

    # Select data for endogenous variables
    y_nl    <- endog_data

    # Make exogenous lagged data
    if(specs$lags_endog_nl == 0){
      x_nl <- data.frame(x = rep(Inf, nrow(endog_data)))
                 } else {
      x_nl <- create_lags(endog_data, specs$lags_endog_nl)
    }

    # Save names of exogenous variables
    linear_names <- names(x_nl)

    # Create tibbles with exogenous regime data and combine them to one data set
    x_nl_s1      <- x_nl %>%
                        dplyr::mutate_all(list(s1 = ~.*(1 - fz)))  %>%
                        dplyr::select(-one_of(linear_names))

    x_nl_s2      <- x_nl %>%
                        dplyr::mutate_all(list(s2 = ~.*fz))  %>%
                        dplyr::select(-one_of(linear_names))

    x_nl         <- cbind(x_nl_s1, x_nl_s2)


    # Check whether model type is 'iv'.
    # 0 = Normal model, 1 = IV type model
    # Prepare identified shock variable and add to exogenous data
    if(specs$model_type == 1){
    # Prepare shock variable
    shock           <- specs$shock
    colnames(shock) <- 'shock'
    shock_name      <- colnames(shock)

    # Make states of shock

    shock_s1        <- shock %>%
                          dplyr::mutate_all(list(shock_s1 = ~.*(1 - fz)))  %>%
                          dplyr::select(-one_of(shock_name))

    shock_s2        <- shock %>%
                          dplyr::mutate_all(list(shock_s2 = ~.*fz))        %>%
                          dplyr::select(-one_of(shock_name))

    # Add shock variable
    x_nl              <- cbind(shock_s1, shock_s2,  x_nl)
                      }

    # Include no trend, trend or quadratic trend
    if(specs$trend == 0){

      # Only constant
      x_nl             <-   x_nl

                    } else if (specs$trend == 1){

      # Constant and trend
      x_nl            <-   x_nl %>%
                            dplyr::mutate(trend = row_number())

                      }  else {

      x_nl             <-   x_nl %>%
                             dplyr::mutate(trend = row_number())     %>%
                             dplyr::mutate(sq_trend = trend^2)
                      }

    # Construct (lagged) exogenous data and merge it with lagged endogenous data
    if(!(is.null(specs$exog_data))){

      # Create lagged exogenous data and merge with endogenous data
      x_exog    <- create_lags(specs$exog_data, specs$lags_exog)     %>%
                                                        `rownames<-`(NULL)


      x_nl      <- cbind(x_nl, x_exog)

    }

    # Add contemporaneous data if supplied
    if(!(is.null(specs$contemp_data))){

      x_nl                         <- cbind(x_nl , specs$contemp_data)

    }


    # Combine endogenous and exogenous data
    yx_all    <- cbind(y_nl, fz, x_nl) %>%
                 stats::na.omit()

    yx_all    <- yx_all[, !(colSums(yx_all) == Inf)]

    y_nl      <- yx_all[, 1:ncol(endog_data)]  %>%
                 as.matrix()

    fz        <- yx_all[, (1 + ncol(endog_data))]

    x_nl      <- yx_all[, (2 + ncol(endog_data)): dim(yx_all)[2]]   %>%
                 as.matrix()



################################################################################
                                } else {
################################################################################

    # Create list to store matrices
      y_nl_store     <- rep(list(NaN), specs$max_lags)
      x_nl_store     <- rep(list(NaN), specs$max_lags)
      fz_store       <- rep(list(NaN), specs$max_lags)

    # Prepare data outside of loop

    # Create list with endogenous variables
      y_nl      <- endog_data

      # Get transition probabilities by logistic function?
      if(isTRUE(specs$use_logistic)){

        fz      <- get_vals_switching(specs$switching, specs)

                         } else {

        fz      <- specs$switching
                         }

        # Use first lag of value from switching function?
        if(isTRUE(specs$lag_switching)){

          fz            <-   dplyr::lag(fz, 1)

        }


    # Prepare instrument variable
      # Check whether model type is 'iv'.
      # 0 = Normal model, 1 = IV model
      # Prepare instrument variable and add to exogenous data
      if(specs$model_type == 1){
      shock           <- specs$shock
      colnames(shock) <- 'shock'
      shock_name     <- colnames(shock)
      }


    for(ii in 1:specs$max_lags){

  # Create lagged variables
      x_nl_temp      <- create_lags(endog_data, ii)

      linear_names   <- names(x_nl_temp)

      x_nl_s1        <- x_nl_temp %>%
                        dplyr::mutate_all(list(s1 = ~.*(1 - fz))) %>%
                        dplyr::select(-one_of(linear_names))

      x_nl_s2        <- x_nl_temp %>%
                        dplyr::mutate_all(list(s2 = ~.*fz))        %>%
                        dplyr::select(-one_of(linear_names))

      x_nl            <- cbind(x_nl_s1, x_nl_s2)


  # Prepare states of instrument
      if(specs$model_type == 1){
      shock_s1      <- shock %>%
                         dplyr::mutate_all(list(shock_s1 = ~.*(1 - fz)))  %>%
                         dplyr::select(-one_of(shock_name))

      shock_s2      <- shock %>%
                         dplyr::mutate_all(list(shock_s2 = ~.*fz))        %>%
                         dplyr::select(-one_of(shock_name))

  # Add state instruments
        x_nl          <- cbind(shock_s1, shock_s2,  x_nl)
      }

  # Add trend if set
     if(specs$trend == 0){

       x_nl            <-   x_nl


                    } else if (specs$trend == 1){

       x_nl            <-   x_nl                                    %>%
                             dplyr::mutate(trend = row_number())


                       }  else {

       x_nl            <-   x_nl                                       %>%
                              dplyr::mutate(trend = row_number())      %>%
                              dplyr::mutate(sq_trend = trend^2)
    }


  # Construct (lagged) exogenous data and merge it with lagged endogenous data
        if(!(is.null(specs$exog_data))){

          x_exog    <- create_lags(specs$exog_data, specs$lags_exog)     %>%
                       `rownames<-`(NULL)
          x_nl      <- cbind(x_nl, x_exog)

      }

  # Add contemporaneous data if set
        if(!(is.null(specs$contemp_data))){

          x_nl      <- cbind(x_nl , specs$contemp_data)

        }

  # Merge endogenous and exogenous data
        yx_all               <-  cbind(y_nl, fz, x_nl)         %>%
                                 stats::na.omit()

        y_nl_store[[ii]]      <-  yx_all[, 1:ncol(endog_data)]  %>%
                                 as.matrix()

        fz_store[[ii]]        <-  yx_all[, (1 + ncol(endog_data))]

        x_nl_store[[ii]]      <-  yx_all[, (2 + ncol(endog_data)): dim(yx_all)[2]]   %>%
                                 as.matrix()




    }
      # Save values te return in list
        y_nl <- y_nl_store
        x_nl <- x_nl_store
        fz   <- fz_store
  }

  return(list(y_nl = y_nl, x_nl = x_nl, fz = fz))
}