## ----include = FALSE----------------------------------------------------------
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 6,
  fig.height = 3.5
)
set.seed(7)
library(bridgr)

## ----data---------------------------------------------------------------------
gdp_growth <- suppressMessages(tsbox::ts_na_omit(tsbox::ts_pc(gdp)))
gdp_nowcast <- gdp_growth |>
  dplyr::slice_head(n = nrow(gdp_growth) - 1)

baro_ragged <- baro |>
  dplyr::slice_head(n = nrow(baro) - 2)

tail(gdp_growth)
tail(gdp_nowcast)
tail(baro_ragged)

## ----compare-methods----------------------------------------------------------
predict_methods <- c("last", "mean", "auto.arima", "ets")

models <- lapply(
  predict_methods,
  function(method) {
    mf_model(
      target = gdp_nowcast,
      indic = baro_ragged,
      indic_predict = method,
      indic_aggregators = "mean",
      target_lags = 1,
      h = 1
    )
  }
)
names(models) <- predict_methods

forecast_df <- dplyr::bind_rows(
  lapply(predict_methods, function(method) {
    fc <- forecast(models[[method]])
    dplyr::tibble(
      method = method,
      forecast_time = fc$time,
      forecast = as.numeric(fc$mean)
    )
  })
)

forecast_df

## ----extension-plot-----------------------------------------------------------
last_obs <- max(baro_ragged$time)
last_obs_value <- baro_ragged$values[baro_ragged$time == last_obs]
future_months <- seq(last_obs, by = "month", length.out = 3)[-1]

observed_tail <- baro_ragged |>
  dplyr::slice_tail(n = 9) |>
  dplyr::transmute(
    time = .data$time,
    value = .data$values,
    method = "observed"
  )

extended <- dplyr::bind_rows(
  lapply(predict_methods, function(method) {
    indicator_tbl <- models[[method]]$indic
    extension <- indicator_tbl |>
      dplyr::filter(.data$time %in% future_months) |>
      dplyr::transmute(
        time = .data$time,
        value = .data$values,
        method = method
      )
    # join to the last observed point for a continuous line
    dplyr::bind_rows(
      dplyr::tibble(time = last_obs, value = last_obs_value, method = method),
      extension
    )
  })
)

ggplot2::ggplot(
  dplyr::bind_rows(observed_tail, extended),
  ggplot2::aes(x = .data$time, y = .data$value, color = .data$method)
) +
  ggplot2::geom_line(linewidth = 0.8) +
  ggplot2::geom_point(size = 1.6) +
  ggplot2::labs(
    title = "Indicator completion at the ragged edge",
    x = NULL,
    y = "KOF barometer",
    color = NULL
  ) +
  theme_bridgr()

## ----direct-------------------------------------------------------------------
direct_model <- mf_model(
  target = gdp_nowcast,
  indic = baro_ragged,
  indic_predict = "direct",
  h = 1
)

forecast(direct_model)

## ----missing-impute, warning = TRUE-------------------------------------------
baro_internal_na <- baro_ragged
internal_gap <- seq(nrow(baro_internal_na) - 14, nrow(baro_internal_na) - 13)
baro_internal_na$values[internal_gap] <- NA_real_

invisible(tryCatch(
  mf_model(
    target = gdp_nowcast,
    indic = baro_internal_na,
    indic_predict = "last",
    indic_aggregators = "mean",
    h = 1
  ),
  error = function(e) message("Error captured: ", conditionMessage(e))
))

imputed_model <- mf_model(
  target = gdp_nowcast,
  indic = baro_internal_na,
  indic_predict = "last",
  indic_aggregators = "mean",
  h = 1,
  missing = "impute"
)

forecast(imputed_model)