Add recession bars/shading to chart

[grantmcdermott]

I needed this for work today.

Here's simple recbar function for manually adding recession bars to an existing plot:

recbar = function(
  start,
  end,
  ymin = par('usr')[3],
  ymax = par('usr')[4],
  col = "hotpink",
  alpha = 0.3,
  border = NA
) {
    rect(start, ymin, end, ymax, col = adjustcolor(col, alpha), border = border)
}

library(tinyplot)
plt(
    GNP ~ Year,
    data = longley,
    type = "l",
    draw = recbar(c(1948, 1953, 1957), c(1949, 1954, 1958))
)

For a more sophisticated version, we could combine this simple recbar function with a sister function that actually calculates the recession dates (e.g, based on some window logic like two consecutive quarters of negative growth). Proof of concept:

recessions = function(gdp, dates, window = 2) {
  streaks = rle(c(NA, 0 > diff(gdp)))
  rec_flag = streaks$lengths>=window & streaks$values==TRUE
  end = cumsum(streaks$lengths)
  start = end - streaks$lengths #+ 1
  start = dates[start[rec_flag]]
  end = dates[end[rec_flag]]
  return(data.frame(start = start, end = end))
}

gdp = read.csv("https://fred.stlouisfed.org/graph/fredgraph.csv?id=GDPC1&cosd=2000-01-01") |>
  setNames(c("date", "gdp"))
gdp$date = as.Date(gdp$date)
rec_dates = recessions(gdp$gdp, gdp$date)

plt(
    gdp ~ date,
    data = gdp,
    type = "l",
    draw = recbar(rec_dates$start, rec_dates$end),
    main = "US GDP (2000 to present)",
    sub = "Note: Shaded regions denote recessions"
)

It shouldn't be too tricky to roll these into a dedicated type_recbar() tinyplot type if that sounds useful?

[vincentarelbundock]

This is cool.

Isn't there a way to use type_rect() for that?

"Recession-bar" seems application-specific. Maybe there's a viz-related word to use to describe these. And maybe we can generalize a bit? It feels a bit narrow to add a new type.

[zeileis]

Nice! I think this is something that especially time series people will find useful!

However, the most code would probably go into the computation of start and end dates for which different heuristics/algorithms would be used. So I'm not sure whether tinyplot would be the right place to package this functionality.

And just in case it is useful for someone, below is the code for the same graphic using type_rect() instead of the custom draw = recbar(...).

tinyplot(
  gdp ~ date,
  data = gdp,
  type = "l",
  main = "US GDP (2000 to present)",
  sub = "Note: Shaded regions denote recessions"
)
tinyplot(
  xmin = rec_dates$start,
  xmax = rec_dates$end,
  ymin = par('usr')[3],
  ymax = par('usr')[4],
  type = "rect",
  fill = adjustcolor("hotpink", 0.3),
  col = "transparent",
  add = TRUE
)

P.S.: I posted before I saw Vincent's response. 🤓

[grantmcdermott]

Ha, fair enough. Perhaps it is too narrow and specific. (Would a type_shade/type_swatch be enough to give it generality?)

One nice thing about the draw approach, as opposed to plt_add, is that shaded regions are plotted below the actual time series lines. But you are both correct that we could just as well call rect directly.

Perhaps this is just one for the "Tips & tricks" vignette.

[zeileis]

Good idea. And regarding the plotting order: If necessary you can also do

tinyplot(y ~ x, data = d, type = "l", ...)
tinyplot_add(xmin = ..., xmax = ..., ymin = ..., ymax = ..., type = "rect", ...)
tinyplot_add()

or

tinyplot(y ~ x, data = d, type = "n", ...)
tinyplot_add(xmin = ..., xmax = ..., ymin = ..., ymax = ..., type = "rect", ...)
tinyplot_add(type = "l")

It is not as efficient as the draw = ... approach but has the advantage that you only combine standard types.