ggplot2: remove blank space for weekends and holid

问题:

running into issues while plotting stock data in ggplot2 and with an x-axis that contains gaps from weekends and holidays. this post has been very helpful, but i run into a variety of issues when trying to use ordered factors.

library(xts)
library(grid)
library(dplyr)
library(scales)
library(bdscale)
library(ggplot2)
library(quantmod)

getSymbols("SPY", from = Sys.Date() - 1460, to = Sys.Date(), adjust = TRUE, auto.assign = TRUE)

input <- data.frame(SPY["2015/"])
names(input) <- c("Open", "High", "Low", "Close", "Volume", "Adjusted")

# i've tried changing rownames() to index(), and the plot looks good, but the x-axis is inaccurate
# i've also tried as.factor()
xaxis <- as.Date(rownames(input)) 
input$xaxis <- xaxis

p <- ggplot(input)
p <- p + geom_segment(aes(x = xaxis, xend = xaxis, y = Low, yend = High), size = 0.50)           # body
p <- p + geom_segment(aes(x = xaxis - 0.4, xend = xaxis, y = Open, yend = Open), size = 0.90)    # open
p <- p + geom_segment(aes(x = xaxis, xend = xaxis + 0.4, y = Close, yend = Close), size = 0.90)  # close
p <- p + scale_y_continuous(scale_y_log10())
p + ggtitle("SPY: 2015")

The plot above (sans red boxes) is generated with the above code segment. And the following charts are some of the issues when attempting some solutions. First, if I try using the data frame's index, I will generate I nice looking graph, but the x-axis is inaccurate; the data currently ends in October, but in the plot below it ends in July:

xaxis <- as.Date(index(input))

Second, if I try coercing the rownames to an ordered factor, I lose my horizontal tick data (representing the open and the close).

xaxis <- factor(rownames(input), ordered = TRUE) 

The same issue of removing the horizontal ticks happens if I use the package bdscale, but the gridlines are cleaner:

p <- p + scale_x_bd(business.dates = xaxis)

回答1:

If you'd like to use bdscale for this, just tell it to use more gridlines:

ggplot(input) + 
  geom_segment(aes(x = xaxis, xend = xaxis, y = Low, yend = High), size = 0.50) +           # body 
  geom_segment(aes(x = xaxis - 0.4, xend = xaxis, y = Open, yend = Open), size = 0.90) +    # open
  geom_segment(aes(x = xaxis, xend = xaxis + 0.4, y = Close, yend = Close), size = 0.90)  + # close
  ggtitle("SPY: 2015") +
  xlab('') + ylab('') +
  scale_x_bd(business.dates=xaxis, max.major.breaks=10, labels=date_format("%b '%y")) # <==== !!!!

It should put October on the axis there, but it's not that smart. Womp womp. Pull requests welcome!

回答2:

You'll probably need to treat the dates as discrete values rather than continuous. This approach with a slightly simplified version of your code might look like:

getSymbols("SPY", from = Sys.Date() - 1460, to = Sys.Date(), adjust = TRUE, auto.assign = TRUE)
SPY <- SPY["2015/"]
colnames(SPY) <- sub("SPY.","", colnames(SPY))
month_brks <- c(1,endpoints(SPY, "months")[-1])

p <- ggplot(data.frame(xaxis=seq(nrow(SPY)), SPY))
p <- p + geom_linerange(aes(x=xaxis, ymin=Low, ymax=High), size=.5)
p <- p + geom_text(aes(x = xaxis,  y = Open), size = 4., label="-", hjust=.7, vjust=0)  # Open
p <- p + geom_text(aes(x = xaxis,  y = Close), size = 4., label="-", hjust=-.1, vjust=0)  # close
p <- p + scale_x_continuous(breaks=month_brks, labels=format(index(SPY)[month_brks], "%d %b %Y"))
p <- p + labs(title="SPY: 2015", x="Date", y="Price")

UPDATE

Updated treatment of axis labels.

回答3:

Well, you can tweak it manually, but it's kind of hacky. First, you should use index, so that your observations are numbered 1 to 188.

   input$xaxis <-index(as.Date(rownames(input)))

Then your own plot code:

p <- ggplot(input)
p <- p + geom_segment(aes(x = xaxis, xend = xaxis, y = Low, yend = High), size = 0.50)           # body
p <- p + geom_segment(aes(x = xaxis - 0.4, xend = xaxis, y = Open, yend = Open), size = 0.90)    # open
p <- p + geom_segment(aes(x = xaxis, xend = xaxis + 0.4, y = Close, yend = Close), size = 0.90)  # close
p <- p + scale_y_continuous(scale_y_log10()) + ggtitle("SPY: 2015") 

And finally, I looked in input where the breaks should be made, and supplied the labels manually:

p + scale_x_continuous(breaks=input$xaxis[c(1,62,125,188)], labels=c("jan","apr","jul","oct"))

NOTE HERE that I was lazy and just took the closest date for 1-jan, 1-apr 1-jul and 1-oct, because 1 jan is a holiday, the label "jan" stands below 2-jan. And I put the label "oct" below below 30-sep, the last entry in input. You can off course adjust this as you wish.

Off course, you could automate the label add a label field with date and extract the month.

回答4:

The method below uses faceting to remove spaces between missing dates, then removes white space between facets to recover the look of an unfaceted plot.

First, we create a grouping variable that increments each time there's a break in the dates (code adapted from this SO answer). We'll use this later for faceting.

input$group = c(0, cumsum(diff(input$xaxis) > 1))

Now we add the following code to your plot. facet_grid creates a new facet at each location where there was a break in the date sequence due to a weekend or holiday. scale_x_date adds major tick marks once per week and minor grid lines for each day, but you can adjust this. The theme function gets rid of the facet strip labels and the vertical spaces between facets:

p + facet_grid(. ~ group, space="free_x", scales="free_x") +
  scale_x_date(breaks=seq(as.Date("2015-01-01"),max(input$xaxis), "1 week"), 
               minor_breaks="1 day",
               labels=date_format("%b %d, %Y")) +
  theme(axis.text.x=element_text(angle=-90, hjust=0.5, vjust=0.5, size=11),
        panel.margin = unit(-0.05, "lines"),
        strip.text=element_text(size=0),
        strip.background=element_rect(fill=NA)) +
  ggtitle("SPY: 2015")

Here's the resulting plot. The spaces for weekends and holidays are gone. The major breaks mark each week. I set the weeks in thescale_x_date breaks argument to start on a Thursday since none of the holidays fell on a Thursday and therefore each facet has a major tick mark for the date. (In contrast, the default breaks would fall on a Monday. Since holidays often fall on a Monday, weeks with Monday holidays would not have a major tick mark with the default breaks.) Note, however, that the spacing between the major breaks inherently varies based on how many days the market was open that week.

回答5:

Haven't been able to get the OHLC to work - think you'd need a custom geom.

I know it isn't exactly what you asked for, but may I tempt you with a delicious candle chart instead?

library(dplyr)
library(bdscale)
library(ggplot2)
library(quantmod)
library(magrittr)
library(scales)

getSymbols("SPY", from = Sys.Date() - 1460, to = Sys.Date(), adjust = TRUE, auto.assign = TRUE)

input <- data.frame(SPY["2015/"]) %>% 
  set_names(c("open", "high", "low", "close", "volume", "adjusted")) %>%
  mutate(date=as.Date(rownames(.)))

input %>% ggplot(aes(x=date, ymin=low, ymax=high, lower=pmin(open,close), upper=pmax(open,close), 
                     fill=open<close, group=date, middle=pmin(open,close))) + 
  geom_boxplot(stat='identity') +
  ggtitle("SPY: 2015") +
  xlab('') + ylab('') + theme(legend.position='none') +
  scale_x_bd(business.dates=input$date, max.major.breaks=10, labels=date_format("%b '%y"))