R：根据条件拆分时间序列数据

Question

I am looking to split a time-series (price data) based on whether the following data-point shows an increase, decrease or no change.

Example:

In the set as.ts(c(1,2,3,4,5,5,4,3,2,1)) the datapoints followed by up would be 1, 2, 3, 4 .

I suspect this can be neatly solved with an if function, but am not sure how to specify the following row, nor am I sure as to how the syntax would look.

Any help would be greatly appreciated!

Answer 1

You can use one of two approaches:

1) Using diff as follows:

ifelse(diff(ts) > 0, 'up', ifelse(diff(ts) < 0, 'down', 'none'))

Output is:

Time Series:
Start = 2 
End = 10 
Frequency = 1 
[1]   up   up   up   up none down down down down

2) Using the lag function from dplyr package:

ifelse(ts > lag(ts), 'up', ifelse(ts < lag(ts), 'down', 'none'))

Output is as follows:

Time Series:
Start = 1 
End = 10 
Frequency = 1 
 [1] <NA>   up   up   up   up none down down down down

You can replace the first NA by using a default in the lag or separately.

EDIT: Based on additional output desired per comments below, I am editing the answer:

You can save the result of the above in a newts - for example, and extract the original values that belong to each of the 'moves' as follows:

newts <- c('up', ifelse(diff(ts) > 0, 'up', ifelse(diff(ts) < 0, 'down', 'none')))
sapply(unique(newts), function(x) ts[newts == x])

Output is as follows:

$up
[1] 1 2 3 4 5

$none
[1] 5

$down
[1] 4 3 2 1

These are three lists. You can do whatever else you want with them, or combine them into other data structures as desired.

NOTE: I am pre-pending the newts with up since the first element is an orphan with no move. Depending on how you want things to output, you may want to adjust this value to your taste.