Is There an R Function to Map a Date Index? - r

Question

I have a data set with a lot of countries and currency data, like this:
iso date SPOT
<chr> <date> <dbl>
1 AUD 2000-01-03 0.658
2 AUD 2000-01-04 0.655
3 AUD 2000-01-05 0.658
4 AUD 2000-01-06 0.653
5 AUD 2000-01-07 0.655
6 AUD 2000-01-10 0.656
7 AUD 2000-01-11 0.658
8 AUD 2000-01-12 0.659
9 AUD 2000-01-13 0.668
10 AUD 2000-01-14 0.666
and I want to create an exact date index where the data for each is mapped to the day of one year ago, so the mapping the data to "LAG1" like this, where LAG1 = date - years(1):
iso date SPOT LAG1
<chr> <date> <dbl> <date>
1 AUD 2000-01-03 0.658 1999-01-03
2 AUD 2000-01-04 0.655 1999-01-04
3 AUD 2000-01-05 0.658 1999-01-05
4 AUD 2000-01-06 0.653 1999-01-06
5 AUD 2000-01-07 0.655 1999-01-07
6 AUD 2000-01-10 0.656 1999-01-10
7 AUD 2000-01-11 0.658 1999-01-11
8 AUD 2000-01-12 0.659 1999-01-12
9 AUD 2000-01-13 0.668 1999-01-13
10 AUD 2000-01-14 0.666 1999-01-14
This was my solution:
df %>%
mutate(LAG1=date-years(1)) %>%
select(iso,LAG1=date,LAG1_SPOT=SPOT) %>%
right_join(.,df,by=c("iso", "LAG1")) %>% as_tibble()
but I don't like it because it's a bunch of lines for something I think should be simpler, and I want to make it into a function.
Is there a better way to do this?

Answer 1

I think your intent of merging/joining is the right way to go. In fact, it's "right" because it will naturally deal with data anomalies better. I also think there are a couple of small logic errors in your code.
Since your data doesn't have enough to look at past years, here is some fake data. I'm making SPOT just a sequence to help visualize the sequence, but otherwise it doesn't matter much. I'm also going to introduce two anomalies in the data to demonstrate how they will show in the end.
library(dplyr)
library(lubridate)
dates <- seq.Date(as.Date("2020-03-15"), by = "day", length.out = 5)
df <- tibble(
iso = rep(c("AUD", "USD"), each = 10),
date = rep(c(dates - years(1), dates), times = 2),
SPOT = 1:20
)
# data missingness
df <- df[-3,]
# repeated date
df$date[12] <- df$date[13]
df
# # A tibble: 19 x 3
# iso date SPOT
# <chr> <date> <int>
# 1 AUD 2019-03-15 1
# 2 AUD 2019-03-16 2
# 3 AUD 2019-03-18 4
# 4 AUD 2019-03-19 5
# 5 AUD 2020-03-15 6
# 6 AUD 2020-03-16 7
# 7 AUD 2020-03-17 8
# 8 AUD 2020-03-18 9
# 9 AUD 2020-03-19 10
# 10 USD 2019-03-15 11
# 11 USD 2019-03-16 12
# 12 USD 2019-03-18 13
# 13 USD 2019-03-18 14
# 14 USD 2019-03-19 15
# 15 USD 2020-03-15 16
# 16 USD 2020-03-16 17
# 17 USD 2020-03-17 18
# 18 USD 2020-03-18 19
# 19 USD 2020-03-19 20
Using your code from above, we see this:
df %>%
mutate(date = date - years(1)) %>%
rename(LAG1_SPOT = SPOT) %>%
right_join(., df, by = c("iso", "date"))
# # A tibble: 19 x 4
# iso date LAG1_SPOT SPOT
# <chr> <date> <int> <int>
# 1 AUD 2019-03-15 6 1
# 2 AUD 2019-03-16 7 2
# 3 AUD 2019-03-18 9 4
# 4 AUD 2019-03-19 10 5
# 5 AUD 2020-03-15 NA 6
# 6 AUD 2020-03-16 NA 7
# 7 AUD 2020-03-17 NA 8
# 8 AUD 2020-03-18 NA 9
# 9 AUD 2020-03-19 NA 10
# 10 USD 2019-03-15 16 11
# 11 USD 2019-03-16 17 12
# 12 USD 2019-03-18 19 13
# 13 USD 2019-03-18 19 14
# 14 USD 2019-03-19 20 15
# 15 USD 2020-03-15 NA 16
# 16 USD 2020-03-16 NA 17
# 17 USD 2020-03-17 NA 18
# 18 USD 2020-03-18 NA 19
# 19 USD 2020-03-19 NA 20
Since I believe your intent is to compare this year's data with last year's data, then the above shows that we have paired them, but the date of reference is last year. I suggest that you should be using +:
df %>%
mutate(date = date + years(1)) %>%
rename(LAG1_SPOT = SPOT) %>%
right_join(., df, by = c("iso", "date"))
# # A tibble: 20 x 4
# iso date LAG1_SPOT SPOT
# <chr> <date> <int> <int>
# 1 AUD 2019-03-15 NA 1
# 2 AUD 2019-03-16 NA 2
# 3 AUD 2019-03-18 NA 4
# 4 AUD 2019-03-19 NA 5
# 5 AUD 2020-03-15 1 6
# 6 AUD 2020-03-16 2 7
# 7 AUD 2020-03-17 NA 8
# 8 AUD 2020-03-18 4 9
# 9 AUD 2020-03-19 5 10
# 10 USD 2019-03-15 NA 11
# 11 USD 2019-03-16 NA 12
# 12 USD 2019-03-18 NA 13
# 13 USD 2019-03-18 NA 14
# 14 USD 2019-03-19 NA 15
# 15 USD 2020-03-15 11 16
# 16 USD 2020-03-16 12 17
# 17 USD 2020-03-17 NA 18
# 18 USD 2020-03-18 13 19
# 19 USD 2020-03-18 14 19
# 20 USD 2020-03-19 15 20
This also shows how data anomalies will present. First, in AUD we see that 03-17 is missing data from last year, so we have nothing to compare the 8 spot against. This is just the fact that we are missing data. Unavoidable, but a lag here would have given us data, likely from the wrong date. Second, we see that our dupe-data (acquisition systems are imperfect!), we now have two rows for USD on 2020-03-18, which is certainly suspect (but outside the scope of your question). But we have compared both of 2019's values with the single 2020 value.
If the data anomalies never show up in your data, I still think join is the correct method for dealing with this, as if there is ever a time that lag will find the wrong row (leap-years?), you will never know that it failed: you'll get data and use it with no indication.
BTW: if you are just looking to reduce the four lines of code, this is perfectly equivalent:
transmute(df, iso, date = date + years(1), LAG1_SPOT = SPOT) %>%
right_join(., df, by = c("iso", "date"))