I'm trying to extract the forecast residuals using fabletools package. I know that I can extract the fitted model residuals using the augment() function but I don't know how that works for the forecasted values and I get the same results as the fitted model residuals. Here is an example:
library(fable)
library(tsibble)
lung_deaths <- as_tsibble(cbind(mdeaths, fdeaths))
## fitted model residuals
lung_deaths %>%
dplyr::filter(index < yearmonth("1979 Jan")) %>%
model(
ets = ETS(value ~ error("M") + trend("A") + season("A"))) %>%
augment()
# A tsibble: 120 x 7 [1M]
# Key: key, .model [2]
key .model index value .fitted .resid .innov
<chr> <chr> <mth> <dbl> <dbl> <dbl> <dbl>
1 fdeaths ets 1974 Jan 901 837. 64.0 0.0765
2 fdeaths ets 1974 Feb 689 877. -188. -0.214
3 fdeaths ets 1974 Mar 827 795. 31.7 0.0399
4 fdeaths ets 1974 Apr 677 624. 53.2 0.0852
5 fdeaths ets 1974 May 522 515. 7.38 0.0144
6 fdeaths ets 1974 Jun 406 453. -47.0 -0.104
7 fdeaths ets 1974 Jul 441 431. 9.60 0.0223
8 fdeaths ets 1974 Aug 393 388. 4.96 0.0128
9 fdeaths ets 1974 Sep 387 384. 2.57 0.00668
10 fdeaths ets 1974 Oct 582 480. 102. 0.212
# ... with 110 more rows
## forecast residuals
test <- lung_deaths %>%
dplyr::filter(index < yearmonth("1979 Jan")) %>%
model(
ets = ETS(value ~ error("M") + trend("A") + season("A"))) %>%
forecast(h = "1 year")
## defining newdata
Data <- lung_deaths %>%
dplyr::filter(index >= yearmonth("1979 Jan"))
augment(test, newdata = Data, type.predict='response')
# A tsibble: 120 x 7 [1M]
# Key: key, .model [2]
key .model index value .fitted .resid .innov
<chr> <chr> <mth> <dbl> <dbl> <dbl> <dbl>
1 fdeaths ets 1974 Jan 901 837. 64.0 0.0765
2 fdeaths ets 1974 Feb 689 877. -188. -0.214
3 fdeaths ets 1974 Mar 827 795. 31.7 0.0399
4 fdeaths ets 1974 Apr 677 624. 53.2 0.0852
5 fdeaths ets 1974 May 522 515. 7.38 0.0144
6 fdeaths ets 1974 Jun 406 453. -47.0 -0.104
7 fdeaths ets 1974 Jul 441 431. 9.60 0.0223
8 fdeaths ets 1974 Aug 393 388. 4.96 0.0128
9 fdeaths ets 1974 Sep 387 384. 2.57 0.00668
10 fdeaths ets 1974 Oct 582 480. 102. 0.212
# ... with 110 more rows
Any suggestions would be greatly appreciated.
I think you probably want forecast errors --- the difference between what is observed and what was predicted. See https://otexts.com/fpp3/accuracy.html for a discussion. To quote that chapter:
Note that forecast errors are different from residuals in two ways. First, residuals are calculated on the training set while forecast errors are calculated on the test set. Second, residuals are based on one-step forecasts while forecast errors can involve multi-step forecasts.
Here is some code to compute forecast errors on your example.
library(fable)
library(tsibble)
library(dplyr)
lung_deaths <- as_tsibble(cbind(mdeaths, fdeaths))
## forecasts
fcast <- lung_deaths %>%
dplyr::filter(index < yearmonth("1979 Jan")) %>%
model(
ets = ETS(value ~ error("M") + trend("A") + season("A"))
) %>%
forecast(h = "1 year")
## defining newdata
new_data <- lung_deaths %>%
dplyr::filter(index >= yearmonth("1979 Jan")) %>%
rename(actual = value)
# Compute forecast errors
fcast %>%
left_join(new_data) %>%
mutate(error = actual - .mean)
#> Joining, by = c("key", "index")
#> # A fable: 24 x 7 [1M]
#> # Key: key, .model [2]
#> key .model index value .mean actual error
#> <chr> <chr> <mth> <dist> <dbl> <dbl> <dbl>
#> 1 fdeaths ets 1979 Jan N(783, 8522) 783. 821 37.5
#> 2 fdeaths ets 1979 Feb N(823, 9412) 823. 785 -38.4
#> 3 fdeaths ets 1979 Mar N(742, 7639) 742. 727 -14.8
#> 4 fdeaths ets 1979 Apr N(570, 4516) 570. 612 41.7
#> 5 fdeaths ets 1979 May N(461, 2951) 461. 478 16.9
#> 6 fdeaths ets 1979 Jun N(400, 2216) 400. 429 29.5
#> 7 fdeaths ets 1979 Jul N(378, 1982) 378. 405 27.1
#> 8 fdeaths ets 1979 Aug N(335, 1553) 335. 379 44.5
#> 9 fdeaths ets 1979 Sep N(331, 1520) 331. 393 62.1
#> 10 fdeaths ets 1979 Oct N(427, 2527) 427. 411 -15.7
#> # … with 14 more rows
Created on 2020-11-03 by the reprex package (v0.3.0)