tidy-rolling-regression.r

library(dplyr)
library(slider)
library(lubridate)
library(tsibbledata)

# Google, Apple, Facebook, Amazon stock
gafa_stock <- as_tibble(gafa_stock)
gafa_stock <- select(gafa_stock, Symbol, Date, Close, Volume)

head(gafa_stock, 2)
#> # A tibble: 2 x 4
#>   Symbol Date       Close   Volume
#>   <chr>  <date>     <dbl>    <dbl>
#> 1 AAPL   2014-01-02  79.0 58671200
#> 2 AAPL   2014-01-03  77.3 98116900

# Insert complex model here...
model <- function(df) {
  lm(Close ~ Volume, df)
}

# Arrange and group by `Symbol` (i.e. stock)
gafa_stock <- gafa_stock %>%
  arrange(Symbol, Date) %>%
  group_by(Symbol)

# - `cur_data()` to get the current group's data frame
# - 10 day rolling regression per group (current day + 9 days before)
gafa_stock %>%
  mutate(
    regression = slide_index(
      .x = cur_data(), 
      .i = Date,
      .f = model, 
      .before = days(9),
      .complete = TRUE
    )
  )
#> # A tibble: 5,032 x 5
#> # Groups:   Symbol [4]
#>    Symbol Date       Close    Volume regression
#>    <chr>  <date>     <dbl>     <dbl> <list>    
#>  1 AAPL   2014-01-02  79.0  58671200 <NULL>    
#>  2 AAPL   2014-01-03  77.3  98116900 <NULL>    
#>  3 AAPL   2014-01-06  77.7 103152700 <NULL>    
#>  4 AAPL   2014-01-07  77.1  79302300 <NULL>    
#>  5 AAPL   2014-01-08  77.6  64632400 <NULL>    
#>  6 AAPL   2014-01-09  76.6  69787200 <NULL>    
#>  7 AAPL   2014-01-10  76.1  76244000 <NULL>    
#>  8 AAPL   2014-01-13  76.5  94623200 <lm>      
#>  9 AAPL   2014-01-14  78.1  83140400 <lm>      
#> 10 AAPL   2014-01-15  79.6  97909700 <lm>      
#> # … with 5,022 more rows