For loop regression estimations (ECM model) - r

Question

I am estimating a time series Error Correction Model on my data (with package 'ecm'). In below code you can see that I specify the short and long term variables with xeq and xtr.
These variables are independent variables and estimate on the dependent variable: Sales.
In this case, it is a pooled model but I want to estimate this model unit by unit (so separate for every brand). Since my dataset is rather large and consists of 360 product categories, each having 3 brands (brand 2, brand 3 and brand 4).
xeq <- DatasetThesisSynergyClean[c('lnPrice', 'lnAdvertising', 'lnDisplay', 'IntrayearCycles', 'lnCompetitorPrices', 'lnCompADV', 'lnCompDISP' , 'ADVxDISP', 'ADVxCYC', 'DISPxCYC', 'ADVxDISPxCYC')]
xtr <- DatasetThesisSynergyClean[c('lnPrice', 'lnAdvertising', 'lnDisplay', 'IntrayearCycles', 'lnCompetitorPrices', 'lnCompADV', 'lnCompDISP', 'ADVxDISP', 'ADVxCYC', 'DISPxCYC', 'ADVxDISPxCYC')]
model11 <- ecm(DatasetThesisSynergyClean$lnSales, xeq, xtr, includeIntercept=TRUE)
summary(model11)
What I want is to generate an output for every brand of every category. To give you glimpse of my data, please run this code:
structure(list(Week = 7:17, Category = c("2", "2", "2", "2",
"2", "2", "2", "2", "2", "2", "2"), Brand = c("3", "3", "3",
"3", "3", "3", "3", "3", "3", "3", "3"), Display = c(0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0), Sales = c(0, 0, 0, 0, 13.440948, 40.097397,
32.01384, 382.169189, 2830.748779, 4524.460938, 1053.590576),
Price = c(0, 0, 0, 0, 5.949999, 5.95, 5.950003, 4.87759,
3.787015, 3.205987, 4.898724), Distribution = c(0, 0, 0,
0, 1.394019, 1.386989, 1.621416, 8.209759, 8.552915, 9.692097,
9.445554), Advertising = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0), lnSales = c(11.4945151554497, 11.633214247508, 11.5862944141137,
11.5412559646132, 11.4811122484454, 11.4775106999991, 11.6333660772506,
11.4859819773102, 11.5232680456161, 11.5572670584292, 11.5303686934256
), IntrayearCycles = c(4.15446534315765, 3.62757053512638,
2.92387946552647, 2.14946414386239, 1.40455011205262, 0.768856938870769,
0.291497141953598, -0.0131078404184544, -0.162984144025091,
-0.200882782749248, -0.182877633924882), `Competitor Advertising` = c(10584.87063,
224846.3243, 90657.72553, 0, 0, 0, 2396.54212, 0, 0, 0, 40343.49444
), `Competitor Display` = c(0.385629, 2.108133, 2.515806,
4.918288, 3.81749, 3.035847, 2.463194, 3.242594, 1.850399,
1.751096, 1.337943), `Competitor Prices` = c(5.30989, 5.372752,
5.3717245, 5.3295525, 5.298393, 5.319466, 5.1958415, 5.2941095,
5.296757, 5.294059, 5.273578), ZeroSales = c(1, 1, 1, 1,
0, 0, 0, 0, 0, 0, 0)), .Names = c("Week", "Category", "Brand",
"Display", "Sales", "Price", "Distribution", "Advertising", "lnSales",
"IntrayearCycles", "Competitor Advertising", "Competitor Display",
"Competitor Prices", "ZeroSales"), row.names = 1255:1265, class = "data.frame")
As you can see, I have all the categories and brands stored in rows. To get an estimation on every single brand I want to write a for loop, but I don't really know how to specify the right category and brand in order to save this output separately.
Eventually want to store the coefficients, std. error, t-values and p-values, of all brands in 4 separate dataframes. But first I need to obtain the output, can you guys help me out?

Answer 1

You could use dplyr like this:
f <- function(.) {
xeq <- as.data.frame(select(., lnPrice, lnAdvertising, lnDisplay, IntrayearCycles, lnCompetitorPrices, lnCompADV, lnCompDISP, ADVxDISP, ADVxCYC, DISPxCYC, ADVxDISPxCYC))
xtr <- as.data.frame(select(., lnPrice, lnAdvertising, lnDisplay, IntrayearCycles, lnCompetitorPrices, lnCompADV, lnCompDISP, ADVxDISP, ADVxCYC, DISPxCYC, ADVxDISPxCYC))
print(xeq)
print(xtr)
summary(ecm(.$lnSales, xeq, xtr, includeIntercept = TRUE))
}
Models <- DatasetThesisSynergyClean %>%
group_by(Category, Brand) %>%
do(Model = f(.))
Models$Category
[1] "2" "3"
Models$Brand
[1] "3" "3"
Models$Model
[[1]]
Call:
lm(formula = dy ~ ., data = x)
# ... and so on
You end up with a list of 3 items (the categories, brands and model summary objects) and length equal to unique category/brand combinations. Could not try it properly, since I do not have the complete data. Model summary for Category 3, Brand 3:
Models$Model[[which(Models$Category == 3 & Models$Brand == 3)]]
Update:
If you want standalone object for each model you can give them corresponding names and use list2env():
names(Models$Model) <- paste0("C", Models$Category, "B", Models$Brand)
list2env(Models$Model, .GlobalEnv)

Answer 2

I would suggest you take a look at some of the tidyverse packages, and consider using a vectorised approach combining split(df, list(df$Category, df$Group)) and purrr's map() function to apply a function to each of your smaller datasets. The code would be something like this:
df %>%
split(f = list(.$Category, .$Brand)) %>%
map(a_function_for_each_group) %>%
bind_rows()
I hope i have understood your question correctly.