I have aggregated retail weekly data with seasonal periods of 52.2 (a 53rd week every five years). I want to use this aggregated data to calculate a seasonal index that can be applied to each item within the category to derive its de-seasonalised demand.
Using stl, I would calculate the seasonal index as "seasonal" / "trend" + 1 (normalised to 52). I switched to tbats because my seasonality was not an integer and I have multiple seasonal periods (52.2 and 261)
I am using tbats with seasonal.periods = 52.2 and extract the components using tbats.components. The components are "observed", "level" and "season". Google has not revealed much in terms of what these components are and how to consume them. I also extracted the residuals
I noticed that "observed" is the log of my data. I also notice that season is changing over time (which is exactly what I want)
My questions are:
1.Is "season" a natural log too? 2.How can I extract the future "season" values? I can run a forecast on the data so I am assuming that there must be a projected "season" 3. What would be the best approach to calculating an "index" considering that it will be divided into the granular data. I am currently using: exp("season") / centered moving average(exp("season"))
My Data:
weeklyu <-structure(list(V1 = c(8L, 5L, 7L, 3L, 1L, 2L, 3L, 1L, 1L, 1L,
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 3L, 8L, 2L,
4L, 8L, 6L, 7L, 8L, 9L, 15L, 15L, 13L, 9L, 16L, 19L, 16L, 16L,
10L, 31L, 45L, 90L, 185L, 34L, 8L, 19L, 11L, 19L, 21L, 8L, 5L,
7L, 6L, 3L, 10L, 2L, 2L, 4L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L,
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L,
2L, 2L, 16L, 22L, 18L, 23L, 11L, 5L, 8L, 21L, 18L, 11L, 26L,
28L, 9L, 3L, 6L, 3L, 6L, 1L, 5L, 3L, 3L, 2L, 1L, 4L, 1L, 1L,
3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 5L,
19L, 11L, 17L, 23L, 50L, 52L, 23L, 18L, 22L, 44L, 37L, 22L, 30L,
32L, 47L, 34L, 30L, 26L, 25L, 44L, 87L, 65L, 30L, 17L, 12L, 2L,
16L, 14L, 17L, 6L, 7L, 3L, 6L, 7L, 8L, 11L, 12L, 4L, 1L, 3L,
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 1L, 2L, 1L, 1L, 1L, 1L,
1L, 1L, 1L, 1L)), .Names = "V1", class = "data.frame", row.names = c(NA,
-188L))
My Code:
wklytbat <- tbats(msts(weeklyu, seasonal.periods = 52.2, ts.frequency=52.2), use.parallel=FALSE)
extract season:
seasu <-data.table(exp(as.numeric(tbats.components(wklytbat)[,'season'])))
