R 中按组的时间序列预测

Question

I have time series data grouped by US state from 2008-2018 and I need to forecast values for 2019-2020. I tested the forecasting on Arizona and wrote something like this:

az <- subset(full_df, full_df$state == "AZ")

az$year <- lubridate::ymd(az$year, truncated = 2L)
az <- xts(az$variable, az$year)

forecast <- forecast(az, level = c(95), h = 2)

This works, producing two point estimates and their CIs.

My only issue is looping this over the entire original dataframe and producing the estimates for each state. Does anyone know how I would go about this?

Answer 1

You can create a function, split and apply. I have not tested this as you did not include any data,

my_forecast_fun(df){
  df$year <- lubridate::ymd(df$year, truncated = 2L)
  df <- xts(df$variable, df$year)
  my_forecast <- forecast(df, level = c(95), h = 2) # avoid naming objects the same as functions (i.e. forecast)
  return(my_forecast)
}

my_list <- split(full_df, full_df$state)

res <- lapply(my_list, my_forecast_fun)

res is a list with the forecast for each state