```{r, echo = TRUE, message = FALSE, warning = FALSE}
library(xts)
library(tseries)
library(fpp2)
library(ggfortify)
library(forecast)
library(readr)
library(tidyverse)
library(nortest)
library(CombMSC)
#a---------
dat <- readr::read_csv(file.choose())
price_ts <- ts(dat, frequency = 12, start = c(1991,1))

#b------
price <- splitTrainTest(price_ts, numTrain = length(price_ts) - 24)
price.train <- price$train
price.test <- price$test
#c--------
plot(price_ts,ylab='price',xlab='month',type='o')
adf.test(price_ts)
#tidak stasioner karena p value > 0.05
#maka deferensiasi akan ditambah supaya stasionier:
diff_price_ts<-diff(price_ts,1)
adf.test(diff_price_ts)
plot(diff_price_ts,ylab='price',xlab='month',type='o')

#maka dilakukan split lagi
price <- splitTrainTest(diff_price_ts, numTrain = length(diff_price_ts) - 24)
price.train <- price$train
price.test <- price$test

#d-----
arima_price_ts<-auto.arima(price.train)

#e----
fit.train <- price.train %>% Arima(order = c(0,1,0), seasonal = c(1,0,0))
summary(fit.train)
fit.train %>% checkresiduals()
fit.train %>% residuals() %>% nortest::ad.test()
ggtsdiag(fit.train)

#f---
fit.train %>% forecast(h = 24) %>%
  autoplot() + ggtitle(fit.train)

#g------
model.test <- Arima(price.test, model = fit.train)
model.test %>% forecast::accuracy()