Forecasting monthly wine sales series using naive method

Loading the packages

library(forecast)
library(knitr)

Reading the data

wines <- read.csv(file="AustralianWines.csv")

Excluding the missing data.

wines <- na.omit(wines)

Data partitioning

Data partitioning for red wine

redwine.ts <- ts(wines$Red, start=c(1980,1), end=c(1994,12), frequency = 12)
time <- time(redwine.ts)
n.valid <- 2
n.train <- length(redwine.ts) - n.valid
redwine.train.ts <- window(redwine.ts, start=time[1], end=time[n.train])
redwine.valid.ts <- window(redwine.ts, start=time[n.train+1], end=time[n.train+n.valid])

Data partitioning for rose wine

rosewine.ts <- ts(wines$Rose, start=c(1980,1), end=c(1994,12), frequency = 12)
time <- time(rosewine.ts)
n.valid <- 2
n.train <- length(rosewine.ts) - n.valid
rosewine.train.ts <- window(rosewine.ts, start=time[1], end=time[n.train])
rosewine.valid.ts <- window(rosewine.ts, start=time[n.train+1], end=time[n.train+n.valid])

Naive method

The R package forecast has the function naive and snaive which can be used to generate forecasting using naive and seasonal naive methods respectively. We are going to generate naive and seasonal naive forecasts for red and rose wine in the validation data.

Naive forecast for the validation period in red wine

naive.red <- naive(redwine.train.ts, h=2)
naive.red$mean

##       Nov  Dec
## 1994 2226 2226

Seasonal naive method for the validation period in red wine

snaive.red <- snaive(redwine.train.ts, h=2)
snaive.red$mean

##       Nov  Dec
## 1994 2374 2535

Naive forecast for the validation period in rose wine

naive.rose <- naive(rosewine.train.ts, h=2)
naive.rose$mean

##      Nov Dec
## 1994  57  57

Seasonal naive method for the validation period in rose wine

snaive.rose <- snaive(rosewine.train.ts, h=2)
snaive.rose$mean

##      Nov Dec
## 1994  56  78

Comparing naive and seasonal naive forecast

Red wine

indexs <- c(2,3,5) ## Only RMSE, MAE, and MAPE.
acc.naive.red <- accuracy(naive.red, redwine.valid.ts)
acc.naive.red <- acc.naive.red[, indexs]
kable(acc.naive.red)

	RMSE	MAE	MAPE
Training set	447.4490	312.0904	22.61781
Test set	411.9248	409.0000	15.49260

acc.seasonal.naive.red <- accuracy(snaive.red, redwine.valid.ts) 
acc.seasonal.naive.red <- acc.seasonal.naive.red[, indexs]
kable(acc.seasonal.naive.red)

	RMSE	MAE	MAPE
Training set	271.6625	206.2771	12.327351
Test set	183.2280	180.5000	6.874702

The tables above show that the seasonal naive method performs better than the simple naive method.

plot(redwine.ts,  main="Forecast for monthly red wine sales", xlab="Time", ylab="Thousand of liters")

lines(naive.red$mean, col=4)

lines(snaive.red$mean, col=2)

legend("topright", lty=1, col=c(4,2), 
       legend=c("Naive method", "Seasonal naive method"))

The graph above shows two forecast method applied to the red wine sales data in the period of November and December of 1994. It is clear that the seasonal naive method fits better to the validation data.

Rose wine

indexs <- c(2,3,5) ## Only RMSE, MAE, and MAPE.
acc.naive.rose <- accuracy(naive.rose, rosewine.valid.ts)
acc.naive.rose <- acc.naive.rose[, indexs]
kable(acc.naive.rose)

	RMSE	MAE	MAPE
Training set	34.94224	22.98305	141.63404
Test set	19.69772	18.00000	23.12534

acc.seasonal.naive.rose <- accuracy(snaive.rose, rosewine.valid.ts) 
acc.seasonal.naive.rose <- acc.seasonal.naive.rose[, indexs]
kable(acc.seasonal.naive.rose)

	RMSE	MAE	MAPE
Training set	32.573735	19.62651	141.3526
Test set	8.544004	8.00000	11.2210