Forecasting Seasonal Sales Amount in R

In this article, we will explore the concept of forecasting seasonal sales amount using R. We will delve into the details of how to prepare and forecast seasonal data using popular libraries such as dplyr, lubridate, and forecast.

Understanding Seasonality

Seasonality refers to the regular fluctuations in a time series that occur at fixed intervals, often due to external factors such as weather, holidays, or economic cycles. In the context of sales forecasting, seasonality is essential to accurately predict future sales amounts.

Preparing Data for Forecasting

Before we dive into forecasting, it’s crucial to prepare our data by exploring its underlying characteristics and patterns.

# Load necessary libraries
library(dplyr)
library(lubridate)
library(forecast)

# Create a sample dataset with fake sales data
set.seed(4)
amount_2014 <- c(sample(3000:3500, 6), sample(4000:5000, 6))
set.seed(5)
amount_2015 <- c(sample(3000:3500, 6), sample(4000:5000, 6))
set.seed(6)
amount_2016 <- c(sample(3000:3500, 6), sample(4000:5000, 4))

sales <- data.frame(year = c(rep(2014, 12), rep(2014, 12), rep(2016, 10)),
                   month = c(1:12, 1:12, 1:10),
                   amount = c(amount_2014, amount_2015, amount_2016))

# Convert the data to a date format
sales <- sales %>% 
  mutate(Month = ymd(paste(year, month)), 
         truncated = 2) %>% 
  arrange(Month)

# Create a time series object from the sales data
sales_ts <- ts(sales$amount, start = c(sales$year[1], sales$month[1]),
               frequency = 12)

Exploring Seasonality using STL Decomposition

To identify seasonality in our data, we can use the seasonal trend decomposition (STL) method. This method breaks down the time series into three components: a trend, a seasonal component, and a residual.

# Perform STL decomposition on the sales data
stl_decomposition <- stl(sales_ts, s.window = 12)

# Plot the resulting decomposition components
plot(stl_decomposition)

In this example, we observe that the seasonality is relatively weak. However, there is a unit root in the series, which indicates that the time series may be non-stationary.

Identifying Unit Root using AutoARIMA

To determine if our data has a unit root, we can use the autoARIMA function from the forecast package.

# Perform autoARIMA on the sales data
auto_arima_result <- auto.arima(sales_ts)

# Print the results of the autoARIMA model
print(auto_arima_result)

The output indicates that the series is ARIMA(0,1,0), which means it has a single differencing term.

Using Holt-Winters Method for Seasonal Forecasting

To address the weak seasonality and unit root in our data, we can use the Holt-Winters method. This method is suitable for forecasting seasonal data that exhibits strong trends and seasonality.

# Perform Holt-Winters model on the sales data
holt_winters_model <- HoltWinters(sales_ts, 
                                 h = 13)

# Print the results of the Holt-Winters model
print(holt_winters_model)

The output shows that the forecasted values for each month are very close to the actual values.

Conclusion

In this article, we explored how to forecast seasonal sales amounts in R using popular libraries such as dplyr, lubridate, and forecast. We covered the steps involved in preparing data for forecasting, exploring seasonality using STL decomposition, identifying unit roots using autoARIMA, and applying the Holt-Winters method for seasonal forecasting. By following these steps, you can develop a robust forecasting model to predict future sales amounts with accuracy.

Recommendations

Always explore your data before performing any forecasting techniques.
Use STL decomposition to identify seasonality in your time series data.
Be cautious when dealing with non-stationary time series and consider using differencing or other transformations to make the data stationary.
The Holt-Winters method is suitable for forecasting seasonal data that exhibits strong trends and seasonality.