If future sales are a black box for you, you might hardly make profitable decisions right now. How many items to purchase? Which products require aggressive advertising? Where are the best markets for us for the next year? Luckily, statistics might help if you have enough historical data. This tool let you generate sales by periods and apply statistical methods to forecast further periods.
It is the simplest but still widely used statistical method for time series forecast. Using the method you consider sales trends being linear without seasonal effects, without a purely defined trend, and without smoothing abnormal observation.
Moving Average (MA) and Autoregressive Moving Average (ARMA)
The moving average method takes into account 'errors' in previous observations, and in comparison to the AR method smooths abnormal data.
The autoregressive moving average method is a combination of both AR and MA methods. To apply the ARMA method use the MA method with auto regression coefficient (P coefficient) as 2
Autoregressive Integrated Moving Average (ARIMA)
The method which also combines the methods AR and MA, but beside that it tries to make data stationary. It is appropriate to use for historical data with pure trend but without seasonal changes.
Seasonal Autoregressive Integrated Moving-Average (SARIMA)
The SARIMA method enriches the ARIMA method with considering seasonal changes. It is one of the most complex and wide spread methods utilized for forecasting time series now
Simple Exponential Smoothing (SES)
The SES model usage is similar to the AR method, but instead of relying upon linear function, it exploits exponential one
Holt Winter’s Exponential Smoothing (HWES)
The HWES method enriches the SES method to work with time series trends and seasonal effects.
When this tool should be used
- You have enough historical sales data, since it is senseless to make forecast based on last 5 days of sales
- Your sales are regular and they are not chaotic, meaning that your decisions do not have 100% impact on your sales and there is at least some correlation between market demand and your sales
- You have some seasonal changes and/or trends, which you noticed but can't fully analyse
To guarantee tool correct work you would need a number of Python libraries: pandas, numpy, statsmodels, scipy, xlsxwriter. To install those packages execute the command:
pip install pandas numpy statsmodels scipy xlsxwriter