Sales forecasting is an essential step in business. It helps a company to achieve its goal and help in decision making. A good forecasting will help the company to earn a good profit with a good marketing strategy that fulfil the demand of the customer.
Sales forecasting is a very crucial part of a business. Inaccurate forecasting can lead to severe damages to the company. Forecasting can be divided into three types; short-term forecast, medium-term forecast, and long-term forecast. There are two types of forecasting methods are widely used in industries which is quantitative and qualitative techniques. Qualitative forecasting techniques include the jury of executive opinion, customer expectation, Delphi method and Bayesian decision theory. Quantitative techniques usually involve mathematical theories and computer software. Some of the common methods are regression analysis, exponential smoothing, moving average, Box-Jenkins, trend line analysis, neural networks and many more. The qualitative forecasting techniques are used in the unpredictable environment and when the source of data is not enough. Qualitative technique usually used when there is need to introduce a new product to the market. Quantitative forecasting techniques are used when an environment is predictable and with the presence of data from past period about sales. The quantitative technique is used for an existing product (Jelena & Vesna, 2006).
In the study of Furseth & August (2017), they have stated the importance of sales forecasting for companies. In this paper, the authors focused on quarterly forecasting. There is several importance of sales forecasting stated by authors which are gauging the demand of a product, managing inventories, financial and strategic planning, setting expectation and marketing strategies. The authors also stated several methods of sales forecasting, but this paper mainly focused on pipeline forecast. A forecast accuracy is defined by the difference between the forecast value and the actual sales. A good forecast is within 10% of actual outcome while an excellent forecast is within 5%. A good way to measure the accuracy of your forecast is to look at two important metrics: price variance and volume variance. It is very important to check on the forecast accuracy because to improve the sales forecast, it is important to first reflect on how and why the companies did not achieve the forecast accuracy.
In another study, conducted by Kurzak (2012), the author has stated about the importance of forecasting but in enterprise management. The author has stated that a good forecasting can point out the future goals, means and method of operation to achieve company’s goal. Forecasting is used in decision making processes and it could improve the accuracy of decision-making. The author has an emphasis on the importance of data in this paper. Forecasting can be divided into two categories based on data which is qualitative and quantitative data. Qualitative data describe the pattern and importance of the factors that affect events and its largely depend on the subjective judgment of an expert. This type of data usually based on intuition and experience from the past. There are several qualitative methods used for forecasting which is Delphi, decision tree and Monte Carlo methods. Quantitative data are time series. Quantitative methods are based on forecasting models constructed based on time series. This includes the model of trends, simple regression, economic science and analogue models
2.3 ARIMA Model
ARIMA model introduced by Box and Jenkins in 1978 had successfully been applied in forecasting social, economic, engineering, foreign exchange, and stock problems. ARIMA model is one of the most important and widely used time series models. The popularity of the ARIMA model was due to its statistical properties as well as the well-known Box–Jenkins methodology within the model building method (Robert and David, 2006). It consists of 3 main components namely, Autoregression (p), Differences (d), and Moving Average (q). There is three-stage method aimed at selecting an appropriate ARIMA model, which are identification, estimation and diagnostic checking. It used as a benchmark to evaluate many new modelling approaches (Zhang, 2003). Many researches had been carried out on the returns of time series data using ARIMA model.
On the other hand, Cholette and Lamy (1986) applied multivariate ARIMA models to deal with the presence of irregularity in leading or explanatory variables in time series. In the proposed ARIMA models with filtering, the series were smoothed before modelling. The forecasts were then compared with smoothed data, which allowed a more relevant assessment of the forecasting performance. The smoothing also prevented irregular fluctuations in leading or explanatory series from migrating to the forecasts of the coincident or dependent series. As a result, an application of the proposed univariate and bivariate ARIMA models with filtering to the Canadian Index of Industrial Production and to the Canadian Composite Leading Indicator yields better forecasts than the standard ARIMA models without filtering.
In the study of David (1990), he argued that, conceptually, ARIMA models were more readily expanded than simpler method’s models to represent real world time series that contain interventions, calendar variation, outliers and variance changes. He insisted that ARIMA modelling had the ability to go beyond the basic univariate model. The motivation of his study comes from the perspectives that a number of empirical studies published in the forecasting literature in the 1970’s and 1980’s which had come to the conclusion that univariate ARIMA time series modelling (Box-Jeans) was less accurate as compare to some simpler and older alternatives, including various exponential smoothing methods. In his point of view, ARIMA model method should be given a chance to demonstrate its maximum potential in any empirical accuracy investigation.