Three Forecasting Techniques
Qualitative technique: This forecasting process uses the qualitative data i.e. expert opinion, information about special event and may or may not take the past sales data into consideration . For example, an expert in anticipation of an Apple’s 25th anniversary can predict the increase in sales from loyal apple fan base and make such prediction.
This technique is often used when the data is scarce and most commonly, during the first product introduction to market. This process often relies on human judgement and rating schemes. Some of the most popular qualitative methods commonly used today are Delphi, Market Research, Panel Consensus, Visionary Forecast and Historical Analogy. 
Causal models: This forecasting process uses highly refined model and specific information about the relationship between system elements, to make forecasts. For example, if temperature drops below 15 degree then increase in the sales of the heater can be forecasted. The causal model aims to extract such cause and effect patterns. Some of the most commonly used causal models are Regression models, Econometric Models, Intention-To Buy & Anticipation Surveys, Input- Output Model, Economic Input-Output Model, Diffusion Index, Leading Indicator and Life-Cycle Analysis.
Statistical forecasting: This forecasting process is based on time series analysis and projections. It focusses entirely on finding statistical patterns and change in patterns, as observed from the historical data and use it to make the future prediction. This process works with the assumption that the existing patterns in the data will continue and it works towards finding past statistical pattern, to be used later to make forecasts. The process works by developing the mathematical model (formula) out of the past patterns and trends and tests the mathematical model against the held-out test data for reasonableness and confidence. Since the mathematical model is used, it is exactly why this process is known as statistical forecasting.
In this mathematical model formulation process, when the errors are found against the test data, then often such errors are used as the basis for refining the model to yields less error / more confidence. This process is often repeated, until the final forecasting model that generates the least error is formulated. 
Some of the common Statistical forecasting methods are Moving Average (MA), Exponential Smoothing (ETS), Box- Jenkins (BJ), X-11 and Trend Projections (TP). These techniques in a way or another model the historical data into statistics, either by taking the average of several consecutive data points to find the larger trend (MA), or by averaging with more weight given to new data to older data(ETS) or finding out the mathematical model for the data (BJ) or decomposing the trend into seasonal, trend cycle and irregular elements (X-11) or fitting a trend line to a mathematical equation and projecting(TP).
One of the major challenge often faced by these statistical model is that, the prediction based on the raw data is extremely difficult due to the presence of the all trends (seasonal, longer trends and irregular trends) combined into a single trend. And unfortunately, these existing methods have difficulty dealing with them. Most of them either identify only the seasonal or the trend or the irregularity component and fail to encounter the combined effect of trend and cycles and irregular components effectively, which leads to poor forecast.
|1.”Harvard Business Review Article : How to Choose the Right Forecasting Technique,” [Online]. Available: https://hbr.org/1971/07/how-to-choose-the-right-forecasting-technique.|
|2. “Statistical Forecasting, John Galt, the forecast experts,” [Online]. Available: http://johngalt.com/galt_university/learning-resources/business-forecasting-glossary/statistical-forecasting/.|
|3.”Statistical Forecasting, Introduction: Eyeon Forecasting Experts,” [Online]. Available: http://www.eyeon.nl/documenten/Pid11/pid11_10_masterclass_statistical_forecasting.pdf.|