The output of forecasting serves the needs of many functions including finance for capital expansion, operations for resource planning, or marketing for sales plan. Forecasting is a fundamental method for making future predictions so the business can position better in responding to predictable challenges or opportunities.
Forecasts are never 100% accurate – the very reason you forecast is to predict the future. Due to future uncertainty, there should be an established forecast error value to achieve rational estimates.
Forecasts require constant updating – forecast must be updated with new demand data to improve accuracy. The forecast must be modified with fresh demand information obtained from time to time.
Aggregate forecasts are generally more accurate than detailed forecasts – the degree of aggregation affects forecast accuracy. It is easy to forecast a company’s total revenue than the sales for each product line. Accuracy is even difficult for individual materials. Forecasting in totality is more accurate than the detailed forecast.
Forecasts are more accurate in short-term than long term forecasts – just like the weather, forecasting the weather today is more accurate than the weather next week. The closer the forecast to actual demand, the more accurate it gets.
Forecasts are more accurate in downstream than upstream supply chain – where you are in the supply chain impacts your forecast accuracy. A retailer’s forecast is more precise than the supplier’s forecast because a retailer is closer to the customer. Demand data get distorted gradually as it moves farther up the supply chain.
Before running a forecasting activity, you need to determine sources of demand. The demand comes from a variety of sources. They can be obtained from;
– Customer orders
– Inter-company orders
– Replenishment orders
– Marketing sales plan
There have been many forecasting methodologies you can use to predict future demand. The technique and the quality of information you use affect the accuracy of your forecast. Market demand variability is influenced by inevitable factors, therefore, identifying and understanding how these factors affect consumer behavior in the future is critical to forecasting accuracy. These factors are divided into two categories; external and internal.
While forecasting generally fueled by numbers, a human judgment must be involved to ensure behavioral insights are incorporated in the forecast.
External factors are the things you cannot control including;
– Market demographics
– Global economic trends
– Disruptive technology
– Consumer preference
– Political climate
Internal factors are the things you can control such as;
– Historical data
– Product life cycle
– New product
– Marketing promotions
– Management input
Figure 9. Forecasting techniques.
Qualitative technique – this method is subjective and relies heavily on human intelligence. Forecasting inputs may come from expert opinion, survey, panel consensus, management estimate, or product life cycle analysis. Qualitative forecasting is useful when a company enters a new market or when historical data is limited.
Quantitative technique – this is a data-driven forecast. Information is collected from a variety of sources to calculate future demand. The quantitative technique has the following procedures;
Time series – this method utilizes historical data to predict future demand. This is appropriate for a demand that does not deviate significantly from year to year. Time series forecasting has the following approaches.
Simple average – this is the simplest approach in forecasting. The forecast for the next period is taken from the average of past and current month.
Calculation: the forecast for July is the average of past and current demand. In this case, 500 in May and 550 in June.
= 500 + 550
Moving average – the moving average forecasting works like a simple average method, except that it takes several recent periods to make a forecast. You can have 3-month, 4-month, and so on moving average forecast.
Calculation: the forecast for July is taken from recent periods. For 3-month moving average, the demand for July is the average of April to June demand, and so on.
3-month moving average (April, May, and June)
= 400 + 500 + 550
4-month moving average (March, April, May, and June)
= 450 + 400 + 500 + 550
5-month moving average (February, March, April, May, and June)
= 550 + 450 + 400 + 500 + 550
Exponential smoothing – in this approach, weight or smoothing constant denoted as (α) is assigned to previous actual demand and the latest forecast. More weight is allocated to the latest data as it is believed to be relevant than previous data which is given with less weight.
Determining the weight depends on the demand pattern. When demand is erratic with low variability, less smoothing constant is appropriate. When demand trend reveals volatility, low to moderate weight is appropriate. When demand trend has low variability, moderate to high smoothing constant can be used. The formula of exponential smoothing is;
Next period forecast = (α) (past actual demand) + (1-α) (previous forecast)
Calculation: the forecast for July is based on the actual demand and forecast from June. While August forecast takes actual demand and forecast from July. A smoothing factor is .2.
= .2 x 550 + (1-.2) x 600
= 110 + 480
= .2 x 580 + (1-.2) x 590
= 116 + 472
Time series decomposition – time series has three major components including trend, seasonal, and random. This forecasting approach uses historical data by decomposing its components to make a forecast. There are many ways to use this approach using complex statistical calculations. However, just utilizing Excel, you would be able to make a forecast easily.
Calculation: below are the step-by-step calculation of time series decomposition
Step 1. Arrange the data by quarterly sales per year
Step 2. Get the slope or trend for each quarter using excel function
Step 3. Get the adjusted annual trend
Step 4. Get the seasonal index by dividing the trend by adjusted annual trend
Step 5. Get the seasonal adjusted time series by dividing sales by seasonal index
Step 6. Get the intercept and slope to obtain the trend component of the time series
Intercept calculation: obtain the average of the entire sales minus the slope times the average of the total number of quarters. In this case, there are 16 quarters from 2014 to 2017. Simply divide 16 by 2 to get the average.
Slope calculation: divide the adjusted annual trend by 4 (since there is 4 quarters in a year).
Step 7. To get the forecast, multiply seasonal index by the sum of intercept plus slope times the time you are forecasting
Step 8. Graph the forecast. Add the forecasted values to historical data to visualize future demand
Causal – causal forecasting analyzes the correlation between factors and demand. Oftentimes it is referred to as cause and effect type of forecasting. It assumes that the dependent variable is affected by the independent variable. For instance, income (independent variable) influences the rate of consumption (dependent variable). Regression is a technique used in causal forecasting.
Regression is best used for non-stationary demand characterized by the upward or downward trend over time.
Calculation: follow the steps below to make a forecast using regression
Step 1. Calculate the intercept
Step 2. Calculate the slope
Step 3. Calculate the forecast by adding the intercept by the sum of slope times the time period you are forecasting
Step 4. Graph the results for visualization
The correlation value denoted by R² suggests the magnitude of the association between time (quarter) and sales. The correlation value ranges between -1 and +1 and determines the direction and strength between two variables. -1 has a strong negative correlation while +1 has a strong positive correlation.
In our case, the correlation value of 0.8355 indicates a strong positive (at least close to +1) correlation. In basic terms, the period correlates or influences sales.
Here is how to get the correlation value.
Simulation – simulation forecasting is a hybrid method where different quantitative techniques are used to make a forecast. Simulation is heavily used in econometric forecasting where different statistical equations are used to model relationships among economic variables to predict future developments. Forecasting the future economic condition of the country uses simulation technique by utilizing statistical models with inputs of economic indicators such as interest rates, national production, consumer confidence, and inflation rates, to name a few.
A dynamic pricing model used by airline companies to forecast ticket sales is also an example of simulation forecasting.