Forecasting

Forecasting is the art of predicting future sales or demand of a product. The survival of any organization depends upon how well they are able to project the demand in future

Types of Demand variation

Time Horizon in Forecasting

1. Short term forecasting               -  1 to 3 months
2. Intermediate term forecasting   -   3 to 12 months
3. Long term forecasting              -    More than a year.

Classification of Forecasting

Types of forecasting

Judgmental Technique

This technique is relies on the art of human judgment i.e. how well a human being can predict the demand of a product in future.

1. Opinion survey
2. Market trial
3. Market research
4. Delphi technique

Time Series Method

In this method, we project the future demand for the product based on the historical pattern of demand.

1) Past Average

In this method forecast is equal to average actual demand of a product for the previous period.

2) Moving Average Method

This method uses past data and calculate a moving average for a constant period. Fresh average is calculated at the end of each period by adding the actual demand data for the most recent period and deleting the data for older period.

Moving average   =  Sum of demands for given period  / chosen number of period

3) Weighted Moving Average

In Moving average method equal weightage is given to demand of each period but this method gives unequal weight to each demand data in such a manner that the summation of all weights always equals to one. More weightage is given to recent demand and least weightage assigned to oldest demand.

Method to find the weightages

if N = Number of periods and have to calculate (N+1)th period forecast;

Sum = N(N+1)/2

Weightage to given recent demand as

N/Sum,  (N-1)/Sum,  (N-2)/Sum, .....   ........  1/Sum

4) Exponential Smoothing Method

When the value of N become very large this method is used. This method assigns weight to all the previous data and the weight assigned are in exponentially decreasing order.

Where,  Ft = Forecast for the period 't'

              Ft-1 = Forecasted demand for the last period 

         

              Dt-1  = Demand for the last period

             

In general, If smoothing constant is not given and is equivalent to

Responsiveness and stability

Responsiveness indicates that the calculated forecast have a fluctuating and swinging pattern where as stability means that the forecast pattern is flat or has less fluctuation. 

Responsiveness is preferred for new product and for that number of period is kept small and stability is preferred for old exiting product and for that number of periods is kept large.

 

Causal Forecasting

It tries to identify the factors which causes the variation of demand and try to establish a relationship between the demand and their factors.

Linear Regression

y = a + bx

yx = ax + bx^2

We get, 

Error Analysis

• Bias

• Bias also called MFE and  it is a measure of over estimation or under estimation.
• A positive Bias means underestimated forecast
• A negative Bias means overestimated forecast

• Mean Square Error(MSE) 

• Mean Absolute Deviation (MAD)

• Mean absolute percentage error (MAPE)

• Tracking Signal(Ts)

Ts tells how well the forecast is predicting the actual value. If value of Ts goes beyond 3*sqrt(MSE) then it indicates that model needs to be revised.