PROJECT ON: QUANTITATIVE TECHNIQUES OF
         FORECASTING DEMAND
DEMAND FORECASTING
  • Demand forecasting is a scientific and analytical estimation of
    demand for a product (service) for a particular period of time.
  • It is the process of determining how much of what products is
    needed when and where.
  • “An estimate of sales in dollars or physical units for a specified
    future period under a proposed marketing plan.” – American
    Marketing Association
  • It is used for planning and decision making in operations research.



    Categorization of Demand
                     Forecasting
By nature of goods
  • Capital Goods: Derived demand
        o Demand for capital goods depends upon demand of
          consumer goods which they can produce.
  • Consumer Goods: Direct demand
        o durable consumer goods: new demand or
          replacement demand
        o Non-durable consumer goods: FMCG
By Time Period
  • Short Term (0 to 3 months): for inventory
    management and scheduling.
  • Medium Term (3 months to 2 years): for production
    planning, purchasing, and distribution.
  • Long Term (2 years and more): may extend up to 10 to
    20 years.
       o For capacity planning, facility location, and
        strategic planning, long term capital requirement,
        and investment decisions.



           HOW TO CHOOSE A DEMAND
            FORECASTING TECHNIQUE
Consider the following factors affecting demand:
       o Imminent objectives of forecast, whether it is for
         a new product, or to gauge impact of a new
         advertisement, etc.
       o Cost involved, cost of forecasting should not be
         more than its benefits, and here opportunity cost
         of resources will also be important.
o Time perspective, whether the forecast is meant
         for the short run or the long run
       o Complexity of the technique, vis-à-vis availability
         of expertise; this would determine whether the
         firm would look for experts “in house” or
         outsource it
       o Nature and quality of available data, i.e. does the
         time series show a clear trend or is it highly
         unstable.

      Quantitative Methods of
          Demand Forecasting
     Trend Projection
     This statistical tool is used to predict future values of a
variable on the basis of time series data.
  • Time series data are composed of:
       o Secular trend (T): change occurring consistently
         over a long time and is relatively smooth in its
         path.
       o Seasonal trend (S): seasonal variations of the
         data within a year
o Cyclical trend (C): cyclical movement in the
      demand for a product that may have a tendency to
      recur in a few years
    o Random events (R): have no trend of occurrence
      hence they create random variation in the series.
  Additive Form: Y = T + S + C + R………..(1)
  Multiplicative Form: Y = T.S.C.R………….(2)
           Log Y= log T + log S + log C + log R………….(3)

Methods of Trend Projection
• Graphical method
    o Past values of the variable on vertical axis and
      time on horizontal axis and line are plotted.
    o Movement of the series is assessed and future
      values of the variable are forecasted.
    o simple but provides a general indication and fails
      to predict future value of demand
Least squares method
• Based on the minimization of squared deviations
  between the best fitting line and the original
  observations given.
• Estimates coefficients of a linear function.
•             Y=a+bX where a =intercept
•                       and b =slope
• The normal equations:
•      ΣY=na + bΣX
•      ΣXY= aΣX+ bΣX2
• Once the coefficients of the trend equation are
  estimated, we can easily project the trend for future
  periods.
• Solving the normal equations:


     (Y Y )( X     X)
A=       (X     X )2
                         Y bX
QUANTITATIVE METHODS
OF TREND PROJECTION
ARIMA method: also known as Box Jenkins method
  • It is considered to be the most sophisticated technique
    of forecasting as it combines moving average and auto
    regressive techniques.
       o Stage One: trend in the series is removed with
         help of „differencing‟, i.e. the difference between
         values at adjacent period of time.
       o Stage Two: Various possible combinations are
         created on basis of:
                                  i. order of involvement of
auto regressive terms;
ii. the order of moving average terms
iii. the number of differences of the original series.
Combinations are selected which provide an adequate fit to
the series.


       o Stage Three: Parameter estimation is done using
         Least Squares.
o Stage Four: „Goodness of fit‟ is tested and if it is
         not a good fit then the whole process is repeated
         from Stage Two.
       o Stage Five: Once a „good fit‟ is attained, its
         coefficients can be used to forecast future
         demand.

       Quantitative Methods
  • Simple (or Bivariate) Regression Analysis:
       o deals with a single independent variable that
         determines the value of a dependent variable.
       o Demand Function: D = a+bP, where b is negative.
       o If we assume there is a linear relation between D
         and P, there may also be some random variation
         in this relation.
Sum of Squared Errors (SSE): a measure of the predictive
accuracy
Smaller the value of SSE, the more accurate is the
regression equation.
  • Nonlinear Regression Analysis
       o Log linear function log D =A + B log P + e
where A and B are the parameters to be estimated and e
represents errors or disturbances.
Linear form of log linear function D* = a + b P* + e Where
D*= log D and P*=log P
       o Multiple Regression Analysis:
         D = a1+a2.P+a3.A+e
       (Where A = advertising expenditure incurred).
       D^ = a^1 + a^2P + a^3A, (where a1, a2 and a3 are the
       parameters and e is the random error term (or
       disturbance), having zero mean).
Similar to simple regression analysis, multiple regression
analysis would aim at estimation of the parameters a1, a2
and a3.
       o Choose such values of the coefficients that would
         minimize the sum of squares of the deviations.



        Problems Associated with
             Regression Analysis
  • Multicollinearity: when two or more explanatory
    variables in the regression model are found to be
highly correlated the estimated coefficients may not be
  accurately determined.
• Heteroscedasticity: Classical regression models
  assume that the variance of error terms is constant
  for all values of the independent variables in the model;
  i.e. variables are homoscedastic.
• Specification errors: Omission of one or more of the
  independent variables, or when the functional form
  itself is wrongly constructed or estimates a demand
  function in linear form, though the function should have
  been nonlinear.
• Identification problem: where the equations have
  common variables, like a demand supply model.

Demand forecasting

  • 1.
    PROJECT ON: QUANTITATIVETECHNIQUES OF FORECASTING DEMAND
  • 2.
    DEMAND FORECASTING • Demand forecasting is a scientific and analytical estimation of demand for a product (service) for a particular period of time. • It is the process of determining how much of what products is needed when and where. • “An estimate of sales in dollars or physical units for a specified future period under a proposed marketing plan.” – American Marketing Association • It is used for planning and decision making in operations research. Categorization of Demand Forecasting By nature of goods • Capital Goods: Derived demand o Demand for capital goods depends upon demand of consumer goods which they can produce. • Consumer Goods: Direct demand o durable consumer goods: new demand or replacement demand o Non-durable consumer goods: FMCG
  • 3.
    By Time Period • Short Term (0 to 3 months): for inventory management and scheduling. • Medium Term (3 months to 2 years): for production planning, purchasing, and distribution. • Long Term (2 years and more): may extend up to 10 to 20 years. o For capacity planning, facility location, and strategic planning, long term capital requirement, and investment decisions. HOW TO CHOOSE A DEMAND FORECASTING TECHNIQUE Consider the following factors affecting demand: o Imminent objectives of forecast, whether it is for a new product, or to gauge impact of a new advertisement, etc. o Cost involved, cost of forecasting should not be more than its benefits, and here opportunity cost of resources will also be important.
  • 4.
    o Time perspective,whether the forecast is meant for the short run or the long run o Complexity of the technique, vis-à-vis availability of expertise; this would determine whether the firm would look for experts “in house” or outsource it o Nature and quality of available data, i.e. does the time series show a clear trend or is it highly unstable. Quantitative Methods of Demand Forecasting Trend Projection This statistical tool is used to predict future values of a variable on the basis of time series data. • Time series data are composed of: o Secular trend (T): change occurring consistently over a long time and is relatively smooth in its path. o Seasonal trend (S): seasonal variations of the data within a year
  • 5.
    o Cyclical trend(C): cyclical movement in the demand for a product that may have a tendency to recur in a few years o Random events (R): have no trend of occurrence hence they create random variation in the series. Additive Form: Y = T + S + C + R………..(1) Multiplicative Form: Y = T.S.C.R………….(2) Log Y= log T + log S + log C + log R………….(3) Methods of Trend Projection • Graphical method o Past values of the variable on vertical axis and time on horizontal axis and line are plotted. o Movement of the series is assessed and future values of the variable are forecasted. o simple but provides a general indication and fails to predict future value of demand
  • 6.
    Least squares method •Based on the minimization of squared deviations between the best fitting line and the original observations given. • Estimates coefficients of a linear function. • Y=a+bX where a =intercept • and b =slope • The normal equations: • ΣY=na + bΣX • ΣXY= aΣX+ bΣX2 • Once the coefficients of the trend equation are estimated, we can easily project the trend for future periods. • Solving the normal equations: (Y Y )( X X) A= (X X )2 Y bX
  • 7.
    QUANTITATIVE METHODS OF TRENDPROJECTION ARIMA method: also known as Box Jenkins method • It is considered to be the most sophisticated technique of forecasting as it combines moving average and auto regressive techniques. o Stage One: trend in the series is removed with help of „differencing‟, i.e. the difference between values at adjacent period of time. o Stage Two: Various possible combinations are created on basis of: i. order of involvement of auto regressive terms; ii. the order of moving average terms iii. the number of differences of the original series. Combinations are selected which provide an adequate fit to the series. o Stage Three: Parameter estimation is done using Least Squares.
  • 8.
    o Stage Four:„Goodness of fit‟ is tested and if it is not a good fit then the whole process is repeated from Stage Two. o Stage Five: Once a „good fit‟ is attained, its coefficients can be used to forecast future demand. Quantitative Methods • Simple (or Bivariate) Regression Analysis: o deals with a single independent variable that determines the value of a dependent variable. o Demand Function: D = a+bP, where b is negative. o If we assume there is a linear relation between D and P, there may also be some random variation in this relation. Sum of Squared Errors (SSE): a measure of the predictive accuracy Smaller the value of SSE, the more accurate is the regression equation. • Nonlinear Regression Analysis o Log linear function log D =A + B log P + e
  • 9.
    where A andB are the parameters to be estimated and e represents errors or disturbances. Linear form of log linear function D* = a + b P* + e Where D*= log D and P*=log P o Multiple Regression Analysis: D = a1+a2.P+a3.A+e (Where A = advertising expenditure incurred). D^ = a^1 + a^2P + a^3A, (where a1, a2 and a3 are the parameters and e is the random error term (or disturbance), having zero mean). Similar to simple regression analysis, multiple regression analysis would aim at estimation of the parameters a1, a2 and a3. o Choose such values of the coefficients that would minimize the sum of squares of the deviations. Problems Associated with Regression Analysis • Multicollinearity: when two or more explanatory variables in the regression model are found to be
  • 10.
    highly correlated theestimated coefficients may not be accurately determined. • Heteroscedasticity: Classical regression models assume that the variance of error terms is constant for all values of the independent variables in the model; i.e. variables are homoscedastic. • Specification errors: Omission of one or more of the independent variables, or when the functional form itself is wrongly constructed or estimates a demand function in linear form, though the function should have been nonlinear. • Identification problem: where the equations have common variables, like a demand supply model.