Forecasting

Introduction    What: Forecasting Techniques Where: Determine Trends Why: Make better decisions

What is Forecasting?

 The art and science of predicting future events

Time Horizon    Short Range – 3 – 12 months Medium Range – 3 months – 3 years Long Range – 3+ years

Where do we Use Forecasts?

   Economic Forecast   Inflation rate Exchange Rate Technological Forecast  Probabilities of new discoveries  Time to commercialize technologies Demand Forecast

Impact of Forecasts    Human Resources: forecast gives warning of need to hire or lay off Production Capacity: forecast gives warning of need for more or less capacity Supply Chain: forecast gives warning of need for more or less inputs to production

How to Make a Forecast        Determine use of forecast Select variable to be forecasted Determine time horizon Select forecasting model Gather data Make forecast Implement results and review model

Qualitative Methods     Jury of Executive Opinion Sales Force Composite Delphi Consumer Marketing Survey

Quantitative Methods   Time Series Associative

Time Series Methods    A sequence of evenly spaced data points (weekly, monthly, quarterly, etc) Future values predicted only from past values X axis is always time

Example of a Time Series

Seasonal peaks Trend component Year 1 Random variation Year 2 Average demand over four years Year 3 Year 4 Actual demand line

Trend    Upward or downward pattern Due to changes in income, population, technology, etc Several years duration

Demand Time

© 1984-1994 T/Maker Co.

Seasonality    Repeating pattern over a period Could be quarterly, monthly, weekly Due to weather or customs

Summer Demand

© 1984-1994 T/Maker Co.

Time

Cycles   Pattern that occurs over several years Affected by political events or international turmoil

Demand Cycle Time



Random Variations    Erratic, unsystematic Caused by random chance and unusual situations Short duration, non-repeating

Naïve Approach  Forecast for next period is the same as demand in most recent period

Moving Average Approach

MA

  Demand in Previous

n

Periods

n

Weighted Moving Average

WMA =

Σ (Weight for period n ) (Demand in period n ) Σ Weights

Exponential Smoothing  F t = F t -1 +  ( A t -1 F t -1 )

MAD

MAD

 

forecast errors n

MSE

MSE

  

forecast errors

 2

n

Exponential Smoothing With Trend Adjustment

F t =



(A t ) + (1-



)F t-1 + T t-1 T t =



(F t - F t-1 ) + (1-



)T t-1

Linear Trend Projection

Equation:

ˆ i  a  bx i

Slope:

b  i n    x i y i n  i   x i    n x n x  y

Y-Intercept:

a  y  b x

Seasonal Variations      Calculate average historical demand for each season Compute average demand over all periods Compute a seasonal index – historical demand / average demand Estimate next year’s total demand Divide estimate by number of seasons, multiply by seasonal index

Regression Analysis    An associative method Find the relationship between an independent variable and a dependant variable Independent variable is a variable other than time

Regression Analysis

Equation: Slope: Y-Intercept:

ˆ i  a  bx i b  i n    x i y i n  i   x i    n x n x  y a  y  b x

Standard Error of Estimate S y , x  i n     y i n    i    i n    y i   a i n    n y  i   b i n    x i y i

What Does Standard Error Mean?

  Standard Deviation of data forming the regression line.

If error becomes large, regression data is widely dispersed and less reliable

Correlation Coefficient

r

 

n



x

2

n

 

xy

   2  

n



y

2

y



y

 2

What Does Correlation Coefficient Mean?

  Strength of linear relationship between independent variable and dependant variable A number between +1 and -1

What Does Correlation Coefficient Mean?

Perfect Negative Correlation No Correlation Perfect Positive Correlation -1.0

-.5

0 +.5

+1.0

Increasing degree of negative correlation Increasing degree of positive correlation

Coefficient of Determination   r 2 Percent of variation in dependant variable that is explained by the regression equation

Evaluating the Forecast     Monitor the forecast with a tracking signal = RSFE / MAD Small deviations are ok and should cancel each other out over time A consistent tendency for the forecast to be higher or lower than actual values is called a bias error

Tracking Signal Limits    +/- 2, 3 or 4 MAD’s Smaller range = less tollerance of error But smaller range = higher costs

Other Ways to Forecast   Adaptive Smoothing – Exponential smoothing constants adapted when tracking signal outside limits Focus Forecasting – Computer tries all forecast methods and selects best fit for next month’s forecast