Forecasting

Chapter 4

M. Verma: September 13-21, 2006

Faculty of Business Administration, Memorial University of Newfoundland

1

Forecasts
• It is a statement about the future.
• Help managers: plan the system & use of the system.

Types of Forecasts
• Economic forecasts:
– Address business cycle by predicting inflation rates, money
supply, housing starts, and other planning indicators.

• Technological forecasts:
– Concerned with rates of technological progress.
– Predict technological change and new product sales.

• Demand forecasts:
– Predict existing product sales (Sales forecast).
– Most important for our us.
M. Verma: September 13-21, 2006

Faculty of Business Administration, Memorial University of Newfoundland

2

Forecasting Time Horizons
1. Short-range forecast.


Generally less than 3 month time span.
Day to day operations.
More accurate than longer-term forecasts.

2. Medium-range forecast.

Spans from 3 months to 3 years.
Used for resource allocation.

3. Long-range forecast.

More than 3 years.
Strategic decisions: new products, capital expenditures, etc.

M. Verma: September 13-21, 2006

Faculty of Business Administration, Memorial University of Newfoundland

3

g. e. Forecast accuracy is higher for shorter time horizons. Forecasts are rarely perfect. daily Vs yearly stock price. • • Randomness precludes perfect forecast. 2. Cushions should be built-in for inaccuracies. (-vely correlated). Verma: September 13-21. Past causal system will exist in the future. – Fluctuations not as drastic. – Fewer uncertainties. Group forecasts (aggregation) is more accurate than individual forecasts. Memorial University of Newfoundland 4 . • • Forecasting errors have opposing canceling effect. 3.Principles of Forecasting 1. If +vely correlated. 4. M. the forecasting errors will be magnified. 2006 Faculty of Business Administration.

Elements of a Good Forecast 1. All functions should use the same forecast. Method should be reliable and consistent. Degree of accuracy should be stated. 2. 4. 2006 Faculty of Business Administration. Must be expressed in meaningful units. 6. Technique should be simple to understand and use. Memorial University of Newfoundland 5 . Verma: September 13-21. Forecasting horizon must be longer than time to implement changes. M. 5. 3.

Memorial University of Newfoundland 6 . Verma: September 13-21. 2006 Faculty of Business Administration.Steps in the Forecasting Process “The forecast” Step 6 Monitor the forecast Step 5 Prepare the forecast Step 4 Gather and analyze data Step 3 Select a forecasting technique Step 2 Establish a time horizon Step 1 Determine purpose of forecast M.

human factors. Quantitative Methods. etc. Used primarily when no data is available. Verma: September 13-21. Associative Models. • These factors are difficult to be quantified and hence are excluded from quantitative methods.g. Permits inclusion of soft information e. • • • Forecasts generated subjectively by the forecaster. Qualitative Methods. Memorial University of Newfoundland 7 . hunches. launch of new product. 2. M. • Forecasts generated through mathematical modeling. personnel opinions. 2006 Faculty of Business Administration.g. • • Time Series Models. e.Approaches to Forecasting 1.

• • Solicit input from customers regarding future purchasing plans. Useful for technological forecasting. Delphi Method (Expert Opinions). Long-term.Qualitative Approaches 1. Panel of experts. Verma: September 13-21. M. Some executives can influence decisions. • • • 2. single-term forecasts. Are often aware of customer’s future plans. Useful for long-range planning and new-product development. Executive Opinions. Helps prepare a forecast. 2006 Faculty of Business Administration. Reduces the risk that one person’s opinion will prevail. • • • 3. • • 4. Consumer Market Surveys. Estimates from individual salespersons are reviewed and then aggregated. Pool opinions of high-level executives. Memorial University of Newfoundland 8 . Sales force Composite. queried iteratively until consensus is reached. improve product design and plan new products.

• Weaknesses: – Forecaster bias can reduce the accuracy of the forecast. – Limited to human processing abilities. – Particularly useful when the future is expected to be very different from the past.Qualitative Methods • Strengths: – Incorporates inside information. M. 2006 Faculty of Business Administration. Verma: September 13-21. Memorial University of Newfoundland 9 .

Trend. 5. Seasonality. Cycles. Data pattern that repeats itself after a period of days. Follow no discernible pattern.Quantitative Methods Basic demand characteristics for Time Series Models (TSM) are: 1. Memorial University of Newfoundland 10 . months. Verma: September 13-21. Affected by political events. • Patterns in the data that occur every several years. • 4. • • Unforecastable variation in demand. M. Systematic increase (decrease) in the mean of the series over time. Fluctuation of data around a constant mean. 2006 Faculty of Business Administration. Random variations. – 2. Level (average). Tied into the business cycles. – 3. etc. weeks.

5. 4. Trend-adjusted Exponential Smoothing (Double). Weighted Moving Average. 2006 Time-Series Models Overview of Quantitative Methods Associative Model Faculty of Business Administration. Exponential Smoothing. Memorial University of Newfoundland 11 . Linear Trend Equation. Moving Average. Verma: September 13-21. 6. 3. 2. Linear Regression. 7.1. Naïve Approach. M.

• Advantages: – – – – Simple to use. – Traces actual data. Virtually no cost. with one-period lag. Verma: September 13-21. 2006 Faculty of Business Administration. M. – If May sales were 48. Memorial University of Newfoundland 12 . then June sales will be 48. • Disadvantages: – Cannot provide highly accurate forecasts. Easily understandable.Naïve Approach • Assumes demand in next period is the same as demand in the most recent period. Data analysis is non-existent.

i  age index.e. n  number of periods. Memorial University of Newfoundland 13 . Verma: September 13-21. period t ). 2006 Faculty of Business Administration. – Uses a number of historical actual data values.Moving Average (MA) Method • MA is a series of arithmetic means. M. Ai  actual value in period i MA  moving average Ft  forecast for next period (i. n  Ai Ft  MAn  i 1 n ( sum of demand in previous n periods ) number of periods where .

2006 Faculty of Business Administration. Verma: September 13-21. 1993: 4 1994: 6 1995: 5 1996: 3 1997: 7 1998: ?? M. • E. You want to forecast sales (000) for 1998 using a 3-period moving average.: – You are the manager of a museum store that sells historical replicas.g. – Slow to react.Moving Average (MA) • Advantages: – It is easy to compute and understand. Memorial University of Newfoundland 14 . • Disadvantages: – All values are weighted equally.

M. • Differs from simple moving average that weighs all periods equally. 2006 Faculty of Business Administration. • Allows the forecaster to emphasize one period over others. wt  weight for period t. • Similar to Moving Average. Memorial University of Newfoundland 15 .0. • All weights must add to 100% or 1.Weighted Moving Average n Ft 1   wt At t 1 where . Verma: September 13-21. – Makes it more responsive to changes. At  actual demand in period t. except weight assignment. Ft 1  forecast for next period.

Weight selection is arbitrary. but it makes the method less sensitive to real changes in the data. Verma: September 13-21. – Cannot pick up trends very well. They lag the actual values. – Require extensive records of past data. Memorial University of Newfoundland 16 . 2006 Faculty of Business Administration. • Disadvantages: – Increasing n (# of periods) does smooth out fluctuations better. • Weights can be changed to achieve this.Weighted Moving Average • Advantages: – Is more reflective of the most recent occurrences. Since they are averages. M. hence will stay within past levels.

30 for the next most recent. and . forecast demand for period 7 using the same weights as in part a. Memorial University of Newfoundland 17 . . Verma: September 13-21.20 for the next. b) If the actual demand for period 6 is 39. a) Compute a weighted average forecast (for period 6) using a weight of . Period 1 2 3 4 5 M.10 for the last. 2006 Demand 42 40 43 40 41 Faculty of Business Administration. .40 for the most recent period.Example Given the following demand data.

Memorial University of Newfoundland 18 . Verma: September 13-21.Previous forecast) Forecast Error Ft  Ft 1   ( At 1  Ft 1 ) Ft  forecast for period t Ft 1  forecast for the previous period   smoothing constant At 1  actual demand for the previous period M.Exponential Smoothing • Is a sophisticated weighted average method. Next forecast  Previous forecast   (Previous actual . 2006 Faculty of Business Administration.

5).Exponential Smoothing • • More versatile than weighted moving average. The actual demand that occurred for that forecast period. Places more weight on recent data. High α values: • • Generate forecasts that respond quickly to recent data. Needs only 3 pieces of data: 1. A smoothing constant α. Verma: September 13-21. 2. Appropriate for less volatile data. Forecast quality is highly dependent on selection of α: – – Underlying pattern stable: Low α. 2006 Faculty of Business Administration. Memorial University of Newfoundland 19 . M.05 and 0. • • Value of smoothing constant between 0 and 1 (generally between 0. Most recent forecast. 3.

M. 2. 5. Has a lagging effect. • Are surprisingly accurate.Exponential Smoothing • Advantages: 1. Very little computation required to use the model. Memorial University of Newfoundland 20 . Easily understandable. Disadvantages: 1. Cannot capture trend. 4. 3. Verma: September 13-21. 2006 Faculty of Business Administration. 2. Formulation is relatively easy. Limited use of historical data.

4 Forecast Error Faculty of Business Administration. Memorial University of Newfoundland 21 .Example • Forecast demand for period 12 using α values of 0.1 and 0.1 Forecast Error Alpha = 0. Period 1 2 3 4 5 6 7 8 9 10 11 12 Actual 42 40 43 40 41 39 46 44 45 38 40 M. 2006 Alpha = 0. Verma: September 13-21.4.

High values mean more responsive to changes in recent trends. – β: To capture trend in data. Is between 0 and 1 as well. 2006 Faculty of Business Administration. • Contains two smoothing constants: – α : As before to react to changes. Verma: September 13-21. Forecastincluding trend ( FITt )  Exponentially smoothedforecast( Ft )  Exponentially smoothed trend (Tt ) Ft  α(Actual demand last period)  ( 1-α)(Forecastlast period  Trend estimate last period) Ft  α ( At 1 )  (1   )( Ft 1  Tt 1 ) Tt   (Forecastthis period . Also called double exponential smoothing.Trend-Adjusted Exponential Smoothing • Is a variation of exponential smoothing. Memorial University of Newfoundland 22 .Forecastlast period)  ( 1- )(Trend estimate last period) Tt   ( Ft  Ft 1 )  (1   )Tt 1 M.

Assume the initial forecast for month 1was 11 units and the trend over the past period was 2 units. 2006 Faculty of Business Administration. Memorial University of Newfoundland 23 . It appears that an increasing trend is present. M.4. Verma: September 13-21.Example A large Portland manufacturer uses exponential smoothing to forecast demand for a piece of pollution-control equipment.2 and β=0. Month (t) Actual Demand (At) Month (t) Actual Demand (At) 1 12 6 21 2 17 7 31 3 20 8 28 4 19 9 36 5 24 10 ?? Smoothing constants are assigned the values of α= 0.

Memorial University of Newfoundland 24 .Linear Trend Equation yt  a  bt ty   t  y  b n  t 2  ( t ) 2 y  b t  a n t  specified number of time periods from t  0 yt  forecast for period t a  value of yt at t  0 b  slope of the line n Example: Cell phone sales for a firm over the last 10 weeks are shown in the following table. and predict sales for weeks 11 and 12. Week 1 2 3 4 5 M. 2006 Unit Sales 700 724 720 728 740 Week 6 7 8 9 10 Unit Sales 742 758 750 770 775 Faculty of Business Administration. Determine the equation of the linear trend. Verma: September 13-21. The data appear to have a linear trend.

minimizes sum of squared deviations around the line.intercept) b  slope of the line M. yc  a  bx yc  predicted (dependent ) variable x  predictor (independe nt) variable a  value of yc when x  0 (y .technique for fitting a line to a set of points • Least squares line .Associative Forecasting Techniques • Predictor variables . 2006 xy )  ( x)( y )  b n( x 2 )  ( x ) 2 y  b x  a n( n Faculty of Business Administration. Memorial University of Newfoundland 25 .used to predict values of variable interest • Regression . Verma: September 13-21.

Months Units Sold Unemployment % 1 20 7.3 4 35 5.6 9 15 8. Memorial University of Newfoundland 26 .Associative Forecasting Technique • Correlation: – Measures the strength of relationship between two variables.0 7 38 5. derive the correlation coefficient and the predictive equation.4 8 50 3. M. 2006 Faculty of Business Administration.0 Predict the number of color TV sets sold.0 11 14 9.0 to +1.2 2 41 4.2 for month 12. and if so. if the unemployment % is 5.0 3 17 7.8 6 31 6. – Can range from -1. Verma: September 13-21. r  xy )  ( x y) n(  x 2 )  (  x ) 2  n (  y 2 )  (  y ) 2 n( Example: Determine if the unemployment levels below can be used to predict the number of color TV sets sold.5 5 25 6.4 10 19 7.0.

M. Sources include: a) b) c) d) e) • Positive: when forecast is lower than actual demand. Negative: when forecast is higher than actual demand. forecast is either too high or too low.Forecast Accuracy • • Forecasts are rarely perfect. Existence of some undetected trend. Random error: “Noise” in the model. Error = Actual – Forecast. Employing the wrong trend line. That cannot be explained by the forecast model being used. 2006 Faculty of Business Administration. Failure to include the right variables. Error can be either bias or random. Et = At – Ft – – – • Bias error: when a consistent mistake is made i. Verma: September 13-21. Memorial University of Newfoundland 27 . Mistakenly shifting the seasonal demand from where it occurs. Using the wrong relationship among variables.e.

Verma: September 13-21. – Sum of absolute error. – A good measure of actual error.Measurement of Error n • Mean Absolute Deviation (MAD). • Tracking Signal (TS). – Penalizes extreme errors. M. Memorial University of Newfoundland MAD 28 . – Is the number of MADs that the forecast value is above or below the actual occurrence. – Weighs all errors evenly. – Weighs according to squared values. – Tends to magnify large errors. 2006 n n  MSE  t 1 ( At  Ft ) 2 n n  ( At  Ft ) TS  t 1 Faculty of Business Administration. | At  Ft | MAD  t 1 • Mean Square Error (MSE). No canceling effect.

Compute MAD and MSE for the following data. Faculty of Business Administration. 2006 Forecast 215 216 215 214 211 214 217 216 Error (A-F) |Error| Error Sq. Period 1 2 3 4 5 6 7 8 Actual 217 213 216 210 213 219 216 212 M.Example 1. Verma: September 13-21. Memorial University of Newfoundland MAD = MSE = 29 .

Verma: September 13-21. Memorial University of Newfoundland 30 .Choosing a Forecasting Technique Not all techniques are suitable for every situation. Time horizon. Availability of qualified people. M. Availability of computers. Accuracy. Availability of historical data. The determinants of which technique to use are: – – – – – – Cost. 2006 Faculty of Business Administration.

10 for June. Memorial University of Newfoundland 31 . sells can openers.000. Apr. The naïve approach.20. Exponential smoothing with a smoothing constant equal to 0. A 5-month moving average. Monthly sales for a seven-month period were as follows: Month Feb. Which method seems least appropriate? Why? What does use of the term sales rather than demand presume? M. A weighted average using 0. Forecast September sales volume using each of the following: 1) 2) 3) 4) 5) c) d) Sales (000 units) 19 18 15 20 18 22 20 A linear trend equation. 0. Aug. Jul. Verma: September 13-21. National Mixer Inc. Jun. 2006 Faculty of Business Administration.30 for July and 0.60 for August. a) b) Plot the monthly data on a sheet of graph paper.Examples 1. assuming a March forecast of 19. Mar. May.

2006 Faculty of Business Administration.3 and β=0. a) Develop a linear trend equation for the following data on bread deliveries. and use it to predict deliveries for periods 16 through 19. Memorial University of Newfoundland 32 .2 to model the bread delivery data in part a.Examples 2. Verma: September 13-21. What is the forecast for period 16? M. Period 1 2 3 4 5 Dozen Deliveries 200 214 211 228 235 Period 6 7 8 9 10 Dozen Deliveries 232 248 250 253 267 Period 11 12 13 14 15 Dozen Deliveries 281 275 280 288 310 b) Use trend-adjusted smoothing with α= 0.