Forecasting

True/False

1. Forecasting techniques generally assume an existing causal system that will continue to exist in the
future.
Ans: TRUE
Forecasts depend on the rules of the game remaining reasonably constant.

2. For new products in a strong growth mode, a low alpha will minimize forecast errors when using
exponential smoothing techniques.
Ans: FALSE
If growth is strong, alpha should be large so that the model will catch up more quickly.

3. Once accepted by managers, forecasts should be held firm regardless of new input since many
plans have been made using the original forecast.
Ans: FALSE
Flexibility to accommodate major changes is important to good forecasting.

4. Forecasts for groups of items tend to be less accurate than forecasts for individual items because
forecasts for individual items don’t include as many influencing factors.
Ans: FALSE
Forecasting for an individual item is more difficult than forecasting for several items.

5. Forecasts help managers plan both the system itself and provide valuable information for using the
system.
Ans: TRUE
Both planning and use are shaped by forecasts

6. Organizations that can respond quickly to changing requirements can use a shorter forecast horizon
and therefore benefit from more accurate forecasts.
Ans: TRUE
If an organization can react more quickly, its forecasts need not be so long term.

7. The purpose of the forecast should be established first so that the level of detail, amount of
resources, and accuracy level can be understood.
Ans: TRUE
All these considerations are shaped by what the forecast will be used for.

8. Forecasts based on time series (historical) data are referred to as associative forecasts
Ans. FALSE
Forecasts based on time-series data are referred to as time-series forecasts.

9. Time series techniques involve identification of explanatory variables that can be used to predict
future demand.
Ans: FALSE
Associative forecasts involve identifying explanatory variables

10. A consumer survey is an easy and sure way to obtain accurate input from future customers since
most people enjoy participating in surveys.
Ans: FALSE
Most people do not enjoy participating in surveys.

1
11. The Delphi approach involves the use of a series of questionnaires to achieve a consensus
forecast.
Ans: TRUE
A consensus among divergent perspectives is developed using questionnaires.

12. Exponential smoothing adds a percentage (called alpha factor) of last period’s forecast to estimate
next period’s demand.
Ans: FALSE
Exponential smoothing adds a percentage to the last period's forecast error.

13. The shorter the forecast period, the more accurately the forecasts tend to track what happens.
Ans: TRUE
Long-term forecasting is much more difficult to do accurately.

14. Forecasting techniques that are based on time series data assume that future values of the series
will duplicate past values.

15. Trend adjusted exponential smoothing uses double smoothing to add twice the forecast error to
last periods actual.
Ans: FALSE
Time-series forecasts assume that future patterns in the series will mimic past patterns in the series.

16. Forecasts based on an average tend to exhibit less variability than the original data.
Ans: TRUE
Averaging is a way of smoothing out random variability.

17. The naive approach to forecasting requires a linear trend line.

Ans: FALSE
The naive approach is useful in a wider variety of settings.

18. The naive forecast is limited in its application to series that reflect no trend or seasonality.
Ans: FALSE
When a trend or seasonality is present, the naive forecast is more limited in its application.

19. The naive forecast can serve as a quick and easy standard of comparison against which to judge
the cost and accuracy of other techniques.
Ans: TRUE
Often the naive forecast performs reasonably well when compared to more complex techniques.

20. A moving average forecast tends to be more responsive to changes in the data series when more
data points are included in the average.
Ans: FALSE
More data points reduce a moving average forecast's responsiveness.

21. In order to update a moving average forecast, the values of each data point in the average must be
known.
Ans: TRUE
The moving average cannot be updated until the most recent value is known.

22. Forecasts of future demand are used by operations people to plan capacity.
Ans: TRUE
Forecasts of future demand are used by operations people to plan capacity.

2
23. An advantage of a weighted moving average is that recent actual results can be given more
importance than what occurred a while ago.
Ans: TRUE
Weighted moving averages can be adjusted to make more recent data more important in setting the
forecast.

24. Exponential smoothing is a form of weighted averaging.

Ans: TRUE
The most recent period is given the most weight, but prior periods also factor in.

25. A smoothing constant of .1 will cause an exponential smoothing forecast to react more quickly to
a sudden change than a smoothing constant value of .3.
Ans: FALSE
Smaller smoothing constants result in less reactive forecast models.

26. The T in the model TAF = S+T represents the time dimension (which is usually expressed in
weeks or months).
Ans: FALSE
The T represents the trend dimension.

27. Trend adjusted exponential smoothing requires selection of two smoothing constants.
Ans: TRUE
One is for the trend, and one is for the random error.

28. An advantage of "trend adjusted exponential smoothing" over the "linear trend equation" is its
ability to adjust over time to changes in the trend.
Ans: TRUE
A linear trend equation assumes a constant trend; trend-adjusted smoothing allows for changes in the
underlying trend.

29. A seasonal relative (or seasonal indexes) is expressed as a percentage of average or trend.
Ans: TRUE
Seasonal relatives are used when the seasonal effect is multiplicative rather than additive.

30. In order to compute seasonal relatives, the trend of past data must be computed or known which
means that for brand new products this approach can’t be used.
Ans: TRUE
Computing seasonal relatives depends on past data being available.

31. Removing the seasonal component from a data series (de-seasonalizing) can be accomplished by
dividing each data point by its appropriate seasonal relative.
Ans: TRUE
Deseasonalized data points have been adjusted for seasonal influences.

32. If a pattern appears when a dependent variable is plotted against time, one should use time series
analysis instead of regression analysis.
Ans: TRUE
Patterns reflect influences such as trends or seasonality that go against regression analysis
assumptions.

33. Curvilinear and multiple regression procedures permit us to extend associative models to
relationships that are non-linear or involve more than one predictor variable.
Ans: TRUE
Regression analysis can be used in a variety of settings.

3
34. The sample standard deviation of forecast error, is equal to the square root of MSE.
Ans: TRUE
The MSE is equal to the sample variance of the forecast error.

35. Correlation measures the strength and direction of a relationship between variables.
Ans: TRUE
The association between two variations is summarized in the correlation coefficient

36. MAD is equal to the square root of MSE which is why we calculate the easier MSE and then
calculate the more difficult Mean Absolute Deviation.
Ans: FALSE
MAD is the mean absolute deviation.

37. In exponential smoothing, an alpha of 1.0 will generate the same forecast that a naïve forecast
would yield.
Ans: TRUE
With alpha equal to 1 we are using a naive forecasting method.

38. A forecast method is generally deemed to perform adequately when the errors exhibit an
identifiable pattern.
Ans: FALSE
Forecast methods are generally considered to be performing adequately when the errors appear to be
randomly distributed.

39. A control chart involves setting action limits for cumulative forecast error.
Ans: FALSE
Control charts set action limits for the tracking signal.

40. A tracking signal focuses on the ratio of cumulative forecast error to the corresponding value of
MAD.
Ans: TRUE
Large absolute values of the tracking signal suggest a fundamental change in the forecast model's
performance

41. The use of a control chart assumes that errors are normally distributed about a mean of zero.
Ans: TRUE
Over time, a forecast model's tracking signal should fluctuate randomly about a mean of zero.

42. Bias exists when forecasts tend to be greater or less than the actual values of time series.
Ans: TRUE
A tendency in one direction is defined as bias.

43. Bias is measured by the cumulative sum of forecast errors.

Ans: TRUE
Bias would result in the cumulative sum of forecast errors being large in absolute value.

44. Seasonal relatives can be used to de-seasonalize data or incorporate seasonality in a forecast.
Ans: TRUE
Seasonal relatives are used to deseasonalize data to forecast future values of the underlying trend, and
they are also used to reseasonalize deseasonalized forecasts.

45. The best forecast is not necessarily the most accurate.

Ans: TRUE
More accuracy often comes at too high a cost to be worthwhile.

4
46. A proactive approach to forecasting views forecasts as probable descriptions of future demand,
and requires action be taken to meet that demand.

47. Simple linear regression applies to linear relationships with no more than three independent
variables.
Ans: FALSE
Simple linear regression involves only one independent variable.

48. An important goal of forecasting is to minimize the average forecast error.

Ans: FALSE
Regardless of the model chosen, so long as there is no fundamental bias average forecast error will be
zero.

49. Forecasting techniques such as moving averages, exponential smoothing, and the naive approach
all represent smoothed (averaged) values of time series data.
Ans: FALSE
The naïve approach involves no smoothing.

50. In exponential smoothing, an alpha of .30 will cause a forecast to react more quickly to a large
error than will an alpha of .20.
Ans: TRUE
Larger values for alpha result in more responsive models.

5
Taking Stock:
1. Explain the trade-off between responsiveness and stability in forecasting systems that uses
time series data.
2. Who needs to be involved in preparing forecasts?
3. How has technology had an impact on forecasting?

Critical thinking Exercises:

1. What capability would an organization have to have to need forecasts?

Experiential Learning Exercises:

1. Obtain data on daily weather forecast of high temperature for 7 days from a local newspaper
or a radio or television channel. Compare the accuracy of local one-day forecast with one-day
naïve forecasts.

Essay:

1. Develop a forecast for the next period, given the data below, using a 4-period moving average.

Solution:
Period 2 + Period 3 + Period 4 + Period 5 = 20 + 18 + 19+ 17 = 18.5
4 4

Answer: 6th period = 18.5 ~ 19

2. Consider the data below:

Using exponential smoothing with alpha = .1, and the Forecast for period 12 = 80, what would the
forecast for period 14 be?

Solution:

Period Demand EXPONENTIAL SMOOTHING

(α=0.1)
11 81  

6
 12 80 81 ( Naïve Forecast)
13 82 80.9
14   81.0

Actual demand + α x (forecast - Actual demand )

14 period = 80.9 + 0.1 x (82- 80.9)

Answer: 81

3. A manager is using exponential smoothing to predict merchandise returns at a suburban branch of a

department store chain. Given a previous forecast of 140 items, an actual number of returns of 148
items, and a smoothing constant equal to .15, what is the forecast for the next period?

Solution:
Actual demand + α x (forecast - Actual demand )
=140 + 0.15 x (148 – 140)
=141.2~ 141

Answer: 141

4. A manager is using the equation below to forecast quarterly demand for a product:
Yt = 6,000 + 80t where t = 0 at Q2 of last year
Quarter relatives are Q1 = .7, Q2 = .7, Q3 = 1.2, and Q4 = 1.3.
What forecasts are appropriate for the last quarter of this year and the first quarter of next year?

Solution:

Forecas
Period Trend Qtr Rel t
6 6480 1.3 6960
7 6560 0.7 7040

Note: use quarter seasonal relatives

Q4 = Yt = 6,000 + 80t
Q4 = Yt = 6,000 + 80 (6) x 1.3 = 6960
Q1 = Yt = 6,000 + 80 (7) x 0.7 = 7040

Answer: Q4 = 6,960 and Q1 =7,040

5. Over the past five years, a firm's sales have averaged 250 units in the first quarter of each year, 100
units in the second quarter, 150 units in the third quarter, and 300 units in the fourth quarter. What are
appropriate quarter relatives for this firm's sales? Hint: Only minimal computations are necessary.

Note : Quarter Ave

Solution and Answer:

Q1 = 250 /200 = 1.25

7
Q2 = 100 / 200 = 0.5
Q3 = 150 / 200 = 0.75
Q4= 300 / 200 = 1.5

6. Given the data below, develop a forecast for the following period using a three-period weighted
moving average and weights of .6, .3 and .1

Answer:

Period Demand Three-period weighted

1 19  
2 20 19 (Naïve Forecast)
3 18 20 (Naïve Forecast)
4 19 18.7
5 17 16.8
6   15.9

7. A manager has been using a certain technique to forecast demand for liters of ice cream for the past
six periods. Actual and predicted amounts are shown below. Would an exponential forecast have
produced better results?

Answer:

Exponential Forecast
Period Demand Current Forecast Error @ alpha =0.1 Error
1 90 87 3    

8
 2 85 88 3 90 5
3 91 87 4 90 2
4 92 89 3 90 2
5 95 90 5 90 5
6 88 92 4 90 2
Sum of Error 22 16
MAD 3.67 3.27

Based on the MAD value, Exponential Forecast is better to use with value of 3.27 compare to
Current Forecast with MAD value of 3.67

8. A new car dealer has been using exponential smoothing with an alpha of .1 to forecast weekly new
car sales. Given the data below, would a naive forecast have provided greater accuracy? Explain.
Assume an initial exponential forecast of 60 units in period 2 (i.e., no forecast for period 1).

9. A CPA firm has been using the following equation to predict annual demand for tax audits:
Yt = 45 + 4t
Demand for the past few years is shown below. Is the forecast performing as well as it might?
Explain.

10. Use linear regression to develop a predictive model for demand for burial vaults based on sales of
caskets.

A) Develop the equation.

11. Given the following data, develop a linear regression model for y as a function of x.

12. Given the following data, develop a linear regression model for y as a function of x.

9
13. Develop a linear trend equation for the data on bread deliveries shown below. Forecast deliveries
for period 11 through 14.

10