Professional Documents
Culture Documents
Slide 1
Chapter 15 Forecasting
Qualitative Approaches to Forecasting Quantitative Approaches to Forecasting The Components of a Time Series Using Smoothing Methods in Forecasting Measures of Forecast Accuracy Using Trend Projection in Forecasting
2003 Thomson/South-Western
Slide 1
We will cover the above-mentioned forecasting methods, covering sections 1 through 3 and section 6. Read over section 5 (Regression analysis which you must have covered in statistics class) Slide 2 Defining forecasting
What is Forecasting?
2003 Thomson/South-Western
4-2
Slide 3
Short-range forecast Up to 1 year; usually less than 3 months Job scheduling, worker assignments Medium-range forecast 3 months to 3 years Sales & production planning, budgeting Long-range forecast 3+ years New product planning, facility location
2003 Thomson/South-Western
4-3
Slide 4
Forecasting Approaches
Qualitative Methods
New products New technology
Quantitative Methods
2003 Thomson/South-Western
4-4
Forecasting methods fall under two main umbrellas: quantitative based and qualitative based. The slide provides the main difference between the two approaches and their applicability.
Slide 5
Qualitative Approaches to Forecasting
Delphi Approach A panel of experts, each of whom is physically separated from the others and is anonymous, is asked to respond to a sequential series of questionnaires. After each questionnaire, the responses are tabulated and the information and opinions of the entire group are made known to each of the other panel members so that they may revise their previous forecast response. The process continues until some degree of consensus is achieved.
2003 Thomson/South-Western
Slide 5
Qualitative approaches to forecasting can take on a variety of different methods. In the Delphi Approach a panel of experts is used to determine what the forecast will be. A good example is the forecasted price of stocks. Stock price forecasts are based on expert opinions from various fund managers and industry experts. Slide 6
Qualitative Approaches to Forecasting
Scenario Writing Scenario writing consists of developing a conceptual scenario of the future based on a well defined set of assumptions. After several different scenarios have been developed, the decision maker determines which is most likely to occur in the future and makes decisions accordingly.
2003 Thomson/South-Western
Slide 6
Scenario writing is another approach to qualitative forecasting. Here a variety of scenarios are written and the decision maker determines what he / she feels is most likely. A business manager might use this to gather input from each of the different business areas. For example, production might write a scenario, sales might write another scenario, and finance might write a scenario. The decision maker would then evaluate each of the potential scenarios and determine the forecast.
Slide 7
Qualitative Approaches to Forecasting
Subjective or Interactive Approaches These techniques are often used by committees or panels seeking to develop new ideas or solve complex problems. They often involve "brainstorming sessions". It is important in such sessions that any ideas or opinions be permitted to be presented without regard to its relevancy and without fear of criticism.
2003 Thomson/South-Western
Slide 7
Subjective approaches typically involve brainstorming to determine what the committee or group views as the likely forecast. A great example is when the sales manager asks each employee to estimate their likely sales. The group then discusses their potential as a whole and from this data the sales manager will forecast his sales.
Slide 8
Quantitative Approaches to Forecasting
Quantitative methods are based on an analysis of historical data concerning one or more time series. A time series is a set of observations measured at successive points in time or over successive periods of time. If the historical data used are restricted to past values of the series that we are trying to forecast, the procedure is called a time series method. If the historical data used involve other time series that are believed to be related to the time series that we are trying to forecast, the procedure is called a causal method.
2003 Thomson/South-Western
Slide 8
Although qualitative methods have their purpose, quantitative methods are rooted in data and should provide a better forecast. It is important to use good, historical data that depicts what has happened and can likely predict what will happen. We will study a number of time series methods for quantitative forecasting and we will use regression analysis to develop a causal forecast.
Slide 9
Moving Average
Exponential Smoothing
Trend Projection
Linear Regression
2003 Thomson/South-Western
Quantitative methods in forecasting fall in turn into 2 main streams: time series based and Associative or Causal Model. Linear regression is the most popular causal method used (read over section 5). We will focus on times series models.
Slide 10
The trend component accounts for the gradual shifting of the time series over a long period of time. Any regular pattern of sequences of values above and below the trend line is attributable to the cyclical component of the series. The seasonal component of the series accounts for regular patterns of variability within certain time periods, such as over a year. The irregular component of the series is caused by short-term, unanticipated and non-recurring factors that affect the values of the time series. One cannot attempt to predict its impact on the time series in advance.
2003 Thomson/South-Western Slide 10
Time series implies that the historical data have been gathered over a defined period of regular time intervals. This could be weeks, months, days, etc. Plotting the data is always the best step. A simple line chart will show if there is any pattern to the data. Patterns might be the result of seasonal increases in sales or regular, cyclical occurring influencers. I always think about my husbands company who purchases a significant number of cars for their employees. The car purchasing is cyclical every three years new cars are bought. If you are the lucky dealership from whom these cars are purchased you would show a cyclical (every 3 years) pattern for this customer. Car dealerships might also show seasonal patterns, for example near the end of the year when they have significant sales in an effort to reduce their inventory (they do this for tax purposes, so I am told ~ not really because of the volume of incoming new models!)
Slide 11
Trend
Cyclical
Seasonal
Random
2003 Thomson/South-Western
4-11
Slide 12
Trend Component
Persistent, overall upward or downward pattern Due to population, technology etc. Several years duration
Response
4-12
Slide 13
Seasonal Component
Regular pattern of up & down fluctuations Due to weather, customs etc. Occurs within 1 year
Summer Response
1984-1994 T/Maker Co.
Mo., Qtr.
2003 Thomson/South-Western
4-13
Slide 14
Cyclical Component
Repeating up & down movements Due to interactions of factors influencing economy Usually 2-10 years duration
Cycle Response
4-14
Slide 15
Random Component
Erratic, unsystematic, residual fluctuations Due to random variation or unforeseen events Union strike 1984-1994 T/Maker Co. Tornado Short duration & nonrepeating
2003 Thomson/South-Western
4-15
Slide 16
Three time series methods are: smoothing trend projection trend projection adjusted for seasonal influence
2003 Thomson/South-Western
Slide 16
These are the three time series methods that we will review.
Slide 17
Smoothing Methods
In cases in which the time series is fairly stable and has no significant trend, seasonal, or cyclical effects, one can use smoothing methods to average out the irregular components of the time series. Three common smoothing methods are: Moving averages Weighted moving averages Exponential smoothing
2003 Thomson/South-Western
Slide 17
Our goal in forecasting is to smooth out the irregular patterns and plan for the known seasonal or cyclical patterns. Three common smoothing methods are: Moving averages, weighted moving averages, and exponential smoothing. Each of these methods has a different approach to smoothing out the patterns in the data.
Slide 18
Smoothing Methods
Moving Average Method The moving average method consists of computing an average of the most recent n data values for the series and using this average for forecasting the value of the time series for the next period.
2003 Thomson/South-Western
Slide 18
The moving average method is the least complex. Basically you use the average of some period of recent data to predict future data. Lets take a look through an example. Slide 19
Example: Roberts Drugs
During the past ten weeks, sales of cases of Comfort brand headache medicine at Robert's Drugs have been as follows: Week Sales 1 110 2 115 3 125 4 120 5 125 Week 6 7 8 9 10 Sales 120 130 115 110 130
If Robert's uses a 3-period moving average to forecast sales, what is the forecast for Week 11?
Slide 19
2003 Thomson/South-Western
Roberts Drugs shows their sales data for the last 10 weeks. They would like to use a 3-period moving average to forecast week 11. To do this, the forecaster will use the historical data for weeks 8, 9, and 10 and determine the average. This value is the forecast for week 11.
Slide 20
Forect+1
2003 Thomson/South-Western
Slide 20
Slide 21
Steps to Moving Average Using Excel Step 1: Select the Tools pull-down menu. Step 2: Select the Data Analysis option. Step 3: When the Data Analysis Tools dialog appears, choose Moving Average. Step 4: When the Moving Average dialog box appears: Enter B4:B13 in the Input Range box. Enter 3 in the Interval box. Enter C4 in the Output Range box. Select OK.
Slide 21
2003 Thomson/South-Western
Excel has a function that easily computes the moving average. See instructions in slide above. For Excell 2007, need to upload the ToolPack Analysis tool. Will show in class and will upload instructions on shell.
Slide 22
Forect+1 #N/A #N/A 116.7 120.0 123.3 121.7 125.0 121.7 118.3 118.3
Slide 22
2003 Thomson/South-Western
Excel has a function that easily computes the moving average. In the above slide you can see that excel has computed the forecasted 3-period moving average for weeks 4-14. It is helpful to have a forecast for historical data ~ that is, we did not need to forecast week 4, because we already knew the sales figures (Remember from the previous slide we had data through week 10). But by doing this, we can determine the accuracy of our forecast methodology. We will look at measures of forecast accuracy later in this lecture. Take a minute and notice how Excel outputs the data. The forecast that is horizontal with week 3 is 116.7. This is actually the forecast for week 4 so, if we wanted to see how well our model would have predicted the forecast for week 4, we can compare the forecast values for week 4 (116.7) with the actual values for week 4 (120).
Slide 23
Smoothing Methods
Weighted Moving Average Method In the weighted moving average method for computing the average of the most recent n periods, the more recent observations are typically given more weight than older observations. For convenience, the weights usually sum to 1.
2003 Thomson/South-Western
Slide 23
The weighted moving average is similar to the moving average except we assign weights to each of the time periods. Remember our goal: to smooth out variations in the data that are irregular. We might find that our sales show an increasing pattern. In this case, using a moving average of 3-periods will not account for the increasing pattern. Instead, we might assign a high weight to the most recent period and a lower weight to the less recent periods. Lets look at an example.
Slide 24
Example: Augers Plumbing Service
Forecast for December (Month 10) using a threeperiod (n = 3) weighted moving average with weights of .6, .3, and .1. Month March April May June July Aug Sept Oct Nov
2003 Thomson/South-Western
Jobs 353 387 342 374 396 409 399 412 408
Slide 24
Lets use a 3-period weighted moving average to forecast Augers plumbing service. The weights are given above. Slide 25
Example: Augers Plumbing Service (B)
Three-Month Weighted Moving Average The forecast for December will be the weighted average of the preceding three months: September, October, and November. F10 = .1YSep. + .3YOct. + .6YNov. = .1(399) + .3(412) + .6(408) = 408.3
2003 Thomson/South-Western
Slide 25
The forecast for period 10 (December) uses the previous 3 months of data (Sept, Oct, Nov). November is the most recent month so it is assigned a weight of .6. October is assigned a weight of .3 and September is assigned a weight of .1. We will multiple each of the previous months actual data times the weight that has been assigned to that month. The resulting value, 408.3, is the forecast for December. Although this is not difficult to do in Excel, you must manually set up the equations. Excel does not have a function that computes the weighted moving average.
Slide 26
Smoothing Methods
Exponential Smoothing Using exponential smoothing, the forecast for the next period is equal to the forecast for the current period plus a proportion () of the forecast error in the current period. Using exponential smoothing, the forecast is calculated by: [the actual value for the current period] + (1- )[the forecasted value for the current period], where the smoothing constant, , is a number between 0 and 1. OR Ft = Y t-1 (1- ) F t-1 +
2003 Thomson/South-Western Slide 26
The third smoothing method that we will consider is exponential smoothing. This method is very useful for companies with little historical data. Exponential smoothing only requires one month of historical data. The formula is shown above in a yellow box. The forecast for period t is equal to the value of the smoothing constant times the actual sales for the previous month plus 1 minus the smoothing constant times the forecast for the previous month. Exponential smoothing uses a smoothing constant and the accuracy of the previous months forecast to determine the forecast for the next period. The smoothing constant is determined by the decision maker. It is a value between 0 and 1. The closer the value is to 1 the more the emphasis is on the previous months forecast (and more recent periods). Lets look at an example.
Slide 27
If Robert's uses exponential smoothing to forecast sales, which value for the smoothing constant , = .1 or = .8, gives better forecasts?
2003 Thomson/South-Western Slide 27
Roberts Drugs is not convinced that his moving average forecast is good. He would like to use exponential smoothing to forecast his sales. Robert is going to use an exponential smoothing constant of .1 and .8. He plans to compare the accuracy of each forecast and determine the best one.
Slide 28
Example: Roberts Drugs
Exponential Smoothing ( = .1, 1 - = .9) F1 F2 = .1Y1 + .9F1 = .1(110) + .9(110) F3 = .1Y2 + .9F2 = .1(115) + .9(110) F4 = .1Y3 + .9F3 = .1(125) + .9(110.5) F5 = .1Y4 + .9F4 = .1(120) + .9(111.95) F6 = .1Y5 + .9F5 = .1(125) + .9(112.76) F7 = .1Y6 + .9F6 = .1(120) + .9(113.98) F8 = .1Y7 + .9F7 = .1(130) + .9(114.58) F9 = .1Y8 + .9F8 = .1(115) + .9(116.12) F10= .1Y9 + .9F9 = .1(110) + .9(116.01)
2003 Thomson/South-Western
= 110 = 110 = 110.5 = 111.95 = 112.76 = 113.98 = 114.58 = 116.12 = 116.01 = 115.41
Slide 28
The forecast for period 1 is 110. This is because the forecast for the first period is always set equal to the actual for that period (unless otherwise given in the problem). This does force the forecast for period 2 to be equal to the first months actual data, but after the first 2 periods the forecast works fine. So to forecast period 2, Robert multiplies the smoothing constant of .1 times the actual sales for period 1 (110). He adds to this the multiplication of .9 (1 - .1) to the forecast for period 1. This process continues Lets focus in on period 10s forecast. The forecast for period 10 is .1 times the actual for period 9 (110) plus .9 times the forecast for period 9 (116.01). If you look closely you can see that a smoothing constant close to 1 will put a lot of emphasis on the actual value for the previous month. So the smoothing constant that you choose is dependent upon how much you want to rely on the previous months data versus the previously forecasted data.
Slide 29
Exponential Smoothing ( = .8, 1 - = .2) F1 = 110 F2 = .8(110) + .2(110) = 110 F3 = .8(115) + .2(110) = 114 F4 = .8(125) + .2(114) = 122.80 F5 = .8(120) + .2(122.80) = 120.56 F6 = .8(125) + .2(120.56) = 124.11 F7 = .8(120) + .2(124.11) = 120.82 F8 = .8(130) + .2(120.82) = 128.16 F9 = .8(115) + .2(128.16) = 117.63 F10= .8(110) + .2(117.63) = 111.53
2003 Thomson/South-Western Slide 29
Using an exponential smoothing constant of .8, we can determine another forecast for Roberts Drugs. In this example, the smoothing constant is close to 1, so a lot of emphasis is being placed on the actual sales data for the previous month. Exponential smoothing can be done easily in Excel. The next slides show how to set up in Excel and an excel poutput for the example.
Slide 30
2003 Thomson/South-Western
Slide 31
Steps to Exponential Smoothing Using Excel Step 1: Select the Tools pull-down menu. Step 2: Select the Data Analysis option. Step 3: When the Data Analysis Tools dialog appears, choose Exponential Smoothing. Step 4: When the Exponential Smoothing dialog box appears: Enter B4:B13 in the Input Range box. Enter 0.9 (for = 0.1) in Damping Factor box. Enter C4 in the Output Range box. Select OK.
Slide 31
2003 Thomson/South-Western
Slide 32
2003 Thomson/South-Western
Slide 33
Repeating the Process for = 0.8 Step 4: When the Exponential Smoothing dialog box appears: Enter B4:B13 in the Input Range box. Enter 0.2 (for = 0.8) in Damping Factor box. Enter D4 in the Output Range box. Select OK.
2003 Thomson/South-Western
Slide 33
Slide 34
2003 Thomson/South-Western
Slide 35
Select the items to be forecast Determine the time horizon of the forecast Collect Data Plot and analyze Select the forecasting model(s) Make the forecast Validate and implement results (monitoring) Accuracy (magnitude of error) and bias
2003 Thomson/South-Western
Slide 35
Slide 36
Mean Squared Error The average of the squared forecast errors for the historical data is calculated. The forecasting method or parameter(s) which minimize this mean squared error is then selected. Mean Absolute Deviation The mean of the absolute values of all forecast errors is calculated, and the forecasting method or parameter(s) which minimize this measure is selected. The mean absolute deviation measure is less sensitive to individual large forecast errors than the mean squared error measure.
Slide 36
2003 Thomson/South-Western
Now that Robert has two different forecasts using exponential smoothing, we need to evaluate which forecast is more accurate. There are two approaches to forecast accuracy. If we subtract the forecast value from the actual value we have an indication of how well the forecast performed for each period. But lets get an overall measure of accuracy. One method is the mean square errors. Here we square the forecast errors for each period and then take their average. The goal is to minimize your MSE. The issue with MSE is that your MSE is in squared terms, so it is not always easy to interpret. One other approach is the mean absolute deviation. To compute the MAD, we take the absolute value of the forecast errors for each period and average these values. The MAD is easier to interpret than the MAD because the units are not squared. The MAD provides an average amount of deviation (over or under) that is given with the forecast. Again, the goal is to minimize the MAD.
Slide 37
Example: Roberts Drugs
Mean Squared Error In order to determine which smoothing constant gives the better performance, calculate, for each, the mean squared error for the nine weeks of forecasts, weeks 2 through 10 by: [(Y2-F2)2 + (Y3-F3)2 + (Y4-F4)2 + . . . + (Y10-F10)2]/9
2003 Thomson/South-Western
Slide 37
Using Roberts Drugs we can determine the MSE for both the smoothing constant of .1 and .8 Slide 38
Example: Roberts Drugs
Week 1 2 3 4 5 6 7 8 9 10 Yt 110 115 125 120 125 120 130 115 110 130 MSE Ft = .1 (Yt - Ft)2 25.00 210.25 64.80 149.94 36.25 237.73 1.26 36.12 212.87 Ft = .8 (Yt - Ft)2
110.00 25.00 114.00 121.00 122.80 7.84 120.56 19.71 124.11 16.91 120.82 84.23 128.16 173.30 117.63 58.26 111.53 341.27 Sum 847.52 Sum/9 94.17
Slide 38
2003 Thomson/South-Western
The above table shows that the smoothing constant of .8 provides a lower MSE, thus it is preferred to the smoothing constant of .1.
Slide 39
Trend Projection
If a time series exhibits a linear trend, the method of least squares may be used to determine a trend line (projection) for future forecasts. Least squares, also used in regression analysis, determines the unique trend line forecast which minimizes the mean square error between the trend line forecasts and the actual observed values for the time series. The independent variable is the time period and the dependent variable is the actual observed value in the time series.
2003 Thomson/South-Western
Slide 39
Trend projection is a time series method that uses the least squares to determine a trend line. Least squares method is also used in regression. Rather than have you labor through computing a trend line by hand, I prefer to use regression to demonstrate both trend projection and causal method forecasting. For regression, an independent variable is used to predict a dependent variable. In forecasting the dependent variable is what we are trying to forecast (i.e. sales, jobs, etc.). The independent variable is what we are using to predict the forecast. For trend projection the independent variable is always time; period 1, 2, 3, etc. For causal regression the sky is our limit with respect to independent variables. For example, we can predict season ticket sales for the Astros based on their previous seasons record. Or we can predict our grade on final exam based on our grades on tests 1 and 2. As you can see, in casual regression, we can use any other variable to help us forecast. A great example is the interest rate and new home sales! Lets look further.
Slide 40
Trend Projection
Using the method of least squares, the formula for the trend projection is: Tt = b0 + b1t. where: Tt = trend forecast for time period t b1 = slope of the trend line b0 = trend line projection for time 0 b1 = ntYt - t Yt nt 2 - (t )2 where: Yt = observed value of the time series at time period t
b0= Y b1 t
Y = average of the observed values for Yt t = average time period for the n observations
2003 Thomson/South-Western Slide 40
I have provided you the equations that are used to determine the least sum of squares trend projection line. As you can see there are a lot of opportunities for error, so I suggest we use Excel.
Slide 41
Forecast the number of repair jobs Auger's will perform in December using the least squares method.
2003 Thomson/South-Western
Slide 41
We will use Augers Plumbing service to demonstrate trend projections. We will see how to use Excel to compute Trend projection.
Slide 42
Y = 3480/9 = 386.667
= (9)(17844) - (45)(3480) (9)(285) - (45)2 = 7.4
2003 Thomson/South-Western
Slide 42
Slide 43
2003 Thomson/South-Western
Slide 43
Steps to Trend Projection Using Excel Step 1: Select an empty cell (B13) in the worksheet. Step 2: Select the Insert pull-down menu. Step 3: Choose the Function option. Step 4: When the Paste Function dialog box appears: Choose Statistical in Function Category box. Choose Forecast in the Function Name box. Select OK. more . . . . . . .
2003 Thomson/South-Western
Slide 44
Slide 45 Continue
Example: Augers Plumbing Service
Steps to Trend Projecting Using Excel (continued) Step 5: When the Forecast dialog box appears: Enter 10 in the x box (for month 10). Enter B4:B12 in the Known ys box. Enter A4:A12 in the Known xs box. Select OK.
2003 Thomson/South-Western
Slide 45
Slide 46
Projected
Slide 46
2003 Thomson/South-Western
Slide 47
2003 Thomson/South-Western
Slide 47
Slide 48
Three-Month Weighted Moving Average The forecast for December will be the weighted average of the preceding three months: September, October, and November. F10 = .1YSep. + .3YOct. + .6YNov. = .1(399) + .3(412) + .6(408) = 408.3 Trend Projection F10 = 423.7 (from earlier slide)
2003 Thomson/South-Western
Slide 48
Slide 49
Conclusion Due to the positive trend component in the time series, the trend projection produced a forecast that is more in tune with the trend that exists. The weighted moving average, even with heavy (.6) placed on the current period, produced a forecast that is lagging behind the changing data.
2003 Thomson/South-Western
Slide 49
Slide 50
End of Chapter 15
2003 Thomson/South-Western
Slide 50