Applied Business Forecasting and Planning

MOVING AVERAGES AND EXPONENTIAL SMOOTHING

Introduction

This chapter introduces models applicable to time series data with seasonal, trend, or both seasonal and trend component and stationary data. Forecasting methods discussed in this chapter can be classified as:

Averaging methods.

Equally weighted observations Unequal set of weights to past data, where the weights decay exponentially from the most recent to the most distant data points. All methods in this group require that certain parameters to be defined. These parameters (with values between 0 and 1) will determine the unequal weights to be applied to past data.

Exponential Smoothing methods.

Introduction

Averaging methods

If a time series is generated by a constant process subject to random error, then mean is a useful statistic and can be used as a forecast for the next period. Averaging methods are suitable for stationary time series data where the series is in equilibrium around a constant value ( the underlying mean) with a constant variance over time.

Introduction

Exponential smoothing methods

The simplest exponential smoothing method is the single smoothing (SES) method where only one parameter needs to be estimated Holts method makes use of two different parameters and allows forecasting for series with trend. Holt-Winters method involves three smoothing parameters to smooth the data, the trend, and the seasonal index.

Averaging Methods

The Mean

Uses the average of all the historical data as the forecast

1 t Ft 1 yi t i 1

When new data becomes available , the forecast for time t+2 is the new mean including the previously observed data plus this new observation.
Ft 2 1 t 1 yi t 1 i 1

This method is appropriate when there is no noticeable trend or seasonality.

Averaging Methods

The moving average for time period t is the mean of the k most recent observations. The constant number k is specified at the outset. The smaller the number k, the more weight is given to recent periods. The greater the number k, the less weight is given to more recent periods.

Moving Averages

A large k is desirable when there are wide, infrequent fluctuations in the series. A small k is most desirable when there are sudden shifts in the level of series. For quarterly data, a four-quarter moving average, MA(4), eliminates or averages out seasonal effects.

Moving Averages

For monthly data, a 12-month moving average, MA(12), eliminate or averages out seasonal effect. Equal weights are assigned to each observation used in the average. Each new data point is included in the average as it becomes available, and the oldest data point is discarded.

Moving Averages

A moving average of order k, MA(k) is the value of k consecutive observations.

Ft 1 yt 1 ( yt yt 1 yt 2 yt k 1 ) K 1 t Ft 1 1yi k i t k

K is the number of terms in the moving average.

The moving average model does not handle trend or seasonality very well although it can do better than the total mean.

Example: Weekly Department Store Sales

The weekly sales figures (in millions of dollars) presented in the following table are used by a major department store to determine the need for temporary sales personnel.

Period (t) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Sales (y) 5.3 4.4 5.4 5.8 5.6 4.8 5.6 5.6 5.4 6.5 5.1 5.8 5 6.2 5.6 6.7 5.2 5.5 5.8 5.1 5.8 6.7 5.2 6 5.8

Example: Weekly Department Store Sales

Weekly Sales
8

Sales

Sales (y)

0 0 5 10 15 Weeks 20 25 30

Example: Weekly Department Store Sales

Use a three-week moving average (k=3) for the department store sales to forecast for the week 24 and 26.
y24 ( y23 y22 y21 ) 5.2 6.7 5.8 5.9 3 3

The forecast error is

e24 y24 y24 6 5.9 .1

Example: Weekly Department Store Sales

The forecast for the week 26 is

y 26 y25 y24 y23 5.8 6 5.2 5.7 3 3

Example: Weekly Department Store Sales

RMSE = 0.63
Weekly Sales Forecasts

Period (t) 1 2 3 4 5 6 7 8 9 10 11 12 13
Sales (y) forecast

Sales (y) 5.3 4.4 5.4 5.8 5.6 4.8 5.6 5.6 5.4 6.5 5.1 5.8 5 6.2 5.6 6.7 5.2 5.5 5.8 5.1 5.8 6.7 5.2 6 5.8

forecast

5.033333 5.2 5.6 5.4 5.333333 5.333333 5.533333 5.833333 5.666667 5.8 5.3 5.666667 5.6 6.166667 5.833333 5.8 5.5 5.466667 5.566667 5.866667 5.9 5.966667 5.666667

Sales

14 15 16 17 18

19 20 21 22 23
0 5 10 15 Weeks 20 25 30

24 25

Exponential Smoothing Methods

This method provides an exponentially weighted moving average of all previously observed values. Appropriate for data with no predictable upward or downward trend. The aim is to estimate the current level and use it as a forecast of future value.

Simple Exponential Smoothing Method

Formally, the exponential smoothing equation is

Ft 1 yt (1 ) Ft

Ft 1 forecast for the next period.

= smoothing constant. yt = observed value of series in period t. Ft = old forecast for period t.

The forecast Ft+1 is based on weighting the most recent observation yt with a weight and weighting the most recent forecast Ft with a weight of 1-

Simple Exponential Smoothing Method

The implication of exponential smoothing can be better seen if the previous equation is expanded by replacing Ft with its components as follows:
Ft 1 yt (1 ) Ft yt (1 )[ yt 1 (1 ) Ft 1 ] yt (1 ) y t 1 (1 ) 2 Ft 1

Simple Exponential Smoothing Method

If this substitution process is repeated by replacing Ft-1 by its components, Ft-2 by its components, and so on the result is: Therefore, Ft+1 is the weighted moving average of all past observations.

Ft 1 yt (1 ) y t 1 (1 ) 2 y t 2 (1 )3 y t 3 (1 ) t 1 y1

Simple Exponential Smoothing Method

The following table shows the weights assigned to past observations for = 0.2, 0.4, 0.6, 0.8, 0.9
0.2 0.2 0.2(1-0.2) 0.2(1-0.2) 0.2(1-0.2) 0.2(1-0.2) 0.2(1-0.2)
2 3 4 5

Weight assigned to Yt Yt-1 Yt-2 Yt-3 Yt-4 Yt-5

0.4 0.4 0.4(1-0.4) 0.4(1-0.4) 0.4(1-0.4) 0.4(1-0.4) 0.4(1-0.4)

2 3 4 5

0.6 0.6 0.6(1-0.6) 0.6(1-0.6) 0.6(1-0.6) 0.6(1-0.6) 0.6(1-0.6)

2 3 4 5

0.8 0.8

0.9 0.9

Simple Exponential Smoothing Method

The exponential smoothing equation rewritten in the following form elucidate the role of weighting factor .
Ft 1 Ft ( yt Ft )

Exponential smoothing forecast is the old forecast plus an adjustment for the error that occurred in the last forecast.

Simple Exponential Smoothing Method

The value of smoothing constant must be between 0 and 1 can not be equal to 0 or 1. If stable predictions with smoothed random variation is desired then a small value of is desire. If a rapid response to a real change in the pattern of observations is desired, a large value of is appropriate.

Simple Exponential Smoothing Method

To estimate , Forecasts are computed for equal to .1, .2, .3, , .9 and the sum of squared forecast error is computed for each. The value of with the smallest RMSE is chosen for use in producing the future forecasts.

Simple Exponential Smoothing Method

To start the algorithm, we need F1 because

F2 y1 (1 ) F1

Since F1 is not known, we can

Set the first estimate equal to the first observation. Use the average of the first five or six observations for the initial smoothed value.

Example:University of Michigan Index of Consumer Sentiment

University of Michigan Index of Consumer Sentiment for January1995December1996. we want to forecast the University of Michigan Index of Consumer Sentiment using Simple Exponential Smoothing Method.

Date Observ ed Jan-95 97.6 Feb-95 95.1 Mar-95 90.3 Apr-95 92.5 May-95 89.8 Jun-95 92.7 Jul-95 94.4 Aug-95 96.2 Sep-95 88.9 Oct-95 90.2 Nov -95 88.2 Dec-95 91 Jan-96 89.3 Feb-96 88.5 Mar-96 93.7 Apr-96 92.7 May-96 94.7 Jun-96 95.3 Jul-96 94.7 Aug-96 95.3 Sep-96 94.7 Oct-96 96.5 Nov -96 99.2 Dec-96 96.9 Jan-97

Example:University of Michigan Index of Consumer Sentiment

Since no forecast is available for the first period, we will set the first estimate equal to the first observation. We try =0.3, and 0.6.

University of Michigan Index of Consumer Sentiment

100 98 96 94 92 90 88 86 Sep-94 Apr-95 Oct-95 Date May-96 Dec-96 Jun-97

Consumer Sentiment Index

Example:University of Michigan Index of Consumer Sentiment

Date Jan-95 Consumer Sentiment 97.6 95.1 90.3 92.5 89.8 92.7 94.4 96.2 88.9 90.2 88.2 91 89.3 88.5 93.7 92.7 89.4 92.4 94.7 95.3 94.7 96.5 99.2 96.9 97.4 99.7 100 101.4 103.2 104.5 107.1 104.4 106 105.6 107.2 102.1

Alpha =0.3 #N/A 97.60 96.85 94.89 94.17 92.86 92.81 93.29 94.16 92.58 91.87 90.77 90.84 90.38 89.81 90.98 91.50 90.87 91.33 92.34 93.23 93.67 94.52 95.92 96.22 96.57 97.51 98.26 99.20 100.40 101.63 103.27 103.61 104.33 104.71 105.46

Alpha =0.6 #N/A 97.60 96.10 92.62 92.55 90.90 91.98 93.43 95.09 91.38 90.67 89.19 90.28 89.69 88.98 91.81 92.34 90.58 91.67 93.49 94.58 94.65 95.76 97.82 97.27 97.35 98.76 99.50 100.64 102.18 103.57 105.69 104.92 105.57 105.59 106.55

Note the first forecast is the first observed value. The forecast for Feb. 95 (t = 2) and Mar. 95 (t = 3) are evaluated as follows:

Feb-95 Mar-95 Apr-95 May-95 Jun-95 Jul-95 Aug-95 Sep-95 Oct-95 Nov-95 Dec-95 Jan-96 Feb-96 Mar-96 Apr-96 May-96 Jun-96

yt 1 yt ( yt yt ) y2 y1 0.6( y1 y1 ) 97.6 0.6(97.6 97.6) 97.6 y3 y2 0.6( y2 y2 ) 97.6 0.6(95.1 97.6) 96.1

Jul-96 Aug-96 Sep-96 Oct-96 Nov-96 Dec-96 Jan-97 Feb-97 Mar-97 Apr-97 May-97 Jun-97 Jul-97 Aug-97 Sep-97 Oct-97 Nov-97 Dec-97

Example:University of Michigan Index of Consumer Sentiment

RMSE =2.66 for = 0.6 RMSE =2.96 for = 0.3

University of M ichigan Index of Consumer sentiments

120

100

80

Sentiment Index

Consumer Sentiment 60 SES (Alpha =0.3) SES(Alpha=0.6)

40

20

0 Jun-94 Oct-95 Mar-97 Months Jul-98 Dec-99 Apr-01

Holts Exponential smoothing

Holts two parameter exponential smoothing method is an extension of simple exponential smoothing. It adds a growth factor (or trend factor) to the smoothing equation as a way of adjusting for the trend.

Holts Exponential smoothing

Three equations and two smoothing constants are used in the model.

The exponentially smoothed series or current level estimate.

L t yt (1 )( Lt 1 bt 1 )

The trend estimate.

bt ( Lt Lt 1 ) (1 )bt 1

Forecast m periods into the future.

F t m Lt mbt

Holts Exponential smoothing

Lt = Estimate of the level of the series at time t = smoothing constant for the data. yt = new observation or actual value of series in period t. = smoothing constant for trend estimate bt = estimate of the slope of the series at time t m = periods to be forecast into the future.

Holts Exponential smoothing

The weight and can be selected subjectively or by minimizing a measure of forecast error such as RMSE. Large weights result in more rapid changes in the component. Small weights result in less rapid changes.

Holts Exponential smoothing

The initialization process for Holts linear exponential smoothing requires two estimates:

One to get the first smoothed value for L1 The other to get the trend b1.
b y 2 y1 1 or b 1 or b 0 1 y 4 y1 3

One alternative is to set L1 = y1 and

Example:Quarterly sales of saws for Acme tool company

The following table shows the sales of saws for the Acme tool Company. These are quarterly sales From 1994 through 2000.

Year 1994

Quarter 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4

t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

1995

1996

1997

1998

1999

2000

sales 500 350 250 400 450 350 200 300 350 200 150 400 550 350 250 550 550 400 350 600 750 500 400 650 850 600 450 700

Example:Quarterly sales of saws for Acme tool company

Examination of the plot shows:

Sales of saws for the Acme Tool Company: 1994-2000

900

Saws

A non-stationary time series data. Seasonal variation seems to exist.

800

700

600

500

400

300

Sales for the first and fourth quarter are larger than other quarters.

200

100

0 0 5 10 15 Year 20 25 30

Example:Quarterly sales of saws for Acme tool company

The plot of the Acme data shows that there might be trending in the data therefore we will try Holts model to produce forecasts. We need two initial values

The first smoothed value for L1 The initial trend value b1.

We will use the first observation for the estimate of the smoothed value L1, and the initial trend value b1 = 0. We will use = .3 and =.1.

Example:Quarterly sales of saws for Acme tool company

Year 1994 Quarter 1 2 3 4 1995 1 2 3 4 1996 1 2 3 4 1997 1 2 3 4 1998 1 2 3 4 1999 1 2 3 4 2000 1 2 3 4 t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 sales 500 350 250 400 450 350 200 300 350 200 150 400 550 350 250 550 550 400 350 600 750 500 400 650 850 600 450 700 Lt 500.00 455.00 390.35 385.88 398.18 378.34 318.61 303.23 307.38 266.55 220.98 261.95 339.77 340.55 311.38 379.12 431.67 427.00 407.92 467.83 558.73 553.10 517.56 564.16 659.35 656.71 608.16 644.43 bt 0.00 -4.50 -10.52 -9.91 -7.69 -8.90 -13.99 -14.13 -12.30 -15.15 -18.19 -12.28 -3.27 -2.86 -5.49 1.83 6.90 5.74 3.26 8.93 17.12 14.85 9.81 13.49 21.66 19.23 12.45 14.83 Ft+m 500.00 500.00 450.50 379.84 375.97 390.49 369.44 304.62 289.11 295.08 251.40 202.79 249.67 336.50 337.69 305.89 380.95 438.57 432.74 411.18 476.75 575.85 567.94 527.37 577.65 681.01 675.94 620.61

Example:Quarterly sales of saws for Acme tool company

RMSE for this application is: = .3 and = .1 RMSE = 155.5 The plot also showed the possibility of seasonal variation that needs to be investigated.

Quarterly Saw Sales Forecast Holt's Method

900

800

700

600

500
Sales

sales Ht+m 400

300

200

100

0 0 5 10 15 Quarters 20 25 30

Winters Exponential Smoothing

Winters exponential smoothing model is the second extension of the basic Exponential smoothing model. It is used for data that exhibit both trend and seasonality. It is a three parameter model that is an extension of Holts method. An additional equation adjusts the model for the seasonal component.

Winters Exponential Smoothing

The four equations necessary for Winters multiplicative method are:

The exponentially smoothed series:

Lt yt (1 )( Lt 1 bt 1 ) St s

The trend estimate:

bt ( Lt Lt 1 ) (1 )bt 1

The seasonality estimate:

St yt (1 ) St s Lt

Winters Exponential Smoothing

Forecast m period into the future:

Ft m ( Lt mbt ) St m s

Lt = level of series. = smoothing constant for the data. yt = new observation or actual value in period t. = smoothing constant for trend estimate. bt = trend estimate. = smoothing constant for seasonality estimate. St =seasonal component estimate. m = Number of periods in the forecast lead period. s = length of seasonality (number of periods in the season) Ft m = forecast for m periods into the future.

Winters Exponential Smoothing

As with Holts linear exponential smoothing, the weights , , and can be selected subjectively or by minimizing a measure of forecast error such as RMSE. As with all exponential smoothing methods, we need initial values for the components to start the algorithm. To start the algorithm, the initial values for Lt, the trend bt, and the indices St must be set.

Winters Exponential Smoothing

To determine initial estimates of the seasonal indices we need to use at least one complete season's data (i.e. s periods).Therefore,we initialize trend and level at period s. Initialize level as:
1 Ls ( y1 y2 ys ) s

Initialize trend as Initialize seasonal indices as:

S1

y ys 1 y y y y2 bs ( s 1 1 s 2 ss ) s s s s
y y1 y , S 2 2 ,, S s s Ls Ls Ls

Winters Exponential Smoothing

We will apply Winters method to Acme Tool company sales. The value for is .4, the value for is .1, and the value for is .3. The smoothing constant smoothes the data to eliminate randomness. The smoothing constant smoothes the trend in the data set.

Winters Exponential Smoothing

The smoothing constant smoothes the seasonality in the data. The initial values for the smoothed series Lt, the trend bt, and the seasonal index St must be set.

Example: Quarterly Sales of Saws for Acme tool

Year 1994 Quarter 1 2 3 4 1995 1 2 3 4 1996 1 2 3 4 1997 1 2 3 4 1998 1 2 3 4 1999 1 2 3 4 2000 1 2 3 4 t 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 sales 500 350 250 400 450 350 200 300 350 200 150 400 550 350 250 550 550 400 350 600 750 500 400 650 850 600 450 700 375 396.9667 372.3747 296.7938 287.3869 302.1219 252.9623 201.4173 268.2504 373.5062 363.8087 317.4823 406.7605 465.9614 444.9496 410.5851 487.3071 597.7855 570.255 510.9496 570.7076 689.6728 667.561 591.6084 640.1658 -12.5 -9.05333 -10.6072 -17.1046 -16.3348 -13.2278 -16.821 -20.2934 -11.5807 0.102908 -0.87713 -5.42206 4.047961 9.563264 6.505758 2.418728 9.84905 19.91199 15.16774 7.720431 12.92419 23.52829 18.96428 9.472591 13.38107 Lt bt St 1.333333 0.933333 0.666667 1.066667 1.273412 0.935307 0.668827 1.059833 1.23893 0.891905 0.691596 1.189227 1.309011 0.912946 0.720351 1.238103 1.270414 0.908756 0.759978 1.236049 1.265679 0.899169 0.766841 1.206915 1.255716 0.899057 0.764981 1.172881 483.3333 362.0524 241.1783 298.3352 345.161 270.2048 157.9377 191.9611 317.9958 333.2237 251.002 371.1103 537.7528 434.1286 325.2062 511.3412 631.5942 561.3363 444.9085 641.1016 738.6906 641.2886 526.4561 725.4539 Ft+m

Example: Quarterly Sales of Saws for Acme tool

RMSE for this application is: = 0.4, = 0.1, = 0.3 and RMSE = 83.36 Note the decrease in RMSE.

Quarterly Saw Sales Forecas:t Winter's Method

900

800

700

600

500 sales Ft+m 400

Sales

300

200

100

0 0 5 10 15 Quarters 20 25 30

Additive Seasonality

The seasonal component in Holt-Winters method. The basic equations for Holts Winters additive method are:
Lt ( yt St s ) (1 )(Lt 1 bt 1 ) bt ( Lt Lt 1 ) (1 )bt 1 St ( yt Lt ) (1 ) St s Ft m Lt bt m St m s

Additive Seasonality

The initial values for Ls and bs are identical to those for the multiplicative method. To initialize the seasonal indices we use
S1 y1 Ls , S 2 y2 Ls ,, S s Ys Ls