Predictive analytics in SC

Qualitative and Judgmental

The Delphi Method
Demand forecasting
Time series forecasting
Moving average
Exponential smoothing model
The Holts method
Linear regression-trend forecast
Week 6
Demand Sensing and Shaping

▪ Demand sensing is a forecasting method that leverages new

mathematical techniques and near real-time information to create an
accurate forecast of demand, based on the current realities of the
supply chain
▪ Demand shaping is an operational supply chain management strategy
where a company uses tactics such as price incentives, cost
modifications and product substitutions to attract customers to
purchase specific items.
Forecasting Process Steps

▪ Determine the use of the forecast

▪ Select the items to be forecasted
▪ Determine the time horizon of the forecast
▪ Select the forecasting model(s) and methods
▪ Gather the data needed to make the forecast
▪ Generate forecasts
▪ Validate and implement the results
Qualitative Models for Forecasting in fact prediction

▪ Knowledge and Insight of the Products

▪ Historical analogy/Market surveys
▪ Delphi Method
▪ Jury of Executive Opinion
Judgmental-Knowledge and Insight of the Products
- It rely on experience and awareness
- It is necessary when historical data are not available or when the
decision maker needs to forecast far into the future
- For example, a prediction of when the next generation of a
microprocessor will be available and what capabilities it might have
will depend greatly on the opinions and expertise of individuals
who understand the technology

5
Historical analogy
- A prediction is obtained through a comparative analysis with a
previous situation
- If a new product is being introduced, the response of consumers to
marketing campaigns to similar, previous products can be used as a
basis to predict how the new marketing campaign might charge

6
The Delphi Method
- It uses a panel of experts, whose identities are kept confidential
from one another, to respond to a sequence of questionnaires
- After each round of responses, individual opinions, are shared,
allowing each to see what the other experts think, anonymously
- Seeing other experts’ opinions helps to strengthen those in
agreement and to influence those who did not agree
- In the next round, the experts revise their estimates, and the
process is repeated, usually for no more than two or three rounds

7
Quantitative methods of forecasting
▪ Time Series Models
▪ Moving average
▪ Weighted moving average
▪ Exponential smoothing Time series components
▪ Associative Models (linear and
multiple regression)
▪ Some Statistical models (Holt’s, Focus
Forecasting, etc.)

linear regression
Time series/statistical forecasting models
- A time series is a stream of historical data, such as weekly sales
- We characterize the values of a time series over T periods as At , t =
1, 2, ….., T.
- Time series that do not have trend, seasonal, or cyclical effects but
are relatively constant called stationary time series
- Trend is a gradual upward or downward movement of a time series

9
Seasonal and cyclic effects
Seasonal effect
- One that repeats at fixed intervals
of time, typically a year, month,
week, or day
- Grocery store-On weekends
Cyclic effect
- Ups and downs over a much longer
time frame, such as several years
- Inflation and recession
10
Example-Identify trend
- A chart of total energy
consumption
- It shows an upward trend
- Consumption was rising quite
rapidly in a linear fashion during
the 1960s, then began increasing
at a slower rate through the
1980s and 1990s
- In the later decade, a slight
downward trend
11
Moving average
- The time series appears to be
relatively stable, without trend,
seasonal, or cyclical effects
- Setting k = 3, the three-period
moving average forecast for
week 18 is:

12
Moving average-Excel

13
Error Metrics and Forecast Accuracy
- To analyze the effectiveness of different forecasting models
- Error metrics needed, which compare the forecast with the actual
observations
- Mean absolute deviation (MAD)
- It is the absolute difference between the actual value and the
forecast, averaged over a range of forecasted values

14
Forecasting Models for Stationary Time Series
Mean square error (MSE)
- Squaring larger numbers has a greater impact than squaring
smaller numbers

- At is the actual value of the time series at time t,

- Ft is the forecast value for time t, and
- n is the number of forecast values (not the number of data points)
15
Forecasting Models for Stationary Time Series
Root mean square error (RMSE)

Mean absolute percentage error (MAPE)

16
Example-Compare Moving Average Forecasts with error metrics

17
Exponential Smoothing Models
- A multipurpose, approach for short-range forecasting is exponential
smoothing

- Ft+1 is the forecast for time t + 1,

- Ft is the forecast for period t,
- At is the observed value in period t,
- a or alpha is a constant between 0-1 called the smoothing constant

18
Example
- The forecast for week 2 is 88 (initial
value needed)
- Choose alpha = 0.7; then the
forecast for week 3 would be

19
Finding the Best Exponential Smoothing Model
for Tablet Computer Sales alpha = 0.6 provides the lowest
error for all three metrics

20
The Holt Forecasting Model
▪ Accounts for trends in time series
▪ Two components
▪ Exponentially smoothed component, Et
▪ Smoothing constant 0 < w < 1
▪ Trend component, Tt
▪ Smoothing constant 0 < v < 1
▪ Close to 0: More weight to past trend
▪ Close to 1: More weight to recent trend
Steps for Calculating Components of the Holt
Forecasting Model

▪ Select an exponential smoothing constant w

between 0 and 1.
▪ Small values of w give less weight to the current
values of the time series and more weight to the
past.
▪ Larger choices assign more weight to the current
value of the series.
Steps for Calculating Components of the
Holt Forecasting Model

▪ Select a trend smoothing constant v between 0 and 1.

▪ Small values of v give less weight to the current
changes in the level of the series and more weight to
the past trend.
▪ Larger values assign more weight to the most recent
trend of the series and less to past trends.
Steps for Calculating Components of the Holt
Forecasting Model

Calculate the two components, Et and Tt, from the time

series Yt beginning at time t = 2 :
E2 = Y2 and T2 = Y2 – Y1
E3 = wY3 + (1 – w)(E2 + T2)
T3 = v(E3 – E2) + (1 – v)T2
Et = wYt + (1 – w)(Et–1 + Tt–1)
Tt = v(Et – Et–1) + (1 – v)Tt–1
…
Holt Example

The closing stock prices on the last day of the

month for Sedan–double cabin in 2005 and
2006 are given in the table. Calculate the
Holt–Winters components using w = .8 and
v = .7.
 Calculate the two components, Et and Tt, from the
time series Yt beginning at time t = 2 :
Holt Solution E2 = Y2 and T2 = Y2 – Y1
E3 = wY3 + (1 – w)(E2 + T2)
T3 = v(E3 – E2) + (1 – v)T2

w = .8 v = .7 Et = wYt + (1 – w)(Et–1 + Tt–1)

Tt = v(Et – Et–1) + (1 – v)Tt–1

E2 = Y2 and T2 = Y2 – Y1
E2 = 46.10 and T2 = 46.10 – 45.51 = 0.59

E3 = wY3 + (1 – w)(E2 + T2)

E3 = 0.8(44.72) + 0.2(46.1 + 0.59) = 45.114

T3 = v(E3 – E2) + (1 – v)T2

T3 = .7(45.114 – 46.10) + 0.3(.59) = –0.5132
Holt Solution
Completed series

Holt exponentially smoothed

(w = .8 and v = .7)
Holt’s Forecasting Methodology

1. Calculate the exponentially smoothed and trend

components, Et and Tt, for each observed value of Yt (t
≥ 2) using the formulas
2. Calculate the one-step-ahead forecast using
Ft+1 = Et + 1*Tt
3. Calculate the k-step-ahead forecast using
Ft+k = Et + kTt
Holt Forecasting Example
Use the Holt series to forecast the closing
price of Sedan–double stock on
1/31/2007 and 2/28/2007.

1/31/2007 is one–step–ahead:
F1/31/07 = E12/29/06 + T12/29/06
= 61.39 + 3.00 = 64.39

2/28/2007 is two–steps–ahead:
F2/28/07 = E12/29/06 + 2T12/29/06
= 61.39 + 2(3.00) = 67.39
Linear Regression-Trend forecast

• Model:
• E(Yt) = β0 + β1t
• Relates time series, Yt, to time, t
• Cautions
• Risky to extrapolate (forecast beyond observed data)
• Does not account for cyclical effects
Example

The data shows the average

undergraduate tuition at all 4–year
institutions for the years 1996–2004
- Use least–squares regression to fit
a linear model.
- Forecast the tuition for year 2005
(t = 11)
Simple Linear Regression Solution
From Excel
$15,000

$14,000

Yˆt = 7997.533 + 528.158t

Tuition $13,000

$12,000

$11,000

$10,000

$9,000

$8,000
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Year

Forecast tuition for 2005 (t = 11):

Yˆ11 = 7997.533 + 528.158(11) = 13807.27
A Blended Approach: Collaboration and Consensus Forecasting
Simple Linear Regression-Another example
By hand