Forecasting

• Prepared by: Dr.Suhad Alnatoor

• Near east university
The role of forecasting in services
• Forecasts form the basis of planning in service enterprises
• Strategic plans (e.g., where to locate stores, warehouses, etc.) are
based on long-term forecasts covering several years.
• Allocation of labor and capital resources for near-term operations are
usually made a few months in advance based on medium-range
forecasts
• forecasting is a key activity that influences many aspects of service
system design, planning, and operations.
Common features of forecasts
• Forecasts are generally wrong —Forecasts are estimates of the mean
of the random variable (demand, price, etc.) and actual outcomes
may be very different.
• Aggregate forecasts are more accurate—Demand forecasts for the
entire product family are more accurate compared to those made for
each member of the product family
• Forecasting errors increase with the forecast horizon—Demand
forecasts for the next month will be more accurate than those for
next year.
• Garbage-In Garbage-Out Principle—Source and accuracy of data used
for forecasting are very important
• Demand forecasts are used by all business functions.
• Given below are a few examples:
• • Accounting: Cost/profit estimates • Finance: Cash flow and
investment decisions • Human Resources: Hiring/recruiting/training •
Marketing: Sales force allocation, pricing, and promotion strategy •
Operations: Production and workforce planning, inventory
management
Forecasting process
• The elements of a good forecast are as follows:
• • Timeliness • Reliability • Accuracy • Regular reviews • Equal chance
of being over and under • Good documentation • Easy to us
• The major steps in the forecasting process are the following:
• Step 1: Determine the purpose of the forecast.
• Step 2: Establish a forecast horizon, namely, how far into the future we
would like to make prediction.
• Step 3: Select the appropriate forecasting technique(s).
• Step 4: Get past data and analyze. Validate the chosen forecasting
technique.
• Step 5: Prepare the final forecast.
• Step 6: Monitor the forecasts regularly and make adjustments when
needed.
• 1. Qualitative or judgmental methods, based on subjective judgments
and past historical data are not needed. They are most useful for new
products or services.
• 2. Quantitative or statistical, require historical data that are accurate
and consistent.
• They assume that past represents the future, namely, history will
tend to repeat itself.
• Most quantitative methods fall in the category of time series analysis
Qualitative forecasting methods
• Based on subjective opinions from company executives and
customers
• no past data or past data are not reliable due to changes in
environment
• Used for Long term forecasts
Qualitative forecasting
• The most commonly used qualitative methods are:
• 1- executive committee consensus: a group of senior executives of the
company from various departments meet, discuss, and arrive at consensus
forecasts for products and services. This is primarily a “top-down” approach
• 2- Delphi method, it uses a panel of experts and facilitator who interacts
directly with the experts and analyze the responses and prepare a statistical
summary which distributed to all panel members for providing the
comments that will be shared again to all members
• Email and online may be used in delphi method,
• The process is continued until a consensus among the experts is reached.
• 3- survey of sales force (“bottom-up” approach):
• the regional sales people, who directly interact with the customers,
are asked to estimate the sales for their regions. They are reviewed at
the managerial level and an aggregate forecast is then developed by
the company
• 4- customer surveys:
• It uses designed customer surveys to determine their needs for
products and services, it is used for brand new product lines or
services
Quantitative forecasting methods
• Time series forecasting
• four different time series patterns with random fluctuations or noise.
Constant level forecasting method
• Last Value Method
• Simple Averaging Method
• Moving Average Method
• Weighted Moving Average Method
• Exponential Smoothing Method
Last value method (Naive Method)
• The next forecast is the last value observed for that variable. In other
words: Under the naive method, the forecast for month 7 = F7 =
Demand for month 6 = 340.
Averaging method
Simple moving average method
Weighted moving average method
Exponential smoothing method
Multi period forecasting constant constant level
Incorporating seasonality in
forecasting
• Seasonal demand: , it has a pattern that repeats after every few
periods. For example, retail sales during Christmas season are usually
higher
• The basic steps to incorporate seasonality in forecasting are as
follows:
• Step 1: Compute the seasonality index for any period given by
• Seasonality Index SI =Average demand during that period/ Overall
average of demand for all periods
• Step 2: Seasonally adjust the actual demands in the time series by
dividing by the seasonality index, to get deseasonalized demand data.
• Step 3: Select an appropriate time series forecasting method.
• Step 4: Apply the forecasting method on the deseasonalized demand
forecast.
• Step 5: Compute the actual forecast by multiplying the
deseasonalized forecast by the seasonality index for that period
• The quarterly sales of laptop computers for 3 years (2014–2016) are
given in Table 2.2. The problem is to determine the forecast for
quarter 1 of year 2017 (period 13).
example
• — Compute the overall average of quarterly demand using the demand
values for 12 quarters
• Overall average =(540+522 + 515 + +  550 + 629 +785)/12 = 7236/12 =
603
• — Compute the quarterly average using the three demand values for
each quarter as follows:
• Quarter 1 average=(540 + 574 + 550)/3 = 555
• Quarter 2 average=(522 + 569 + 550)/3 = 547
• Quarter 3 average=(515 + 616 + 629)/3 = 587
• Quarter 4 average=(674 + 712 + 785)/3 = 724
• Note that the third quarter average is slightly higher primarily due to
“Back-to-School” sales to students. The fourth quarter average is
much higher due to “Christmas Sales.”
• The Seasonality Indices (SI) are then computed as follows:
• SI for quarter 1 = 555/603 = 0.92
• SI for quarter 2 = 547/603 = 0.907
• SI for quarter 3 = 587/603 = 0.973
• SI for quarter 4 = 724/603 = 1.2
• Step 2: Compute the deseasonalized sales data by dividing the
actual sales data (Dt) by its appropriate seasonality index. For example,
deseasonalized sales for 2014 quarter 1 = 540/.92 = 587.
• Step 3: Select any time series forecasting method. For illustration, we
will use the exponential smoothing forecasting method with α = 0.2. For
the initial forecast for quarter 1 of year 2014, we will use 600.
• Step 4: Apply the exponential smoothing method on the
deseasonalized demand to get deseasonalized forecast as shown in
Table 2.3. For example, the deseasonalized forecast for quarter 1 (2017)
is given by
• Step 5: Compute the actual forecast for year 2017, quarter 1 as follows:
Multi-period forecasting with
seasonality
Incorporating trend in forecasting
• The forecasts lag behind the trend (growing or falling) with moving average
and exponential smoothing methods.
• To accommodate trend in forecasting we use the following methods:
• Simple linear trend model:

• Solving the two linear equations, we determine the values of a and b. Then,
the forecast for period t = n + 1 is given by
• Fn+1 = a + b (n+1)
• a and b are constant level and trend
• The main drawback of the simple linear trend model is that it assumes that
Example
• Consider the monthly demand for a product for the past 6 months given in
Table 2.4. The demand data clearly show an increasing trend. The problem
is to determine the forecast for month 7 using the simple linear trend
model.
• SOLUTION
• Table 2.5 gives the necessary computations for the linear regression model.
• Using the last row (sum) in Table 2.5 and substituting them in Equations 2.6
and 2.7, we get the following:
• 1700 = 6a+ 21b
• 6355 = 21 +91b
Holt’s method
• Holt’s method is also known as double exponential smoothing or trend
adjusted exponential smoothing method

• Given the actual demand Dt , forecast Ft and the estimates of level (Lt )
and trend (Tt ) for period t, the forecast for period (t + 1) is given by
• The term (Dt − Dt−1) in Equation 2.11 represents the latest trend
based on the actual values observed for the last two periods,
• while (Ft − Ft−1) is the latest trend based on the last two forecasts.
Once Lt+1 and Tt+1 are determined, using Equations 2.9 and 2.10,
respectively,
• the forecast for period (t + 1) is given by their sum, as in Equation 2.8
Example 2.5 (Holt’s Method)
• Consider again the 6-month time series data on demand given in
Table 2.4. We will apply Holt’s method to determine the forecast for
month 7. In order to get started with Holt’s method, we need the
values of the smoothing constants, α and β, and the initial estimates
for the level and trend for month 1, namely L1 and T1. Let us assume
that α = β = 0.3.
Multi period forecasting with trend
(Using Holt’s method)

where Ln+1 and Tn+1 are the latest estimates of level and trend,
Forecasting error
• The accuracy of the forecast depends on forecast errors, which
measure the differences between the forecasted demands and their
actual (observed) value. We use them to choose the best method of
forecasting, determine smoothing values, α, β , to test for new data
Forecasting errors methods
• Mean absolute deviation (MAD)
• Mean squared error (MSE)
• Standard deviation of forecast errors (STD)
• Bias
• Mean absolute percentage error (MAPE)
• The last two methods are preferred by managers
• Bias has 3 values, zero, negative and positive, zero is best , bias>0,
forecasting is overestimating, bias <0 forecasting is underestimating
example
• Consider the actual demand, its forecast, and the errors for five
periods given in Table 2.9. The forecasts are obtained using
exponential smoothing method with α = 0.1. Compute MAD, MSE,
STD, Bias, and MAPE for measuring the accuracy of the forecasting
method
Remarks
• Large negative value for the Bias indicates that the method is
consistently underestimating the actual demands.
• Large values of MAD, STD, MAPE, and MSE indicate that the method
is not forecasting well.
Selecting the best forecasting
method
• Multiple measures of forecast errors
• Choose a particular method that does well for all measures
• Find the average forecast for the best 2 or 3 methods in case that they
do well in some measures and poorly in others
Monitoring forecast accuracy
• Tracking Signal
• NOTES:
• i. The numerator of Equation 2.20 can be positive, negative, or zero, while the denominator
is always positive. Hence, TSk can be positive, negative, or zero for any k. Note that TS1 is
always equal to +1 or −1.
• ii. Unlike the forecast error defined in Section 2.9, tracking signal is not a single number.
Instead, it is a series of numbers, which can be used to detect changes in the pattern of the
forecast.
• iii. The generally acceptable values of TSk are ±6. When the tracking signals go outside these
limits, the forecaster should be notified to determine the cause of these limit violations.
• iv. Tracking signals outside the limits do not automatically imply that the forecasting method
is not working. Environmental conditions, such as local economy, sales promotions, new
competition, and so on, can cause sudden fluctuations in tracking signals.
Example
• Consider the data given in Table 2.9 (Example 2.11). Compute the
tracking signals and plot, A downward trend of tracking signals points
to forecasting problems ahead.