CHAPTER 10

FORECASTING
DEMAND
Azaria Zhafirah Darmawansyah | 20090321596
Dini Reswari | 20090321611
Pembimbing: Prof. Dr. Muhardi, S.E., M.Si.
 INTRODUCTION AND LEARNING
OBJECTIVES
• Chapters 8 and 9 focused on setting capacity levels and schedules to
meet demand  Demand was predetermined, and the emphasis was
on achieving on-time or fast delivery.
• Because demand is variable, it is difficult to predict.  This chapter
addresses this complexity and introduces you to popular techniques
used to forecast demand.
Correct demand estimates are important not only to set appropriate resource levels for timely delivery, but also to
contain costs.

Cost overruns occur when demand is overestimated (excessive capacity and supplies on hand) or underestimated
(overtime and expedited deliveries to respond to shortages)
INTRODUCTION AND LEARNING
OBJECTIVES
INTRODUCTION AND LEARNING
OBJECTIVES

Qualitative
Forecasting
Techniques EFFECTIVE

Quantitative
INTRODUCTION AND LEARNING
OBJECTIVES
after reading this chapter, you will be able to:
1. List the most popular qualitative approaches and mention their advantages
and disadvantages
2. Identify the various components of a time series
3. Develop forecasts based on time series models: moving averages, weighted
moving averages, exponential smoothing, and trend projections
4. Adjust forecasts for trend and seasonality
5. Modulate the stability and responsiveness of a forecast
6. Assess the accuracy of forecasts
7. Develop forecasts based on associative models
BOX 10.1 – DID YOU KNOW?
Continuous Demand Forecasting Improves Operating Margins
Demand forecasting solutions can improve operating margins by decreasing
labor costs and average length of stay. Unfortunately, many systems are only
capable of producing simple forecasts, often with projections of patient census a
few days out. These projections do not help manage staffing costs effectively.
Robust tools should include a variety of forecasting methods taking all relevant
variables into account. Use of historical data on patient volumes, identification
of factors responsible for those volumes, detection of trends and seasonal
patterns, forecast accuracy checks, and simulation of different
inpatient/outpatient mix and staffing scenarios should all be part of predictive
analytics, which, when integrated with a hospital’s workflows, can optimize the
way resources are deployed (Whitehurst, 2015).
BRADLEY PARK HOSPITAL 10.1
Sheila was the vascular surgery administrator. She knew that job
backwards and forwards and had the complete faith of the surgeons
with whom she worked. When Dr. Vaughn, the division chief, asked her
why stents were not always available before an endovascular stenting
procedure, Sheila was perplexed. Because stents cost over $10,000
each, she had kept a close eye on the inventory, making sure that
orders were placed at regular intervals. Yet, sometimes the stents
stayed in inventory for too long. At other times, there were not enough.
“I think I need to review the number of surgeries more closely,” she
said. “It looks like I might be missing something.”
FORECASTING DEMAND
• A demand forecast is a prediction of future demand  never perfect
• The actual demand is rarely equal to that indicated in the forecast.

learn how to select forecasting techniques that decrease the chances and
magnitudes of errors, but you will always need to make allowances for these
errors when using demand forecasts to make decisions (e.g., setting capacity
levels).
Cont..
Aggregate demand forecasts that focus on combined lines of products
or services tend to be more accurate because the forecasting errors for
single lines of services tend to cancel one another when the service
lines are grouped together.
For example, the prediction of the total demand for hospital beds is
generally more accurate than the demand forecast for beds in the
intensive care unit (ICU).
Furthermore, short-term forecasts (a few months) tend to be more
accurate than long-term forecasts because uncertainties increase and
business conditions change over a long time span.
 Qualitative
Forecasting
Techniques EFFECTIVE

Quantitative
Cont…

Qualitative Quantitative
Qualitative approaches rely on Quantitative approaches project
subjective human inputs and historical demand into the future
judgments. They are most (time series models) or use
appropriate when data on prior multiple variables to predict
demand are unavailable, when demand (associative models).
political and economic conditions are
changing, when available data are
outdated or irrelevant, or when new
products and services are introduced
to the market.
Some organizations favor one type of
approach over another, but a combination of
the two is usually most effective
BOX 10.2 – WORDS OF WISDOM

“The key to making a good forecast is not in

limiting yourself to quantitative information”
(Silver, 2012).
QUALITATIVE
There are qualitative techniques that are relevant for healthcare
services:
Jury of
Delphi
Executive
Method
Opinion

Consumer
Market
Surveys
Jury of Executive Opinion
• mostly used when new products and services are considered or when
existing forecasts need to be revised due to unusual circumstances
• The opinions, judgments, and knowledge of high-level executives
from various functions of the organization are summarized to
formulate demand forecasts.
• The main advantages  incorporates experts’ opinions it can yield
fast results.
• One disadvantage is that some executives exert more power than
others  executives’ biases.
Delphi Method
• used for long-term forecasting (e.g., future demand for harvesting
stem cells).
• pools experts’ demand estimates in an iterative fashion.
• Experts, often in different locations, are asked to develop forecasts
independently  Staff aggregate the results to produce a single
forecast  These aggregated, anonymous forecasts are sent back to
the experts who may then revise their original estimates  The
revised estimates are again aggregated and sent to the experts until a
consensus is reached.
Delphi Method
main advantages Disadvantage
the lack of bias resulting from the that the process may be
experts’ anonymity and the exceedingly time-consuming
strength of a forecast built on
consensus
Consumer Market Surveys
• popular tool used to obtain existing and potential healthcare
consumers’ intentions to buy products and services
• help estimate demand and design or redesign processes to
accommodate customers’ needs and wants
• For example, cancer patients’ opinions about their care experience
may make a compelling argument for building a cancer center with
lab, imaging, surgery, therapy, and counseling services under one roof.
Although this method provides a useful input to forecasting demand,
it may produce overly optimistic estimates of potential as opposed to
actual demand.
QUANTITATIVE:
Approaches project historical demand into the future (time series
models) or use multiple variables to predict demand (associative
models)

Time Series Models Associative Models

Time Serie Models
The primary assumption of time series forecasting is that the behavior
or pattern of this historical demand is likely to continue in the future
Another assumption is that future demand can be predicted exclusively
from past demand, thereby ignoring the potential influence of other
variables, such as price
Components of Time Series Model:
Components of Time Series Model:
Trend: a long-term, upward or
downward movement of demand.
Cycle: a pattern that occurs every
Demand trends occur as a result of
several years
changes in demographics,
competition, or societal changes.
• For example, the increasing • the economic recession starting in
population of aging baby boomers December 2007 resulted in fewer
creates a higher demand for visits at private physician practices,
geriatric care services over time. but it fueled demand at
community health centers (Robert
Wood Johnson Foundation, 2009).
Components of Time Series Model:
Irregular movement: a random
variation. It occurs by chance and
Seasonality: a pattern that occurs
cannot be predicted. It pretty much
every several years
guarantees that your forecast will
never be perfect!
• For example, the number of • Why did 120 patients show up at
patients in the emergency the emergency room (ER) today
department (ED) is lowest between when only 97 showed up the day
midnight and 8:00 a.m. The before? Nobody knows.
number of hip fractures peaks
during the winter months.
BRADLEY PARK HOSPITAL 10.2
Sheila pulled the number of stent surgeries performed over the past 3 years. Because Bradley Park
Hospital’s (BPH’s) cardiac unit had a strong reputation, it attracted patients from all over the state.
Clearly, rescheduling surgeries because stents were not available would tarnish the hospital’s image and
threaten its status as a center of excellence. When looking at the spreadsheet, she did not see much
variation. There was a slight increase from 1 year to the other due to BPH’s growing reputation, but she
had already accounted for that when ordering supplies. She then decided to plot the data (Table 10.1
and Figure 10.3). What she saw puzzled her: The number of stent surgeries was lower in April,
November, and December. Although emergency surgeries were unpredictable, elective surgeries were
often scheduled based on patients’ and physicians’ preferences. On a hunch, she checked the surgeons’
schedules. “There it is,” she thought. “Several of our surgeons attend the Annual Meeting of the
American Association for Thoracic Surgery in April. And of course, with all the holidays, November and
December tend to be slow months. Patients prefer to be at home rather than in the hospital during that
time of year. The problem is that I only know the number of surgeries once they have been scheduled.
That’s a short time window. I can’t really be proactive in placing orders …” Sheila a walked to the
finance department. Those people spent their lives working with spreadsheets and making all kinds of
projections. They should be able to help her.
Stability and Responsiveness
• Random variations in demand • However, if demand changes for
can confuse and mislead the a reason, the forecast should
forecaster reflect the change as soon as
• Therefore, a forecast should not possible.
overreact to simple, random • The ability of a forecast to react
fluctuations. quickly to true changes in
• This property is called stability demand is called
responsiveness.
Naïve Approach & Simple Moving Average

Naïve Approach Simple Moving Average

• the naïve approach is simple. It • forecast averages demand over a
merely assumes that demand in the certain number of past periods to
next time period will be equal to the forecast demand in the next period.
demand in the current period. • A 3-month moving average would
• It works best if the random use the average of the 3 previous
fluctuations are very small. If they months to project demand next
are large, the naïve forecast will month; a 7-day moving average
produce erratic forecasts, which would average the demand over the
cannot be used for planning past 7 days to predict tomorrow’s
purposes. demand, and so on.
Weighted Moving Average
• the weighted moving average
includes weights that place more
emphasis on recent demand.
• There is no rule to determine the
weights. Judgment and trial and
error are the best ways to
choose the weights.
• The weights take on values
between 0 and 1, and their sum
must equal 1.
Weighted Moving Average
Exponential Smoothing
• It is similar to a weighted moving • the exponentially smoothed
average technique in the sense forecast is based on the actual
that it assigns (a) a weight demand and exponentially
between 0 and 1 (called the smoothing forecast for the
smoothing constant), α, to the previous period.
actual demand in the previous • The forecasts for previous
period and (b) a weight (1 – α) to periods are assigned a weight of
the exponential smoothing (1 – α).
forecast for the previous period
Exponential Smoothing
The main advantage of the
exponential smoothing technique
over other averaging techniques is
that it uses only two data points—
the actual demand and forecast
for the previous period—rather
than demand values over multiple
periods. This parsimony has
contributed to its popularity in
business.
FORECAST ACCURACY
• Forecast accuracy is determined Because a forecast predicts future
by how closely the forecast demand, we cannot compare it to
matches the actual demand. Any actual demand.
deviation from the actual However, because we have
demand is a forecast error: assumed that historical demand is
F ERROR= ACTUAL DEMAND - FORECAST representative of future demand,
we can formulate “forecasts of the
past” using a particular model and
compare those forecasts to
historical demand.
FORECAST ACCURACY
Mean Absolute Deviation Mean Squared Error Another
the mean absolute deviation is the measure of forecast error is the
average of absolute forecast mean squared error.
errors. Its formula is The MSE captures the average of
the squared deviations between
the forecast and the actual
demand:
FORECAST ACCURACY
Mean Absolute Percentage Error
With both the MAD and MSE, the
size of the deviations depends on
the volume of the item being
forecast. For example, if the
demand is in thousands of units,
the absolute and squared
deviations could be quite large.
The MAPE remedies this problem
by expressing the deviation as a
percentage of the actual demand.
TECHNIQUES FOR TRENDS
Trend-Adjusted Exponential
Smoothing
Remember that an exponentially
smoothed forecast will always lag
behind an upward or downward
trend in the data. To compensate
for this deficiency, it is possible to
adjust the exponentially
smoothed average for the positive
or negative trend.
Linear Trend Projections
• The least-squares method enables us to develop a trend line that
minimizes the sum of the squared deviations between the line and
the actual data points. That line illustrates theupward or downward
relationship between an independent and a dependent variable.
The regression equation
 The Trend Fits The Data Quite Well.
However, it does not
account for seasonal fluctuations. Before we explain how to
incorporate seasonality into our
forecasts, it is important to keep in mind that the least-squares method
is suitable when:

1. The trend movement is linear. If it is curvilinear, another model is

required.
2. Future projections are not made too far into
the future.
For example, stock
• market projections are more likely to be accurate for the next few
months than for
• the next 10 years. Over long periods of time, new patterns may
emerge. In a
• similar vein, the number of periods for which demand is projected
should never
• exceed the number of periods included in the database.
3.The Deviations actual demand and trend
line distributed normally.
• The deviations between actual demand and the computed trend line
are assumed to be normally distributed with most of the demand
points close to the line, and only a few points scattered further out
TECHNIQUES FOR SEASONALITY

• seasonality does not necessarily refer to a calendar season. It is any

recurring
• variation that may happen daily, weekly, monthly, and so on. In a time
series, seasonality is expressed as the amount of deviation
between the actual demand values and the averagemvalue of the
series. If demand is fairly constant, the average value is simply the
mean of thedata. When a trend is present, the average value is the
trend value.
Additive And Multiplicative.

Additive model, a certain amount, let us say 20 units, is added to and

subtracted from the trend value or average to reflect the upward and
downward movements of the seasonal pattern.

Multiplicative model, seasonality is a percentage of the average value

or trend. Because the multiplicative model is predominant in business,
we will focus on it exclusively in this chapter. Seasonal
percentages are known as seasonal indices.
Adjusting Forecast for Seasonality
Decomposing a Time Series Using Least-
Squares Regression
• When there is a trend, as is the case in our data set, it is common
to first decompose the time series into its components and then
forecast future values of each component (Chase,Aquilano, & Jacobs,
2006).
• Again, we follow a series of steps to quantify trend and
seasonality and then develop forecasts (Table 10.11)
ASSOCIATIVE MODELS

So far, we have assumed that demand fluctuated as a function of

time only. However, demand may be influenced by a myriad of other
variables such as price, geographic distance from target market, age
distribution of patient population, and so on. If the relationships
between these variables and demand are linear, we can use the
multiple linear regression technique to show the effect of predictor
(or independent) variables on a dependent variable (demand);
 Forecasting Patient Volume in the Emergency Department

• Demand forecasting is important for the allocation of resources in the ED. A study
• explored and evaluated several quantitative forecasting methods to predict daily ED
• patient volumes: seasonal autoregressive integrated moving average, exponential
• smoothing, time series regression with and without climatic variables, and artificial
• neural networks. The benchmark was a multiple linear regression model based on
• calendar variables and holidays. The analysis confirmed annual and weekly seasonal
• patterns in ED visits. Based on the MAPE, the various forecasting methods proved to
• be only marginally better than the simple benchmark model (Jones et al., 2008).
THANKYOU