Summary Chapter 5

Demand Estimation and Forecasting

DEMAND ESTIMATION
I. Introduction
This chapter discusses the concept of elasticity, which measures the sensitivity of quantity
demanded to price changes and demand determinants. It highlights the importance of using
available tools and data to forecast demand, as seen in the shift from cell phones to smart
phones and the shift from console-based games to online ones. The chapter presents two
statistical approaches for estimating and forecasting product demand, emphasizing the
importance of obtaining good data, correctly interpreting and evaluating results, and using them
to make informed management decisions.

The Critical Importance of Good Data

Statistical analyses are only as good as the accuracy and appropriateness of the sample
of information used. Official sources like the U.S. Department of Commerce, Labor, and the
World Bank provide reliable data for economic studies. However, data for specific product
categories may be more difficult and costly to obtain. ACNielsen and Information Resources,
Inc. (IRI) are leading providers of market research services, primarily used by large food and
beverage companies.

Market researchers use various methods to understand consumer behavior, including

direct surveys, focus groups, and consumer panel surveys. Direct surveys involve observing
specific consumer responses and body language. Focus groups involve observing specific
consumer groups and competitors. Consumer panel surveys involve participants recording
store purchases and transmitting the data to the research firm via a home phone modem.

Retail stores are increasingly using technology to collect and process large amounts of
data about consumer behavior, primarily through scanning devices like bar code readers and
point-of-sale terminals. RFID technology has made inventory and sales tracking easier, with
Wal-Mart requiring suppliers to attach these devices to their products. The company has built
a data warehouse with a vast amount of data, second only to that of the U.S. government. Big
data, defined as data too large to handle by one computer, consists of volume, velocity, and
variety. Data analysis can range from simple descriptive statistics to sophisticated inferential
multivariate analyses like regression analysis.

II. Introduction to Regression Analysis

Specifying the Regression Equation and Obtaining the Data
To estimate the demand for a good or service, consider all factors that might influence
it. For example, college students' pizza demand could be influenced by tastes, preferences,
income, prices of related goods, future expectations, and number of buyers. However, not all
variables may be included in a demand analysis, and sudden shifts in tastes and preferences
may be challenging to measure.
Regression analysis should consider all variables that impact demand, but the choice of
variables depends on data availability and the cost of generating new data. Two types of data
used in regression analysis are cross-sectional and time series. For example, a survey of 30
randomly selected college campuses in the United States collected information on pizza
consumption, pizza prices, tuition costs, soft drink prices, campus location, and college type.
The economic theory of demand was used to select variables, such as pizza price and soft drink
price. Tuition was used as a proxy variable, and location was included to determine if pizza
demand is affected by available substitutes. Residential colleges were included to moderate
food purchase behavior off-campus. Using these data, we then express the regression equation
to be estimated in the following linear, additive fashion:

where Y = Quantity of pizza demanded (average number of slices per capita per month)
a = Constant value or Y intercept
X1 = Average price of a slice of pizza (in cents)
X2 = Annual tuition (in thousands of dollars)
X3 = Average price of a 12-ounce can of soft drink (in cents)
X4 = Location of campus (1 if located in a concentrated urban area, 0 otherwise)
X5 = Residential college (1 if college is residential, 0 otherwise)
b1, b2, b3, b4, b5 = Coefficients of the X variables measuring the impact of the variables
on the demand for pizza

The dependent variable is the quantity demanded, while the independent variables are
the units of measurement. The unit of measurement for location and residential variables is
different. The regression equation and measurement scheme allow for the estimation of the
slope coefficients of the independent variables and the intercept term using various software
packages for regression analysis.

Estimating and Interpreting the Regression Coefficients

Economists use software like SPSS, SAS, and R for regression analysis. For pizza demand
estimation, Excel's regression function was used, although it lacks a Durbin-Watson statistic.
Excel is suitable for business research and is more accessible than statistical software packages.

The regression results show that the PSlice coefficient has a negative sign, indicating that as
the price of pizza changes, the quantity demanded for pizza will change in the opposite
direction. The positive sign of the tuition coefficient indicates that tuition costs and pizza
demand are directly related, with higher tuition costs causing greater demand for pizza. The
negative sign of the soft drink price confirms the complementarity between soft drinks and
pizza, with college students tending to buy less pizza as the price of a soft drink increases. The
location dummy variable indicates that students in urban areas buy three quarters of a slice less
per month than those in rural areas. The magnitudes of the estimated regression coefficients
are interpreted as a unit change in each explanatory variable.

Recall that the formula for computing point elasticity is

Estimating elasticity requires an estimate of quantity demanded, such as at a nonresidential

college campus in an urban setting with a tuition of $14,000, average pizza and soft drink prices,
and average prices.
Q slice = 29.16 - 0.093·(100) + 0.060·(14) - 0.082·(110) - 0.747·(1) - 3.049·(0)
= 10.99 (or about 11 slices per month)

To compute the point elasticities for each variable assuming the preceding values, we simply
plug in the appropriate numbers into the point-elasticity formula. The partial derivative of Y
with respect to changes in each variable (i.e., ∂Y/ ∂X) is simply the estimated coefficient of
each variable.
The demand for pizza is price inelastic, with some cross-price elasticity between soft drinks
and pizza. Tuition doesn't significantly impact pizza demand due to a low elasticity coefficient.

Statistical Evaluation of the Regression Results

The regression results are based on a survey of colleges across the US, and the statistical
significance of each coefficient is assessed using the t-test, which divides the coefficient by its
standard error. That is :

The t-test is a statistical tool used to determine the impact of a variable on demand. If
the coefficient passes the t-test, it indicates the variable's impact on the entire population of
college campuses. However, statistical analysts set up degrees of uncertainty, with the rule of
2 generally implying a 5% level of significance. The coefficient of determination (R2) is
another important indicator used to evaluate regression results. R2 can be as low as 0 or as high
as 1.0, with closer R2 indicating greater explanatory power. In the pizza regression, R2 = 0.885,
indicating that 88% of the variation in pizza demand can be accounted for by the variation in
pizza price, tuition, soft drink price, and college location.

Analysts often use the adjusted R2 measure to compare equations with different
independent variables. The F-test, often used alongside R2, measures the statistical significance
of the entire regression equation. The procedure for conducting the F-test is similar to the t-test,
with a critical value determined at the.05 or.01 level. The critical F-values depend on the sample
size and number of independent variables in the equation. For the pizza demand equation, the
critical F-value is 36.8, indicating the entire equation is statistically significant at the.01 level.

Review of Key Steps for Analyzing Regression Results

We now review all the key steps discussed so far in the regression analysis of a demand
equation using the following stylized regression equation:

where Q = Quantity of the product demanded

P = Price of the product (in cents)
PR = Price of a related product (in cents)
I = Per capita income (in thousands of dollars)
n = Sample size
R2 = Adjusted multiple coefficient of determination
The steps is :
1. Check Signs and Magnitude
2. Compute Elasticity Coefficients
3. Determine Statistical Significance
Implication of Regression Analysis for Management Decisions
The bottom line of statistical analysis, including regression analysis, is its ability to help
managers make good decisions. In a pizza example, the price of pizza and its complementary
product, the soft drink, are key factors influencing demand. Price decreases lead to decreased
revenue, so managers may not try lowering prices to increase sales. Tuition and location do not
have statistically significant impacts on pizza demand, and managers of national chains may
not need to consider these factors when deciding where to open pizza franchises.

III. Problems in the use of Regression Analysis

The Identification Problem
The identification problem is a significant challenge for regression analysis in
estimating demand for a good or service. For example, if a scatter plot of pizza consumption
data shows a slope upward, it may indicate irrational behavior in pizza consumers. However,
the positive coefficient of the price variable in the demand equation may be a result of
movement in supply and demand over the past 20 years. Advanced estimation techniques, such
as two-stage least squares and indirect least squares, can help deal with samples with
simultaneous shifting of demand and supply. However, if the identification problem is not
recognized and addressed, ordinary least squares methods may result in biased estimates of
regression coefficients.

FORECASTING
I. Introduction
Forecasting is a challenging task, especially in the future, but it is an integral part of our lives.
We make forecasts when buying lottery tickets, betting on horses, or making decisions about
weather or investments. In business, government, and nonprofit institutions, forecasting
becomes even more important as organizations become more complex and change rapidly.
Decision-makers need help in weighing factors and understanding changing relationships to
make decisions with ever-increasing impacts. Forecasting aims to reduce uncertainty in
organizations, but managers must develop realistic expectations and use forecasts as an aid to
decision-making.

II. Subject of Forecasts

Forecasting is a crucial tool for businesses to anticipate future sales and profits, requiring
various categories of forecasts, from macro to individual series.
- Forecasts of gross domestic product—which describe the total production of goods and
services in a country.
- Forecasts of the components of GDP—for example, consumption expenditure,
producer durable equipment expenditure, and residential construction.
- Industry forecasts—for Global Foods, this would represent forecasts of sales of soft
drinks, bottled water, and its other products.
- Forecasts of sales for a specific product—for instance, diet cola

Demand Estimating and Demand Forecasting

This chapter discusses regression analysis and demand estimation, both similar but different in
their purpose. Demand estimating is used by managers to investigate the impact of changes in
independent variables on demand, while forecasting focuses on predicting future sales activity
based on likely assumptions about independent variables. In some cases, forecasting is
achieved without causal factors, predicting future sales solely by projecting past data.

III. Prerequisites of a Good Forecast

 A good forecast must be consistent with other business aspects, based on relevant past
knowledge, and injected with analysts' judgments when conditions have changed significantly.
It must consider the economic and political environment, potential changes, and be timely, as
an accurate forecast that is too late may be worthless. In some cases, experts' opinions may be
used to form the forecast.

IV. Forecasting Techniques

Forecasting methods vary depending on the subject matter and the forecaster. Factors to
consider include the item to be forecast, the situation's interaction with available methods,
available historical data, and the time for forecast preparation. High accuracy requirements
may require more sophisticated methods, which are typically more costly. However, simplicity
in forecasting methods is not necessarily a negative characteristic or a detriment, and it is
advised against discarding simple methods and replacing them with more complex ones.

Qualitative forecasting is based on individual or group judgments and is not based on

historical data. Quantitative forecasting uses significant prior data for prediction and can be
naive or causal. Time series models are used by 72% of companies, while simple models and
exponential smoothing account for 60% and 30%, respectively. Causal forecasting methods are
used by 17%, while qualitative methods are used in 11%. Surveys are the most popular method
for quantitative forecasting, but many companies review baseline forecasts periodically to
overlay judgment on them. Despite this, judgmental methods remain prevalent in many
business organizations.

Expert Opinion
Expert opinion techniques include the jury of executive opinion, which involves a group of
corporate executives predicting trends and changes, and the Delphi Method, developed by
Rand Corporation in the 1950s. These methods involve a sequential series of written questions
and answers, with experts not meeting. The process is time-consuming, but computers and
email have simplified it. A consensus is obtained, and the method has been successful in
predicting scientific breakthroughs, population growth, automation, space programs, and future
weapon systems. However, it has drawbacks such as insufficient reliability, oversensitivity to
ambiguity, difficulty in assessing expertise, and the impossibility of predicting the unexpected.

Opinion Polls and Market Research

Opinion polling is a method of surveying a population to determine future trends, rather than
seeking experts. It can identify changes in trends that may be missed when using quantitative
methods. The choice of the sample population is crucial, and questions should be simple and
clear. Market research is closely related to opinion polling, as it helps estimate market potential
and market share by understanding consumer behavior and preferences.

Surveys of Spending Plans

Surveys of spending plans are similar to opinion polling and market research, but focus on
macro-type data related to the economy. Two well-known surveys are the Survey of Consumers,
conducted monthly by the University of Michigan, and the Consumer Confidence Survey,
published by The Conference Board. The former, initiated in 1946, collects data on personal
finances, business conditions, and buying conditions, while the latter, published since 1967,
focuses on inventories and sales expectations. These surveys help businesses forecast and plan
their spending, as changes in consumer attitudes significantly impact spending.

Economic Indicators
Economic indicators are a crucial tool for forecasting changes in the direction of activity. They
are used to alert businesses to changes in economic conditions and are used widely in
forecasting general economic activity. The success of the indicator approach depends on
identifying historical economic series that correlate with and precede the series to be predicted.
A composite of leading indicators can be used to predict, as it should exhibit a slowing and
actual decrease before overall economic activity turns down and start to rise while the economy
is still experiencing low activity. Economic indicator data are published monthly by The
Conference Board in Business Cycle Indicators. There are three major series: leading,
coincident, and lagging indicators. Leading indicators represent commitments indicating future
economic activity, while coincident indicators identify peaks and troughs. The leading
indicators are considered good forecasters if they can predict recessions or slowing downs
within a certain period of time.

Projections
Trend projections are a naive form of forecasting that project past data into the future without
considering reasons for change. Three techniques are examined: compound growth rate, visual
time series projection, and least squares time series projection. For more frequent data,
smoothing methods like moving-average and least squares time series projection are necessary.
Time Series Analysis
This section discusses time series forecasting, using the method of least squares instead of
visual estimation. This method, which relies on only one independent variable, time, is
considered naive as it does not explain the reasons for changes. Despite its mechanical nature,
time series analysis is easy to calculate, requires minimal analytical skill, provides the best fit,
and is generally reliable in the short run. However, analysts should consider additional
information about changes in underlying forces and fine-tune conclusions based on potential
changes in the series.

Forecasting with Smoothing Techniques

The naive method of using past observations to predict the future involves simple moving
average or exponential smoothing. These techniques work best when there's no strong trend,
infrequent changes, and random fluctuations. However, they can be employed for large-scale
forecasts and estimates involving only one period into the future.

Recent Forecasting Methods and Results

one of the major findings of this survey were as follows:
• The most familiar forecasting techniques were moving averages followed by
exponential smoothing, regression and straight-line projection.
• The familiarity with qualitative methods (such as jury of executive opinion) decreased
from the previous surveys.
• Exponential smoothing appeared to be the most satisfactory method, followed by trend-
line analysis.
• The satisfaction level with the various techniques appears to be declining from previous
surveys.
• Despite the great improvement in tools such as software that are now available to
forecasters, the overall degree of accuracy of the forecasts is substantially lower than
that of the previous two studies.
The authors conclude that the major reasons for the decline in accuracy are:
• Lack of familiarity by users with techniques, and lack of satisfaction with forecast-
ing systems.
• Most forecasting personnel are not held accountable for performance.
 • Sales forecasting performance does not affect compensation of forecasting personnel.
• The authors state that “sales forecasting will not improve until companies commit
the resources . . . with personnel trained in the use of sales forecasting techniques . . .
properly measured and rewarded for performance.

Econometric Models
This chapter discusses quantitative forecasting techniques, specifically causal or explanatory
models. Regression analysis is an explanatory technique, requiring analysts to select
independent variables that influence the dependent variable. While simple projection models
can provide adequate results, using explanatory variables can enhance accuracy and credibility.
However, no regression equation can explain the entire variation of the dependent variable due
to numerous explanatory variables and complex interrelationships. Single-equation regression
models of demand, which use time series variables, are discussed.

V. Global Application : Forecasting Exchange Rates

Chapter 2 discusses the challenges faced by multinational corporations in forecasting
sales, expenses, and cash flows for operations in different countries, which depend on the
exchange rate between foreign and domestic currencies. MNCs often invest in foreign
operating facilities to obtain cash flows. A common method for forecasting exchange rates
is through the forward rate, which is the market consensus on the future spot rate. The
forward rate is an unbiased forecaster of the spot rate, making it a best-guess forecast.
However, it is not an accurate forecast at any one time, but is probably as good an estimator
for the short run. In addition to its lack of accuracy, other shortcomings must be considered:
1. The present exchange rate system does not permit currencies to float freely.
Governments interfere in the exchange rate markets when they consider it to be
of benefit to their country’s economy.
2. Although forward rates can be established for relatively long periods into the
future (in some cases, they can go out as far as 10 years), by far the largest
volume of forward contracts is for 180 days or less.
3. Reliable forward markets exist only for currencies of the leading industrial
economies of the world.

Longer-term exchange rate forecasts often use econometric models. A major problem in
constructing these multiple regression models is in finding appropriate reliable independent
variables. In most cases, the independent variables are stated in terms of differentials between
the domestic and foreign measures, such as:
1. Growth rates of GDP
2. Real interest rates
3. Nominal interest rates
4. Inflation rates
5. Balance of payments