V T20 LP3 FEB 3 - 7, 2020 WEEK 3/9

IY2593 MANAGERIAL ECONOMICS

CHAPTER OUTLINE CHAPTER 5
1. Regression analysis
2. Limitation of regression analysis
3. The importance of business fore-
casting
4. Prerequisites of a good forecast
5. Forecasting techniques

Demand Estimation

OBJECTIVES
B L E K I N G E T E K N I S K A H Ö G S KO L A

Understand the importance of forecast-

ing in business
Know how to specify and interpret a
regression model
Describe the major forecasting tech-
niques used in business and their
limitations
Explain basic smoothing methods of
forecasting, such as the moving average
and exponential smoothing
IY2593 V T20 LP3 MANAGERIAL ECONOMICS 2

Chapter 5

Data Collection Statistical analyses are only as good

as the accuracy and appropriate-
changes, the variable (Y) chang-
es in the opposite direction
ness of the sample of information » positive coefficient shows that
that is used. as the independent variable (Xn)
changes, the dependent variable
Several sources of data for business (Y) changes in the same direc-
analysis: tion
» buy from data providers (e.g. » The regression coefficients are
ACNielsen, IRI) used to compute the elasticity
» perform a consumer survey for each variable.
» focus groups
» technology: point-of-sale data Statistical evaluation of regres-
sources sion results
t-test: test of statistical signifi-
cance of each estimated coefficient
Regression Analysis Regression analysis: a procedure
commonly used by economists to
(whether the coefficient is signifi-
cantly different from zero).
estimate consumer demand with
bˆ
available data. t
SEbˆ
Two types of regression b̂ = estimated coefficient
» Cross-sectional: analyze several SEb̂ = standard error of estimated
variables for a single period of coefficient
time
» Time series data: analyze a single A zero coefficient means the inde-
variable over multiple periods of pendent variable has no effect on
time the dependent variable. How much
“different” from zero is needed
Regression equation: linear, addi- for the variable to be determined
tive, for example, significant, the t-test answers this
question.
Y = a + b1X1 + b2 X 2 + b3 X 3 » ‘Rule of 2’: if absolute value of t
is greater than 2, estimated co-
Y : dependent variable efficient is significant at the 5%
a : constant value, y-intercept level (for large samples, but for
Xn: independent variables, used to small samples, lookup a t table)
explain Y » if coefficient passes t-test, the
bn: regression coefficients (measure variable has a significant im-
impact of independent variables) pact on demand

Interpreting the regression

results
Coefficients:
» negative coefficient shows that
as the independent variable (Xn)
IY2593 V T20 LP3 MANAGERIAL ECONOMICS 3

R 2 (coefficient of determina-
tion): percentage of variation in the
drop one of the closely related
independent variables from the
Forecasting
variable (Y) accounted for by varia- regression. Techniques
tion in all explanatory variables (Xn)
» R 2 value ranges from 0.0 to 1.0 Autocorrelation problem: Factors in choosing the right fore-
The closer to 1.0, the greater the » Also known as serial correlation, casting technique:
explanatory power of the regression. occurs when the dependent » item to be forecast
variable relates to the Y variable » interaction of the situation with
F-test: measures statistical signif- according to a certain pattern. the forecasting methodolo-
icance of the entire regression as a » Note: possible causes include gy--the value and costs
whole (not each coefficient). omitted variables, or non-line- » amount of historical data avail-
arity; Durbin-Watson statistic is able
Steps for analyzing regression used to identify autocorrelation. » time allowed to prepare forecast
results » Solution: to correct autocorrela-
» check coefficient signs and mag- tion consider transforming the Six forecasting techniques
nitudes data into a different order of 1. expert opinion
» compute elasticity coefficient magnitude or introducing lead- 2. opinion polls and market re-
» determine statistical significance ing or lagging data. search
3. surveys of spending plans
See example in Chapter 5 pp. 150- 4. economic indicators
151. Forecasting 5. projections
6. econometric models
Challenges Approaches to forecasting
» Identification Common subjects of business » Qualitative forecasting is
» Multicollinearity forecasts: based on judgments expressed
» Autocorrelation » gross domestic product (GDP) by individuals or group
» components of GDP, e.g. con- » Quantitative forecasting uti-
Identification problem: sumption expenditure, producer lizes significant amounts of data
» The estimation of demand may durable equipment expenditure and equations
produce biased results due to » industry forecasts, e.g. sales of » Naïve forecasting projects past
simultaneous shifting of supply products across an industry data without explaining future
and demand curves. » sales of a specific product trends
» Solution: use of advanced correc- » Causal (or explanatory) fore-
tion techniques, such as two- A good forecast should: casting attempts to explain the
stage least squares and indirect » be consistent with other parts of functional relationships between
least squares may compensate the business the dependent variable and the
for the bias. » be based on knowledge of the independent variables
relevant past
Multicollinearity problem: » consider the economic and 1. Expert opinion techniques
» Two or more independent political environment as well as » Jury of executive opinion: fore-
variables are highly correlated, changes casts generated by a group of
thus it is difficult to separate the » be timely corporate executives assembled
effect each has on the dependent together
variable. » The Delphi method: a form of
» Solution: a standard remedy is to expert opinion forecasting that
IY2593 V T20 LP3 MANAGERIAL ECONOMICS 4

uses a series of questions and » initial claims for unemployment Economic indicators: draw-
answers to obtain a consensus insurance backs
forecast, where experts do not » manufacturers’ new orders for » Leading indicator index has
meet consumer goods and materials forecast a recession when none
» vendor performance, slower ensued
2. Opinion polls deliveries diffusion index » A change in the index does not
Sample populations are surveyed to » manufacturers’ new orders, non- indicate the precise size of the
determine consumption trends. defense capital goods decline or increase
» may identify changes in trends » building permits, new private » The data are subject to revision
» choice of sample is important housing units in the ensuing months
» questions must be simple and » stock prices, 500 common stocks
clear » money supply, M2 5. Trend projections
» interest rate spread, 10-year A form of naïve forecasting that
2. Market research Treasury bonds minus federal projects trends from past data
Closely related to opinion polling funds without taking into consideration
and will indicate why the consumer » index of consumer expectations reasons for the change
is (or is not) buying, and » Compound growth rate
» who the consumer is Coincident indicators identify trends » Visual time series projections
» how s/he is using the product in current economic activity » Least squares time series projection
» characteristics the consumer » employees on nonagricultural
thinks are most important in the payrolls Compound growth rate
purchasing decision » personal income less transfer Forecasting by projecting the aver-
payments age growth rate of the past into the
3. Surveys of spending plans » industrial production future
Information about ‘macro-type’ » manufacturing and trade sales
data relating to the economy, espe- » provides a relatively simple and
cially: Lagging indicators confirm swings in timely forecast
» consumer intentions past economic activity » appropriate when the variable
» inventories and sales expecta- » average duration of unemploy- to be predicted increases at a
tions ment, weeks constant percentage
» ratio, manufacturing and trade The formula:
4. Economic indicators inventories to sales E
= (1 + i )
n
A barometric method of forecasting » change in labor cost per unit of B
designed to alert business to chang- output, manufacturing (%) E = final value
es in conditions » average prime rate charged by B = beginning value
» Leading, coincident, and lagging banks i = constant growth rate
indicators » commercial and industrial loans n = years in the series
» Composite index: one indicator outstanding
alone may not be very reliable, » ratio, consumer installment Visual time series projections
but a mix of leading indicators credit outstanding to personal Plotting observations on a graph
may be effective income and viewing the shape of the data
» change in consumer price index and any trends.
Leading indicators predict future for services
economic activity
» average hours, manufacturing
IY2593 V T20 LP3 MANAGERIAL ECONOMICS 5

» Quadratic line: Y = a + b(t) + c(t)2 Et +1 = wX t + (1 − w )Et

» Choose one with the best R2
w = weight assigned to an actual
Time series components: Cyclical observation at period t
» isolate cyclical components by
smoothing with moving average 6. Econometric models
Causal or explanatory models of
Time series components: Season- forecasting
ality » Regression analysis
» need to identify and remove » Multiple equation systems
seasonal factors, using moving Endogenous variables: dependent
An example in which the constant averages to isolate those factors variables that may influence other
compound growth rate approach » remove seasonality by dividing dependent variables
would be misleading. data by seasonal factor Exogenous variables: from outside
the system, truly independent varia-
Time series analysis Time series components: Ran- bles
A naïve method of forecasting from dom
past data by using least squares sta- » random factors cannot be pre- Illustration of a model of the
tistical methods to identify trends, dicted and should be ignored relationship between the domestic
cycles, seasonality, and irregular currency and a foreign currency:
movements. Smoothing techniques
Et = a + bI t + cRt + dGt
» easy to calculate » Moving average: The larger the
» does not require much judgment number of observations in the E = Exchange rate of a foreign cur-
or analytical skill average, the greater the smooth- rency in terms of domestic currency
» describes the best possible fit for ing effect. I = Domestic minus foreign infla-
past data » Exponential smoothing: allows for tion rate
» usually reasonably reliable in the the decreasing importance of R = Domestic minus foreign nomi-
short run information in the more distant nal interest rate
Time series data can be represented past. G = Domestic growth rate of GDP
as follows: Moving average a, b, c, d = Regression coefficients
Yt f (Tt , Ct , St , Rt ) Average of actual past results used
to forecast one period ahead.
Yt = actual data value at time t
Tt = trend component at t
Et +1 = ( X t + X t +1 + + X t − N +1 ) / N Summary
Ct = cyclical component at t Et+1 = forecast for next period
St = seasonal component at t Xt, Xt-1 = actual values at their re- » Regression analysis is a primary
Rt = random component at t spective times tool to understand demand.
N = number of observations includ- » Reliable input data and proper
Time series components: Trend ed in the averaging estimation and evaluation are
» To remove trend line, use least needed.
squares method Exponential smoothing » Forecasting is important in many
Allows for decreasing importance organizations. In business, fore-
Possible best-fit line styles: of information in the more distant casting is a necessity.
» Straight line: Y = a + b(t) past, through geometric progres- » This chapter discussed six of the
» Exponential line: Y = abt sion major forecasting technique.