DIRECT TESTIMONY
OF
J. EDWARD SMliH
ON BEHALF OF
THE OFFICE OF THE CONSUMER
MAY 22,2000
ADVOCATE
TABLE OF CONTENTS
Page
I.
STATEMENT
II.
PURPOSE
III.
INTRODUCTION:
VOLUME VARIABILITY
OF MAIL PROCESSING
COSTS .. . .. . . .. . .. . .. . .. . .. .. .. . .. . .. . . .. . . .. . .. . . .. . .. . .. . .. . .. . .. . .. . .. . .. .. .. . .. .. . .. . . .. . .. . .. . .. . . .. . . .. . . . .. . . 5
IV.
V.
OF QUALIFICATIONS
..,..............._,,,..........,,.,.............,....,,.,.,.,.,
AND SCOPE OF TESTIMONY
. .. . .. . ...__._.........................................
1
2
A.
Both Dr. Bradley and Dr. Bozzo Have Presented Analyses of
Segment 3 Mail Processing Costs. . . . .. . .. . .. . .. .. .. .. .. .. . .. .. . .. . . . . .. . .. . . .. . . . .. . . 5
B.
The Estimated Volume Variabilities Presented by Dr. Bradley
and Dr. Bozzo Differ, but the Variabilities
are in General
Substantially Less than 100 Percent . .. .. .. . .. .. . . . . . .. . . . . . . . . . . . . . 6
C.
The Commission Has Identified Criteria and Standards that Can
Serve as a Basis for the Evaluation of an Econometric Study. . . .._._...... 8
DR. BRADLEY’S
STUDY . .. . . .. . . .. . . .. . .. . .. . .. . .. . . .. . .. . .. . .. .. .. . .. . .. .. . .. . . .. . . .. . .. . .. . .. . . .. . . .. . 10
A.
A Review of Dr. Bradley’s Study Highlights Previous and
Potential
Problems Associated
with the Measurement
of
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. . . .. . . . .. 10
Volume Variability.
B.
Dr. Bradley’s Study Was Criticized as Being of a Short-Run
Nature Due to the Use of 4-Week Accounting Periods Coupled
with the Lack of Consideration of Capital and Investment. __,................ 12
C.
The Database for Dr. Bradley’s Study Was Unreliable . .. . .. . . .. . .. . .. . . .. . . .. . . 13
D.
Dr. Bradley’s Fixed Effects Approach Was Criticized by the
Commission . .. .. . . .. . . .. . . . .. . .. . . .. . .. .. . .. . . . . .. .. . .. . .. .. .. . .. .. . . .. . . .. . .. . . . . . . .. . . . .. . . . 14
E.
Dr. Bradley Extrapolated His Econometric Results to a Number
of Other Activities. . . .. . . .. . . .. . . . .. . .. . .. . . . . .. . .. . .. . .. .. . .. . . . .. . . .. . . . . . .. . . . . . . . . 16
DR. BOZZO’S VOLUME VARIABILITY
A.
STUDY .. . .. .. . .. .. .. . .. . . . .. . .. . .. . .. . . .. . . . .. . . . 18
Dr. Bozzo’s Revisions of 10 of the 25 Mail Processing
Modeled by Dr. Bradley Continue to Have Deficiencies
Activities
. . . .. .. . .. . . . . . . 18
Docket No. R2000-1
VI.
OCA-T4
B.
Dr. Bozzo’s Study Needs Substantial
C.
The Commission
Should Recommend
Establishment
of a
Working Group to Resolve the Mail Processing Issues. . . . . . . . . 21
Work for Completion.
DR.
BOZZO’S
METHODOLOGY
IS
EVALUATED
UNDER
ESTABLISHED CRITERIA .,,........,.......,.......,..................................................
A.
Criterion 1: A Study Should Include the Development and Use
of an Adequate Database, Appropriately Verified and Complete..........
1.
2.
3.
4.
5.
6.
B.
19
22
23
The database was not adequately examined and verified
for accuracy. .. .. .. .. .. . .. .. . . . .. . . .. . .. .. . .. . .. .. . . .. . .. . .. . . .. .. .. . . .. . .. . . . 23
Changes in postal investment subjected the investment
data trends to changes during 1994-96; previous data
may be unrepresentative
of operating conditions in the
forthcoming rate effective time period. . . .._._.._.._..__........,,..,.,,..., 26
The continued use of the manual ratio is undesirable. _....__._.._._.
29
The QICAP variable has not been demonstrated
as
appropriate. .. . . . .. ..__....................................................,,..,,,,,,.,
30
(a)
The presentation of the variable QICAP, used to
measure
capital usage
at each facility, is
inadequate.
.. .. . .. . . . . . . .. . . . . . . . .. . .. . . .. . .. . . . . 30
The
variable
QICAP
appears to be deficient from a
(b)
computational viewpoint. . . . .. .. . . . . .. . .. .. .. . .. .. . .. . . . . 31
QICAP is available on a facility basis, not on an
Cc)
activity level basis; this may lead to meaningless
results when including capital investment in the
32
study.
Some of Dr. Bozzo’s computations
illustrate the
(4
nature
of
the
variable
QICAP.
34
dubious
(e)
The approach to equipment depreciation and the
failure to consider
maintenance
efforts also
renders QICAP meaningless. . . . . . .. .. .. .. . .. . .. .. .. . .. .. . .. . .. . . . 34
Capacity
utilization
is another
potentially
important
variable missing from Dr. Bozzo’s database . . . .. .. . .. .. . .. .. . . .. .. .. . 36
In conclusion, there are serious data problems underlying
the foundation of the study. __...........................................,.......,..
36
Criterion 2: Models Should Be Derived from the Appropriate
Economic Theory and Should Fit Correctly Wrthin any System
that Applies Them. . . .. . .. .. . .. .. .. . .. .. . .. . . .. . . .. . .. . . .. . .. . ... .. . .. . . . .. . .. .. .. .. . .. . .. . . . . 37
1.
The economic assumptions and theory for the current
study are not clear; in many cases they appear to be
wrong . . .. . .. . . .. . .. . .. .. . .. .. .. . .. .. . .. . . .. . . .. . . .. . .. . .. . .. . . .. . .. . .. . .. . .. .. .. .. . .. .. . .. . . .. . 37
OCA-T-4
Docket No. R2000-1
2.
3.
4.
5.
6.
7.
8.
9.
C.
Dr. Bozzo and Dr. Bradley do not agree on the type of
function being estimated;
much improvement
in the
presentation of the labor demand function is needed. . . . . .. . . . . . 38
Dr. Bozzo’s study is short run. The proper approach for
examining postal facilities is on a longer-run basis as
related to major investment plans and movement along
the facility expansion path. ..._...... .. .. . .. . .. . . .. . . .. .. . .. . . .. . .. . .. . . . . . .. . . .. 40
Dr. Bozzo addressed Dr. Bradley’s omission of variables
Dr. Bozzo considers
in the regression
equations.
additional variables, but the consideration is still deficient. . . .. . .. . 43
The newly presented information about networks needs
to be fully incorporated in the analysis. . . . . .. . . .. . .. .._...................... 44
Dr. Bozzo estimates mail processing
activities (e.g.,
manual
processing,
OCR,
BCS) as independent
activities; based on witness Degen’s comments
on
networks and facilities, serious consideration needs to be
given to the simultaneous modeling of activities. . . . . . . . . . . . . . . 45
Dr. Bozzo’s treatment of homotheticity appears to lead to
incorrect conclusions. ,.,....,,_...._......: _,...,.,..,.,..,..,........................
47
Dr. Bozzo has raised some important issues about cost
minimization; resolution of the issues may affect the cost
segment 3 analysis ,..,,._...,,._,.._.......,...,.......,..,....,.......................
47
In conclusion, the theory underlying Dr. Bozzo’s model
has not been shown to be derived from the appropriate
economic theory. ,.,.,..,.,_.,_._........,.,...,,........................................
51
Criterion 3: The Study Should Include a Discussion of the
Modeling
Approach
and How It Is Consistent
with the
Underlying Data. .. . . . .. . . .. . .. . .. .._............................................,..,...............
1.
2.
3.
52
Another problem associated with Dr. Bozzo’s work is his
modeling of capital (as opposed to the accuracy of the
QICAP variable itself), The use of capital affects future
... ..
.
. 52
Postal Service costs. . . .. . .
(a)
Dr. Hsiao has useful guidance on the modeling of
capital and investment in economic models. . . . .. . . .. . . . .. .. 52
Dr. Bozzo has not modeled capital in a way that
(b)
would meet the criteria outlined by Drs. Intriligator,
Bodkin, and Hsiao. .. . .. . .. .. .. . .. .. . .. . . .. . . .. . .. . . .. . .. . .. . .. . .. . . . .. . . .. . . 54
Witness Degen’s testimony is a major input to the
understanding and modeling of postal mail processing
55
Witness
Degen’s
testimony
is consistent
with the
application of intuition and common sense that indicates
the volume variability for mail processing approaches
100 percent. ,......,.......,.,..,......,,.....,,...........,....,..........................
66
Docket No. RZOOO-1
OCA-T-4
4.
In terms of identifying a major factor driving cost, intuition
appears reasonable ,.......................................,,..,................,.,.,..
68
In conclusion, Dr. Bozzo’s choice of econometric model
is inconsistent with the economic modeling of the postal
process . . .. .. .. . .. .. . .. . . .. . . .. . . .. .. . .. . .. . .. . .. . .. . .. .. . .. .. .. . .. .. . .. . . .. . .. . .. . . .. . . .. . .. . 68
5.
D.
Criterion 4: A Correct Estimation Procedure which Is Suitable to
the Estimation Needs at Hand Should Be Used. .. .. . . .. .. . . . .. .. . . . . . . . . 69
1.
2.
E.
VII.
The “between” model is currently the “least bad” model
available. .. .. .. . .. . .. .. . . .. . . .. . .. . . .. . .. .. . . .. . .. . .. .. . .. . .. .. .. . .. .. . . .. . . .. .. . . .. . . .. . . .. . 69
In conclusion, Dr. Bouo
has not adopted a correct
estimation procedure. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. .. .. . .. .. . .. . . .. . .. . .. . .. . . . .. . .. . . 70
Criterion 5: Results for Econometric Equations and Alternative
Econometric
Analyses Should Include a Discussion of the
Information
for the
Values,
Signs, and Other Relevant
Variables. . . .. . .. . .. . .. .. .. . .. .. . . .. . . .. . . .. .. . . .. . .. . .. . . .. .. . .. . .. . .. .. .. . .. .. . .. . . .. . .. . .. . . .. . . .. . . . 71
CONCLUSIONS
. . .. . .. . .. . .. .. . .. .. . .. . .. . . .. . . .. . .. . . .. .. . .. . . .. . .. . .. . .. .. . .. .. . . . . . .. .. . . . .._...... 72
A.
Variabilities Were Traditionally Assumed To Be 100 Percent.
The First Study, Performed by Dr. Bradley, Was Seriously
Deficient. . .. . .. . .. . .. . .. .. .. . .. .. . . .. . . .. . . .. . .. . . .. .. . . .. . .. . .. . .. . .. .. .. .. . .. . .. . .. . . .. . .. . .. . . . .. . . .. 72
B.
Dr. Bozzo’s Study Is also Seriously
C.
The Work that Has Been Reviewed Represents the Latest Part
of a Major Modeling Effort. . .. . . .. . . .. . .. . .. . .. . . .. .. . .. . .. . .. .. .. . .. .. . . .. . .. . .. . . .. . . .. . . .. . . 75
Deficient . .
..__.._........................ 73
TABLES
Table 1: Mail Processing
Activity.. ..............................................................................
.......................................................
12
988-l 999 ......................................................
27
Mail Volume and Segment 3 Hours.. ..........................................................
29
of Dr. Bradley’s Variabilities
Table 2:
Summary
Table 3:
Postal Service Investment--l
Table 4:
Table 5: Variabilities--Bradley,
Table 6:
8
OLS Summary
Fixed Effects, Between,
Pooled, and Random.. ..... .65
by Selected Activities.. .......................................................
- iv -
67
UNITED STATES OF AMERICA
Before The
POSTAL RATE COMMISSION
WASHINGTON. D.C. 20268-0001
Postal Rate and Fee Changes, 2000
Docket No. R2000-1
1
DIRECT TESTIMONY
OF
J. EDWARD SMITH
1
I.
STATEMENT
OF QUALIFICATIONS
My name is J. Edward Smith, and I am an econometrician
2
Advocate
of the Postal Rate Commission.
with the Office of the
3
Consumer
4
variety of economic
5
positions.
6
economic
7
planning;
8
are an A.B. from Hamilton College, and an M.S. and Ph.D. from Purdue University.
9
have testified
assignments
My experience
analysis
in industrial,
has focused
related to forecasting,
and rates, prices, marketing,
approximately
before the Postal Rate Commission
11
R97-1.
academic,
consulting,
on the modeling
project
analysis,
and planning
20 times before
10
I have previously
analysis.
regulatory
on mail processing
worked
in a
and governmental
of costs
production
and revenues;
and strategic
My economics
commissions,
volume variability
degrees
I
most recently
in Docket No.
OCA-T-4
Docket No. R2000-1
1
II.
2
PURPOSE
AND SCOPE OF TESTIMONY
The purpose
of my testimony
3
segment
3 mail processing
4
work was a continuation
5
costs variability
6
percentage
7
‘measured the variability
a
the volume of mail, as measured
9
fed (TPF).
costs presented
the volume variability
by Dr. A. Thomas
of Dr. Michael D. Bradley’s pioneering
presented
in Docket
of cost, measured
the
Dr. Bozzo’s
variability
measures
change in volume.
has
assumed
variability
is 100 percent,
12
activities
and found
13
variability
is an important
14
$17 Billion, and the variabilities
15
the activity are used to yield a measure
16
attributable
17
of the costs to various rates, classes and categories.
Dr. Bozzo measured
ranging
that
mail
variabilities
in
from 52 percent
volume
for 10 mail processing
to 95 percent.
issue, for segment 3 mail processing
Volume
costs are in excess of
applied to the various cost pool costs associated
of attributable
costs.
requiring that the Commission
recommend
Dr. Bozzo’s testimony appears in this docket in USPS-T-15, Docket No. R2000-1.
Witness Bradley’s testimony appeared in Docket No. R97-1 as USPS-T-14.
-2-
with
Costs that are not
19
2
Dr. Bozzo
processing
18
1
the
in hours worked, with respect to changes
11
become institutional,
for
in terms of total pieces handled (TPH) or total pieces
Commission
variabilities
Bozzo.’
analysis
work on mail processing
Volume
No, R97-1.’
change in cost with respect to the percentage
Traditionally
10
is to evaluate
assignment
OCA-T-4
Docket No. R2000-1
1
My evaluation
2
following
3
1. A study
4
5
evaluation
appropriately
include
by the Commission
the development
the study meets the
in Docket No. R97-I?
and use of an adequate
database,
verified and complete.
2. The study should
consistent
7
3. An adequate
8
4. A correct
include
a discussion
with the underlying
of the modeling
and how it is
data.
model and analysis of functional properties
estimation
approach
procedure
is necessary.
that is suitable to the estimation
needs
at hand
should be used.
9
5. Results
for econometric
11
include a full explanation
12
variables.
13
The Commission
econometric
15
1. First, the Commission
16
equations
and alternative
analyses
should
of the values, signs, and other relevant information
for the
has also indicated
14
econometric
some of the procedures
by which it reviews
work:
reviews
the econometric
research
using
the criteria
for
evaluation.
17
2. Second, the Commission
18
3. Finally, the Commission
reviews the statistical properties of the estimates.
tries to identify a preferred model to find a result that it can
safely rely upon: a result that is stable and robust.
19
In considering
20
21
is based on whether
study
criteria mentioned
should
6
10
of Dr. Bouo’s
purposes
3
of providing
Dr. Bozzo’s
a background
study
I will first review
and context evaluation).
Dr. Bradley’s
(for
I will then discuss the
Docket No. R97-1, Opinion and Recommended Decision, Volume 2, Appendix F et 1.
-3-
study
OCA-T-4
Docket No. R2000-1
1
degree to which Dr. Bozzo’s study meets the evaluation
2
meet the criteria
cited, the Commission
3
apply
volume
4
alternatives,
5
traditional
variabilities
criteria.
If the research
fails to
may decline to accept the conclusions
or apply
the
best
of several
and
unsatisfactory
pending further analysis,
My analysis
of the database
issue focuses
6
adequacy
of the variables.
7
economic
issues
8
include a variety
9
effects model and possible alternatives.
I address the modeling
as impacting
the modeling
of econometric
models.
I discuss
Estimation
Although
11
Bradley’s study, many of the problems associated
12
found in the revised study.
I also comment
13
concluded
could
14
implementation
Dr. Bouo’s
Dr. Bozzo’s’study
choice
-A-
is a follow-on
can
of the fixed
criteria could
work to Dr.
with the original study continue to be
on how the estimation
be satisfactory
of a working group.
and the
procedures
I comment on how the evaluation
be reviewed by the Commission.
that
process
issues by focusing on the theoretical
process.
10
in a way
on the scrubbing
process
to all participants
could be
through
the
Docket No. R2000-1
1
III.
OCA-T-4
INTRODUCTION:
2
3
A.
4
Volume
VOLUME
VARIABILITY
OF MAIL PROCESSING
Both Dr. Bradley and Dr. Bozzo Have Presented
Mail Processina Costs.
variability
for mail processing
is defined
5
cost that results from a percentage
6
non-volume
7
of person hours of segment
8
terms of total pieces handled
9
pieces fed (TPF) or in some cases total pieces handled
10
estimated
factors constant.
Postal
Service
and distribution
Analyses
of Segment
as the percentage
change
3
in
change in volume, holding delivery points and other
Both Dr. Bradley and Dr. Bozzo measured
3 mail processing
effort,
operates
over
12
processing
13
and delivery.
14
generated,
15
subsequent
16
No. R97-1, Dr. Bradley modeled 25 mail processing
17
mail processing
18
Dr. Bozzo has limited his updated
19
the MODS
volume in terms of total
(TPH).
testimony
38,000
plants,
where
The econometrically
are presented
offices,
stations,
centers providing for mail collection,
The mail processing
costs in terms
Dr. Bradley measured volume in
(TPH), and Dr. Bozzo measured
variabilities of Drs. Bradley’s and Bouo’s
The
11
COSTS
branches,
processing
the segment
in Table I.
and
and sorting,
3 labor costs are
prepare the mail, sort the mail to three or five digits, and dispatch the mail to
destinations
for additional
plants (denoted
operations.
sorting or distribution.
In his testimony in Docket
and handling activities at the major
as MODS facilities) and at Bulk Mail Centers (BMCs).”
study to the analysis of ten mail sorting activities in
As was well documented
in Docket
No. R97-1, there was
MODS oftices perform the various sorting activities and report costs and volumes through the
Management Operating Data System; non-MODS offices tend to be smziller, perform the same types of
functions as do MODS offices, but do not report through the Management Operating Data Systems.
There are over 300 MODS ofices. The number of non-MODS offices is substantially larger. The 21 Bulk
Mail Centers (BMCs) process packages and report their data through the Productivity Information
Reporting System (PIRS).
4
-5-
Docket No. R2000-1
1
significant
2
with data scrubs
3
estimation
4
and the relationship
OCA-T-4
disagreement
and data checking,
approach.
methodology,
disagreement
There was also concern
of the econometric
Mail processing
5
with to Dr. Bradley’s
including serious problems
over the use of the fixed effects
about the lack of explanatory
variables
model to economic theory.
costs comprise
a significant
portion of Postal
Service
costs.
6
Total costs in the Base Year were $59.6 Billion, with segment 3 costs at $17.6 Billion.’
7
According
8
MODS
9
Bradley’s testimony
to witness
offices,
Van-Ty-Smith,
the segment
$0.8 Billion in BMCs, and $4.4 Billion in non-MODS
presented
the first comprehensive
10
his testimony,
11
for segment
12
variability was based on analysts’ judgments
13
He testified
14
previously
15
16
17
B.
18
The estimated
19
Table 1 are generally
5
6
I
3 costs consist of $12.5 Billion in
analysis of volume variability.
Dr. Bozzo traced the history of the assumption
3 costs.
that
limited the econometric
In
of 100 percent volume-
by a task force formed in the late 1960’s.’
computational,
analyses
Dr.
of 100% volume variability
He stated that the era of the assumption
methodological,
facilities.’
and
theoretical
constraints
had
of volume variability
The Estimated Volume Variabilities Presented by Dr. Bradley and Dr.
Bozzo Differ, but the Variabilities are in General Substantially Less than
100 Percent.
volume variabilities
presented
by Dr. Bradley and Dr. Bozzo in
less than 100 percent. 8 The variabilities
are subsequently
used
Direct testimony of Karen Meehan, USPS-T-l 1, Exhibit 1lA at 2 and 9.
Direct testimony of Eliane Van-Ty-Smith. USPS-T-17 at 24-25.
USPS-T-15 at 4, lines 7-18.
8
The discussion is limited to consideration of only those activities for which Dr. Bozzo presented
estimated variabilities. In UPS/USPS-T15-9, Dr. Bozzo indicated that he had omitted 24 observations
(continued on next page)
-6-
OCA-T-4
Docket No. R2000-1
1
My evaluation
2
following
3
1. A study
include
2. The study should
consistent
7
3. An adequate
8
4. A correct
the development
in Docket No. R97-1:’
and use of an adequate
database,
include
a discussion
with the underlying
of the modeling
and how it is
data.
model and analysis of functional
estimation
approach
procedure
properties is necessary.
that is suitable to the estimation
needs
at hand
should be used.
9
5. Results
for econometric
11
include a full explanation
12
variables.
13
The Commission
equations
and alternative
analyses
should
of the values, signs, and other relevant information
for the
has also indicated
14
econometric
15
1. First, the Commission
econometric
some of the procedures
by which it reviews
work:
reviews
the econometric
research
using the criteria
for
evaluation.
16
17
2. Second, the Commission
18
3. Finally, the Commission
reviews the statistical properties of the estimates.
tries to identify a preferred model to find a result that it can
safely rely upon: a result that is stable and robust.
19
In considering
20
21
by the Commission
the study meets the
verified and complete.
6
10
study is based on whether
criteria mentioned
should
appropriately
4
5
evaluation
of Dr. Bouo’s
purposes
3
of providing
Dr. Bouo’s
a background
study
I will first review
and context evaluation).
Docket No. R97-I, Opinion and Recommended
-3-
Dr. Bradley’s
study
(for
I will then discuss the
Decision, Volume 2, Appendix F at 1.
Docket No. R2000-1
OCA-T-4
1
degree to which Dr. Bozzo’s study meets the evaluation
2
meet the criteria
cited, the Commission
3
apply
volume
4
alternatives,
traditional
variabilities
criteria.
may decline
or apply
If the research
fails to
to accept the conclusions
the
best
of several
and
unsatisfactory
pending further analysis.
5
My analysis
6
adequacy of the variables,
7
economic
a
include a variety
9
effects model and possible alternatives.
issues
of the database
issue focuses
the modeling
of econometric
process.
I discuss
models.
Estimation
and the
Dr. Bouo’s
procedures
choice
I comment on how the evaluation
10
be reviewed by the Commission,
11
Bradley’s study, many of the problems associated
12
found in the revised study.
I also comment
13
concluded
could
14
implementation
that
process
I address the modeling issues by focusing on the theoretical
as impacting
in a way
on the scrubbing
Although
of the fixed
criteria could
Dr. Bozzo’s study is a follow-on
-4-
work to Dr.
with the original study continue to be
on how the estimation
be satisfactory
of a working group.
can
process
to all participants
could be
through
the
Docket No. RZOOO-1
1
III.
OCA-T-4
INTRODUCTION:
2
3
A.
4
Volume
VOLUME
VARIABILR-Y
Both Dr. Bradley and Dr. Bouo
Mail Processina Costs.
variability
for mail processing
OF MAIL PROCESSING
Have Presented
is defined
5
cost that results from a percentage
6
non-volume
7
of person hours of segment
a
terms of total pieces handled
9
pieces fed (TPF) or in some cases total pieces handled
10
estimated
11
factors constant.
Postal
Service
and distribution
Analyses
of Segment 3
as the percentage
change
in
change in volume, holding delivery points and other
Both Dr. Bradley and Dr. Bouo
3 mail processing
effort.
measured
operates
over
12
processing
13
and delivery.
14
generated,
15
subsequent
16
No. R97-1, Dr. Bradley modeled 25 mail processing
17
mail processing
ia
Dr. Bozzo has limited his updated
19
the MODS
volume in terms of total
(TPH).
testimony
36,000
plants,
where
The econometrically
are presented
offices,
stations,
centers providing for mail collection,
The mail processing
costs in terms
Dr. Bradley measured volume in
(TPH), and Dr. Bozzo measured
variabilities of Drs. Bradley’s and Bouo’s
The
COSTS
branches,
processing
the segment
in Table I.
and
and sorting,
3 labor costs are
prepare the mail, sort the mail to three or five digits, and dispatch the mail to
destinations
for additional
plants (denoted
operations.
sorting or distribution.
In his testimony in Docket
and handling activities at the major
as MODS facilities) and at Bulk Mail Centers (BMCS).~
study to the analysis of ten mail sorting activities in
As was well documented
4
in Docket
No. R97-1, there was
MODS offices perform the various sorting activities and report costs and volumes through the
Management Operating Data System; non-MODS offices tend to be smaller, perform the same types of
functions as do MODS offices, but do not report through the Management Operating Data Systems.
There are over 300 MODS offices. The number of non-MODS offices is substantially larger. The 21 Bulk
Mail Centers (BMCs) process packages and report their data through the Productivity Information
Reporting System (PIRS).
-5-
Docket No. R2000-1
1
significant
2
with data scrubs
3
estimation
4
and the relationship
5
OCA-T-4
disagreement
with to Dr. Bradley’s
and data checking,
approach.
methodology,
disagreement
There was also concern
of the econometric
Mail processing
including serious problems
over the use of the fixed effects
about the lack of explanatory
variables
model to economic theory.
costs comprise
a significant
portion of Postal
Service
costs.
6
Total costs in the Base Year were $59.6 Billion, with segment 3 costs at $17.6 Billion.’
7
According
a
MODS offices,
9
Bradley’s testimony
to witness
Van-Ty-Smith,
the segment
3 costs consist of $12.5 Billion in
$0.6 Billion in BMCs, and $4.4 Billion in non-MODS
presented the first comprehensive
10
his testimony,
11
for segment
12
variability was based on analysts’ judgments
13
He testified
14
previously
15
16
17
B.
ia
The estimated
19
Table 1 are generally
analysis of volume variability.
Dr. Bozzo traced the history of the assumption
3 costs.
that
limited the econometric
In
of 100 percent volume-
by a task force formed in the late 1960’s.’
computational,
analyses
Dr.
of 100% volume variability
He stated that the era of the assumption
methodological,
facilities.6
and
theoretical
constraints
had
of volume variability.
The Estimated Volume Variabilities Presented by Dr. Bradley and Dr.
Bozzo Differ, but the Variabilities are in General Substantially Less than
100 Percent.
volume variabilities
presented
less than 100 percent.’
by Dr. Bradley and Dr. Bozzo in
The variabilities
are subsequently
5
Direct testimony of Karen Meehan, USPS-T-l 1, Exhibit 11A at 2 and 8.
6
Direct testimony of Eliane Van-Ty-Smith, USPS-T-17 at 24-25.
7
USPS-T-15 at 4, lines 7-18.
used
B
The discussion is limited to consideration of only those activities for which Dr. Bozzo presented
estimated variabilities. In UPS/USPS-T15-9, Dr. Bozzo indicated that he had omitted 24 observations
(continued on next page)
-6-
Docket No. R2000-1
OCA-T-4
1
by USPS Witness Van-Ty-Smith
2
Volume-Variable
3
variabilities
4
Bradley would lead to the conclusion
5
In comparison,
6
conclusion
7
volume
a
attribution
proposal
9
institutional
costs by a similar amount.
Cost.
estimated
in conjunction
with Pool Total Cost to compute
Of the segment 3 Total Pool Cost of $5.4 Billion relevant to the
by Dr. Bouo,
the application
of the variabilities
the use of the variabilities
then $5.4 Billion would
would
10
is of such a magnitude
11
the Commission.
reduce
developed
by Dr.
that $4.4 Billion of cost would be volume variable.
developed
that $4.1 Billion would be volume variable.
variable,
Pool
be directly
attributable
costs
by Dr. Bouo
would lead to the
If the costs were 100 percent
assigned.9
by $1.3
Thus Dr. Bozzo’s
Billion and
increase
This transfer of costs between accounting
that it will most certainly
influence
the rates recommended
pools
by
from the data set and reran the estimation of variabilities. However, the changes to the results were very
minimal. Since the results were not statistically significant, he did not subsequently refile Appendix E.
Accordingly, because the changes are de minimis and since the original numbers are clearly set forth in
his testimony and can be considered statistically accurate, I am working with his written testimony as filed
and adopted by him. None of my comments would change based on the information he has presented.
9
USPS-T-17, Docket No. R2000-1 at 24, (Van-Ty-Smith).
-7-
Docket No. R2000-1
OCA-T-4
Table 1
Mail Processing Activity
Variabilities
Variabilities
Total Cost
Dr.BCS Sorting
OCR Sorting
FSM Sorting
LSM
SPBS Non Priority
SPBS Priority
Manual Flats
Manual Letters
Manual Parcels
Manl. Priority Mail Srtg
Cancel. And Mail Prep.
Subtotal
Composite Variability
0.945
0.786
0.918
0.905
0.469
0.802
0.866
0.797
0.395
0.448
0.654
1
2
C.
3
The Commission
discussed
the evaluation
5
witnesses
Bradley,
6
adequacy
of an econometric
7
1.
9
3.
986,430
172,189
956,895
71,282
132,856
66,122
398,302
1.246,479
23,934
116,373
193,556
4.364,418
0.81
934,238
164,522
851,615
75,142
181,579
52,849
355,068
1,149,513
31,630
135,596
162,480
4.094.231
0.76
1.043.841
219,070
1.042,369
78,765
283,275
82,447
459,933
1563,963
60,593
259,762
295,957
5.389.975
0.895
0.751
0.817
0.954
0.641
0.641
0.772
0.735
0.522
0.522
0.549
analysis. I0 The Commission
Neels, and Smith.
should
include
that Can Serve as
in Docket No. R97-1 the standards
and criteria for
reviewed
comments
The relevant criteria for the evaluation
by
of the
study are well understood:
the development
and use of an adequate
database,
verified and complete.
The study should include a discussion
consistent
10
11
A study
of an econometric
appropriately
2.
Attributable
Cost
per Dr. Bouo
LQOQ
The Commission Has Identified Criteria and Standards
a Basis for the Evaluation of an Econometric Studv.
4
a
Attributable
Cost
per Dr. Bradley
UQQ
with the underlying
An adequate
of the modeling
approach
and how it is
data.
model and analysis of functional properties
is necessary.
Docket No. R97-I, Appendices to Opinion and Recommended Decision, Volume 2, Appendix F,
-8-
Docket No. RZOOO-1
1
4.
2
3
QCA-T-4
A correct estimation
procedure
to the estimation
needs at hand
should be used.
5.
Results for econometric
4
include a full explanation
5
the variables.
6
The Commission
7
econometric
8
criteria for evaluation.
9
estimates.
10
that is suitable
work.
equations
and alternative
econometric
analyses
should
of the values, signs, and other relevant information
has also indicated some of the procedures
First, the Commission
Finally, the Commission
by which it reviews
reviews the econometric
Second, the Commission
using the
of the
model to find a result
that it can safely rely upon; that is, a result that is stable and robust.
-9-
research
reviews the statistical properties
tries to identify a preferred
for
Docket No. R2000-1
1
IV.
OCA-T-4
DR. BRADLEY’S
A.
STUDY
A Review of Dr. Bradley’s Study Highlights Previous and Potential
Problems Associated with the Measurement of Volume Variabilitv.
There were significant
with Dr. Bradley’s
original study.
general, to Dr. Bouo’s
some detail.
the traditional
9
l
11
and estimating
problems associated
Unfortunately,
these problems
have carried over, in
study, so it is appropriate
to first examine
Dr. Bradley’s study in
Dr. Bradley’s testimony
8
10
data, methodological
assumptions
There are differences
presented two major conclusions
that differed from
about volume variability:
in volume
variabilities
for mail processing
across activities;
and
.
The estimation
of mail processing
variabilities
generally
produces
a number less
12
than 100 percent
13
Both UPS witness Neels and I disputed the results, focusing on the variety of issues
14
related to databases,
variables,
model specification,
15
estimation
of mail processing
16
processing
activity.
17
variabilities.
18
specific cost pools, Dr. Bradley estimated
19
(which he designated
was
performed
Table 2 summarizes
and other factors.”
at the
Dr. Bradley’s
Based on total mail processing
Dr. Bradley’s
level of the individual
25 estimated
labor costs disaggregated
cost elasticities
by modeling
mail
mail processing
into activityhours of labor
as a measure of cost) as a function of total pieces handled (TPH),
UPS-T-l, Docket No. R97-1 (Neels); OCA-T-500, Docket No. R97-1 (Smith)
-IO-
OCA-T-4
Docket No. R2000-1
to be a measure
of output.‘2
Additional
1
deemed
2
segmented
3
to the sum of all manual letter TPH, mechanized
4
He also used seasonal
5
for the ebbs and flows of mail throughout
time trend, and a manual ratio (computed
dummy variables
explanatory
variables
included
a
as the ratio of manual letter TPH
letter TPH, and automated
to denote the accounting
letter TPH).
periods to account
the year.
I2
This summary of Dr. Bradley’s work is not comprehensive or complete, focusing only on the
essential highlights of his work. For example, Registry and Encoding were separately estimated.
-ll-
Docket No, R2000-1
OCA-T-4
Table 2
Summary of Dr. Bradley’s Variabilities
Variabilities
Estimated by
Dr. Bradley
Activity
Comparable Activities
Estimated on the Basis
of Proxies
MODS Offices
Proxy Variability
General Support Activities
BCS Sorting
OCR Sorting
LSM Sorting
FSM Sorting
Manual Letter Sorting
Manual Flat Sorting
Manual Parcel Sorting
Manual Priority Mail Sorting
SPBS Priority Mail Sorting
SPBS Non Priority Mail Sorting
Cancellation and Mail Prep
0.945
0.766
0.905
0.916
0.797
0.666
0.395
0.446
0.602
0.469
0.664
MODS Allied Acfivities
Opening Pref Mail
Opening Bulk Business Mail
Pouching
Platform
Remote Encoding
Registry
0.720
0.741
0.629
0.726
1.000
0.150
BMC Offices
Sack Sorting
0.991
0.654
0.969
0.754
0.716
0.672
Primary Parcel Sorting
Secondary Parcel Sorting
irregular Parcel Post
Sack Opening Unit
Non Machinable Outsides
Mail Processing Support
Miscellaneous Processing
Empty equipment
Damaged Parcel Rewrap
System Variability
System Variability
System Variability
System Variability
Piece Handlings Unavailable
Mechanized Sack Sorting
Mechanized Parcel Sorting
Bulk Presort
Manual Sack sorting
Mailgram Sorting
Express Mail Sorting
ACDCS (Scanning)
Business Mail Reply
BMC Mech. SS
BMC Mech. PS
Opening Units
BMC Platform
Manual Ltr Sorting
Manual Pri. Sorting
Pouching
Manual Ltr Sorting
Customer Service Activities
Automated Sorting/Stations
Mechanized Sorting/Stations
Manual Sorting/Stations
Box Section Sorting/Stations
Express Mail SortingCSOMan
Special Service Activities
Mist Activities at CSO
Mail Markup and Forwarding
Business Mail Entry
OCR 8 BCS
LSM and FSM Activities
Man1Lrt. and Manl. Flat
Manl Lrt. and Manl. Flat
Manual Pri. Sorting
Registry Activity
Registry Activity
Avg. Mech. Activities
Platform Activity
BMC Allied Activities
Platform
Floor Labor
0.533
0.605
Data Sources
USPS-T-14, Docket No. R97-1. page 9
Dr. Bradley’s Study Was Criticized as Being of a Short-Run Nature Due to
the Use of 4-Week Accounting
Periods Coupled with the Lack of
Consideration of Capital and Investment.
1
2
3
B.
4
The Commission
5
which
postal
rates
has indicated that the postal rate cycle, the period of time over
are fixed,
is the appropriate
- 12-
time period
for the purposes
of
Docket No. R2000-1
OCA-T-4
1
determining
the relationship
between costs and mail volume.13 In contrast, Dr. Bradley’s
2
study focused
3
time frames. There was no longer-run
4
expansion
5
Commission
6
that the relationship
7
periods
will reflect,
a
volume
across accounting
9
accounting
on 4 week accounting
path,”
periods along with some consideration
consideration
which is the relevant
indicated
periods,
approach
of longer
of costs as related to the facility
to the measurement
of costs.
The
that the cyclical nature of mail volume over a rate cycle implied
between
primarily,
input use and mail volume across adjacent
seasonal
periods
variation
in mail volume.
can occur with little change
leading to a low variability
estimate.
accounting
Large changes
in
in labor hours across
I will subsequently
show that
10
Dr. Bozzo’s study is also short run: the use of quarterly data, and even a “same period
11
last year” analysis, does not change its short-run
nature.
The Database for Dr. Bradlev’s Studv Was Unreliable.
12
C.
13
The MODS and PIRS databases
provided
observations
period
14
(AP) and site for the years 1966-1996.
15
continuity,
16
data by site, accounting
17
25,000
16
terms of quantity of data, the major relevant data generated from a field site and used in
and adequacy,
observations
resulting
periods,
or more.‘*
Dr. Bradley scrubbed
by accounting
in the establishment
and activities.
Although
the data for accuracy,
of a database
consisting
of
The data sets were large, with up to
the database
was large when measured
13
Docket No, R97-1, Opinion and Recommended Decision, Volume 2, Appendix F at 13.
1.
The expansion path is the equilibrium point of costs as facility size changes.
in
IS
Data sets were typically in the 17,000-25,000 observations range after scrubbing. A few data sets
were significantly smaller.
-13-
Docket No. R2000-1
OCA-T-4
the study (exclusive
periods) consisted
of information
relating to facility identification,
scrubbing
was
necessary.
Bradley’s scrubbing
scrubbing
Furthermore,
only of two variables: hours and TPH.
the MODS data was substantially
criticized.
Substantial
Dr. Bradley concluded
argument
concerning
appeared
to be largely statistically
the accuracy
that extensive
the deficiencies
process generally focused on the elimination
process
review of the data with field personnel.
facility characteristics
and a variety of other data relevant
data
of Dr.
of relevant data.
The
Information
to
on capital,
to the analysis
of mail
were not included in the data set.
10
D.
11
Dr. Bradley
Dr. Bradlev’s Fixed Effects Approach
estimated
the relationship
Was Criticized bv the Commission
between
hours and TPH with a translog
12
function, using a fixed effects approach for the econometric
estimation.
13
of a specific
intercept
14
account for differences
15
translog
16
choices:
17
l
ia
of
based; there did not appear
have been a detailed
processing
activity type, and time
activity,
function,
he asserted
between
Dr. Bradley
Pooled:
If this approach
approach
would
have
that the fixed effects
In selecting
facilities.‘6
considered
three
had been used,
been
based
on
16
then
the
was adequate
the estimation
estimation
approaches
according
assumption
In the analysis
method for the
as possible
to Dr. Bradley
that
to
the
facility-specific
An issue that was not considered was whether some degree of segmentation into data subsets for
the facilities would have improved the estimation process. Instead, Dr. Bradley assumed that the fixed
effects approach would account for the differences.
-14-
Docket No. R2000-1
OCA-T-4
1
characteristics
2
pooled model approach
3
(GNR).
4
effects were important
5
were not appropriate.
6
9
Fixed
were not important.17
Dr. Bradley
for this reason,
indicated
that he rejected the
relying on the Gauss-Newton
Regression
He stated that in every case the GNR tests indicated that the facility-specific
Effects:
and that both the pooled and the cross sectional
The reasons
cited for the differences
in hours between
models
facilities
7
included the age of the facility, the quality of the local work force, and the quality of
a
the mail that the facility must process.18 Dr. Bradley indicated that his experience
9
studying
mail-processing
Regression.19 The fixed effects approach
attempts to capture
12
facilities
not captured
in the equations,
13
intercept.
However,
14
when the fixed effects never change.
Effects:
by the variables
the approach
Dr. Bradley
16
participating
party advocated
17
the assumption
ia
across facilities are non-stochastic.
as
that there were significant
11
Random
facilities
suggested
non-volume
.
across
strongly
10
15
variations
activities
indicated
works only in measuring
rejected
such a model.
the
random
by
a
Gauss-Newton
differences
as measured
between
by the
fixed effects at a site
effects
Such an approach
that the facility specific characteristics
in
model,
and
no
would be based on
that cause productivity
to vary
11
To the degree that data modeling the characteristics of a facility could be developed, such data
could be included in the study as exogenous variables.
‘8
USPS-T-14, Docket No. R97-1 at 40, lines 1 through 4.
I9
This is a key point. Subsequent testimony will disagree with some of the findings, and this has a
key impact on wnclusions, Dr. Bouo also used a fixed effects approach. He appears to have provided
inadequate explanation and response to the Commission’s comments on fixed effects.
-15-
Docket No. R2000-1
1
OCA-T-4
The Commission
found that the fixed effects in Dr. Bradley’s study may represent
2
effects that are both related and unrelated
3
the facilities,
4
Accordingly,
5
the computed
6
effects estimating
7
a
E.
9
Dr. Bradley
included
to postal volumes; for example,
in the fixed effects,
the Commission
can be a function
the size of
of the volume
of mail.
found that if the fixed effects were volume variable,
volume variabilities
were incorrect.
then
Dr. Bozzo has again used the fixed
procedure.
Dr. Bradley Extrapolated
Activities.
performed
for a limited number
Degen’s needs, for Mr. Degen was required to f&m cost pools for certain activities that
12
had no recorded
13
variabilities
14
activities for which he had computed
15
had been
unable
16
variabilities
for mail handling activities to non-MODS
17
that cost variabilities
ia
change
19
understanding
20
.
could
the existence
measures.
not be measured
to compute
the
The results. did not entirely
of
11
workload
and BMCs.
of mail sortation
of Other
activities
from
offices
the analysis
Results to a Number
10
20
at MODS
His Econometric
Since workload
econometrically.
variabilities
for mail processing
traditional
of why variabilities
100
activities
percent
Therefore,
tie extrapolated
Dr. Bradley
used
the results
for
conclusion
are less than one was a major
are less than one:
of relatively fixed functions within the activity,
-16-
unavailable,
offices.2o Dr. Bradley’s
assumption.
USPS-T-14, Docket No. R97-1. Section Vat 86-90
were
as proxies for activities for which he
Finally,
variabilities.
measures
meet witness
He commented
on
his
OCA-T-4
Docket No. R2000-1
1
the division
l
2
3
l
and specialization
of labor (leading
to the conclusion
activities should have increased
efficiency), and
technological
in machine paced activities
change,
resulting
4
speed having a high variability.*’
5
He indicated that gateway
operated
that manual
at the same
activities (e.g., OCR and platform) would run at both low
6
and high levels depending
on the time of day.
7
would tend to have lower variabilities.n
21
USPS-T-14, Docket No. R97-1 at 56.
22
USPS-T-14, Docket No. R97-1 at 58.
-17-
Finally, he assumed
backstop activities
Docket No. R2000-1
1
V.
OCA-T-4
DR. BOZZO’S VOLUME VARIABILITY
2
3
A.
4
Dr. Bozzo made a number of changes
Dr. Bozzo’s Revisions of 10 of the 25 Mail Processing
bv Dr. Bradlev Continue ta Have Deficiencies.
5
the approach
6
(1) Dr. BOUO’S approach
previous
a
to volume changes
9
quarterly
continues
to Dr. Bradley’s methodology;
the short
study, the mail processing
run approach
elasticities
on an eight weeks basis.
Modeled
however,
to estimation.
In the
only reflected the response
of costs
Dr. Bozzo has modified the data to a
basis, but the analysis is still based on short run costs, measuring
10
in cost with respect to volume
11
utilization
12
through
13
path in response
14
longer-term
15
expansion
path is the hyperplane
16
hours/lPF
relationship.
(2) There
Activities
continues to be fatally flawed.
7
17
STUDY
and investment--which
but not adequately
can have a significant
their effects on facility expansion.
to volume changes
basis
is less
as a result
data
better.
scrubbing,
addressing
changes
issues of capacity
impact on longer-run
Movements
costs
along a facility expansion
will occur when capital and labor vary on a
of the Postal
Service’s
that should
but the
be measured,
for the data
The
not the short
scrubbing
are
run
not
significantly
19
of the data on a site by site basis for data items suspect (unless required to answer
20
an interrogatory).
(3) Microeconomic
22
interspersed
23
of other
no discussion
plans.
ia
21
There was apparently
rules
investment
theory related to cost, production,
with comments
sophisticated
on non-cost
concepts.
-1.8-
and factor demand
minimization,
However,
with field based personnel
homotheticity,
the theory
functions
is
and a variety
is not presented
in an
Docket No. R2000-1
OCA-T4
1
organized
2
trivial issue.
3
the conclusions,
4
variable and whether
5
needed.
6
form.
(4) Variables
There appears to be a number of theoretical
The treatment
of capital could potentially
errors.
have a significant
but it is not clear whether capital is an exogenous
assumed
some type of reduced form simultaneous
non-volume
variable
that are actually
a
be endogenous.
realities.
11
and the alternative
(6) Dr.
There
Bozzo
measurement
14
(7) The econometric
16
deficiencies
(6) There
is a major difference
incorporated
the’model
variable:
capital
in
methodology
been
17
consideration
ia
is missing.
19
B.
20
The analysis
some
of networks.
The previously
continues
when it may in fact
the
analysis;
however,
the
actual
in detail in the previous case.
of additional
variables,
for example,
key variable--capacity
the
utilization--
manual ratio continues to be used.
Dr. Bozzo’s Studv Needs Substantial
of mail processing
by Dr. Bozzo
or inapplicable.
However, a potentially
discredited
estimated
field
to be fixed effects, even though the major
were discussed
introduction
the
from Mr. Degen’s testimony.
of capital appears to be inaccurate
of this approach
has
between
model that can be developed
has
13
15
system is
theory does not appear to be well tied to the mail processing
10
12
or endogenous
volume
manual ratio is still present, and capital is treated as exogenous
(5) The economic
effect on
equations
7
9
This is not a
Work for Completion.
facilities is a complex,
21
issue. The volume variability analysis has consumed
22
five years for the initial work presented
intellectually
major resources,
challenging
apparently
up to
by Dr. Bradley, and another five person years of
-19-
Docket No. R2000-1
OCA-T-4
1
work for the work presented
2
more limited scope of activities.
3
required to complete the work.
4
Furthermore,
5
variabilities
6
and non-MODS
7
significant
9
additionally
variabilities
methodological
performed
which
10 of the previously
actually changed
12
data scrubbing
have not yet been estimated.
Finally, there are
that volume
variability
issues
be thoroughly
As can be seen from Table 1, the proposed
and methodological
variabilities
that the tone of my testimony
testimony
of Dr. Bradley and the follow-on
15
been
16
carried to their ultimate conclusions,
17
possible
18
apparently
19
recommending
20
variability
to present
within
require
in excess
to the Commission
is negative,
work of Dr. Bozzo.
new econometric
a four month
have
due to changes in
changes.
14
satisfying
and
The current estimators
over the short course of several years, apparently
I recognize
25
In addition there are a large number of MODS
explored before being adopted by the Commission.
11
estimated
issues in dispute over the work.
it is important
appear to be tentative.
on a much
Possibly another five person years of effort would be
that Dr. Bradley estimated.
10
13
which was, however,
Dr. Bozzo has only estimated
Accordingly,
6
by Dr. Bouo,
methodologies
as related to both the
Although
it would have
and economic
I have found that such an accomplishment
time frame--particularly
of five person
years
since
of work.
such
-2o-
is not
an effort would
Accordingly,
the following approach to a resolution
issues.
theories
I am
of the volume
OCA-T-4
Docket No. R2000-1
C.
The Commission Should Recommend Establishment
to Resolve the Mail Processina Issues.
The
resolution
implications,
significant
and extensions
amount
atmosphere
9
of the
resolved
in a non-adversarial
technical
issues regarding
could be substantially
recommending
11
intervernors
variability
and improvements
of additional
of a working
10
volume
effort.
group
atmosphere.
narrowed.
major
cost
allocation
to the work appear likely to require a
technical
issues
can be discussed
in the
and
In this way, I believe many of the more
of the data and variables
Accordingly,
that the Postal Service establish
and other interested
has
That effort can best be accomplished
in which
the handling
issue
of a Working Group
the Commission
and the estimators
may wish to consider
an ongoing working group of interested
groups for the review, analysis, and conclusion
12
-2l-
of the
OCA-T-4
Docket No. R2000-1
VI.
DR.
BOZZO’S
METHODOLOGY
IS EVALUATED
I have listed, above, the several deficiencies
USPS modeling of mail processing
5
the context of an overall evaluation
6
difficult for the Commission
7
laundry list of problems
8
may appear to be concerned
9
of the problem.
variabilities,
in the context of a full-blown
11
whether
12
my testimony
13
Commission
14
measuring
15
important
16
of Dr. Bozzo’s work in terms of the criteria discussed
17
opinion in Docket No. R97-1.
meets generally
discussing
established
“The blueprint
criteria.
five important
as appropriate
issues in a
That is, certain issues
.n23 An econometric
application
in a format
analysis,
has stated,
way, it may be
of individual
with minutia, of little overall significance
is well-understood...
are present in the
in a structured
to weigh the relative significance
of econometrics
ESTABLISHED
Standing alone, without placing them in
10
recognized
that I conclude
of the methodology
As the Commission
it successfully
UNDER
to the resolution
for a successful
study is judged by
I am therefore
presenting
criteria similar to that which the
for evaluating
econometric
methodology.
Dr. Bozzo’s study against these criteria, I have found the study deficient
respects in each of the areas,
The following
Docket No. R97-1, Opinion and Recommended
- 22 -
In
in
sections
present an evaluation
in Appendix
F of the Commission’s
Deci.sion,Volume 2. Appendix F at 1
Docket No. R2000-1
1
2
A.
OCA-T-I
Criterion 1: A Study Should Include the Development and Use of an
Adeauate Database, Aoorooriatelv Verified and Complete.
3
4
1.
5
A review
The database
accuracv.
was
of the data scrubbing
6
provides
7
The Commission
8
usable
9
identify erroneous
issues
some insight into the inadequacy
concluded
data and ineffective
10
a selection
11
not adequately
examined
associated
observations.
the rules applied
The Commission
the estimated
indicated
that, “It is the Commission’s
understanding
12
requires
that when data are removed
13
econometrician
14
erroneous.“25
has investigated
because
for
work
for both studies.
they eliminated
in the scrubs did not reliably
concluded
bias by unduly affecting
verified
Dr. Bradley’s
of the underlying databases
that the scrubs were excessive
because
with
and
that the scrubs produced
variabilities.24
The Commission
that good econometric
from a sample, they are removed
and found good cause for believing
practice
because
the
that the data are
Dr. Bradley’s initial data review appears to have been based on the application
15
16
of statistical
17
used in the current
18
current
19
inaccuracies;
20
due to a major data discontinuity
21
25
analysis.
study
The differences
between
study are actually
in lieu of four week
Dr. Bradley’s data set and the data set
quite minor.
accounting
Quarterly
data are used in the
period data in order to smooth
out
the rejection criteria are relaxed; and the overall time period is changed
at the time of the Postal reorganization.
Id. at31.
Id. et 28.
-23-
Docket No. R2000-1
1
OCA-T-4
The underlying
data bases from which
Dr. Bradley
obtained
the data for the
2
study are unreliable.
3
Inspection
4
unmistakable
5
study apparently
6
have been initially
subjected
7
verification appears
to be required to provide a sound basis for the analysis.
8
Dr. Bouo’s
9
fact.”
As the Commission
Service,
a conscientious
internal evidence
continues
responses
One response
indicated,
examination
of the data
of serious errors.“26
sets would
to minimal
to interrogatories
discussed
actual
appear
The data set used in the current
data errors due to commingling
11
also discussed
12
facilities or the introduction
13
should be performed
data questions
field verification.
reporting
Several of
“after the
of manual and SPBS
at a site.*’
The response
related to 13 sites, largely involving reclassifications
of new facilities.
to
Field level data
to focus on data checking
parcels, and a gap in the manual priority volume
This is the type of data verification
of
that
prior to beginning the analysis,
In view of the known
deficiencies
of the MODS
data base, as well as the
15
changing nature of the data as verified by questions
raised in interrogatories,
16
that the database
to substantial
17
accuracy
18
sampling basis to verify the degree of accuracy.
19
with the people responsible
20
gather
should
and completeness.
information
disclose
to be drawn from the same data source and appears
10
14
“Even without the report of the
have been subjected
Such verification
could
for data collection to determine
on site specific
26
Id. at 26.
27
UPS/USPS-T15-13, Tr. 15/6387-8.
circumstances.
-24.
field verification
be performed
Follow-up
I conclude
initially
for
on a
efforts would involve contact
data accuracy as well as to
The
actual
examination
and
Docket No. R2000-1
OCA-T-4
1
verification
2
been performed to any significant
3
of data from sites with input from field personnel
Statistical
data
4
verification.
5
items and perform
6
implemented
7
degree.
scrubbing
A proper approach
is not an adequate
to the verification
for on-site
data
reliability.
Procedures
must then be
to upgrade the data set if the data prove to be unreliable.
8
segmentation
9
comparisons.
the
data
review,
of the database
Clusters
there
into subsets
of sites
could
was
no discussion
of similar
technology
11
clustering
12
similar sites, much of the fixed effects problem identified
13
avoided.
14
statistical estimates;
and automation
of processing
(thereby
activities,
A smaller number
avoiding
the meaningless
and probably
16
response
17
priority variabilities
18
reasonable
19
and that the conclusions
that indicated
manual
accurate
of
ratio), the
By grouping
by the Commission
might produce
however, the tradeoff might be increased
possible
by size, degree
other classifications.
of sites based on clustering
An example of the importance
of the
sites to facilitate
have been considered
10
28
substitute
of data is to select a sample of data
a field check to determine
In performing
15
does not appear to have
could be
less precise
accuracy.
of the data issue was provided in an interrogatory
there were large upward revisions
due largely to the application
to the manual parcel and
of tighter sample selection rulesz8 It is
to conclude that the study is deficient in terms of its underlying
may be tentative, depending
AAPIUSPS-TIS-5, Tr. 1516227.
-25-
significantly
database,
on data scrubbing.
OCA-T-4
Docket No. R2000-1
2.
The
Changes in postal investment subjected the investment data trends
to changes during 1994-96; previous data may be unrepresentative
of operatina conditions in the forthcomina rate effective time period.
history
of Postal Service
investments
in Table 3 and the accompanying
in mail processing
5
summarized
6
Postal Service’s
7
It remained,
8
previous
9
years included in Dr. Bozzo’s study differ significantly
investment
in mail processing
graph.29
equipment
equipment
Table 3 indicates
changed
1993 through
years.
1995.
Moreover,
Thus, the investment
plans
Postal
Service
10
for the later
11
delineated
12
project a high level of investment
13
that part of the data relied upon by Dr. Bozzo is not representative
14
which the rates will be in effect. According
15
unrepresentative
in the annual investment capital plans,”
in the three
expenditures
from the investment
for future
that the
during 1994-1996.
on average, at a level much higher than the level of investment
years,
in the early
expenditures
investments
and the Postal Service continues
in mail processing
equipment.
It therefore
23
30
ANMLISPS-TIO-17, Tr. 21408.
31
OCA/USPS-T15-6, Tr. 196298.
-26-
to
of the period for
to Dr. Bozzo, the potential
impact
data is important:
ANMIUSPS-T9-47-49, Tr. 21199-202
are
appears
My main motivation for employing data over a shorter time period
was the desire to balance the potentially competing aims of efficient
estimation
of
the
labor
demand
accurate
estimation
and
functions....However,
extending the sample period back in time does not
hold other things equal. It raises the possibility of introducing
nonsampling errors in the estimates to the extent the earlier data are
unrepresentative
of current operations.“3’
16
17
18
19
20
21
22
23
is
of
Docket No. R2000-1
OCA-T4
in the investment
1
Fluctuations
2
analysis.
3
data irrelevant to current
4
graph derived from Table 3.
The investment
data may make them unrepresentative
88
89
90
91
92
93
94
95
96
97
98
99
of
data will impact the values for capital, possibly making earlier
practices.
The investment
data are plotted in the following
Table 3
Postal Service Investment-1988-1999
Year
for purposes
Total Postal Service
Investment
Mail Processing Equipment
Investment
623.9
1,987.5
2,436.4
1.883.1
1,924.8
1.309.6
I,6355
21284.9
3,306.Q
3,202.6
3,947.0
3,817.3
91 .Q
560.0
466.4
397.7
201.1
634.5
326.9
866.8
1,220.5
808.2
1,204.l
1.158.1
Source: ANM/USPS-TQd7-49,Attachment
-27-
Docket No. R2000-1
OCA-T-4
Postal Service
+
1
Total USPS Imestment +
Accordingly,
investment
in examining
the
Investment
Investment in Mail Processing Equipment
Hours/TPF
relationships,
series that may be unrepresentative
Dr. Bozzo
2
underlying
3
changing
4
aggregate
5
was a major change
6
discontinuity
with a dummy variable, but Dr. Bozzo has not addressed
7
conclusions
of the changing trends
nature of segment 3 data for segment
has an
of current operations.
The
3 hours and total mail is shown on an
basis in Table 4 in terms of payroll hours for segment 3 and total mail. There
in trend in the 1997 time frame.
-28-
Dr. Bradley
treated
a similar
the impact on his
Docket No. R2000-1
OCA-T-4
Table 4
Mail Volume and Segment
Cost Segment 3
Payroll Hours
Work
Volume
Total
AlLMail
201,578,279
1998
1997
694J45.627
694,686,240
196,904,690
190.888.059
183,439,474
179,932.615
1995
693.945.735
680.293,834
667,448,113
654,575,064
617.449,610
615,041,369
631.555,134
633,771,319
64lB45.471
638,779,872
626,078,466
1778177.362
170,312,972
165,654,138
165,057,806
165,502,505
161,603.263
1993
1991
1990
1989
160.953,625
1987
153.152.758
1461578,077
1985
140.097.958
131,544,622
118,476.588
113,121.664
110,130.400
1983
603,546.949
582,351,682
560,064.472
524,770.256
518,265.Oll
525,640,282
528.221.756
527.506,828
517,087.887
116.451.141
991828,883
96,913,154
1979
1978
1
3.
2
The continued
Dr. Bozzo continues
3
automation.
4
nevertheless
5
determine
6
comparable
32
Recognizing
maintains
the manual
3 Hours
use of the manual ratio is undesirable.
to use the manual
that the manual
ratio as a measure
of the degree
of
ratio can be affected
by volume,
he
that the mail processing
technology
ratio.J2 He maintains that a computed
rather than mail volumes
manual ratio number
is
from site to site, even though the size of the sites may range from small to
USPS-T15 at 24, line 11
-29-
OCA-T-4
Docket No. R2000-1
1
large.‘”
2
mail flows and automation
3
he appears
4
would not affect the manual ratio, other things equal, but since the TPF are likely to be
5
related to network
6
equal.JS
However,
8
measures
to believe that the size of the mail processing
characteristics
9
example,
of the degree
4.
11
14
15
The regression
for an activity
Other
need to be developed;
for
of machines for an activity at a site could be used
capability
as appropriate.
The presentation of the variable QICAP, used to measure
capital usaae at each facilitv. is inadeauate.
equations,
as outlined on pages 117 and 118 of Dr. Bozzo’s
16
testimony,
use a variable denoted as “CAP”.
17
referenced
in LR-I-107.36 QICAP is denoted as a quantity index for facility capital.
18
value of the capital items at a facility are depreciated,
IQ
transformed
33
in TPF
is inappropriate.
The QICAP variable has not been demonstrated
(a)
12
13
facility as measured
ratio in the analysis
of automation
of automation
Finally,
one would expect that other things are not, in fact,
the capacity and numbers
as a measurement
affect local
usage, they may affect the manual ratio variable.”
In my view, use of the manual
7
10
he also admits that to the extent network characteristics
Apparently,
this is the QICAP variable
adjusted for inflation, and
into a capital flow. The details of the procedure were apparently
OCAIUSPS-T15-8, Tr. 1516301.
3d
OCAIUSPS-T15-11, Tr. 1516305.
u
OCA/USPS-T15-15. Tr. 1516309.
H
USPS-LR-I-107, Docket No. R2000-1
- 30 -
The
presented
Docket No. R2000-1
OCA-T-4
1
in the previous case.)’
The derivation of QICAP was discussed during an informal
2
technical conference
3
However, the presentation
4
discussed in Dr. Bouo’s
5
previously presented
6
deficiencies
with Dr. Bouo
of the derivation
testimony,
and it is impossible
QICAP is not even
to determine the relevance
of
There are a number of
with the QICAP variable.
The variable
computational
(b)
The use of the variable
9
of QICAP is inadequate;
information to the current use of QICAP.
associated
7
6
and was also the subject of interrogatories,
QICAP appears
viewooint.
QICAP in a regression
equation
deficient
from
a
might yield spurious
10
results.
11
to site.”
12
computations
13
QICAP is a cardinal number although on a practical basis it may be possible to perform
14
sufficient computations
to adjust the number for adequacy
15
Regression
are based on the addition
16
matrices that define the regression
17
when added or multiplied may result in the wrong conclusionsSg
18
be a mathematical
31
?a
Dr. Bozzo indicates that the QICAP numbers
to be
He indicates
that they
need to be made.
equations
are not strictly additive from site
are approximately
Accordingly,
equation.
additive,
but that
additional
Dr. Bozzo has not demonstrated
under certain circumstances,
and multiplication
Numbers
problem in using QICAP in a regression
that
of numbers
that yield inaccurate
Accordingly,
in the
results
there may
equation.
USPS-LR-H-272, Docket No. R97-1.
OCAIUSPS-T15-45, Tr. 1516341-2.
39
A very simple example will illustrate this: if the price of food rises by 3percent and the price of
clothing rises by 2 percent, then prices are not up by 5 percent.
-31-
Docket No. R2000-1
OCA-T-4
(c)
QICAP is available on a facility basis, not on an activity level
basis; this may lead to meaningless results when including
capital investment in the studv.
The variable QICAP is available only on a facility basis.
5
the capital used at a facility rather than for an activity.
6
various
types of automated
7
sorters,
optical character
8
parcels or letters), only one number is available:
9
the facility.
or mechanized
readers)
Furthermore,
operations
and manual
QICAP is a measure of
For example,
(e.g., cancellation,
operations
(e.g., manual
bar code
sorting
of
the overall amount of capital used at
capital used in activities that are not even being modeled
10
also included in QICAP as long as the capital is present at the facility.
11
modeling
12
facility, regardless
13
minimally.
14
at a site with
Accordingly,
is
the
of any activity at a facility is based on the overall usage of capital at the
of whether
Dr. Bozzo essentially
the particular
activity is capital intensive or uses capital
maintains that the QICAP variable in its current state is the
He maintains that it is not possible to classify
15
best estimate of capital usage available.
16
all equipment
17
level capital measures
18
cost pool would not represent
19
represent
20
pool using the Property
21
space, which he alleges to be an important non-equipment
at a site by cost pool.
which would
According
to Dr. Bozzo, the resulting cost pool
result from segmenting
available
data by activity
the cost pools of capital per se, but rather, they would
the portion of the cost pools capital that could be associated
Code Number
(PCN).
- 32 -
with the cost
He further notes that data on facility
component
of a hypothetical
Docket No. R2000-1
OCA-T-4
1
cost pool capital index, are not available
2
not obvious
3
primary economically
4
the effect of including
5
demand
6
as the capital services employed in other pools.
7
that a cost-pool-level
capital
relevant measure
the’facility
by cost po01.~~ He further maintains
measure
of capital.
would
He has indicated
8
the installation
9
level of automation
even the
that, in his view,
capital index is to capture the fixed effect on labor
in a given cost pool of the capital services employed
An example
be the sole--or
that it is
illustrates the deficiency
of Flat Sorting Machines
than currently
of QICAP.
in detail.
exists.
Witness Kingsley has discussed
Such machines
Apparently
machines
will provide a higher
of significantly
10
capital value, sophistication,
11
facilities.
12
major mail processing
13
Reader (OCR) and Bar Code Sorter (BCS) machines.
14
FSM’s at a given site, the QICAP variable appears
15
degree
16
activity, one would be using a value for capital strongly influenced by other activities.
facilities
the investments
example
18
considering
19
of how sophisticated
have sophisticated,
testimonies,
the automated
Accordingly,
Optical
Character
in any analysis
of
likely to reflect to a disproportionate
a potentially
flat sorting machines, one considers
it is clear that most
high capability
in OCR and BCS machines.
demonstrates
less
are currently in use at the mail processing
Based on Mr. Degen’s and Ms. Kingsley’s
A further
17
and capability
in that cost pool as well
In analyzing
greater
mismatch,
the flat sorting
if instead
of
the manual casing of mail. Regardless
activities of the plant are, it does not appear that
40
Although square feet of space clearly cost money, Dr. Bouo
associated space affects hours of labor
-33-
has not explained how the
OCA-T-4
Docket No, R2000-1
1
this investment
will have much impact on the manual casing of letters, a technology
2
existence for many years.
(d)
3
4
Turning
Some of Dr. Bozzo’s computations
nature of the variable QICAP.
the capital
6
elasticity of manual flats and manual letters with that of a bar code sorter.
The capital
7
elasticities
8
OCR. The conclusions
9
and there is inadequate
for the manual
operations
testimony,”
the dubious
one can compare
5
to Table 6 of Dr. Bozzo’s
illustrate
in
are greater
than the capital elasticities
for the
that one could draw from Table 6 do not comport with reality,
discussion
of the results.
of the results should be provided.
At the very least, some extensive
10
discussion
11
capital data are needed at the activity level if activities are to be analyzed.
12
that such data are not available does not suffice as a reason for its non-inclusion.
13
14
15
(e)
The approach to equipment depreciation
consider
maintenance
efforts
also
meaninaless.
The Postal Service
16
depreciation
17
mail processing
equipment,
18
percent per year; and buildings,
19
of capital for mail processing
20
viewpoint
the machines
For purp.oses of analysis, it appears
rates, by equipment
8.3 percent
per year;
postal
2.33 percent per year.‘?
machines.
41
USPS-T-15 at 119.
42
OCAIUSPS-T1547, Tr. 15/6344-5
-34-
are as follows:
equipment,
11.5
QICAP is used as a measure
Dr. Bozzo asserts
have useful value consistent
A statement
and the failure to
renders
QICAP
category,
support
that
that from an economic
with the geometric
perpetual
Docket No. R2000-1
OCA-T4
1
inventory equation.‘3
2
based on internal Postal Service studies; however, these are internal studies based on
3
previous,
historical experience.
4
may
quite
5
depreciation
6
to be based on accounting
data rather than operational
7
that an FSM is 8.3 percent
less productive
8
9
be
Dr. Bouo
different
has justified the accelerated
The modern equipment
from
rates meaningless.
that
installed
depreciation
rate as being
that is currently being installed
previously,
rendering
In addition, the depreciation
the
historical
rates being used appear
reality: it is difficult to imagine
after its first year on the job.
In an industrial setting, various vintages of the same machine may be present on
the factory floor.
Regardless
10
the machines
will typically
11
major difference
12
increased
13
be a relationship
14
age (i.e., depreciation)
15
their operability
16
comparing
17
maintenance
the machines
From the viewpoint
between
hours of operation
of the machinery
as they depreciate
through
hours, operating
increased
analysis.U
20
view,
one
would
43
OCAIUSPS-T15-49. Tr. 1516349.
may require
based on the
Older machines will maintain
maintenance.
simultaneously
Accordingly,
to consider
OCAJJSPS-T15-63, Tr. 1516376.
-35-
in
maintenance:
are strongly interrelated.
time is included
as a variable in the
QICAP is correct from a depreciation
need to note that operating
The
there will usually
and levels of maintenance
or maintenance
Even assuming
regression
when operating.
of activities in factories,
hours, and capital equipment
no management
by the accountants,
is that the older machines
after a few years.
vintages of capital it is necessary
19
ad
accrued
have the same level of productivity
(if any) between
maintenance.
However,
18
of the level of depreciation
and maintenance
point of
labor is carried
in
Docket No. R2000-1
OCA-T-P
1
another account but is a complement
2
is seriously deficient without consideration
3
4
5.
5
to machine operating time. Accordingly,
of management
Capacity utilization is another potentially
from Dr. Bozzo’s database.
It is well known that the output, efficiency,
6
operations
are strongly
7
knowledge
that investors, economists,
8
utilization
9
industrial style process, capacity utilization
measure
11
Furthermore,
12
capacity
13
correlated
14
should
15
Accordingly,
16
17
of
capacity
utilization,
hours.
important variable missing
and resource requirements
utilization.
For example,
of factory
it is common
and the financial press examine factory capacity
and
and other economic changes.
is a key number.‘5
this
is
a
For an
Dr. Bozzo’s study has no
potentially
serious
deficiency.
there is no reason to believe that TPF or TPH are approximations
utilization.
Dr. Bozzo treats
with capacity
also be obvious
them
as an output,
under certain circumstances,
that capacity
utilization
so while
is not measured
In conclusion, there are serious
foundation of the studv.
The data problems
associated
they
of
may be
they do not measure capacity.
It
as a fixed effect.
the lack of a capacity utilization variable is a major deficiency
6.
18
related to capacity
as a signal of price, employment,
IO
and maintenance
the study
data problems
of the model.
underlying
the
with the current study include data scrubbing/non
19
verification, problems with specific variables (QICAP, manual ratio), the potentially
20
unrepresentative
nature
of the data
series,
45
and
issues
associated
with
omitted
If capacity utilization were at 100 percent, it would still be possible to increase production in the
short run through extraordinary measures, and in the longer run through the addition of machines and/or
plants.
- 36 -
Docket No. R2000-1
OCA-T-4
1
variables.
2
Finally, a potentially key variable, capacity utilization, is missing
Dr. Bozzo’s database
B.
3
4
1.
Both
8
relationship
9
Bradley’s
The economic assumptions and theory for the current study are not
clear: in manv cases thev appear to be wrona.
Dr. Bradley
and
Dr. Bozzo
used
of labor hours and TPF or TPH.
lack of stated
cost theoretic
confusion
added unnecessary
11
work that Dr. Bozzo has presented.
12
replete
13
underlying
14
it is difficult to follow the logical progression
15
the analysis and the functions being estimated
with references
microeconomics
On a preliminary
microeconomic
throughout
testimony
the
basis I have identified
considered
18
.
Statement of the function being estimated;
19
.
Selection of variables to be estimated;
20
.
Treatment
of Network issues;
21
.
Variables:
Manual Ratio and QICAP;
in the following sections:
USPS-T-15 at 44, lines 18-20
- 37 -
in this proceeding
the presentation.
the following
study
also applies to the
price theory.
of the derivation,
17
46
to estimate
for his mail processing
A similar statement
The econometric
are interspersed
functions
Dr. Bozzo indicated that “....I find that Dr.
to the Docket.“*
to advanced
translog
underpinnings
10
16
of reliability.
Criterion 2: Models Should Be Derived from the Appropriate Economic
Theorv and Should Fit Correctlv Wtthin any Svstem that ADDlies Them.
5
6
7
does not appear to meet the standards
properties,
problems,
However,
is
the
Accordingly,
and logic of
which will be
Docket No. R2000-1
OCA-T-4
1
l
Time Frame: Short run and long run; and
2
.
Cost minimization
2.
Dr. Bozzo and Dr. Bradley do not agree on the type of function
being estimated; much improvement in the presentation
of the
labor demand function is needed.
Dr. Bradley estimated the relationship
cost function.
8
economists
9
obviously
IO
highlights
11
assumptions.
12
of hours and TPH, which he denoted as a
Dr. Bozzo defines the relationship
are estimating
what
as a labor demand
is essentially
the same
function.
The
function.
cannot be both a cost function and a labor demand function.
the
absence
of a clear
economic
exposition
Both
function
This confusion
of economic
theory
Dr. Bozzo indicates that his labor demand function is actually a conditional
13
demand
function
that
14
minimization
15
neither
16
problem of constructing
or from a generalized
derivation,
18
applicable.
19
cost equation,
20
analyst
21
variables
47
48
from
and ultimately
the theories underlying
comment
The Commission
a partial
non-cost minimization
and the reader
The Commission’s
17
can be derived
equilibrium
model.
model
can obtain almost any desired
of cost
the Commission
are left with the
his testimony. 47
in discussing
criticism
Dr. Bradley’s
in Docket
variability
cost function
is again
Bradley’s
No. R87-1 that ‘an imaginative
estimate
by carefully
and the time period to be used in the analysis’ seems to apply.“q
choosing
the
Dr. Bozzo’s
OGVUSPS-T-15-56, Tr. 1516356-g.
Docket No. R97-1, Appendices to Opinion and Recommended Decision, Volume 2. at 6.
- 3% -
labor
However, he performs
said that, “Given the arbitrary nature of witness
the Commission’s
and
Docket No. RZOOO-1
OCA-T-4
I
conditional
2
function
3
appropriate
4
specified
5
mathematical
relationship
6
as Shepard’s
lemma, provides that if the cost function is locally differentiable,
7
demand function is equal to the partial derivative of the cost function with respect to the
a
wage.49
9
Shephard’s
IO
labor demand
is defined
is open to similar criticism.
as xi=xi(w,, w2,,w,, p) for j = l...n.
derivations
First, a labor demand
For estimating
purposes,
from the production function would yield an estimating
in terms of the production
It is possible
function variables.
equation,
As indicated by Dr. Bozzo, the
between the cost function and labor demand function,
that the Postal Service
lemma does not apply.
non-equilibrium”
11
12
13
14
15
16
17
ia
19
function
conditions
operates
Dr. Bozzo responded
under conditions
known
the labor
in which
to a question about “cases of
under which his theory is substantiated:
To the extent that the term refers to situations under which the
relevant theoretical conditions of the cost minimizing (or generalized noncost minimizing) model do not hold, my results would still represent an
empirical analysis of the Postal Service’s demand for labor in mail
processing
operations,
but the mathematical
relationship
(“Shepard’s
lemma”) between the labor demand and cost functions would not
necessarily hold.%
Dr. Bozzo did not fully explain the applicability
20
Bozzo has also indicated
21
theory and operational
22
the cost functions
23
reason.5’
49
that he included variables
reality.
of his labor demand function.
to bridge the gap between
Dr.
generic
He indicated that the labor demand models used, and
implicitly associated
with them, employ additional
In order to verify that Dr. Bozzo’s approach
OCAIUSPS-T-15-17, Tr. 151631l-2.
50
OCAIUSPS-T-15-59(a), Tr. lW6365-6.
51
OCAfUSPS-T-15-56(c), Tr. 15/6358-g.
- 39 -
is grounded
variables
for that
in economic
theory,
’
Docket No. R2000-1
1
the Commission
OCA-T-4
needs an explicit derivation
analysis of the endogenous
impact
on labor demand
competition.
5
whether
6
comments
7
8
9
10
or exogenous
nature of investment,
under conditions
of monopsony,
This would
additional
alleviate
equations
about exogenous
3.
of the labor demand function, an additional
concerns
were
monopoly,
about variables
needed,
and endogenous
and a discussion
particularly
and imperfect
in the equations
in view
of the
and
of Dr. BOZZO’S
variables.
Dr. Bozzo’s study is short run. The proper approach for examining
postal facilities is on a longer-run basis as related to major
investment plans and movement alonq the facilitv expansion path.
The concepts
of the short run and the long run are clear from the viewpoint of
(for example,
theoretical
12
labor) are variable.
In the long run, all of the factors of production
13
Service investments
in capital to reduce operating costs indicate a long run approach is
14
applicable
15
labor and volume, the appropriate
16
expansion
17
equipment.
18
facility as incremental
19
economics.
In the short run some of the factors of production
11
to the analysis.
Instead
path that occurs
of measuring
relationship
when
the
This focus on the expansion
In Docket
labor or incremental
No. R97-1,
expansion
path.
I advocated
However,
the short run relationship
to measure
Postal
Service
between
along the
invests
plant and
path reflects changes
in new
in the scale of the
capital are added
that a pooled equation
could measure the
it has become increasingly
clear that the labor
longer-run
21
hour/TPF
22
at a variety of disequilibrium
points, based on varying capacity
23
levels of mail.
in a subsequent
Accordingly,
Postal
is the movement
20
data points gathered
are variable.
based on field data probably
-4o-
section
measure
mail processing
utilization
I advocate
and varying
that the regression
Docket No. R2000-1
OCA-T-4
1
analysis at this time should be performed
2
set of individual
3
the “least bad” approach,
4
Dr. Bouo
5
estimating
6
measure
7
processing.
that would be used in the pooled case.
even though various statistical deficiencies
states that, “Since capital
longer-run
functions.
is treated
He is only measuring
and management
This is probably
have been noted.
as a quasi-fixed
Dr. Bozzo’s approach
‘short run’ functions.“v
The Postal Service witnesses
8
9
observations
on data means rather than on the larger data
factor,
I am
is wrong; there is a need to
transitory
changes
in mail
appear to have a time frame of
as little as one year to as much as five years in mind when they discuss the longer run,
10
the period over which capital investment varies,
11
two to three year range.
Dr. Bozzo recognizes
12
13
clerk and mail handler
14
fluctuate in the short run:
The time frame seems to center on the
that there are short-run
labor mail processing
and longer-run
demands
aspects
and that labor
of
can
My review of witness Moden’s testimony (Docket No. R97-I, USPS-T-4)
and discussions with Postal Service operations experts revealed that
there are two main staffing processes.
One process assigns the existing
complement to various operations to meet immediate processing needs,
and operates on time scales on the order of hours (let alone eight weeks).
However, the longer term process of adjusting the clerk and mail handler
complement operates more slowly-our operational discussions suggested
up to a year.53
15
16
17
18
19
20
21
22
23
52
53
OCAIUSPS-T15-61, Tr. 1516373.
USPS-T-15 at 16, lines 6-13.
-41-
Docket No. R2000-1
1
OCA-T-4
In conclusion,
it would appear that there are several time periods relevant to the
2
estimation
3
term adjustments
4
appears
5
Both of these time frames have little relevance to the longer-run
6
seem to drive mail processing
7
capacity utilization, and may represent
8
mode that is significantly
9
which would appear to approximate
of postal costs.
One time period is a day, the period over which very short-
to labor are made on an operational
A second time frame
to be the 4 week or 3 month time frame used by Dr. Bradley and Dr. Bozzo.
10
neighborhood
11
personnel,
12
capital investment
13
time frame.
costs,
have little relevance
of two years,
and equipment
decisions
without
plans that
information
on
unreliable data readings for plants operating
in a
time period,
the length of the rate effective time period in the
seems
to be the time frame
decisions
are realized.
over which
Given the increasing
investment,
importance
of
to the Postal Service, this would appear to be the relevant
the ongoing length of the investment
15
initial proposal to project completion,
16
new plant on line.
17
take 5-7 years and actual
18
Service sites new plants to adjust to the network on a continuing
19
increasing
Site acquisition,
Postal flows.
expansion
different from equilibrium. y Finally, a longer-run
Mr. Degen also recognizes
14
basis.
it may take anywhere
planning,
construction
Accordingly,
for a new plant can easily
l-2 years.“”
the actual longer-run
Y
“From
from 6 to 9 years to bring a
and approval
another
process:
Apparently
the Postal
basis, in recognition
of
time frame in which an
Apparently, the set of mail-processing plants is under continuous modifications as plants are
added, subtracted, and modified in the network. In some cases, the data,generated by the plants may be
of a transitory nature and irrelevant to the analysis.
55
USPS-T-16, at 15. lines 4-7,
-42-
Docket No. R2000-1
OCA-T-4
1
investment
decision
is made and implemented
after a relatively
2
framework
appears
to be in the neighborhood
of two years,
3
recognized
that investment
4
deployments
5
is an ongoing
indicating
planning
Dr. Bozzo
has also
that major equipment
usually take more than one year.%
It appears
that a longer term model would
6
sectional
analysis
7
testimony
as outlined in his Figure 3.
8
9
10
process,
protracted
4.
as modeled
by the “between”
best be approximated
model,
based
by a cross
on Mr. Degen’s
Dr. Bozzo addressed Dr. Bradley’s omission of variables in the
regression equations. Dr. Bozzo considers additional variables, but
the consideration is still deficient.
11
12
13
14
15
16
17
18
19
20
21
22
23
Dr. Bozzo indicated that:
The problem of which variables
24
major problem in applied econometrics.
25
lacking
26
measurements
27
additive,
Since the additional explanatory variables--particularly
wages and
network variables--are statistically significant, my results indicate that Dr.
Bradley’s Docket No. R97-1 mail processing models for the operations I
studied were under specified. As a result, Dr. Bradley’s results appear to
exhibit omitted-variables
biases to some degree.
However, since the
revised variabilities accounting for these factors are lower, contrary to the
expectations set forth in the Commission’s Docket No. R97-1 analysis, the
direction of the omitted variables biases in Dr. Bradley’s results were
mainly upwards, not downwards.57
in important
variables:
are to be included in a regression
I am concerned
a measurement
equation
that the work presented
of capacity
utilization,
other variables.
56
OCAJJSPS-Tl5-13, Tr. 156307.
57
USPS-T-15 at 127. lines 10-17.
The analysis
-43-
of network
is still
specific capital
relating to activities rather than facilities, capital measurements
and possibly
is a
effects,
that are
and the
Docket No. R2000-1
OCA-T-4
1
variables
2
section of my testimony.
considered,
3
4
5.
5
is also, in my opinion,
deficient;
The newly presented information
incoroorated in the analvsis.
There are repeated
references
this is discussed
about networks
to mail processing
networks
in another
needs to be fully
in both Dr. Bozzo’s
6
and Mr. Degen’s testimonies.
7
in regards to segment
8
the literature in at least some form since at least 1986.% Mail processing
9
sites do not stand alone in terms of the network
of originating
10
There seem to be three types of network issues.
First, there is the intra-plant
11
of activities that feed mail to each other.
12
could change based on a variety of factors, including
13
of network effect is apparently
14
Bozzo measures
15
possible deliveries.
16
processing
17
interrogatory
18
depends
19
Service’s network.59
58
Although
networks
3 mail-processing
this network
have not been previously
referenced
costs, the concept of the network has been in
and destination
One gets the impression
configuration
nodes.
network
that this network
network volumes.
the delivery configuration
activities and
A second type
of the service territory.
with a variable
measuring
Dr.
the number
Finally, the position of the plant in the mail flow between other mail
plants also seems to be a type of network
response,
the size of facilities
and their
not only on the volume of mail processed,
relationship.
According
mail processing
to an
operations
but also their position in the Postal
Laurits R. Christensen Associates, United States Postal Service Quartedy Real Output, Input, and
1982 Quarter 1 Through 1986 Quarter 1, March ~1986; “A Report to Charles Guy,
Total Factor Productivity,
Director, Office of Economics, United States Postal Service,” in USPS-LR;H-272, Docket No. R97-1.
59
of
USPS-T-15 at 26, lines 4-6.
-44-
Docket No. R2000-1
OCA-T-4
1
The analysis conducted
2
did not address the networking
3
between
4
demand.”
5
length of processing
6
amount of labor required for set up and take down activities, the operation’s
7
gateway
8
variability.
9
concept.
plants.
by Dr. Bozzo addressed
of activities at the plant level or the interchange
windows,
or backstop,
Accordingly,
the complexity
other indicators
The Commission
11
not appear to have been developed
processing
19
performed
20
investment.
21
that the operation
M
schemes,
the relative
of the level of costs, and the degree
networks
the
role as a
of volume
an important
in the recent past in evaluating
and this concept, which is new’to the segment 3 analysis, does
adequately.
Dr. Bozzo estimates
mail processing
activities (e.g., manual
processing,
OCR, BCS) as independent
activities; based on
witness Degen’s comments on networks and facilities, serious
consideration
needs to be given to the simultaneous modeling of
activities.
Dr. Bozzo’s approach
18
appear to determine
of mail processing
has not reviewed
Dr. Bradley’s testimony,
6.
with volumes,
both Mr. Degen and Dr. Bozzo have introduced
10
17
of mail
Both of these types of network effects might have an impact on labor
These factors, often in conjunction
12
13
14
15
16
only the possible deliveries; he
activities
is focused on single activities at a time: he treats the mail-
as separable.
However,
alone; this is partly recognized
Based on my experience
mail-processing
by Dr. Bozzo in his discussion
with batch production
of one mail processing
activities
processes
activity is not independent
USPS-T-15 at 47.
-45-
are not
of capital
I would expect
of another.
Dr.
Docket No. R2000-1
1
Bozzo
2
They advocate
3
asked if he considered
4
“Yes. First, to characterize
5
employ
6
equation
7
Second,
8
Bozzo apparently
9
another.
10
referenced
OCA-T4
Freight
the specification
ten equations
includes
with ten output
in his estimation
Friedlaender
variables;
in addition
of the Postal Service’s
and
to piece
‘distribution
Spady
seem
to advocate
of processing
patterns,
key’.“62
Dr.
from
simultaneous
volumes of mail, and the interaction
13
cross-examination,
14
site has an impact on the hours per TPF relationship.63 He maintained
15
the manual
16
measure of the degree of automation
17
possibly even existing) relationship
18
The issue requires additional exploration.
appears
to be inadequately
addressed
Dr. Bozzo acknowledged
ratio captured
the effect.
in Dr. Bozzo’s analysis.
of
During oral
that the mix of activities in operation
Although
at a
that the use of
the use of the manual
ratio as a
is subject to serious criticism, there is no clear (or
between ihe manual ratio and the activities at a site.
Ann F. Friedlaender, Richard H. Spady. Freight Trampod Regulation, Cambridge, MIT Press,
1961.
63
each
handling%
of each activity as being separable
activities
62
results, I
additionally,
12
61
When
efforts, Dr. Bozzo indicated,
for which I report econometric
‘cost drivers’
the operation
and Spady.61
in terms of multiple outputs.
(piece handling)
non-volume
is an element
by Friedlaender
of activities.
The relationship
11
of a cost function
the set of operations
considered
However,
Regulation
such an approach
other
my analysis
consideration
Transportation
OCAIUSPS-T15-Gl(e), Tr. 1516373-4.
Tr. 1516417
-46.
Docket No. RZOOO-1
OCA-T-4
1
2
7.
3
In his testimony
Dr. Bozzo’s treatment
conclusions.
of homotheticity
Dr. Bozzo asserts that “...capital
4
identical,
in equilibrium,
under the assumption
5
cost) functions are ‘homotheW...Homotheticity
6
of the operation will not alter relative factor demands
7
equilibrium
8
with examples of the implementation
9
costs.
(and other things equal).‘W
appears to lead to incorrect
and labor variabilities
that the cost-pool-level
will be
production
(or
implies that changing the level of output
such as the capital/labor
ratio, in
However, the Postal Service testimony is replete
of major investment
This concept was further developed
programs
in the Postmaster
designed to reduce
General’s
recent speech
10
in Nashville.65 The focus is on the elimination of major labor costs via capital investment
11
to achieve
12
homotheticity
13
14
15
an overall
assumption
a.
16
Dr. Bozzo
17
facilities
18
testimony
19
information
M
reduction
minimize
of total costs.
Accordingly,
appears to be an inappropriate
Dr. Bozzo
minimization;
3 analysis.
has stated
in this proceeding
of a
assumption
has raised some important
issues about cost
resolution of the issues may affect the cost segment
that his theory
costs and, in support,
does not discuss
the application
is independent
cites a publication
of whether
by Toda.=
the Postal
Dr. Bozzo’s
QICAP and he has provided only a limited amount of useful
on the development
of the variable QICAP.
Accordingly,
I
USPS-T-15 at 40, lines lo-14
65
Prepared remarks at the National Postal Forum, Nashville, Tennessee, March 20, 2000, See
OCWJSPS-96, Tr. 2119152.
65
The article introduced by Dr. BOZZOon the topic of non cost minimization appears to ba by
Yasushi Toda, “ESTIMATION OF A COST FUNCTION WHEN THE COST IS NOT MINIMUM: THE
CASE OF SOVIET MANUFACTURING INDUSTRIES, 19561971,” The Review of Economics and
Statistics, Vol. LVIII, August 1976, 259-268.
-47-
Docket No. R2000-1
OCA-T-4
1
have relied on the library reference that he has mentioned?’
2
QICAP are filled with references
3
measurements
4
cost minimizing
5
depend
on a measurement
6
appears
to use in other circumstances).
7
article would be; however, this is an issue that needs to be reviewed.
8
In
9
associated
to Total Factor Productivity.
of Total Factor Productivity
firms.
reviewing
Dr. Bouo
the
indicates
of Total
associated
Factor
library
of QICAP does not
(which
the Postal
inappropriate
measure
of the value
12
depreciation
13
not suitable for productivity
reference,
two
potential
reported
The Moody’s composite
National
Consolidated
of owned
Trial
capital.
deficiencies
in the NCTB is based on accounting
Balance
is an
To be specific,
the
period conventions
accounts.68
of average yields on corporate
bonds is used in arriving
15
at the USPS cost of capital.69 OCA witness Dr. Edwin Rosenberg
16
indicated
17
of money plus 118 percent.‘O
67
Service
with QICAP were found:
11
(2)
Productivity
for non-
It is not clear what the impact of the Toda
in the
14
Toda’s article shows that
that his measurement
reported
(1)
referencing
may be incorrect when computed
Depreciation
10
The documents
(OCA-T-3)
that the Postal Service can borrow from the U.S. Treasury
has
at the cost
USPS-LR-H-272.
68
USPS-LR-H-272. “USPS Quarterly Total Factor Productivity Methodology, A Report to Charles
Guy, Director, Office of Economics, USPS,” L.R. Christensen Associates, January 1998.
69
Ibid. at 47.
70
OCA-T-3, Docket No. R2000-1.
-40-
Docket No. R2000-1
1
OCA-T-4
In discussing
operating
procedures
approach
to operations
Postal
Service
cost
minimization,
one
from the point of view of economic
and pricing in terms of whether
is addressing
theory,
The Postal Service
or not it maximizes
can result in a very different situation than one in which efficient competitive
5
certain
its output
equilibrium
is sought.
From classical
6
economic
theory, an output maximizing
company (in comparison
7
to a profit maximizing/cost
8
the equality of marginal cost with demand under different conditions
than would occur
9
under
work
pure
competition.
IO
organizations
11
responsiveness
12
tendency
13
community
Dr. William
and responsibilities.”
to over-invest.
He indicated
achieving
on
public
charged with public
that such an organization
has a
to note that a refrain in the Postal Service
is the need to grow volume and increase investment.
of corporate practices has been a major goal in the corporate
sector in recent years as companies
16
Witness Tayman,
17
of any benchmarking
18
Also,
19
documents
20
spending
72
pioneering
theory for ai enterprise
It is interesting
15
71
does not operate efficiently,
Niskannen’s
provided the microeconomic
The benchmarking
14
minimizing company)
when
in commenting
on investment
studies on investment
requested,
prepared
have attempted
the
Postal
by or for the Postal
by its counterparts,
was
Service
either in other advanced
Niskannen, William A., Bureaucracy and Representative
Tr. 2600-l.
-49-
efficient.
policies, indicated that he was unaware
standards
Service
to become increasingly
relating to equipment
unable
to produce
evaluating
in place.72
any
internal
the level of capital
industrial nations or by its major
Government,
Chicago, Aldine, 1971
Docket No. RZOOO-1
~CA-T-~
1
competitors
2
witness
Kingsley
3
Service
since the beginning
4
compared
5
nations.73 These responses tend to confirm that there are no benchmarking
6
7
in the United States, such as FedEx or UPS.
has stated that there are no studies produced
of 1998 evaluating
to the automation
In a response
achieved
Postal Service
by or for the Postal
its flat processing
by its counterparts
to the interrogatory
(AAP), the USPS recites Professor
8
9
10
11
12
13
14
Moreover,
automation
in other advanced
of the Association
Panzar’s direct testimony
as
industrial
studies.
of American
Publishers
in Docket No. R97-1:
However, the efficiency of the Postal Service operating plan is not an
issue for the analyst. As long as it is given that postal services will be
produced following Postal Service practices and procedures, the relevant
marginal and incremental costs for pricing purposes are those calculated
based on the Postal Service operating plan.”
It is clear that, on occasion,
15
(apparently
failing to meet plans)
16
whether
17
introduced
18
function
19
marginal
20
minimization
such
to this proceeding
that is not based
productivity
Toda’s
21
22
businesses
23
addition,
73
74
an investment
pricing
the USPS does not achieve
and has very limited,
budget
is efficient.
by Dr. Bouo,
is apparently
assumptions
conclusions
for the analysis
appear to have been operated
the industries
Accordingly,
a very different
case. The impact on Dr. Bouo’s
work was developed
if any, analyses
are relevant.
on the theoretical
its investment
Toda’s
AAPIUSPS-1, Tr. 21/8611.
-5O-
comments,
The behavior
of a cost
of cost minimization
assumption
and
from the cost
needs further explanation.
under an output-maximizing
ANMIUSPS-Tl O-27, Tr. 5/l 578
verifying
of the Soviet economy.
were under various governmental
budget
regulations
Soviet
objective.
In
in acquiring the
Docket No. R2000-1
OCA-T-4
1
factors of production,
2
not set on a shadow
3
improper pricing could theoretically
4
an efficient competitive
5
and the prices of finished goods and intermediate
price basis. Accordingly,
a mixture of operating
is a major purchaser
7
specialized
8
services.
Therefore,
Postal Service may, through
9
operating
decisions,
affect factor prices.
10
inefficient
factor allocation,
11
that result in the distortion
12
prices different from those that would normally occur in a competitive
13
Bozzo did not address the implications
16
and
of monopsonistic
of goods and services,
even has some degree
9.
inefficiencies
arrive at a situation different from that obtained from
6
14
15
were
equilibrium.‘5
The Postal Service
machinery
products
power
as well as monopolistic
in the purchase
competitive
input, production,
in achieving
the USPS may make purctiasing
of factor prices, resulting
of some types
of
power in the sale of certain
its resource
Accordingly,
and possibly
and
an economically
and investment
in the generation
decisions
of factor input
environment.
Dr.
for the labor demand function
In conclusion, the theory underlying Dr. Bouo’s model has not
been shown to be derived from the appropriate economic theorv.
In my opinion, the Postal Service has not demonstrated
economic
theory.
I have also noted deficiencies
17
supported
by appropriate
18
statement
of the function
19
the network, the time frame, and cost minimization.
being estimated,
that Dr. Bozzo’s model is
the selection
of variables,
in the
the treatment
of
7s
Toda, op.cit at 264. Dr. Toda actually found that some of the Soviet industries operated
efficiently (a result he did not expect to find) and that some industries operated ineflciently. Regardless of
the empirical findings. the theory is applicable insofar as it applies to firms that do not minimize costs. A
partial explanation of Dr. Toda’s empirical findings would be that the Soviet economy actually did, in some
cases, operate efficiently.
-51-
Docket No. R2000-1
1
2
C.
OCA-T-4
Criterion 3: The Study Should Include a Discussion of the Modeling
Approach and How It Is Consistent with the Underlvina Data.
3
4
5
1.
6
Previous,
Another problem associated with Dr. Bozzo’s work is his modeling
of capital (as opposed to the accuracy of the QICAP variable itself).
The use of capital affects future Postal Service costs.
current,
7
Service and are focused
8
projection
9
by the Postal Service:
deployment
11
us to increase
12
additional
13
mail.” 76
14
15
and mechanization
18
Econometric
19
Michael
20
econometric
76
throughout
The use of capital and the
to be created has been highlighted
Service
equipment
barcodes
work on fixed effects
the analysis
Models,
lntriligator
gains.
Techniques,
and
estimation
to the Postal
continued
its accelerated
and software.
This allowed
on letter mail, while deploying
letter, flat, and package
Dr. Hsiao has useful guidance on the modeling
and investment in economic models.
Dr. Hsiao’s pioneering
indirectly
1999, the Postal
are important
to sort the higher volumes of automated
(4
17
productivity
our ability to place accurate
equipment
efforts
budget and efficiencies
“During
of automation
investment
on achieving
of the investment
10
16
and future
Ronald
has been referenced
of volume variability.”
and Applications,
Bodkin,
addresses
of capital
directly
or
A quote from the textbook
co-authored
the
issue
by Dr. Hsiao with
of capital
in the
process:
United States Postal Service, f999 Comprehensive Statement on Postal Operations at 50
77
Cheng Hsiao, Analysis of Panel Data, Cambridge University Press, 1986. Another book
referenced is Econometric Mode/s, Techniques, and Applications, with Michael D. Intriligator, Ronald G.
Bodkin, and Cheng Hsiao, Prentice Hall, 1996, Second edition.
-52-
Docket No. R2000-1
OCA-T-4
1
2
3
4
5
6
7
8
9
IO
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
. ..The inputs should, in theory, be measured in terms of services of the input
per unit of time, but such data are generally not available, so they are instead
typically measured by the amount of the input utilized or available in the
production process. Labor input is typically measured as labor hours employed
per year, but it is also sometimes measured as number of employees.
Capital
input is typically measured by the net capital stock (net of depreciation), but it is
also sometimes measured by the gross capital stock and by certain direct
measures (e.g., number of tractors in use for agriculture)....
Of these variables, the one that creates the most problems is the capital
input. While data on output and labor are generally available, data on capital are
either not available or of questionable validity. Enormously complex problems of
measurement arise with respect to capital as an input to the production process.
First, capital generally represents an aggregation of very diverse components,
Even
including various types of machines, plant, inventories, and so on.
machines of the same type may cause aggregation problems if they are of
different vintages, with different technical characteristics,
particularly different
levels of productivity or efficiency.
Second, some capital is rented but most is
owned. For the capital stock that is owned, however, it is necessary to impute
rental values to take account of capital services. Such an imputation depends, in
pat-t, on depreciation of capital. Depreciation figures are generally unrealistic,
however, since they entail both tax avoidance by the firm and the creation by the
tax authorities of incentives to invest via accelerated depreciation.
Third there is
the problem of capacity utilization. Only capital that is actually utilized should be
treated as an input, so measured capital should be adjusted for capacity
Accurate data on capacity utilization are, however, difficult or
utilization.
impossible to obtain.?+ Other problems could be cited as well, but all these
suggest that, if at all possible, the use of an explicit measure of the capital stock
should be avoided, since it is virtually impossible to find data adequately
representing capital stock.78
i:
Ei
11 An early approach to capacity utilization was to assume that the percentage of capital utilized
was the same as the percentage of labor utilized and thus to reduce the total capital available by
the (labor) unemployment rate, as in Solow (1957). More recently, there are various methods
used to adjust capital for the degree of utilization which are independent of the unemployment
rate, For example, the Wharton capacity utilization rate method assumes 100% utilization at local
peaks of the industry output series, with capacity assumed to grow linearly from peak to peak.
Capacity utilization is then obtained as the percentage of output relative to the value obtained on
the linearly interpolated capacity series.
35
36
:;
Intriligator, Bodkin, and Hsiao. op.cit. at 284-85.
- 53 -
Docket No. R2000-1
1
2
OCA-T-4
Dr. Bozzo has not modeled capital in a way that would meet
the criteria outlined bv Drs. Intriliaator. Bodkin, and Hsiao.
0))
3
Dr. Bozzo’s approach
does not meet the criteria outlined in the above quote.
4
Bozzo has no measure
of capacity
utilization
5
factory batch processing/job
6
capacity utilization
7
important variable omitted from the analysis.
8
is appropriately
9
or as an endogenous
in his equations,
shop type of process.
In analyzing
has a strong impact on cost performance.
modeled
as an exogenous
Mail processing
In addition,
factory operations,
This is a potentially
it is not clear whether
On the subject of the capital variable, Dr. Bozzo indicates that:
11
12
13
14
15
16
17
18
19
20
21
With respect to the capital variable, my inclusion of the capital quantity
rather than price is appropriate for a treatment of capital as a “quasi-fixed”
factor. While I would expect capital costs to be volume-variable to some
degree (possibly to the same degree as labor costs as discussed in
USPS-T-15 at pages 3941) I would nevertheless expect that the nature
of the Postal Service’s capital planning and deployment processes is such
that capital and labor are not simultaneously determined, but rather that
the available capital is taken as a “given” when labor work assignments
are made.79
situation
23
econometric
79
eo
is impossible.
estimating
that capital is neither exogenous
Accordingly,
very
capital
equation system.
10
22
is a
variable (as I believe Dr. Bozzo has done),
variable in a simultaneous
Dr. Bozzo indicates
Dr.
some
model is needed.
OCAIUSPS-T-15-56(b), Tr. 15/6359.
Tr. 1516414.
-54-
review
nor endogenous;m
of the
specification
such a
of the
Docket No. RZOOO-1
2.
Witness
4
mail processing
5
6
7
8
9
10
11
12
13
14
15
OCA-T-4
Witness Degen’s testimony is a major input to the understanding
and modelina of postal mail processina.
Degen presents
information
on the physical and operational
nature of
as related to volume variability:
. ..I show that the structure of mail processing operations does not
support the assumption that volume-variability
factors should uniformly
equal 100 percent.
My analysis of the structure of mail processing
operations also reveals that the pooled regression approach advocated by
OCA witness Smith and the cross-sectional
analysis favored by UPS
witness Neels, in Docket No. R97-1, potentially ignores (sic) features of
the Postal Service network and operations that are vital to distinguishing
the cost effects of volume changes from the effects of non-volume
factorsa
Mr. Degen raises two important issues in his testimony?
16
l
Mail processing
operations
have cost causing
characteristics
17
location, service area, and role within the Postal Service’s
18
change as a result of a small, sustained
19
l
For a small, sustained,
20
factors
21
plants.
22
Witness
remaining
increase
in national
the same, volume will increase workload
23
processing
plants interact-suggests
24
be evaluated
25
activity costs on the mail processing
of the postal network--the
that volume variability
at the plant or inter-plant
network that will not
increase in volume,
and representative
Degen’s discussion
related to their
RPW, all other
in all, or nearly all,
ways in which the mail
should more appropriately
facilities network level, rather than in terms of
plant floor. In examining the current Postal Service
81
USPS-T-16 at 4, line 23 through 5, line 6.
82
USPS-T-16 at 6, lines 18-23.
- 55 -
Docket No. R2000-1
OCA-T-4
1
network, he notes in Section 2 of his testimony
2
delivery units are networked,
3
well as other
4
intermediate
5
offices.
with mail processing
He indicates
transshipment
In addition,
that
processing
facilities
and processing
for specialized
7
Standard
8
and also sort outgoing
9
parcel Standard
being performed
plants
can sort mail as well serve
(BMCs) constitute
Standard
parcels to other BMCs.
network
of
BMCs sort incoming
The role of BMCs in processing
indicates
of pallets (no piece sortation
12
addition of nodes as the nation has grown and its population
that the network
of processing
Mr. Degen concludes
plants
affects the workload
in the entire network.
15
increase
volume,
16
processing
plants,
17
processing
labor costs and mail volumes.”
18
additional
in national
of letters and flats).
influenced
by the interaction
21
of the relationship
between
reinforces
has changed.
workload
He continues
of mail processing
- 56 -
growth,
of
for mail
between
“...I must conclude
mail
that the
the network.“a3
that postal costs are strongly
plants and that the longer-run
cost and volume is appropriate-&.
USPS-T-16 at 15, lines 9-12 and at 15, lines 20-21
distribution
of the relationship
my conclusion
the
that national volume growth
and hence of volume-related
is a key element of my analysis
testimony
distribution
He states, “The geographic
volumes will cause workload growth throughout
Mr. Degen’s
Mr. Degen
is not static, but has involved
in Section 3 of his testimony
14
20
non-
Mail (A) varies, but it usually involves sack, tray, and bundle sorting
11
83
as
Mail parcels to 5 digit ZIP codes for delivery units in their service territories,
and the cross-docking
19
a separate
Mail (A) and (B).
10
13
in large plants as
points for various sections of the network.
the 21 Bulk Mail Centers
6
that over 30,000 post offices and other
considering
analysis
volume, not
Docket No. R2000-1
OCA-T-4
1
in terms of its behavior
2
volume is adjusted:
3
on total cost.
4
the arithmetic means of cost data appears to be more appropriate
5
approach.
6
costs (the types of costs that would vary as the nodes of the network
7
delineated
8
Bozzo.
10
such an approach
Accordingly,
Theoretically,
plant, but rather on an overall basis as
would look at the effect of a change in volume
the “between”
analysis presented
by Dr. Bouo,
rather than the short- run cost estimation
5 of his testimony,
analysis of the impact of volume growth.
Mr. Degen
extensively
based on
than is a fixed effects
one strives to more closely attain the estimation
by Mr. Degen),
In Section
9
in any one processing
of longer-run
changed
presented
presents
as
by Dr.
a graphical
To quote Mr. Degen:
11
12
13
14
15
16
17
18
19
20
In questioning Dr. Bradley on his testimony in Docket No. R97-1, the
Commission used a plot of TPH and hours from the manual letter cost
pool to imply that visual inspection of the plot indicated 100 percent
volume-variability for that cost pool. Dr. Bozzo thoroughly addresses the
issue of graphical representation
and analysis of the MODS data in his
testimony, but I would also like to discuss it here because the pictures
succinctly illustrate how ignoring non-volume characteristics of plants can
lead to a biased,
misleading
understanding
of the hours-volume
relationship. u
.
21
Mr. Degen maintains that a graph of hours against volume can result in the erroneous
22
conclusion
23
opinion, is caused by the absence
24
However, the argument
25
be discussed
ed
that hours will vary in direct proportion to volume.
of information
on network and plant characteristics.
for 100 percent volume variability
subsequently.
USPS-T-16 at 24, lines 6 through 13
- 57 -
The error, in Mr. Degen’s
is visually compelling,
as will
Docket No. R2000-1
OCA-T-4
The issue of the correct estimation
examining
Mr. Degen’s graphsa
three techniques
of volume variability
is best addressed
Mr. Degen’s graphs can be used to justify any of the
under consideration
in this case-fixed
As will be shown, the fixed effects approach
effects, pooled, or “between.”
is unsuitable:
a simple review of the data
shows that the eye (and economic logic) suggests the fixed effects approach
Figure
structure
“true”
1 of Mr. Degen’s
of a mail processing
or
component
“underlying”
cost
of the hours/pieces
by
testimony,
reproduced
operation
for a hypothetical
structure
he
means
here, shows
the
the “true”
mail processing
systematic,
is wrong.
plant.
cost
By
non-stochastic
relationship.
Figure 1
The Underlying Cost Structure for a Plant
10
85
I do not imply that Mr. Degen would agree with any of my analysis; I would expect him to
disagree. I use his graphs to show that a convincing argument can be made for the possibility of
essentially 100 percent volume variability.
- 58 -
Docket Non R2000-1
1
OCA-T-4
Volume variability
is less than 100 percent for the hypothetical
plant operation,
plant in Figure 1 ,86
2
At some times during
the plant will be operating
3
volume (suggesting
4
be at a lower volume of TPH (with a lower level of capacity utilization).
a high level of capacity utilization),
at relatively
high
and at other times the plant will
Figure 2
Observable
Data from the Underlying
Cost Structure
with Random Noise for One Plant
Total
In Figure 2 of his testimony,
7
8
same plant generated
9
relationship
by adding
Piece
Handlings
Mr. Degen shows
random
simulated
noise to the underlying
sample data for the
hours and pieces
plotted in Figure 1.
86
This is exactly what one would expect, given that this is a short-run diagram relating small
changes in hours and TPH.
-59-
Docket No. RZOOO-1
OCA-T-4
Figure 3
Data for Ten Plants with Similar Cost Structures
but Different Levels of Efficiency
Illustrating
True Cost Structure
.
:
. *...: *..‘.E *: .
*.
.
Ia .‘... .- **-.
H . .- . .
..*
.*
. . ..
Total
1
Piece
Handlings
Mr. Degen’s Figure 3 presents ten plants with cost structures
2
in Figure 2. but with different levels of efticiency.87
3
line analogous
4
ten lines, all of them short run.
81
similar to the plant
For each plant, Mr. Degen plotted a
to that plotted in Figure 1. Accordingly,
there are ten sets of points and
Mr. Degen and Dr. Bozzo attribute the differences in efticiency~~to differences in networks and
other factors not associated with volume Of mail. Nevertheless, the Postal Service has extensive
testimony and comments on investment and efforts to achieve lower costs. Treating these fixed effects
factors as exogenous rather than endogenous to the capital investment process seems to be wrong.
Docket No. R2000-1
OCA-T-4
Figure 4
Data for Ten Plants with Similar Cost Structures
but Different Levels of Efficiency
Illustrating Misinterpretation of Cost Structure
Total Piece Handlings
1
In Figure 4, the lines presented
2
examined
3
suppressed.
4
data as in Figure 3--ten separate
5
combined
88
short-term
hypotheses
in Figure 3, which
of the relationship
represented
between
An overall trend line is added to the diagram.
lines for ten facilities--the
the formerly
hours and pieces, are
Instead of visualizing the
data are considered
on a
basis.”
Mr. Degen’s title for Figure 3: referencing
cost structure is short run. Similarly, the title
Structure,” was included in the reproduced figure.
structure is the line he has labeled “100% Volume
the “true cost structure;” is correct in the sense that the
for Figure 4, referencing a “Misinterpretation of Cost
but, in contrast to Mr. Degen, I believe that the true cost
Variability”.
-61-
Docket No. R2000-1
1
Whatever
2
lines one looks
3
alternatively
4
that additional
5
inadequate
6
OCA-T-4
interpretation
at-i.e.
one wishes to give to the data is dependent
one could derive
a fixed effects
on which
model from Figure
3; or
one could define a pooled model from a review of Figure 4, recognizing
variables
would be needed
and that any two variables
approach
is
insofar as it may omit important information.
From a review of the graphs, two distinctly
from the underlying
determined
are possible.
The conclusion
6
specified the choice of model; all that then remains is the model estimation.
9
is then the selection
Analysis
is essentially
alternatives
,7
of the appropriate
model
different
line for estimation.
Figure 5
of Response to Volume
rplHt
Total Pg,“, Handli in gs
-62.
Growth
once one has
The issue
Docket No. R2000-1
OCA-T-4
Turning to Mr. Degen’s Figure 5, two plants are examined:
Assume that Plant A is designed
the mail processing
network.
and sized correctly,
operates
operation.
Assume that Plant B is designed
in the short
run, and the line shows
TPH, and that the optimal capacity
IO
11
designed,
these
and sized correctly
along the line. The two most important
operating
various
levels
for a higher level of
points in the diagram
the real labor costs of processing
at plant design
capacity--the
based on the evolving mail processing
mail at each of
level for which they were
network as described
by Mr. Degen.
Figure 5 has two types of plots in it. The facility by facility plots (labeled “Plant A”
12
and “Plant B”) are the types of plots that both Dr. Bradley and Dr. Bozzo generate
13
estimate.
These are short-term
14
treatment
of the data on a pooled
15
would estimate the line AC.
16
data and an underlying
17
of
is at point C. Again, on a short-run basis the plant
are points A and C. They represent
the plants when
based on Mr. Degen’s theory of
Assume that optimal capacity is at “A”, but that the plant
frequently
may operate anywhere
Plant A and Plant 6.
plants,
Alternatively,
one could allow for the
basis or cross sectional
basis. In that case, one
Such a modeling approach
longer-term
The mail-processing
sized
plots of data.
capacity expansion
would be consistent
variously
and
in a real world
19
properties.
20
plants, but based on confirmed
21
cross sectional
22
generated.
23
accounted
A pooled regression
regression
The appropriate
environment
line could be generated.
data from approximately
line based on the arithmetic
econometric
techniques
Accordingly,
costs
exhibit
there are
stochastic
It would be based, not on two
300 plants.
Alternatively,
a
means of the plants could be
and variables
for in order to avoid problems of omitted variables.
-63.
with the
path.
network consists of over 300 plants.
18
and
would need to be
The results could be 100
Docket No. R2000-1
OCA-T-4
1
percent
volume variable, or some other number either greater than (or less than) 100
2
percent
volume
3
variables
4
been performed
5
and “between”
6
the best currently available.
7
variable.
were used.
sectional
in this case.
regressions
approach
9
are averaged
results
would
not be known
until the appropriate
Such an analysis correctly using all relevant variables has not yet
Of the approaches
8
The
However,
on a preliminary
in Dr. Bozzo’s testimony,
presented
(i.e., a mean is determined);
it would
that the cross
For each mail processing
plant, the data
a regression
approach,
analysis is then performed
the sites.
11
appears
and based on Mr. Degen’s
12
as plant size varies, based on the plants sized for the postal network.
to be superior to either the fixed effects or pooled models.
The results from the various
14
Table 5.
I have indicated
15
generated
16
data are assumed
17
i.e. all of the cross sectional
18
aspects.
19
documented,
20
the Commission
21
find this to be the “least bad” model.
22
model in view of the underlying
23
adoption
models considered
that the “between”
data-thereby
The major statistical
problems
It examines
In general,
and longer-
term
with the model have been well
that some action is necessary
I do not recommend
problems
model
and the line has all of the plants--
having both short-term
associated
costs
cross sectional
However, at this point, it is the only model “left standing.”
should conclude
testimony,
model, a type of cross sectional
equilibrium,
on
by Dr. Bozzo are set forth in
by Dr. Bozzo, is the “least bad” of the models.
to show a longer-run
but also
appear
IO
13
This is a cross sectional
which are unsatisfactory
by Dr. Bozzo,
may be the “least bad.”
basis, there are the pooled
Accordingly,
in adopting
adoption
a model, I
of the “between”
with the data and the study.
if
I recommend
of 100 percent variability until a different approach is shown to be reasonable.
Docket No. R2000-1
OCA-T-4
Variabilities-Dr.
Bradley,
Table 5
Fixed Effects, Between,
Pooled, and Random
Variabilities
Using Different Methods
Dr. Bouo
Fixed
BCS Sorting
OCR Sorting
FSM Sorting
LSM
SPBS Non Priority
SPBS Priority
Manual Flats
Manual Letters
Manual Parcels
Man1 Priority Sorting
Cancl. And Mail Prep
0.945
0.895
0.786
0.918
0.905
0.469
0.802
0.866
0.797
0.395
0.448
0.654
0.751
0.817
0.954
0.641
0.641
0.772
0.735
0.522
0.522
0.549
1.044
1.101
1.026
0.913
0.889
0.889
0.963
0.906
0.730
0.748
0.845
Random
Total Cost
0.916
0.821
0.880
0.918
0.662
0.662
0.803
0.790
0.615
0.627
0.569
1.043.841
219,070
lJ42.369
78,765
283.275
82,447
459,933
1,563,963
60,593
259,762
295,957
0.931
0.862
0.913
0.922
0.724
0.724
0.842
0.845
0.645
0.642
0.643
Total
5,389,975
Attributable
Costs Based on Various
Fixed
Effects
Total Cost Dr. Bradley
&QaQszQQQ~~m~
BCS Sorting
OCR Sorting
FSM Sorting
LSM
SPBS Non Priority
SPBS Priority
Manual Flats
Manual Letters
Manual Parcels
Man1Pri Mail Sorting
Cancl and Mail Prep
Total
Variabilities
Dr. Bouo
Between
Pooled
Random
Effects
1,043,841
219,070
1,042,369
78,765
283,275
82,447
459,933
1,563,963
60,593
259,762
295,957
986,430
172.189
956,895
71,282
132.856
66,122
398,302
1,246.479
23,934
116,373
193,556
934,238
164,522
851.615
75,142
181,579
52,849
355,068
1,149,513
31,630
135,596
162,480
1,089,770
241,196
1,069,471
71,912
251,831
73,295
442,915
1,416,950
44,233
194,302
250,084
971,816
188,838
951,683
72,621
205,091
59,692
387,264
I.321549
39,082
166,767
190,300
956,158
179,856
917,285
72,306
187.528
54,580
369,326
1,235,531
37,265
162,871
168,400
5.389,975
4,364,418
4,094.231
5145,960
4.554,704
4J41.106
-65-
Docket No. R2000-1
OCA-T-4
3.
Witness Degen’s testimony is consistent with the application of
intuition and common sense that indicates the volume variability for
mail processina aporoaches 100 percent.
The above analysis of Mr. Degen’s
5
looked at from simply the perspective
In addressing
6
testimony
is substantiated
if the problem is
of intuition and common sense.
the issue of data and modeling, Dr. Bouo
states in his testimony:
7
8
9
During the hearings on the Postal Service’s direct case in Docket
No. R97-1, Chairman Gleiman asked Dr. Bradley to confirm the
intuition
IO
11
12
13
. ..that if costs vary 100 percent with volume, the graph of
those costs and the volume data points should resemble a
straight line with a l-to-l
slope. Docket No. R97-1, Tr.
1115578 at 4-6.
14
15
16
17
18
19
Dr. Bradley agreed, and even added that the line should go
through the origin (Id., at 8-9, 11)~~ In my opinion, Dr. Bradley
should not have confirmed Chairman Gleiman’s intuition.
It has
been understood since Docket No. R71-1 that to measure “volumevariability,” it is necessary to hold constant the non-volume factors
that affect costsaP
Dr. Bradley’s statement that the line should additionally pass through the
21
origin was in error. As a general matter, the cost surface passing through the
origin is neither necessary nor sufficient for the 100 percent volume-variability
result.
2
24
Dr. Bozzo
apparently
believes
makes
the bivariate
graphs
25
process
26
compelling
27
contains
plots of the number
28
activities
studied by Dr. Bozzo.
29
graphs are based on Dr. Bradley’s
89
in showing
that hours
that the multivariate
nature
irrelevant.
the graphs
However,
and TPH vary together
of hours and TPH for some
of the modeling
closely.
The Appendix
of the mail processing
Dr. BOZZO has referred to Dr. Bradley’s
data.
are visually
data, so the
The graphs are open to the same criticisms
USPS-T-15 at 59, lines 4 through 13.
-66-
Docket No. R2000-1
OCA-T-4
1
voiced in Docket No. R97-1.
Only two pieces of data are plotted.
2
information
in Dr. Bradley’s data set and which was actually collected
3
from the field operations
4
that data are not denoted
5
data set in order to be consistent
6
concluded
7
even 100 percent
8
conclusion.
9
ordinary
actually contained
(and remaining after his scrubbing)
by accounting
period.
Data were obtained from Dr. Bradley’s
with Dr. Bozzo’s
volume variability.
This would
least squares
associated
11
omitted variables,
appear
comments.
I have previously
This is a simple and intuitively
not consider
lack of variables,
etc.) has the characteristics
OLS Summary
A!a!ltY
ocs
Table 6
by Selected
0.77
0.94
0.97
0.96
0.90
0.90
13
not consider the myriad of issues that contribute
14
relationships.
15
through
indefensible
insofar as the regression
to the understanding
the lines do show that a simple visualization
a high level of volume
variability,
The regression runs are provided in Library Reference OCA-LR-I-2.
- 67 -
of issues
Activities
RSauare
lines are econometrically
An
in Table 8.%
0.19
1.oi
0.98
1.01
1.05
1.09
The regression
initial
times series nature of the data,
presented
12
the data suggests
plausible
any of the myriad
Rearessor
BCS
LSM
FSM
MANL
MANF
However,
possibly
to be the case for a number of the activities.
line (which does
with serial correlation,
all of t.he
is also present, recognizing
that the plots are consistent with a high degree of volume variability,
10
90
However,
does
of the TPHlhours
of a straight line
resulting
in a high R
Docket No, R2000-1
OCA-T-4
1
square.
2
or the pooled model.
A modeling approach
3
4
4.
5
and TPH.
7
amount
8
scrubbed
9
dummy variables.
10
11
analysis included
There is, however,
of information
presented.
model
a major factor driving cost, intuition appears
a large number of variables
a difference
between
the number
in addition to hours
of variables
and the
All of the variables
were either derived from the
data of hours and TPH via cross products,
or were simply time trend or
Except for time trend and seasonal
information,
the actual data show
that TPH and hours vary together closely.
The issues under consideration
12
appropriate
13
yield 100 percent variability.
14
compelling
15
probably
16
might also reach such a conclusion.
17
18
19
with the data would be the “between”
In terms of identifying
reasonable,
Dr. Bradley’s
6
consistent
variables,
the underlying
and suggest
are the correct
methodology,
and whether
of the relationship,
such estimation
The graphs derived from the application
the existence
at or approaching
estimation
100 percent.
would
of intuition are
of a relationship
for high volume variability,
A correctly
econometric
constructed
model
In conclusion,
Dr. Bouo’s
choice of econometric
model
inconsistent with the economic modelinq of the postal process.
5.
The level of econometric
20
currently
by Dr. Bozzo is clear.
21
work is the inappropriate
22
more appropriate
23
Bozzo’s econometric
sophistication
evidenced
The major concern
model at this time. The microeconomic
-68-
by Dr. Bradley and
with their econometric
choice of a model for estimation.
model are at best muddled.
previously
is
The “between”
assumptions
estimation
model is the
underlying
Dr.
We are faced with analyses of non
Docket No. R2000-1
Cost minimizing
OCA-T-4
firms, cost functions
that have become labor demand
state of the art price theory which is not organized
Progression.
Furthermore,
omitted and variables
5
6
D.
8
The deficiencies
fundamentally
11
unsuitable
12
The “between”
of the fixed effects approach
identical to that of Dr. Bradley.
in the availability
13
pooled
14
(among other variable deficiencies),
model unsatisfactory.
Without
The use of cross sectional
Dr. Bozzo’s overall approach
Accordingly,
is
the fixed effects model is
The “between”
17
model is subject to deficiencies
18
number of criticisms of the econometric
19
as outlined by Dr. Bradley.
20
demand
21
relevance
based
also render
an application
a measure
of capacity,
capital,
model has data available
estimation
deficiencies
However, the “between”
on size of the facilities.
In addition to deficiencies
need potential
available.
of the
and networks
error.
of costs as facilities
on a cross sectional
in the set of variables
to the current proceeding
the methodology
of variables
models allows for an analysis
vary.
23
to the
as it has been applied have been
the pooled model is subject to specification
16
22
Is Suitable
at this point.
Deficiencies
15
which
model is currently the “least bad” model available.
in Docket No. R97-1, where it was rejected.
10
or logical
data are deficient, both in terms of variables
Criterion 4: A Correct Estimation Procedure
Estimation Needs at Hand Should Be Used.
1.
outlined
fashion
and
included (such as QICAP and the manual ratio).
7
9
the underlying
in a coherent
functions,
basis, but the
There have been a
of the “between”
model,
model permits an analysis of labor
Accordingly,
the “between”
model
has
and is the “least bad” model.
in Dr. Bozzo’s current models, several major areas of
improvement.
-69-
First, at the activity level, investment
Docket No. RZOOO-1
OCA-T-4
1
almost certainly
2
function of TPF or TPH, so in a sense investment
3
handling process.
4
labor hours.
5
each activity as if it were independent
6
plant.
7
requirements,
and network
a
interrelated.
This potential
9
workforce,
10
It may be appropriate
Dr. Bouo
However,
one
would
Dr. Bozzo’s
the
efficiency,
of the ten activities
Dr. Bouo’s
and
model treats
labor
modeled
usage,
investment
to be significantly
or the facilities.
It is difficult to imagine that the cost of
has not adopted
is not acceptable.
could be improved,
15
could be stated, one could perform the modeling
16
such an approach
17
“between”
case as the “least bad.”
ia
discussed,
I do not view its adoption as appropriate,
the appropriate
variables
is not appropriate
developed,
a correct
Assuming
estimation
that the data
and a clear economic
Accordingly,
both in terms of the interrelationships
21
estimation
22
of costs as a result of First Handling
modeling formations
of activities
and cost is appropriate.
Pieces
-7o-
I have
need to be considered,
and whether
In modeling
but
we are left with the
However, in view of the many uncertainties
20
theory
effort using a pooled approach;
at this time.
basis, alternative
of the
of other work performed in the plant.
Dr. Bozzo
fixed effects approach
of investment
both investment
could be due to some sharing
In conclusion,
procedure.
On a longer-term
variable to the mail
interrelationship
14
19
Second,
is, in turn, a
of every other activity in the mail processing
expect
aspects
investment
is an endogenous
has not examined this area.
the management,
2.
However,
to model simultaneously
performing work in one activity is independent
11
12
13
has a major impact on the costs.
some simultaneous
activities, the incidence
rather than TPF or TPH should
be
Docket No. R2000-1
OCA-T4
1
examined.
2
be, but it is clear that there are a number of options that need to be explored.
In summary, it is not yet even clear what the best modeling
3
4
5
E.
6
Dr. Bouo
approach would
Criterion
5:
Results for Econometric
Equations
and Alternative
Econometric Analyses Should Include a Discussion of the Values, Signs,
and Other Relevant Information for the Variables.
on
presents a variety of alternative
7
variants
his preferred
a
approaches
9
Bozzo has ignored Criterion
methods.
econometric
Fundamental
have not been explored.
Accordingly,
5, strictly speaking;
analyses,
changes
11
be a rework of the economic
12
likely an exploration
13
output and investment,
14
to be a significant
15
to be the application
16
analyses
17
or other relevant information
product
an improvement
production,
and an improved microeconomic
upgrading
have not yet been performed
simultaneous
analysis.
techniques.
and any discussion
for variables
is moot.
-71-
First, there needs to
in presentation
of the quality and availability of data.
of suitable estimating
modeling
it is also clear that this requirement
needs to be applied to the study after the study has been redone.
of multiple
new
while it is difficult to say that Dr.
IO
theory-with
and
but they are all
Therefore,
and more
determination
of
Second, there needs
Finally, there needs
the most important
at this time of values, signs,
Docket No. R2000-1
1
VII.
OCA-T-4
CONCLUSIONS
2
3
A.
4
The Commission
Variabilities Were Traditionally Assumed To Be 100 Percent.
Studv. Performed bv Dr. Bradlev. Was Seriouslv Deficient.
costs.
has always applied a variability of 100 percent when attributing
5
mail-processing
6
presented
a witness,
Dr. Bradley, who proposed
7
processing
operations
to measure volume variability.
a
the change in the estimated volume of mail processed
9
required
In Docket No. R97-1, the Postal Service reviewed the policy and
to process that volume.
labor hours for mail processing
11
arriving at an estimate
12
variabilities
13
witness
14
percent, and the Postal Service contended
15
than the traditional
16
of volume variability.
Dr. Bradley’s
objections
18
regressions.
19
Bradley’s
approach
20
necessary
before the Commission
The
techniques,
23
in Appendix
change
that his variabilities
change
in
in the volume
of mail,
that the resulting
volume
by Postal
were significantly
on several
to Dr. Bradley’s
and his handling
Commission’s
The Commission
Bradley’s
hours of labor
the percentage
cost pools could be applied
were raised
choice of regression
22
to analyze
Service
lower than
100
should be applied rather
100 percent variability used by the Commission.
17
21
with the estimated
He concluded
volume variabilities
model for mail
That model purported
for each percentage
for each of the several
Numerous
a new econometric
From this, he calculated
10
Degen.
The First
recommended
grounds
and
F to the Opinion.
decision
indicated
deficiencies
these problems
The Commission
- 72 -
his
of the data prior to running
his
specifically
rejected
Dr.
that additional
study
was
revised its approach to mail processing
found fundamental
model and discussed
model specifications,
variability.
ins the specifications
for Dr.
in both its opinion and in greater detail
recognized
that Dr. Bradley’s
model
Docket No. R2000-1
OCA-T-4
failed to consider the impact of capital.
The Commission
noted that Dr. Bradley did not
base his analysis upon a correctly specified cost function as indicated
production.
The Commission
for analysis,
citing
also faulted Dr. Bradley’s
his several
seemingly
scrubbing
of the data prior to running
alternative
approaches
the fixed-effects
arbitrary
regressions.
and further analysis.
model selected
models such as the pooled or the “between”
in fact, had numerous
method of preparing the data
restrictions
and over-zealous
The Commission
Finally, the Commission
by Dr. Bradley
by the theory of
recommended
clearly indicated
in lieu of other possible
model was not sufficiently
supported
and,
infirmities.
Dr. Bozzo’s Studv Is also Seriouslv Deficient.
10
B.
11
The
Postal
Service
12
analyzes
13
Commission’s
14
and that of other witnesses
15
Bradley’s
16
Bradley.
mail processing
has now presented
costs, critiques
R97-1 Opinion.
work were valid.
Significantly,
Dr. Bozzo’s
Dr. Bradley’s
testimony
study,
ia
the Commission
19
the
20
Commission.
21
regression
22
opinion.
Postal
prior
Dr. Bozzo
in Docket No. R97-1 and found that some criticisms of Dr.
In response,
presentation
continues
model that the Commission
Dr. Bozzo dresses
to the
Dr. Bozzo reviewed the work of Dr. Bradley
Dr. Bozzo modified
in Docket No. R97-1, closer inspection
Service’s
that further
and responds
the methodology
While Dr. Bozzo purports to present a study meeting the objections
17
regression
that
has
to ground
essentially
expressed
by
indicates a startling similarity to
been
soundly
rejected
by the
the analysis
on the fixed effects
rejected.‘in
the Docket No. R97-1
up Dr. Bradley’s defective cost function,
-73-
of Dr.
renaming it a labor
Docket No. R2000-1
OCA-T-4
1
demand
2
heretofore
3
has ever been utilized by the Commission
4
Postal
5
Commission
6
variability
7
function.
He adds at least two variables affected by volume, “QICAP,” and the
unmodeled
Service
“network” characteristic,
model is essentially
According
in considering
Dr. Bradley
is faced with continuing
to Dr. Bozzo, neither variable
segment
revisited.
3 costs.
Thus, without
to apply the traditional
The new
more, the
100 percent
volume
to the ten cost pools.
The Commission
may wish to attribute
mail-processing
costs for the ten cost
a
pools on the basis of a variability analysis other than that in Docket No. R71-1 on which
9
the Commission
has based its traditional
10
mail processing
information
11
appropriate
12
approach.
and data supplied
by the‘Postal
classical
economic
theories,
available,
the cross
sectional
“between”
13
presented,
although
14
in costs that are 95 percent attributable;
15
subject to error is, therefore, hardly worth the effort.
16
I recommend,
to apply
I do not advocate
instead,
Having independently
I conclude
model
its adoption.
Service and applied the
that upon the information
is the “least
bad”
variability
reject
Dr. Bozzo’s
continue
ia
Alternatively,
19
presented
20
which should be modified to reflect a volume variability of 100 percent.
by Dr. Bozzo.
model results
the use of the model, which is known to be
17
I recommend
now
of the models
In fact, the “between”
that the Commission
the traditional
reviewed the
the “between”
to the ten cost
model as the “least
I provided OCA witness Thompson
- 74 -
pools
study
in the
and
study.
bad” of the models
the list of those cost pools
Docket NO. R2000-1
C.
OCA-T-4
The Work that Has Been Reviewed
Modeling Effort.
Apparently,
approximately
Dr. Bradley
and Dr. Bouo
economic
additional
on a combined
effort involved
basis
has projected
in the completion
theoretical
modifications
spent
that there
of the work.
The
in the work as well as the extension
a
of the work
9
additional and/or improved data, and different estimating
to additional
I have discussed
activities,
additional
types
of mail processing
facilities,
approaches.
the work in terms of some of the criteria for evaluation
11
Appendix
12
work is not yet complete.
13
substantial
14
a final study.
I recommend
that the Commission
15
establishment
of a working
group to discuss,
16
data, and modeling
17
Obviously such a group would require the honest and effective participation
ia
parties involved.
19
have
theory is not set forth as clearly as is desired, so it is possible that
there would be substantial
10
the Latest Part of a Major
ten person years on the issue, and Dr. Bouo
would be a significant
underlying
Represents
F of the Commission’s
opinion in Docket No. R97-1.
Nevertheless,
set out in
By those standards,
we are faced with the distressing
the
fact that
effort as well as significant elapsed time has occurred with no production
Whether
approaches
through
evaluate,
and comment
in an effort to bring these
a working
20
deficiencies
21
need to be presented
22
focused
23
nature of the function being estimated.
group
in the work to be addressed.
on assumptions
and the Postal Service consider
or otherwise,
about homotheticity,
- 75 -
manner,
economic
the
on theoretical,
issues to a conclusion.
there
First, the underlying
in a more comprehensible
of
of all of the
are a number
economic
assumptions
with particular
efficiency,
of
networks,
emphasis
and the
Docket No. R2000-1
1
Second, there need to be improvements
2
variables
3
relationship
4
considered
5
meaningless,
6
current state.
QICAP,
of the
manual
in the data, particularly
ratio, and capacity
investment
(if the activity approach
data
to the
utilization,
activity
is pursued).
explanatory
variables
as related to the
It is important
being
estimated
that the
is carefully
Even if the QICAP variable were not
it would not measure the level of capital associated
Third, additional
7
a
OCA-T4
may be needed,
with an activity in its
particularly
in terms of
run aspects
of the mail
the network.
Fourth,
9
10
processing
11
longer-run
recognizing
process,
the short-run
aspects
and longer
fixed effects analysis
presented
is unsuitable;
a
analysis is needed.
Finally, my comments
12
the network
13
work.
This is the necessary
14
not adequate
15
Commission
16
in detail.
are based on a four-month
consequence
from a scientific
to recommend
examination
of the time constraints
analysis
point
the establishment
-76-
of view.
of Dr. Bozzo’s
of a rate case, but is
Accordingly,
I urge the
of a working group to consider this issue
APPENDIX
OCR OPERATIONS/ HOURS ON TPH
USING ONLY CONTINUOUS DATA FROM 8801-9613
INCLUDING OFFICES @ LEAST 39 OBS/LAG MODEL
USES 12 AP DUMMIES TO CAPTURE SEASONAL EFFECTS
Plot
of HRS'TPH.
Legend:
A = 1 obs,
B = 2 obs,
etc.
HRS ,
25000
20000
'
!
15000
!
10000
!
5000 !
,
0 ^
A
A
A
AA
A
AAAA
AA
AAA
AAAABRA
AA AAA
A
AA
AA
A
A
A
C
AB
A
BAA
A
A
AAAAA
A
AA
AABDB
BABBC AB
B AA BCDFBACABBA AA A A ADCCAB AA
A BAA
ACECFB ACA B ECCAACDDBAA A A
CE CDIDBHACCBBBC AABAB BAAFCCCCAA
ADBACDFIONOSHILHGBBCAACFCCDABCCA A
AFJliiDCLNTXQVPMYQMICDHBHDHBC
A A
HLPTXZYYSVZRVQPUQOMNODFBBDB
A
AACCOZZZZZZZZZZZZZZZZZSRNECFEBC
A
BHVSZZZZZZZZZZZZZZZXXIJKEABAB
BKLMWZZZZZZZZZZZZZZZVLJCDBAAA
KXZZZZZZZZZZZZZZZZZOEBABA
AKZZZZZZZZZZZZZZZTPJGBAABAA
FZZZZZZZZZZZZZZHFIA
A
RZZZZZZZZZZZJFAA
ZZZZZZZZTD
CZZZA
100000
40000
60000
TPH
NOTE: 13796 obs hidden.
A
A
A
80000
Docket No. R2000-1
OCA-T-4
Appendix
BCS
OPERATIONS/ HOURS ON TPH
ONLY
FROM 8801-9613
USING
CONTINUOUS
DATA
INCLUDING OFFICES @ LEAST 39 OBS/LAG MODEL
USES 12 AP DUMMIES TO CAPTURE SEASONAL EFFECTS
Plot of HRS'TPH. Legend: A = 1 obs, B = 2 obs, etc
HRS ,
50000 !
B
AA
A
40000 I
30000 !
20000 !
10000 ^
,
0 !
ABBD
A
AB A
AAAAA
A BAA
AAA
A
AA
B
BBB
A
ABBC B ACA AA
AAFABCAB BBH AB
A ADAED B B BB
A BHEECB BEE A A
A CDAEFBBB BC BC
ABBDCDCDIBBDEEC
BAD DCEEFCB DBBC A
ADHIIIHOPIFHCABFB A
AEFINQKQVRTNKEDGECA
CEGFLNFQTNMFBDC A
ABA
ELNHKLWSSHABAE A
CAOQPXVZZRLEBCA
ADINUWZYZZIDAFFGHA
COVZZZZZZNKDCCEFAAA A
AGZZZZZZZZQFFFDAB
DIZZZZZZZQVACFBBCA
A
CJZZZZZZZZLNB AAA
EWZZZZZZZZJDDAA
RZZZZZZZNGIC
A
NZZZZZZZOADEA A
NZZZZZZXDA ECA
ZZZZZZDB A AA
QZZZZZJA
RZZZZB
ZZP
TPH
NOTE: 19005 obs hidden.
A-2
A
Docket No. R2000-1
OCA-T-4
Appendix
FSM OPERATIONS/ HOURS ON TPH
USING ONLY CONTINUOUS DATA FROM 8801-9613
INCLUDING OFFICES 6 LEAST 39 OBS/LAG MODEL
USES 12 AP DUMMIES TO CAPTURE SEASONAL EFFECTS
Plot
of HRS'TPH.
Legend:
A = 1 obs,
B = 2 obs,
etc.
HRS ,
10000
-
AA
A
60000 '
50000 !
40000
-
30000
I
20000
!
10000 !
0 '
AAAA
A
AA
A
RA B
BA
A AD
AAAA
A
C
A
A
A A BAC
AAA A AB
AAB CBA
ABBBA
AABBAAA B
AA ACA
CBB DA
C DAAB
CB ACBBA
ARWlA AAABCBA
A ?&AA BA AAB AA
RAAAAA
BBAAAA
A BA AABCDDDCADC
ACCDEBDEHBHGFECDA
BACCGEFFGDBCBC
A BCIMGFFEGEBCBCAA
ABFHPUXPMCGAABDACGBB
CJSZZZTTHHBBCDC A
CHWZZZZZZOLCDA
DHZZZZZZZWLIDA
A EZZZZZZZZZPGCA
APZZZZZZZZZQD
GZZZZZZZZZUEA
EZZZZZZZZK
czzzzzzzv1
ZZZZZZZF
LZZZZZD
ZZZZP
CZA
A-3
A
Docket No. R2000-1
OCA-T-4
Appendix
LSM OPERATIONS/ HOURS ON TPH
USING ONLY CONTINUOUS DATA FROM 8801-9613
INCLUDING OFFICES @ LEAST 39 OBS/LAG MODEL
USES 12 AP DUMMIES TO CAPTURE SEASONAL EFFECTS
Plot of HRS'TPH. Legend: A = 1 obs, B = 2 obs, etc.
HRS ,
120000
-
A
A
A
100000 !
80000 !
60000 !
40000
!
20000
!
A
AAA
A
B
A
B
AAA
A AABAB BAB
A
AA
A
AAAAAAABD
AA
AB BDB AACCAA BAABCAAAA
A
A AEB AGCDC CAB CBAABAA
A
AB BAEADAB CA BAACAE A
B AABDDCCCDECCBBAD A DBAA
A AAC ECDBDBEBCBDDBCADAD
B
BHCCEHCFBBBADCBBCDB
AA
AA
AA BDAEAFABDHECBBBDAA
ABA
ADCCCCBGKEADBAAAA
AAABB ABCCDEJNCHABBBDA
ABAA ADEEEGNKUPIEDGBEE
B A BHDLIROUHJEKF B
AA B AJHFFSONPSMNCGAA
A
BBHEHMGOOVXUSNF
BACSKOLPVUWXZVOGGEB
BGJWVVRZZZZZZOKCAA
AACHPYZZWZZZZWTWHB
CBBJOZZYZZZWWZTJC
CAMVZZZZZZZZZQJA
HTZZZZZZZZZLD
EKZZZZZZZZZJF
JZZZZZZZZZOB
APZZZZZZZTC
AZZZZZZZI
HZZZZZZI
ZZZZZZA
zzzzz
A
ZZZT
0 ! zz
NOTE: 15313 obs hidden.
A-4
A
OCA-T-4
Appendix
Docket No. R2000-1
MANUAL LETTER OPERATIONS/ HOURS ON TPH
USING ONLY CONTINUOUS DATA FROM 8801-9613
INCLUDING OFFICES @ LEAST 39 OBS/LAG MODEL
USES I2 AP DUMMIES TO CAPTURE SEASONAL EFFECTS
Plot
Legend:
of HRS'TPH.
A = 1 obs,
B = 2 obs,
etc.
HRS ,
200000
!
A
AAAA
A AC
A
AAABB
A
A
AA
B AB BA
AAA4A
AA
BACCA
A
150000
^
A
'A
A
A
A
BAABA
AABBAC AA AA
AA BB
ABA BAC B A ARA
B
A ACA
A
AAEC B AA AAA BAA
A
A
CDBBEAA A AAA A C A
A
AAAACCB
DAB
AAA
A
A
100000
A A
^
A
50000
CA BC BD BBAA C CB A
B AE AEDCACCAABACAA
AACBBCEBACBCDDCBACA A
AADC AADBADDHGBCBBCCCABA
A
ABA AAGLDLIOJEECA AA A
BAGFFADCJQNLJDCDBCBBA A A
ACBGFINIXUQGMFGBFCEB AB
A JKMUSTZXXZTULJJFMEBAA
A
AGMVVYZZZZZZRKLKB EC A
DKNXZZZZZZZMTMFFBA A A
EZPZZZZZZZZZSLIKCA
GZZZZZZZZZZZTLHAAA
CZZZZZZZZZZZQCAAAA
UZZZZZZZZYZKE
ZZZZZZZZZPB
!
ZZZZZZWL
0 ! zzzz
Sf-fffffffffff~fffffffffff-fffffffffff-fffffffffff-fffffffffff”fffffffffff~fff
ffffffffAff
20000
0
120000
40000
60000
140000
TPH
NOTE: 21328 ohs hidden.
A-5
80000
100000
Docket No. R2000-1
OCA-T-4
Appendix
MANUAL FLAT OPERATIONS/ HOURS ON TPH
USING ONLY CONTINUOUS DATA FROM 8801-9613
INCLUDING OFFICES @ LEAST 39 OBS/LAG MODEL
USES 12 AP DUMMIES TO CAPTURE SEASONAL EFFECTS
Plot of HRS'TPH. Legend: A = 1 obs, B = 2 ohs, etc.
HRS ,
100000 !
A
B
A
AAA
A
A
AA
ABA
AABAAA
AA
BB C AA
A
B AA B ACA
B
BA
ACA
AA
A B B ACBA
80000 !
A
A
AAAA
A
AA
60000 .
40000 !
20000 !
I
,
0 -
A BABDAAAAB
50000
CAB
BA
20000
TPH
NOTE: 20608 obs hidden.
A-6
A
A
A
A
A
A
A
A
AAA
A
:
60000
AB
A
AABCEBCBAAAEBAA A
AB DEEDIDABBAA
CBGEHFA BEBAA
DCGDCDDBCCCCCA
B AACDBCCECEEFBBABAB A
ABGDFGEEEAICCB AA A
A DBCDEFFGGAEDAAAB A
AABBHADCDHDDCADA
ACCAFAFEFDDFE A
FHFDGFJOLLIDCA
A
AEGDKILKTQIKGDFBD
CJKJORYYXYJHCCHEDFRA
DDIINUZZZZZMIIGLFDFCD
ABHJQUZZZZZZQIGEFBBCACA
FZRZZZZZZZUMJMGDARRA
GZZZZZZZZZZTQGDCC
KZZZZZZZZZZNJCB
FZZZZZZZZZSG
DZZZZZZZZJBA
FZZZZZZZA
CZZZZZE
zzz
Sf-fffffffffffff-fffffffffffff-fffffffffffff~fffffffffffff~fffffffffffff-fffff
ffffffff^ff
0
10000
A
30000
40000