PUC Anderson Direct.pdf

10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
e e
1
2
3
DIRECT TESTIMONY OF D. LYNN ANDERSON
INTERMOUNTAIN GAS COMPANY
CASE NO. U-10M-88
4 Q. Please state your name and business address for the record.
5 A. My naie is D. Lynn Anderson. I am employed asa Utilties Rate
6
7
Analyst for the Idaho Public Utilties Commission located at 472 West
Washington Street, Boise, Idaho.
8 Q. Wil you please outlie your academic and professional background?
9 A. I completed undergraduate coursework at Brigham Young University
and, in 1974, graduated from Idaho State University with a Bachelor
of Science Degree in Government and a Bachelor of Arts Degree in
Sociology. In conjunction with my undergraduate study, I have
completed substantial coursework in engineerig, architecture, mathe-
matics, economics and physical sciences . Currently, I am a candidate
for a Master of Public Administration Degree at Boise State University.
In 1978, I completed a graduate-level engieering prograi at North-
western University in highway safety and traffic stu die s . In addition,
I have attended various professional conferences, workShops and
special programs includig the NARUC Annual Regulatory Studies
Program at Michigan State University.
Prior to joining the Commission Staff, I wasemployed as a
Research Analyst by the Idaho Transportation Department, where my
major duties included cost-effectiveness evaluations of highway safety
projects and various othet statistical analyses.
With the Commission Staff ,my duties have included the review
of specific portions of utity rate applications, and other research
assignments related to utity regulation.
28 Q. What is the purpose of your testimony in this case?
e e
1 A. I was assigned the responsibilty of reviewing Internountai Gas
2 Company's econometric model which predicts a 6.5% decrease per
3 household in residential gas consumption from 1980 to 1981. In this
4 regard, I have reviewed the data, methodology, assumptions and
5 actual results of the inodel.
6 Q. In reference to the input data for Intermountai Gas Company's
7 econometric model, what comments Can you make?
8 A. Upon examination of Intermountai Gas Company's workpapers, I
9 found the service area data on Pages 1-3 to be in error due to
10 apparent confusion of the service boundaries. These errors were
11 substantial for the Utah Power and Light area data, e.g., 1978 gas
12 revenues are overstated by 67.2%.
13 Q. Were there any other significant data errors?
14 A. Yes, the per capita income figures for various service areas are
15 based on erroneous surrogates of service area incomes. Ada County
16 is used to represent Idaho Power Company's service area,butactualy
17 has significantly higher income than most, if not al other Idaho
18 Power counties. Bannock County is used as a Utah Power & Light
19 surrogate, but is actually served mostly by Idaho Power and has a
20 higher per capita income than most of the Utah Power area. Bonne-
21 vile County is used to represent Idaho Falls, which the Bureau of
22 Economic Analysis lists separately and has significantly higher income
23 figures than the rest of the County. And finaly, Twin Fals County
24 is used to represent the municipal and cooperative electric companies
25 of Magic Valley - all of which exist nearly exclusively and alost
26 totaly inclusively within Miidoka and Cassia Counties and exclude
27 Twi Falls County entirely.
28 Q. Do you have any other comments regardig the data?
2 ANDERSON
e e
1 A. In addition to the actual data errors I have inentioned, there is a
2 basic inconsistency of data regardig the v.arious service areas.
3 Only the Idaho Power area data were used to calculate annual gas
4 consumption, gas price and electricity price. However, al service
5 are.a data, including that of Utah Power &: Light and the varous
6 municipals and cooperatives, were used to calculate income averages.
7 This inconsistency causes problems when interpreting the statistical
8 results of the model. In his testimony, Mr. Robinson used the
9 results to predict tota residenti consumption, while the model, by
10 design, is most nearly representative of IGC's customers who receive
11 electric service from Idaho Power Company.
12 Q. Have you estimated what impact these errors and inconsistencies
13 have on the outcome of the model?
14 A. Yes, but only intuitively. Because IGC's servce area
is so heavily
15 influenced by the portion served by Idaho Power (approxiately 78%
16 of customers), and because changes in income appear relatively un-
17 important to the model, the data errors may be faily inconsequential.
18 However, they do make the study results more difficult to interpret.
19 Q. Do you believe that Intermountai Gas Company's model contains the
20 most appropriate independent variables?
21 A. No. Heating degree days, an obviously important factor
in gas con-
22 sumption, was omitted from IGC's modeL. According to Mr. Robinson,
23 inclusion of this variable resulted in ilogical inverse relationships
24 between heating degree days and gas consumption, i.e., as the
25 temperature became colder, gas consumption was predicted to decrease.
26 However, in testing the relationship between heating degree days
27 and gas usage, I have developed a simple liear equation which
28 shows a positive correlation between heating degree days and gas
3 ANDERSON
e e
1
2
3
consumption. This correlation is actualy higher than the detrended
correlations of al of the other independent variables included in
IGC's model, except that of the real price of gas.
4 Q. Isn't it true that statistical modelig requires a balance between
5 logic and statistics and thus, if an independent. variable results in
6 ilogical results, then that variable should be deleted from the model?
7 A. Yes, to an extent, that is true . However, if the variable in question
8
9
10
11
12
13
logicaly should have, and empiricaly does have an importat impact
on the dependent variable, then it behooves the researcher to re-
evaluate the origial set of variables and assumptions and then
attempt to statisticaly fit alternative equations. This process may
result in the exclusion of less important variables which could be the
cause of the ilogical relationship of . the more important variable.
14 Q. Do you have any other comments concerning the model's variables?
15 A. It is noteworthy that the varible of previous year's consumption,
16
17
18
19
by itself, results in a predicted consumption decrease of 30 therms
per customer, or about 60% of the decrease predicted from 1980 to
1981. Thus, if al other varibles remaied constant, this model
would contiue to predict consumption decreases.
20 Q. You have said the IGC's model does not include the most appropriate
21
22
23
24
variables. What explanation can you offer for the rather remarkably
good statistical results of their model which include an explanatory
value (R2) of 93% and a standard error of only 1. 02 therms per
customer?
25 A. First, iac's standard or II average II error of estimate and greatest
26 error figures are in need of correction because of a misapplication of
27 logarithms in I GC' s workpapers. Thus ,in Mr . Robinson's testimony,
28 Page 6, Lines 21-25, he states that the "greatest diference is only
4 ANDERSON
e e
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
1. 07 therms", but the number should be 66 therns. His statement
that "the standard error of the regression is 1. 039 therns per
household" should actualy read as 33 . 20 therms . And, most impor-
tantly, his statement that "a 95% probabilty that actual consumption
of natural gas durig the relevant time frame wi be within approxi-
mately 82,000 therms" should be corrected to approxiately 5.2
milon therns. Thus, the Company's projected decrease of 4.1
mion theris is accurate only within 5. 2 milon therms at a 95%
confidence leveL. It should be emphasized that a confidence level
indicates the probabilty that the actual amount wil fall withi the
estiated range, but only with the assumption that al past relationships
between variables wi remain unchanged durig the predicted tie
period.
Staff Exhibit No. _(DES-I) is a plot of the percentage.
errors of the model's fitted estiates of gas consumption. The
approxiate confidence levels for 2%, 4% and 6% ~rrors are shown as
40%, 70% and 90%, respectively. In addition, the actual number of
data points fallg within each confidence range is shown to agree
very closely to the confidence percent, indicating that the assumption
of a normal distribution of errors is probably valid.
In addition , there are at least three factors extraneous to
statistical validity that contributed to the model's "good" statistical
results. First, all variables in IGC's equation, both dependent and
independent, are related either directly or inversely to time with
correlation coefficients rangig from .62 for the real price of gas to
.97 for both the real price of electricity and real per capita income.
Since the data was not detrended, much of the explanatory value of
the independent variables is due to this cross correlation of variableS
5 ANDERSON
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
e e
to time.
The second factor, closely related to the first, is that each pair
of independent variables has a high correlation coefficient, rangig
from .51 to .96, which indicates a high degree of multicollearity .
The absence of multicollearity is one of the basic requirements for
quantities used as variables in multiple regression. statistics.
Third, three of the variables provide little additional statistical
explanation for residential consumption beyond that of the two most
important variables, namely the real price of gas and the previous
year's consumption. In fact, using only these two independent
variables to perform multiple liear regression results in an equation
with statistical results slightly better than those of IGC's modeL.
Since R 2 values cannot decrease by addition of variables, it is
possible to create artifically high statistical results by inclusion of
relatively unimportant, even irrelevant, variables which also add to
the probabilty of data error. A rule-of-thumb of .multiple regression
allows no more than one independent variable per five data points.
IGC's model more than doubled this approxiate allowance.
These three errors in the basic application of the model's statis-
tical process can each result in exaggerated inferences of relation-
ships and predictive power.
23
22 Q. In extrapolating 1981 consumption from the model, were al necessary
24 A. Mr. Robinson's testimony stressed that the estimates ofination for
assumptions reasonable and straightforward?
25
26
27
28
1980 and I981 of 14% and 9%, respectively, were very conservative,
which, by definition, would result in conservative Consumer Price
Index figures. However, this model translates a lower CPI into a
low predicted level of consumption. Since IGC is seekig an increase
6 ANDERSON
e e
1 in rates due to projected decreased sales, it is clear that "conserva-
2 tive inflation estimates" is a potentially misleadig statement.
3 Q. The multiple regression equation of the model was based on actual
4 data from 1965 to 1978, thus requirg the 1981 consumption estimate
5 to be projected nearly 3 years. Is there now sufficient data for
6 1979 Or 1980 to test the accuracy of the model?
7 A. Yes, from IGC's workpapers, the actual 1979 sales per customer in
8 the Idaho Power service area computes to 884therms, compared to
9 the model's prediction of 754 therms. In compensatig for the abnor-
10 maly cold heatig season, the actual sales should be adjusted to 814
11 therms, which sti leaves the model's prediction 60 therms per
12 customer, or 7.4% less than adjusted actual usage. For the total
13 Company, the error is approxiately 4.8 milon therms.
14 In addition, sufficient 1980 sales data is avaiable to similarly
15 calculate that the model wil underestimate actual usage by6 .4%, or
16 48 therms per customer. The Company's projection of sales for 1980
17 is approxiately 3.8 milon therms less than what can be expected to
18 occur. It should be noted that the Company's error in projectig
19 usage per customer in both 1979 and 1980 is approximately equal to
20 the total amount of decrease now projected for 1981.
21 Q. Do you have any additional comments to make concerning this specific
22 model or modelig in general?
23 A. All predictive models are based on the assumption that past relation-
24 ships between variables wil either remai constant or wil change
25 predictably over time. IGC's model assumes the former, thus pre-
26 dictig future household consumption of gas based on an underlying
27 assumption of insatiable conservation, weatheriation and replacement
28 of gas appliances, but not to the extent of becoming non-customers.
7 ANDERSON
e e
1 It is more probable, however, that there are saturation levels for
2 conservation and weatherization ,and that as these saturation levels
3 are approached, gas customers Wi begi absorbing price increases
4 without reducing consumption- up to the point where prices dictate
5 they eliinate al gas appliances and become a non-customer. Thus,
6 in addition to the methodological shortcomings, it seems inevitable
7 that IGC's model wil fai, as it already has, because of basic assumptio
8 errors.
9 Q. What is your recommendation to ths Commission regardig the projected
10 6.5% residential sales reduction?
11 A. I believe that the Company's projected 6.5% reduction in sales is not
12 sufficiently demonstrated to be used as a known and measurable
13 adjustment to revenues in the present rate case .
14 Q. Are you, in effect, saying that residential sales wi not declie from
15 1980 to 1981?
16 A. No. Sales may decrease further, if decliing sales per customer and
17 the more recent trend of decreasing number of customers continue.
18 However, I do not believe IGC's study has sufficiently quantiied
19 these trends. In addition ,as demonstrated by the model's under-
20 estimation of sales in 1979 and 1980, it seems probable that the 1981
21 sales are also underestiated. It .is quite possible that the trend of
22 decreasing sales may be at, or near, the I1bottoming out" point due
23 to conservation saturation and the probable favorable impact of IGC's
24 recent advertising campaign, both of which are not included as
25 variables in the model.
26 Q. Even if sales do not decrease as much as the model predicts, isn't it
27 true that the adjustment procedure presented by Mr. Lebens wil
28 ensure that neither the Company nor the ratepayers Wi suffer from
8 ANDERSON
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
e e
any financia rik?
A. This is true., but with two exceptions. First, the time-value of
money, includig both interest and inflation, is not accounted for in
the adjustment. This would be very minor to individual ratepayers,
but could be very significant to the Company.
Second, underestimation of sales by IGC would result in a
higher than necessary price which, according to their model, could
result in unnecessary addìtional sales reduction. These reduced
sales would probably not be regaied after the price was re-adjusted,
since, according to IGC, the statistical relatìonship between gas
price and therms sold is probably not reversible.
Q. Regardless of long-run implications of energy price structures ,if
the Commission denies recognition of the projected sales decrease for
1981 wi it not be guaranteeing that IGC wil not earn its authoried
rate of return?
A. No, while the possibilty exists that in the short-run sales may
decrease, the extent of any decrease is not accurately shown by the
Company's sales projection. Furthermore, the likeliood that sales
wi decrease should provide IGC's management with additional incen-
tive to increase innovation and efficiency, thereby reducing or
eliinatig the net revenue impact of fewer therms sold.
It is also important not to emphasize short-term profits to the
exclusion of long-te.rm implications. Even if IGC's model produced
reliable results, it would seem detrimental to the Company to raise
prices in anticipation of decliing sales because the sales, according
to the model, would declie 11.1% for every 10% price increase. The
Company would seemingly be caught in an irreversible downward
spiral of decliing sales and decliing revenues by the very act of
9 ANDERSON
1
2 Q.
3 A.
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
-e
raising the price to prevent loss of revenues.
Does this conclude your direct testimony?
Yes, it does.
10 ANDERSON
-e
A roximate 90%Confidence Ran e
r-A 70%C.R.0\0 rox.0"d .00NA40%C.R.O"....r-..'-.O"r-O"..
in in in.¡.¡.¡i:i:i:00
'M 'M 'M .r-0 0 0 O"0.0.0...
el el el.¡.¡.¡el el el r-0 0 0 .r-
O"tH tH tH ..0 0 0
0\0 0\0 0\0r-~0 \000r-"d .r-
O"
in in in ..i:i:i:'M OM OM
CD el el el..¡.¡.¡i..¡i:i:i:r-el 0 0 0 O"~i=u u U ..
~'M.¡
1- "do.in:E w0 .r-u tH O"0 ..U)~i-~el;:.~z i-r-1-CD O"~.¡..E-i:
Š 1-
~CD N().r-r:i:O"w CD ..E-"CZ'M1-tHi:i-0 r-u O"..
0.r-
O"..
O".\0O"..
00.CD \Di:O"
'M ....
CD.¡r-el \D
i=O"
'M ...¡in
W \0U.\0~O"1-..
INTERMOUNTAIN GAS CO.
CASE U-L034-88
Staff Ex.No.0\0 0\0 0\0 0\0 0\0 0\°0\°0\°0\°0\°0\°0\°0\°0\°
CDLA-1)r-\D i."d ~N ..0 ..N ~"d i.\D+++++++i I I I I I
D .L.ANDERSON PERCENT OF ERROR FROM ESTIMATED TO ACTUAL THERMS