Anne Neumann (DIW Berlin, University Potsdam) Maria Nieswand (DIW - - PowerPoint PPT Presentation

Improving Regulatory Benchmarking Models by Testing Restrictions: An Application to U.S. Natural Gas Transmission Companies

Anne Neumann (DIW Berlin, University Potsdam) Maria Nieswand (DIW Berlin) Torben Schubert (Fraunhofer ISI)

Maria Nieswand 08 October 2011, Berlin, Germany

Neumann, Nieswand, Schubert 2 Improving Regulatory Benchmarking Models by Testing Restrictions

The aim of the paper is to present an approach which makes deriving regulatory benchmarking models more objective, and therefore, improves the efficiency estimation.

Incentive Regulation ARegV DEA vs. SFA Modeling Technology Benchmarking Statistically Based Model Specification

1

Regulatory Modeling of the Technology

2

Data

3

Methodological Issues

4

Theoretical Framework

5

Preliminary Results

6

Conclusions Agenda

Neumann, Nieswand, Schubert 3 Improving Regulatory Benchmarking Models by Testing Restrictions

The Natural Gas Pipeline System

Neumann, Nieswand, Schubert Improving Regulatory Benchmarking Models by Testing Restrictions

Source: U. S. Department of Transportation, Pipeline and Hazardous Materials Safety Administration (PHMSA)

1

4

Describing the Technology

Neumann, Nieswand, Schubert 5 Improving Regulatory Benchmarking Models by Testing Restrictions

1

OPEX

Source: El Paso/ Dominion Corporate

Transformation process Inputs Outputs

CAPEX Natural gas delivered Transportation service Number of connections

“classical” outputs of networks Environmental variables

Deriving the Regulatory Benchmarking Model

Neumann, Nieswand, Schubert 6 Improving Regulatory Benchmarking Models by Testing Restrictions

1

OPEX

Transformation process Inputs Outputs

CAPEX Natural gas delivered Transportation service Number of connections

Measured by

• Total deliveries
• Transmission system peak deliveries
• Length of mains

Environmental variables

Data (unbalanced panel, obs=191, n=43, t=2003-2007)

Neumann, Nieswand, Schubert 7 Improving Regulatory Benchmarking Models by Testing Restrictions

2

Variable min 1st quartile median mean 3rd quartile max Opex [in 1,000 Euro]

315 8,983 24,510 48,620 68,970 285,600

Opex [in Euro, 2003 prices]a

268 7,776 20,590 42,420 60,010 244,300

Total deliveries [in 1,000 Dth]

20,120 143,600 363,000 507,000 742,200 3,138,000

Peak deliveries [in 1,000 Dth]

122 585 1,303 1,614 2,283 7,124

Length of mains [in miles]

80 364 1,402 2,379 3,915 9,627

Note: a As deflator the producer price index for utilities is used (Source: Bureau of Labor Statistics Data). Source: FERC Form No. 2

Table: Descriptive Statistics

Methodological Issues

Neumann, Nieswand, Schubert Improving Regulatory Benchmarking Models by Testing Restrictions

• No statistical noise
• No functional form of

technology

• Curse of dimensionality

(slow rate of convergence)

• Sensitive to extreme values
• Outlier detection (also based on statistical tests)
• Robust nonparametric frontiers (order-m, order-α)
• No misspecification
• No statistical tests on coefficients?
• Restriction tests
• Meaningful results require lots of observations
• Increase number of observations:

pooling data  test on technical progress

• Reduce the number of variables:

exclude/ aggregate variable  restriction tests

• Robust nonparametric frontiers (order-m, order-α)

3

8

Some Drawbacks of DEA??? Implications and Approaches!!!

Linear Program

Neumann, Nieswand, Schubert 9 Improving Regulatory Benchmarking Models by Testing Restrictions

4

Input-oriented DEA with VRS The question

Finding a Statistically-based Answer I

Neumann, Nieswand, Schubert 10 Improving Regulatory Benchmarking Models by Testing Restrictions

4

The hypothesis: Estimate the model under the different model specifications and obtain From theory it holds And therefore

Finding a Statistically-based Answer II

Neumann, Nieswand, Schubert 11 Improving Regulatory Benchmarking Models by Testing Restrictions

4

and

• btain a test statistic
• r critical values

critical values vary with subsample’s size depending

• n the critical

values

Preliminary Test Results

Neumann, Nieswand, Schubert 12 Improving Regulatory Benchmarking Models by Testing Restrictions

5

95%quantile as a function of m

• bserved value of test statistic
• ptimal subsampling size
• Model specification:

x

• Varying subsample size
• Varying 95%-quantile
• Varying critical values
• Varying decisions w.r.t.
• Reject

if test statistic > critical value

• For optimal subsample size of 38
• Critical value is about 0.13
• DO NOT REJECT

Conclusions

Neumann, Nieswand, Schubert 13 Improving Regulatory Benchmarking Models by Testing Restrictions

• Objectify the regulatory benchmarking model selection
• For our sample (U.S. natural gas transmission companies)
• Output variables total deliveries and transmission system peak

deliveries can be included as an aggregate

• Variables are not really individually relevant
• Reduce dimensions of the model specification
• Improve estimation (because the rate of convergence increases with less

dimensions)

• Can easily be adopted in regulatory practice

6

Thank you for your interest. Questions and comments are welcome!

DIW Berlin — Deutsches Institut für Wirtschaftsforschung e.V. Mohrenstraße 58, 10117 Berlin www.diw.de Maria Nieswand mnieswand@diw.de

From Questioning to (Statistically) Answering

Neumann, Nieswand, Schubert 15 Improving Regulatory Benchmarking Models by Testing Restrictions

B

The efficiency estimator under the different model specifications: The consequence: The test statistic: The hypothesis: