Deposing an Econometrics Expert

1
Deposing an Econometrics Expert

• Presentation to
• Boston Bar Association Business Litigation
  Committee
• by
• Roy J. Epstein, PhD
• Expert economic analysis for complex litigation
• Adjunct Professor of Finance, Boston College
• April 9, 2008

2
What is Econometrics?

• Combines economic theory, data, and statistical
  methods
• Mainstream tool in legal proceedings
• Generates formulas to show causation (liability)
  and to estimate damages
• E.g., did release of a pollutant lower property
  values and, if so, by how much

3
Most Common Econometric ModelLinear Regression

• Predicts dependent variable in terms of one or
  more explanatory variables, e.g.
• Crop Yield 5Rain 2Fertilizer
• Coefficients of 5 and 2 best fit the rain and
  fertilizer data to crop yield
• Sorts out individual effects of multiple causal
  factors, e.g.,
• 5 bushels per additional inch of rain
• 2 bushels per additional ton of fertilizer

4
Principal Outputs from Linear Regression

• Estimated value of each coefficient in the
  regression equation
• Test of statistical significance of each
  estimated coefficient
• Not significant means a coefficient is
  statistically indistinguishable from zero,
  regardless of value actually obtained

5
Clash of Models

• For same alleged conduct and facts
• Expert for one side typically finds large and
  statistically significant coefficients
• Expert for other side typically finds small
  and/or statistically insignificant effects

6
How Econometric Experts Reach Opposite Conclusions

• Different results usually due to combination of
• Using different explanatory variables
• Using different data
• Using different statistical procedures
• Deposition must explore each area

7
If You Could Ask Only a Single Question at the
Deposition

• What did you do to establish the reliability of
  your results?

8
Deposition Step 1Discovery

• Opposing experts backup materials
• Raw data and/or identification of exact sources
• Details of all data manipulations
• All regression runs, graphs, and other data
  analyses considered
• Allow adequate time for your expert to
  replicate/review

9
Deposition Step 2Planning Your Questions

• Opposing experts results usually sensitive to
  assumptions involving choice of variables, data,
  and estimation procedures
• Work with your expert in advance
• Identify key assumptions
• Know effect of adopting alternative assumptions
• Questions should probe basis for opposing
  experts choices

10
Deposition Step 3General Topics to Cover
11
Estimated Coefficients

• Algebraic sign
• Effect of explanatory variable in right
  direction?
• Magnitude
• Implausibly large or small?
• Statistical significance
• Did expert use 95 confidence interval?

12
Variables

• Selection of explanatory variables
• How many different models were estimated? How
  were they different? Did any yield contrary
  results?
• What did expert do to establish chosen model was
  more reliable than alternatives considered?

13
Data

• Reliability of data sources
• Procedures used to construct data
• Rationale for grouping of transactions
  (transaction, plaintiff, all customers, product,
  industry)
• Rationale for time period chosen
• Checks/controls for outliers (atypical data
  points)

14
Estimation Procedures

• Ordinary Least Squares (OLS) most widely used
  procedure but inappropriate in certain situations
• Adjustments may be needed for reliable
  coefficient estimates
• Tests exist to assess whether alternative
  procedures should be used
• Did the expert use them?

15
Case Studies
16
1) General Use of Regression Ivy League
Financial Aid Antitrust Litigation
17
Assessing Market Impact of Alleged Conduct

• DOJ sued MIT and Ivy League schools for colluding
  on financial aid awards
• Key issue did challenged practices have
  anticompetitive effect?
• MIT used econometric model to analyze prices
  charged by national sample of schools
• No evidence that alleged conduct raised prices

18

19
The Model

• Dependent variable average price (tuition room
  and board) by school
• 14 explanatory variables to account for different
  school characteristics
• No price effect of alleged collusion
• Controlling for other factors, MIT and Ivys
  charged 322 less than other schools
• But effect not statistically significant,
  therefore indistinguishable from zero

20
2) Assumptions about Explanatory Variables
Estimating Profits in a Damages Claim

• a case last year in which Dr. Epstein was
  involved

21
Different Models for Profit Analysis

• Defendant produced two products, A and B
• Defendant overhead expenses caused by total
  sales (1 explanatory variable)
• Plaintiff separate effects on overhead from
  products A and B (2 explanatory variables)

22
Importance of Choice of Explanatory Variables

• Defendant each 1 increase in total sales adds
  0.40 in overhead (and statistically significant)
• Plaintiff sales of B have no statistically
  significant effect on overhead
• Profitability of product B
• Zero under defendant theory
• Substantial under plaintiff theory

23
3) Data Reliability (or Lack Thereof) the
Conwood Case
24
Conwood v. US Tobacco

• Plaintiff analysis relies on extreme data outlier
• 1 billion claimed damages, after trebling
• Sustained after review by Supreme Court

25
Data Outlier Skews Regression Result
Washington, DC
26
Informative Legal Decisions
27
Selected Cases that Discuss Quality of
Econometric Evidence

• Freeland v. ATT Corp., 238 F.R.D. 130 (S.D.N.Y.
  2006)
• Issues omitted explanatory variables, misuse of
  average prices
• In Re Methionine Antitrust Litigation (West Bend
  Elevator, Inc. v. Rhone-Poulenc), 2003 U.S. Dist.
  LEXIS 14828 (N.D. Cal., August 26, 2003)
• Issues omitted explanatory variables, irrelevant
  data, improper/insufficient time period, improper
  estimation procedure
• Johnson Electric v. Mabuchi Motor America, 103 F.
  Supp. 2d 268 (S.D.N.Y 2000)
• Issues unreliable data, implausible magnitudes
  of coefficients

28
Summary

• Most econometric models sensitive to one or more
  assumptions regarding
• Choice of explanatory variables
• Appropriate data
• Estimation procedure
• Regression results not reliable until
  sensitivities identified and explained
• Deposition must address basis for opposing
  experts assumptions

29
For Further Information

• Roy J. Epstein, PhD
• Expert economic analysis for complex litigation
• 1280 Massachusetts Ave., 2nd Fl.
• Cambridge, MA 02138
• rje_at_royepstein.com
• [phone]
• Adjunct Professor of Finance, Boston College