Oracle Corporation et al v. SAP AG et al

Filing 933

Declaration of Stephen K. Clarke in Support of 851 Defendants' Opposition to Plaintiffs' Motion No. 1 to Exclude Expert Testimony of Stephen K. Clarke [Filed Pursuant to D.I. 915 ] filed by SAP AG, SAP America Inc, Tomorrownow Inc. (Attachments: # 1 Exhibit 1, # 2 Exhibit 2, # 3 Exhibit 3, # 4 Exhibit 4, # 5 Exhibit 5, # 6 Exhibit 6, # 7 Exhibit 7, # 8 Exhibit 8, # 9 Exhibit 9, # 10 Exhibit 10, # 11 Exhibit 11, # 12 Exhibit 12, # 13 Exhibit 13, # 14 Exhibit 14, # 15 Exhibit 15, # 16 Exhibit 16, # 17 Exhibit 17, # 18 Exhibit 18, # 19 Exhibit 19, # 20 Exhibit 20)(Related document(s) 915 ) (Froyd, Jane) (Filed on 10/4/2010) Modified on 10/5/2010 (vlk, COURT STAFF).

Oracle Corporation et al v. SAP AG et al

Doc. 933

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

Robert A. Mittelstaedt (SBN 060359) Jason McDonell (SBN 115084) Elaine Wallace (SBN 197882) JONES DAY 555 California Street, 26th Floor San Francisco, CA 94104 Telephone: [phone] Facsimile: [phone] [e-mail] [e-mail] [e-mail] Tharan Gregory Lanier (SBN 138784) Jane L. Froyd (SBN 220776) JONES DAY 1755 Embarcadero Road Palo Alto, CA 94303 Telephone: [phone] Facsimile: [phone] [e-mail] [e-mail] Scott W. Cowan (Admitted Pro Hac Vice) Joshua L. Fuchs (Admitted Pro Hac Vice) JONES DAY 717 Texas, Suite 3300 Houston, TX 77002 Telephone: [phone] Facsimile: [phone] [e-mail] [e-mail] Attorneys for Defendants SAP AG, SAP AMERICA, INC., and TOMORROWNOW, INC. UNITED STATES DISTRICT COURT NORTHERN DISTRICT OF CALIFORNIA ORACLE USA, INC., et al., Plaintiffs, v. SAP AG, et al., Defendants. OAKLAND DIVISION Case No. 07-CV-1658 PJH (EDL) DECLARATION OF STEPHEN K. CLARKE IN SUPPORT OF DEFENDANTS' OPPOSITION TO PLAINTIFFS' MOTION NO. 1 TO EXCLUDE EXPERT TESTIMONY OF STEPHEN K. CLARKE Date: September 30, 2010 Time: 2:30 p.m. Courtroom: 3, 3rd Floor Judge: Hon. Phyllis J. Hamilton FILED PURSUANT TO D.I. 915
SFI-649475v1 CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

Dockets.Justia.com

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 A.

I, Stephen K. Clarke, declare as follows: 1. I have personal knowledge of the matters discussed herein.

Background and Qualifications. 2. I am a Certified Public Accountant (Accredited in Business Valuation) in the State

of Arizona; a Certified Fraud Examiner; and a Chartered Accountant in England & Wales. A copy of my resume is attached as Exhibit 1. I have been engaged as a testifying economic damages expert in dozens of intellectual property disputes over the last 22 years. Such disputes have related to copyrights, patents, trade secrets, trade dress and unfair competition, and have involved aggregate claims well in excess of $100 billion (prior to this matter). I have provided testimony as an economic expert in many venues including Federal and State Courts, arbitration panels, and bankruptcy hearings in the United States, and the Crown Courts in Great Britain. I have valued over $20 billion worth of businesses in the same 20 year period. My degree is in Management Sciences from the University of Manchester in England. I taught economics at Arizona State University for several years. 3. In December 2007, I was retained by Defendants to address Plaintiffs' alleged

damages. I have been working on this case since then. B. Georgia-Pacific Analysis. 4. I devoted 144 pages (nearly 50%) of my 294 page report to a detailed rebuttal of

Plaintiffs' expert Paul K. Meyer's Georgia-Pacific opinion and an analysis of each of the 15 Georgia-Pacific factors. On the other hand Meyer spent only 76 pages of his 281 page report addressing the Georgia-Pacific factors. I analyzed several critical factors that Meyer failed to consider, including Plaintiffs' prior licensing agreements with other support vendors and partners and Plaintiffs' established relationships with other third-party support vendors who are still partners and offer similar services to TomorrowNow ("TN"). My report addresses the GeorgiaPacific factors in detail and the analysis considers all of the relevant facts in deriving the royalty rate. 5. TN's standard pricing structure was based on 50% of Plaintiffs' price. TN

established its 50% pricing structure by about mid-2004 and, for the most part, continued that
SFI-649475v1 CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

pricing structure through the wind down of its operations in October 2008. TN made a few exceptions to its standard 50% pricing structure. Of its 358 customers, TN provided its services at no charge to less than 4% of its total customers. C. Economic Causation Analysis. 6. The Litigation Services Handbook1 provides an overview of the first steps that an

expert witness takes to calculate lost profits: The first step in a damages study translates the legal theory of the harmful event into an analysis of the economic impact of that event. In most cases, the analysis considers the difference between the plaintiff's economic position if the harmful event had not occurred and the plaintiff's actual economic position. The damages study restates the plaintiff's position `but for' the harmful event; this step is often called the but-for analysis. Damages, then, are the difference between the but-for value and the actual value. I have attached a copy of the relevant excerpt as Exhibit 2. 2 I consider economic causation in every case because I am attempting to identify the damages that arose as a result of the alleged acts. Failure to consider causation results in an inappropriate analysis. Properly applied, a study of economic causation allows the economist to trace the effects of the damage causing acts through to the damages opinion, separating their effects from other factors that may have affected a firm's operations but are unrelated to the damage causing acts. The methodology by which economic causation is applied is dependent on the facts and circumstances of each case and is therefore fact intensive. 7. In this case, my causation analysis involved 358 customers and had to be done one

customer at a time. The analysis involved reviewing over ten million pages of documents for causation related information. The only practical way to organize and categorize such a vast volume of documents is in a database that tracks the reasons each customer terminated its support with Plaintiffs and/or made purchases from SAP. 8. Because of the similarity of characteristics among different customers, I grouped

customers exhibiting similar characteristics, using the term "pools" to describe the grouped Weil, Roman L., et al. Litigation Services Handbook: The Role of the Financial Expert, 3rd Ed., John Wiley & Sons, Inc. (2001), at 5.4. 2 For the Court's convenience, I have identified the portions of certain exhibits that I refer to herein with red boxes outlining the relevant material.
SFI-649475v1 1

2

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

customers for ease of reference. However, the nomenclature is immaterial: the pools could have been given one of many other names, such as groups or classes. It would be impractical at trial for the jury to consider all the relevant evidence for each customer individually. An orderly presentation of pools of customers exhibiting similar characteristics is more efficient and at least possible within the trial schedule. 9. As a damages expert, I routinely analyze economic causation in a wide variety of

industries. For example, I have applied my expertise to the high-tech, aerospace, entertainment, gaming, and real estate industries, and done so in claims ranging from intellectual property infringement and contract disputes to fraud analyses and bankruptcies. In each case, I examined the documents produced, performed appropriate analysis and research, and consulted with industry experts where appropriate to generate an understanding of the relevant market in which the parties operate. D. Analysis of Third Party Support Market. 10. I routinely assess industry markets and competition in the course of my valuation

analyses. I am a Certified Public Accountant, Accredited in Business Valuation, and have 38 years of experience valuing a wide range of businesses. I am required under the AICPA's Statement on Standards for Valuation Services No. 1 to obtain non-financial information, including information on the economic environment, geographical markets, industry markets, and competition, sufficient to understand the subject entity. 11. I have performed approximately 2,000 valuation analyses during my 35 year

career in accounting and economics, and have managed at least 200 valuations in numerous industries in the course of my expert work. During the past 22 years as a litigation consultant, I have also performed hundreds of lost profits damages calculations for all manner of businesses, and have analyzed their competitors' information in numerous cases. 12. In this case, I relied on numerous sources of information, including: company

websites; industry articles; analyst reports, including Gartner and Forrester; documents produced by Plaintiffs, Defendants, and customers that describe the offerings of third party support providers; the TN Wind-Down Report, which tracked where customers went for support after TN
SFI-649475v1

3

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

ceased operations; and Plaintiffs' At-Risk Reports, which tracked losses to third party support providers beginning at least as of January 2005 and continuing at least through the beginning of 2008. E. Regression Analysis. Training and expertise. 13. I did my first study of regression analysis at Manchester University in England as

part of my Bachelor's degree, between 1969 and 1972. The classes were part of a number of mathematics courses which included study of business statistics and other analytical tools. I continued my education on use of the technique at London School of Accountancy where I studied numerous modeling techniques (including regression analysis) in a class called "Elements of Financial Decisions," which was 25% of my final examination to become a Chartered Accountant in England and Wales. The examination is one of the most demanding of any professional qualification. Regression analysis was a significant part of the curriculum for Economics 502, which was the class I taught at Arizona State University. 14. Since I became a Chartered Accountant, and later a CPA, I have run hundreds of

regressions for the purposes of my work, usually in order to quantify variable expenses in the course of computing lost profits. In addition, I also created a multi-variate hedonic regression analysis designed to quantify the effect that the creation of a 36,000 acre park in Scottsdale, Arizona had on the value of surrounding property. The regression analysis considered numerous factors that may have played a role in changing the value of the land in the neighborhood of the park. The regression incorporated several dummy variables (very similar to what Oracle's statistics expert Dr. Levy calls fixed effects) for events such as a new freeway, and included approximately a dozen other variables including lot size, building density, a variety of amenities, and locational effects, such as distance of the property from the park. I presented the regression analysis in court and the jury agreed with my conclusion after vigorous cross examination. I also ran a regression analysis related to the effect that the introduction of a new piece of software had on the sales revenues of a major software development company. I taught graduate level Managerial Economics at Arizona State University for three years. Managerial Economics is a
SFI-649475v1

4

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

branch of economics that applies microeconomic analysis to decision models of management. This includes operations research, risk analysis, production analysis, pricing analysis, and capital budgeting. Finally, although I did not do any forecasting as part of my analysis in this matter, I have applied regression techniques designed to determine the effect on sales or costs of various events, such as a change in selling price or a change in the competitive environment. 15. My report in this case includes a single variable regression analysis (otherwise

known as simple regression) that I have used many times throughout my career, in both the business and litigation context. Plaintiffs' main complaint appears to be that the Court should exclude my regression analysis simply because I first studied the technique a long time ago. However, the technique has not changed since I first studied it. 16. Support for my position is provided by The Litigation Services Handbook,

published by Wiley & Sons (3rd edition), a recent text intended to assist accounting professionals in the context of litigation. Exhibit 2. The book offers extensive guidance on how accountants might compute variable costs and suggests a regression analysis as one way to do so. See pages at 7-11 to 7-25. As Levy admits in his declaration at 5:1 and 8:2-5, my equation estimates the change in cost due to a change in revenue. This is the precise definition of variable cost and is the exact reason I did the analysis. I have attached several descriptions of the variable cost curve as it appears in economic textbooks in Exhibits 7 to 13. Levy's criticisms of my use of R2. 17. Levy presents various charts and regression lines comparing what he calls "R2

Clarke" to an alternative R2 for my zero intercept regressions. However, it appears Levy used the Chart Tool within Excel to calculate his R2 for zero intercept regression lines. To verify that Levy used Chart Tool, I re-computed R2 for my data using the Chart Tool function. As Figure 1 below shows, the calculated R2 using the Chart Tool results in the exact same output Levy quotes in his declaration criticizing my analysis. See, for example, page 9, Figure 2. Therefore, I conclude Levy used Chart Tool to do his work.

SFI-649475v1

5

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

1 2

Figure 1
OEMEA: Real Total Expenses vs. Real Total Revenues

3 4 5 6 7 8 9 10 11
Real Total Expenses (in '000s

$3,000

$2,500

$2,000 Reg res s io n Line Equation & R-squared using Chart Tool y = 0.1803x + 965.74 R2 = 0.6254 $1,500

$1,000

$500

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 18.
$$-

Zero Intercept Regression Line Equation & R-squared using Chart Tool y = 0.3695x R2 = -0.1085 $1,000 $2,000 $3,000 $4,000 $5,000 $6,000 $7,000 $8,000 $9,000

Real Total Revenue (in '000s)

However, the Chart Tool produces incorrect outputs under certain conditions, and

those conditions include those applicable to my analysis. According to Microsoft, the publisher of the Excel program, Chart Tool is always incorrect3 and should not be used to compute R2 for a regression with a zero intercept. 19. I used the Excel Analysis Toolpak (ATP) to calculate my R2 which returns the

correct R2 for a zero intercept regression. If Levy had computed R2 using the ATP he would have derived the same results as I. I verified the accuracy of my work by re-running the analysis using another statistical package called STATA, and R2 was identical. Levy's criticisms of goodness of fit. 20. Levy implies that R2 is not important in determining the strength of a regression

equation. The reality is different. There are numerous authoritative sources that describe R2 as a measure of goodness of fit. The Litigation Services Handbook (Exhibit 2) states at page 7-18 that
3

I have included a printout of the Microsoft Support website as Exhibit 3. 6
CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

SFI-649475v1

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

an R2 equal to 1 means the equation "explains the variation in the dependent variable perfectly..." and that a high value for R2 means the equation "...explains a large portion of the variation in the dependent variable." In addition, Parsons and Schultz, Marketing Models and Economic Research, states that the most common decision rule for choosing among alternative linear models generally is to select the model with the largest Corrected R-squared.4 Moreover, numerous other practitioners including Pappas & Brigham, Managerial Economics (3d Ed.) (Exhibit 5), and Hirschey, Managerial Economics (3d Ed.) suggest that when choosing between alternative equations, strength of relationship is best.5 Levy's criticisms of zero intercept model. 21. As Levy admits in his Declaration, my equation estimates the change in cost due to

a change in revenue. See page 5, line 1 and page 8, lines 2 to 5. This is the precise definition of variable cost and is the exact purpose of my analysis. Levy goes on to suggest that use of a zero intercept is not appropriate for estimating variable costs and states that my "analyses do not conform to generally accepted scientific methods used to measure how costs change as revenue change (sic)". See page 2, lines 6 to 7. However, a thorough knowledge of accounting is required to distinguish variable costs from fixed costs. Levy's lack of understanding of basic accounting causes him to make fundamental errors. Levy seems to believe that I needed to quantify incremental costs and much of his criticism assumed my objective should have been to estimate incremental costs. However, my model is designed to estimate the variable costs, which are not the same as incremental costs, and which may be vastly different. My model achieves that purpose. 22. Fixed, Variable, and Total Cost curves (which are often straight lines in spite of

being called `curves') show the relationship between the revenues a company generates and the types of costs incurred to generate those revenues. Certain of the costs a company incurs are The Corrected R-squared, also known as Adjusted R-squared, takes into account the number of variables and sample size (both of which affect a model's statistical significance). Although I used R-squared in my analysis, I computed Corrected R-squared and gave that result in my analytical output and found there was little difference between Corrected R-squared and Rsquared in my models.
5 4

I have included relevant excerpts from these texts as Exhibits 4-6, respectively. 7
CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

SFI-649475v1

1 2 3 4 5 6 7 8 9 10 11

fixed over a given range of activity (for example, factory rent and annual insurance), while other costs vary with the level of sales (for example, selling commissions and direct manufacturing costs). As I illustrate in the graph below, the fixed cost curve (plotted against revenues) is a straight horizontal line parallel to the revenue axis that intersects the cost axis at a point equal to the company's fixed costs. The variable cost curve will begin at the origin of the graph because, by definition, if there are no revenues there are no variable costs. The origin is where the two axes (X and Y) meet. The variable cost curve will slope upwards to the right (see graph below). By adding the variable cost curve to the fixed cost curve, we derive the total cost curve (see graph below). The following picture illustrates what I am describing: Figure 2
Total Cost Curve

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
SFI-649475v1 Intercept Fixed Cost Curve

C o s t s

A
Variable Cost Curve

B

C Revenues

23.

It is critical to understand that the total cost curve is exactly parallel to the variable

cost curve over a given range of activity. As the graph shows, the total cost curve tracks the variable cost curve but is shifted higher on the graph. It is a mathematical fact that the slope of the total cost curve gives the increase in variable cost as a function of revenue. In my analysis, I

8

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

establish the slope of the variable cost curve according to basic economic principles. My equation matches the variable cost curves illustrated in multiple economic texts6 24. Oracle provided only quarterly accounting data for the relevant period. So I used

the provided quarterly data to do my regression of costs against revenues. All I needed as an output from the regression equation was the slope of the equation at Average Revenue.7 I applied the slope to the Average Revenue (Point C in the graph) to derive the variable cost at Average Revenue (Point B in the graph). I used the slope of the cost curve at Average Revenue to derive the fixed and variable cost ratios. Because the slope of the total cost curve and the slope of the variable cost curve are identical (which is mathematically inarguable) and the slope is derived from the regression equation, it follows that my variable cost analysis is correct. 25. Mathematically the intercept value represents the value of Y (in this case, total

costs) when X (in this case, revenue) is zero. Levy repeatedly insists that the intercept value of the regression equation represents the firm's fixed cost. He is mistaken. Levy presents a number of examples in an effort to show that the calculated intercept value in his alternative model specifications represents fixed cost. Statistics textbooks, including those Levy references in his Declaration, indicate that in most cases the intercept value in a regression equation is nothing more than a mathematical anchor and has no practical meaning unless there are a sufficient number of independent variable observations near zero.8 Additionally, Damodar Gujarti cites in his book Basic Econometrics "cost analysis theory" as an instance in which "regression through

I have included relevant excerpts from various economic text books (Stiglitz and Walsh, Principles of Microeconomics at 136-137; Colander, Microeconomics (4th Ed.) at 209; Case and Fair, Principles of Microeconomics at 158-159 & 162; Ayers and Collinge, Microeconomics: Explore & Apply at 180; McConnell and Brue, Microeconomics: Principles, Problems & Policies (17th Ed.) at 150-151; O'Sullivan and Sheffrin, Microeconomics Principles and Tools (2d Ed.) at 169; Salvatore, Theory and Problems of Managerial Economics at 130) as Exhibits 7 to 13, respectively. 7 Average Revenue in this declaration refers to average quarterly revenue from Oracle's first quarter of fiscal 2006 through the second quarter of Oracle's fiscal 2009 for the OUSA and OEMEA entities (produced in discovery), and for Oracle as a Whole quarterly information from the first quarter of fiscal 1997 through the first quarter of fiscal 2010 (from the publicly available website, oracle.com). 8 Macfie and Nufrio, Applied Statistics for Public Policy at 432, 446. I have included the relevant excerpts as Exhibit 14.
SFI-649475v1

6

9

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

the origin," (functionally the same as my zero intercept regression), is a relevant and useful tool.9 It is also intuitively obvious that Levy is mistaken because if Oracle's revenues were zero the firm would be out of business and it would incur no fixed or variable costs.10 26. To quantify fixed costs, I used a standard statistical formula found in numerous

statistics textbooks that estimates the intercept value from the slope coefficient (b) in the equation: Intercept = average Y minus [b times average X]11 Where: Intercept = estimated fixed costs average Y = Average Total Cost12 b = slope of the cost equation at Average Revenue and average X = Average Revenue 27. In other words, at Average Revenue fixed cost is equal to the Average Total Cost

minus the slope (b) times Average Revenue. My model, therefore, only generates one output; namely the ratio of variable to total costs13 at Average Revenue. I then use the ratio to quantify a variable cost percentage in my subsequent analysis (about which, Levy appears to have no criticism). 28. On the other hand, Levy uses his equation to estimate the total cost function, and

then subtracts a regression-calculated intercept value that has no economic meaning to derive variable costs. That was not my model's purpose and Levy is incorrect to suggest otherwise. 29. My methodology and the estimate it derives are premised on certain fundamental

accounting principles. First, by definition, variable costs are zero when revenues are zero. This Gujarti, Basic Econometrics at 155-157 (Exhibit 17). As a practical matter, the firm would be shedding costs as its business shrunk from billions of dollars to zero until at the time revenues were zero it would have no costs at all. 11 Pindyck and Rubinfeld, Econometric Models and Economic Forecasts (2d Ed.) at 156, and Macfie and Nufrio, Applied Statistics for Public Policy at 430. I have included the relevant excerpts as Exhibits 15 and 14, respectively. 12 Oracle provided quarterly accounting data. Accordingly, Average Total Cost is the average of the quarterly actual total costs incurred by the relevant firm entities.
10

9

13

The variable cost ratio is one minus the fixed cost ratio 10
CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

SFI-649475v1

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

is inarguable. I have included in Exhibits 7 to 13 relevant excerpts from microeconomics textbooks to illustrate this. Second, my analysis estimates relevant variable cost not incremental cost. Meyer calculated lost profits on the basis of revenue minus direct costs (revenue minus direct costs is generally referred to as gross margin). Meyer applied Oracle's published gross margin on support revenue, which is 90% (i.e., if revenues are $100 then direct expenses are $10 and gross margin is $90). Oracle did not provide its accounting data in a form that would allow incremental costs to be computed. Accordingly, only variable costs are at issue and my analysis is the only evidence on what the relevant variable costs are in this case. Furthermore, it is virtually impossible to compute Oracle's incremental costs over a range of revenue activity in the hundreds of millions of dollars because so many of its costs are partially fixed. In fact, many of the firm's costs may be thought of as "sticky" and are sometimes referred to as step variables. Step variables do not vary directly with revenues but rather remain fixed over a modest range of activity (which means the cost curve is flat over that range of activity), then increase in a jump to a higher value. An example of a step variable is space rental, which tends to be fixed over a given range before jumping or shrinking to a new value when a firm takes on new space or relinquishes excess space. Because the total revenues at issue in this case are a few hundred million dollars (of course, the exact amount is at issue), and because of the limited production of Oracle's accounting information, the only rational way to quantify lost profits is to compute allegedly lost revenues then subtract the relevant variable costs. (Of course, the same is true for SAP on the disgorgement of profits computation). 30. Oracle's annual report, 10-K, confirms the gross margin of approximately 90% but

also includes a statement14 that says the reported gross margin does not include all of the costs incurred to generate the revenue. Therefore, the evidence Oracle produced in this case proves Oracle Corporation Form 10-K for the fiscal year ended May 31, 2007, page 103, footnote 2, states, "The margins reported reflect only the direct controllable costs and expenses of each line of business and do not represent the actual margins for each operating segment because they do not contain an allocation of product development, information technology, marketing and partner programs, and corporate and general and administrative expenses incurred in support of the lines of business. Additionally, the margins do not reflect the amortization of intangible assets, restructuring costs, acquisition related costs or stock-based compensation." I have included the relevant portion as Exhibit 18.
SFI-649475v1 14

11

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

that Meyer's estimate of the deductible costs is wrong. Meyer ignores this statement from Oracle's 10-K even though it is an admission by Oracle that the 90% margin he applied in computing Oracle lost profits is overstated, which overstates lost profits. 31. Lost profits in this case are Oracle's lost revenues arising as a result of the action

alleged minus the variable costs incurred to earn those revenues. Therefore, I quantified the variable costs Oracle would have incurred to generate its lost revenue. In addition to a regression analysis, there are other ways to estimate variable costs. However, other methods involve analysis of a company's detailed income statements, which is impractical in large, complex companies like Oracle that have tens of thousands of accounts in their immense general ledgers. Furthermore, Oracle did not produce all of the accounting information needed to perform such an analytical approach. 32. Oracle's and SAP's published financial statements show that they manage the

business such that for all practical purposes all direct and operating expenses are variable. As the graphs below show (Figures 1 and 2), SAP's operating margin has been within a few percentage points of 25% while revenues have increased from about 7 billion up to almost 11.6 billion which is an almost 66% increase. Oracle has exhibited a similar pattern. Although its revenues increased from $9.5 billion in 2002 to $23.3 billion in 2009, a 145% increase, its margins have barely changed and have varied within a few percentage points of 35%. 33. The SAP and Oracle financial statements in their annual reports reveal that no

matter what happens to revenues, operating profit margin remains within a tight range. Therefore, both companies have demonstrated their ability to add or shed fixed expenses rapidly and with almost total freedom of action in order to maintain their net margins. I have not taken the aggressive step of claiming that all of Oracle's expenses should be deducted from their allegedly lost revenues to compute lost profits, but the 90% margin Meyer applied is far too high and admittedly wrong. Interestingly, Levy proves Meyer wrong in his use of the 90% margin. Levy opines that the OUSA variable costs are 35% of revenue which directly contradicts Meyer's opinion that variable costs are 10% of revenue.

SFI-649475v1

12

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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
SFI-649475v1
5, 000 ($, millions) 20, 000 25, 000

Figure 3
SAP Operating Margin
14,000 12,000 10,000 8,000 6,000 4,000 2,000 2003 2004 2005 To t al Revenue 2006 2007 2008 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

Op erat in g Margin

Figure 4
Oracle Operating Margin
100% 90% 80% 70% 15, 000 60% 50% 10, 000 40% 30% 20% 10% 0 2002 2003 2004 2005 2006 2007 2008 2009 Fis c al Year T otal Revenues Operating Margin 0%

13

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

34.

Based on the foregoing accounting analysis, I quantified variable costs by

modeling costs against revenue according to standard microeconomic principles to arrive at a factor measuring the change in cost due to a change in revenue. Using my model output, I applied the variable cost percentage to lost revenues to estimate the lost profits of each Oracle entity at issue in this case over the relevant period. Levy's criticisms of "estimates." 35. In his Declaration, Levy is confused with regard to what I was estimating. On

page four of his declaration, he claims that I attempted to estimate total costs, variable costs, and fixed costs. Then on page five of his declaration, he claims that I estimated average costs. Levy is incorrect and he is misleading the Court. I never use the equation to estimate the actual costs (fixed or variable) for any level of revenue. I am also not estimating average costs. Rather, I use the equation only to identify the slope of the variable cost curve which I then use to quantify the percentage of variable to total cost. My model is valid for my purposes and follows standard microeconomic principles. Levy and Meyer offer no alternative method to quantify the relevant expenses incurred by the Oracle entities at issue in this case, and that determination is absolutely required to properly compute lost profits damages, because Oracle admits that the gross margin Meyer used in his analysis overstates Oracle's gross profits. Levy's criticisms of my understanding of log-log models. 36. Levy states that I do not understand the relationship between variables in a double

log model (which Levy refers to as a log-log model). Levy takes issue with my statement that the intercept in the double log model is meaningless. In my deposition I said, "although there's an intercept embodied in the calculation, that intercept has no meaning. There is no use in my analysis of an intercept value independent of its role in that log function." Exhibit 19 (6/10/10 Clarke Tr. at 962:19-23). 37. In a double log model the intercept cannot be used to calculate fixed costs, so the

intercept has no meaning outside the terms of the function in which it is stated. Although the intercept cannot be used to directly calculate fixed costs, it is still required in the equation. In other words, the equation
SFI-649475v1

14

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

Total Cost = aX

b

would be incomplete without the "a" term. That the "a" is not usable to estimate fixed costs (i.e., costs when revenue is zero) is immediately obvious if you substitute zero for X in the equation because the result is zero which means fixed costs are also zero. Irrelevant hypothetical scenarios using simulated data. 38. None of Levy's hypothetical scenarios are properly specified because they do not

conform to appropriate cost accounting, nor do they fit standard microeconomic principles. In their textbook Managerial Economics, 15 Samuelson and Marks indicate clearly that you must use a model that makes a priori economic sense. None of the hypothetical scenarios Levy presents do so. For example, Levy's Figure 4, Scenario 1 suggests that all costs are fixed and none are variable. Scenario 2 suggests that variable costs decline as revenues increase. Scenario 3 suggests that fixed costs are negative. None of these models pass muster as economically sensible and are actually misleading. Levy's criticisms of bias in regressions. 39. As I said in my deposition, most time-series regression analyses have some degree

of autocorrelation. It is in the nature of time-series analysis for that to happen because so many variables change in a certain manner over time (for example, population tends to grow over time so modeling a set of variables that include population and time frequently results in the model exhibiting autocorrelation). But even if a time-series regression has autocorrelation, the estimated regression coefficients are still unbiased.16 The problem autocorrelation may cause is that the R2, F-statistic, and t-statistics are overstated, making the equation appear stronger than it actually is. But the regression coefficients themselves are unbiased by the autocorrelation. Levy's criticisms of my treatment of autocorrelation. 40. Autocorrelation is present when there is a pattern in the error terms derived by the

regression equation. Such patterns often arise in data that has seasonality such as buying patterns for toys, which show a pattern of peaking in the fourth calendar quarter. Levy criticizes my I have included relevant excerpts in Exhibit 20. 16 Pindyck and Rubinfeld, Econometric Models and Economic Forecasts at 153; Schmidt, Econometrics at 223. I have included relevant excerpts as Exhibits 15 and 16, respectively.
SFI-649475v1 15

15

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

treatment of autocorrelation. Any criticism of my analysis because of alleged inappropriate treatment of autocorrelation is now irrelevant because only the "Oracle as a Whole" exhibited autocorrelation and as a result of this Court's ruling on Defendants' summary judgment motion, Oracle as a Whole is now out of the case. However, I deal with Levy's criticisms as if Oracle as a Whole were still relevant. 41. Levy states that I did not check for autocorrelation. However, for all of my

equations I checked for autocorrelation using a Durbin-Watson statistic and provided the calculation in Appendices M and U to my report. As I wrote in my report, the Durbin-Watson statistics were: 2.07 (no autocorrelation) for SAP, 1.17 (inconclusive for autocorrelation) for OUSA, and 0.86 (autocorrelation present) for Oracle as a Whole. Therefore, only Oracle as a Whole exhibited autocorrelation and Oracle as a Whole is no longer in the case. In addition, as I said in deposition, with such high outputs for Corrected R2 (over 89%), F-statistic (432), and tstatistic (greater than 20), I determined there was no need for an autocorrelation adjustment. Had I adjusted for autocorrelation, the R2 would have been reduced but not to any significant degree. Levy's criticisms of F-test. 42. The F-statistic measures whether variances are different between two or more

populations. During my deposition I was given a document for an F-test that was not necessary in this case. The document referred to a Chow test, which is used to determine whether the slopes of two equations are different. However, I was not comparing the slopes of two different equations in my analysis, so I did not need to do a Chow test. The questioning attorney evidently was confused about which type of F-test he was referring to in his question. I answered regarding an F-test for the significance of an entire equation, a perfectly legitimate response to his question. I stated that the test was not applicable because I already knew the equation was statistically significant (based on R2) and, because there was only one variable, an F-test is unnecessary. Levy's criticisms of lack of fixed effects analysis. 43. Levy criticizes my SAP equation for not considering fixed effects. I am certainly

aware that techniques can be employed to qualitatively account for different variables, although I refer to them as "dummy variables" rather than fixed effects. I have used them extensively. Levy
SFI-649475v1

16

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)

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

specifies an alternative SAP model with 16 additional dummy variables and suggests that the variable cost factor is 16% points (58.2% minus 42.2%) less than the figure I calculated. His analysis is inappropriate and results in a flawed opinion. 44. The accounting data SAP provided were not detailed enough to allow a qualitative

accounting analysis (in such an analysis, each account is considered in turn and classified as being fixed, variable or a blend of both ­ based on the analysis, a variable cost percentage is estimated). Accordingly, I was forced to use regression analysis to quantify variable costs and Meyer should have done something similar. 45. Had Levy checked the SAP accounting data produced in this case, he would have

discarded his fixed effects model because the data are based on geography, not functional area or revenue. According to SAP's 2009 Annual Report (p. 238): "Our internal reporting system produces reports in which business activities are presented in a variety of ways, for example, by line of business, geography and areas of responsibility of the individual Executive Board members (Board areas)." However, SAP only produced geographical information in discovery in this case. 46. SAP AG in Germany is the corporate headquarters for SAP and incurs numerous

costs on behalf of its subsidiaries. In fact, SAP AG alone accounts for over one third of all SAP costs and revenues. Any analysis by country using Levy's suggested approach would be significantly affected by interference from SAP AG cross-charges and its intercompany accounting policies. Once again, therefore, Levy's lack of understanding of standard accounting practices leads him to a flawed conclusion. Separating these entities by including dummy variables to account for countries (or size) would significantly bias both the slope and intercepts of the resulting equations. Accordingly, I did not pursue the dummy variable approach Levy recommends and instead used the straight panel data to estimate the average variable cost factor. // // // //
SFI-649475v1

17

CLARKE DECL. ISO DEFS.' OPP. TO PLFFS.' MOT. TO EXCLUDE CLARKE Case No. 07-CV-1658 PJH(EDL)