Thursday, 30 September 2021

Essentiality Rate Inflation and Random Variability in SEP Counts with Sampling and Essentiality Checking for Top-Down FRAND Royalty Rate Setting

Fair, reasonable and non-discriminatory (FRAND) royalty rates for licensing standard-essential patents (SEPs) are increasingly derived “top-down” by dividing a notional aggregate royalty percentage (e.g., 10% of a smartphone’s selling price for 4G LTE) among patent owners based on the proportions of patents they own that are deemed to be standard essential.[1] Other rate-setting methods include use of comparable licenses and measuring value derived from SEPs. However, as stated by Justice Birss in Unwired Planet: “In assessing a FRAND rate counting patents is inevitable.”[2] He used top-down methodology as a cross check. The FRAND rate Decision in TCL v. Ericsson, that was unanimously and entirely vacated on appeal, also relied on a top-down valuation.[3]

Ball colour, not patent essentiality, can be identified
 with 100% accuracy when sampling

Essentiality checking is also proposed by the European Commission and others to improve transparency for prospective licensees.[4]  But transparency is only legitimate if what is being revealed and counted is reasonably accurate and does not mislead. Otherwise, it will likely do more harm than good.

If patent counting with essentiality checking is going to be used in FRAND-rate determinations, then it is vital we understand its dynamics, failings, how to properly interpret its results, and how to design and size essentiality determination studies that are fit for purpose. This research article contributes to that quest.

Figuring out which patents are true SEPs and then counting them is fundamental to top-down analysis. With hundreds of thousands of patents declared possibly standard essential,[5] and the insuperably huge task in checking them all for essentiality, the European Commission and others propose that assessing only samples of patents declared essential to a standard would suffice.

Institutionalising use of patent counting—with or without sampling—is an under-researched and impetuous leap of faith. Different assessors come up with wildly different patent counts. Comparison of two separate assessors determining essentiality on the same sample of patents indicates that assessors tend to inflate true essentiality rates. These over-estimates result from statistical bias with numbers of false positives (i.e., truly not essential patents being found to be essential) exceeding false negatives when true essentiality rates are rather less than 50%.

I also conclude from my simulation modelling and analysis that:

1. The lower the true essentiality rate and the lower the rate of agreement among different assessors, the larger the differences will be between the essentiality rates determined by assessors and the true essentiality rate. For example:

a. If true essentiality rates are 30% and two different assessors agree with each other on 75% of their determinations, they will tend to estimate 36% essentiality rates and be accurate in 85% their determinations. 

b. If true essentiality rates are only around 10% (e.g., for 4G LTE or 5G), as some experts plausibly argue, and if two assessors agree with each other on 84% of their determinations, they will tend to estimate 17% essentiality rates and be accurate in 91% of their determinations. That means there will be nearly as many false positives as correct determinations of essentiality.

2. Therefore, if true essentiality rates are at the lower end of expectations, for example, at around 10% or less, it is imperative assessors are highly accurate in their determinations, otherwise false positives will swamp their correct determinations and make their overall results meaningless.

Sampling reduces the precision of SEP counts with increased variability, which is exacerbated by erroneous essentiality determinations. Sampling theory and simple simulations using results of patent-counting studies already undertaken—including several with sample sizes below a few hundred—reveal unacceptably large ranges in expected essentiality rate determinations (i.e. the percentage of declared-essential patents that are deemed to be essential)[6] at what various study authors regard as the “well accepted bound” of the 95% confidence level.[7]  This variability is particularly large where patent essentiality rates are at low levels, such as at around 10%. For example, 10% ± 1.5% is actually ± 15% variability as a proportion of that 10% figure. The quantitative analysis I undertook for this article measures the extent of diminutions, which should be properly and fully considered before sample sizes are set, and before the short cut of sampling is blindly adopted at all.

Top-down methodology seemingly enables precise assessments of FRAND royalties, but this is an illusion due to various inconsistencies and inaccuracies in patent selection, sampling and essentiality assessment. Key questions are how much precision is adequate and how can that be obtained? The optimal balance between extensive and costly patent-essentiality assessments, that can take days of work per patent, versus reducing the number of assessments by sampling should be decided with due regard to accuracy and confidence levels required, and based on empirical assessments.

While there are no set bounds for the acceptably accurate range in determinations, I have considered a reasonable proportionate accuracy requirement for essentiality rate determination to be <± 15% (i.e., a 30% range for the determined essentiality rate as a proportion of the true essentiality rate) at the 95% confidence interval level. With my opinion that true essentiality rates are more like 10% than 30% or 40%, I conclude from my analysis that samples including thousands of patents are required in top-down FRAND-royalty rate setting. For example, if the essentiality rate is only 10%, a sample size approaching 3,000 declared-essential patents per standard, at the very least, would be required.

My full article, including detailed statistical analysis, can be downloaded here.

[1] These percentages in mobile phone licensing are typically applied to the wholesale selling prices of finished goods products.

[2]  Approved Judgment in Unwired Planet versus Huawei, 5th April 2017 at 806 (11).

[5] Declaring one’s patents that are possibly standard essential is a requirement for participation in organisations such as 3GPP in the setting of standards such as 4G LTE and 5G.

[6] This is also called the essentiality ratio.

No comments: