KPSS Patent Valuation Data Exploration

Data Source

KPSS Patent Values - Market-based valuations of patents from publicly traded firms

• Paper: Kogan, Leonid, Dimitris Papanikolaou, Amit Seru, and Noah Stoffman. “Technological innovation, resource allocation, and growth.” Quarterly Journal of Economics 132.2 (2017): 665-712.
• GitHub Repo: https://github.com/KPSS2017/Technological-Innovation-Resource-Allocation-and-Growth-Extended-Data
• Alternative: Noah Stoffman’s website: https://kelley.iu.edu/nstoffma/
• Coverage: 1926-2010 (original), possibly extended to 2018+

How KPSS Values Are Calculated

Method: Stock market event study around patent grant dates

1. Identify patent grant date
2. Calculate 3-day abnormal returns around grant (using market model)
3. Multiply abnormal return × firm market cap = patent value
4. This gives us a dollar estimate of what investors think the patent is worth

Why this matters: Gives us a market-based valuation rather than just counting patents

Expected Data Structure

Based on the KPSS paper, we expect:

• Patent number (to merge with our USPTO patents)
• Patent value (in dollars)
• Grant date (year or exact date)
• Firm identifier (PERMNO from CRSP or similar)
• Possibly: Standard errors, significance flags, additional firm characteristics

Data We Need to Extract

Core Fields:

• Patent number (to merge with Full_Patent.parquet)
• Patent value in dollars
• Grant year (for our panel timing)

Optional Fields:

• Firm identifiers (to understand which firms our patents come from)
• Confidence intervals (to understand measurement error)
• Industry classifications (for heterogeneity analysis)

Data Structure & Ontology Questions

Key questions to answer when exploring the data:

1. Patent Number Format

• What format are patent numbers in? (e.g., “1234567”, numeric only?)
• Do they match our USPTO patent IDs in Full_Patent.parquet?
• Will we need to standardize patent number formats for merging?

2. Coverage & Completeness

• How many total patents in KPSS data?
• What time period: 1926-2010? Extended to 2018? To 2020?
• What % of our 92K patents will match?
• Critical: Does KPSS cover our analysis period (2000-2020)?

3. Value Distribution

• How many patents have value = 0? (Market didn’t react)
• How skewed is the distribution? (Mean vs. Median)
• What are the outliers? (Blockbuster patents worth billions)
• Should we winsorize extreme values?

4. Public Firm Coverage

• Only covers patents from publicly traded firms
• What % of pharmaceutical patents are from public firms?
• Bias concern: Are public firm patents systematically different?
• Missing: University patents, private company patents, individual inventors

5. Measurement Issues

• Zero values: No market reaction or measurement error?
• Negative values: Are they possible? (Unexpected patent signals overinvestment?)
• Confounding events: What if other news happened on grant date?

Expected Challenges

1. Time Coverage

• KPSS original data: 1926-2010
• Our panel: 2000-2020
• Gap: 2011-2020 may not be covered
• Options:
  • Use KPSS for 2000-2010 subset only
  • Check if extended data exists (2011-2018)
  • Replicate methodology for recent years (requires CRSP access)

2. Match Rate

• KPSS only includes ~3-4M patents from public firms
• Our data has 92K unique patents
• Expected match rate: 20-40%?
• Question: What % of pharma/biotech patents are from public firms?

3. Selection Bias

• Public firms may be systematically different:
  • Larger R&D budgets
  • More commercially oriented
  • Better patent quality
• If GWAS affects public vs. private firms differently, our estimates are biased

4. Zero Values

• Many patents have zero estimated value
• Reasons:
  • Market already anticipated the patent
  • Patent granted on non-trading day
  • Long-term value not captured in 3-day window
  • Patent genuinely has low value
• Question: Include as zeros or drop from analysis?

Next Steps (Manual Download Required)

You’ll need to:

1. Go to https://github.com/KPSS2017/Technological-Innovation-Resource-Allocation-and-Growth-Extended-Data
2. Clone the repo or download the patent values file
3. Check Noah Stoffman’s website for extended data (2011-2018)
4. Place data files in this directory
5. Run the exploration script (to be created next)

Once downloaded, we can:

• Parse the patent value data
• Check time coverage (critical!)
• Calculate match rate with our patents
• Understand value distribution
• Assess feasibility for our analysis

Alternative: Replicate KPSS Methodology

If KPSS data doesn’t extend to 2020, we could replicate:

Requirements:

• CRSP stock price data (daily)
• Patent grant dates from USPTO
• Patent-firm matching (USPTO assignee data)
• Event study code (standard finance methodology)

Feasibility: Medium difficulty

• Need university CRSP subscription (likely have it)
• Code is relatively standard
• Time consuming: ~1-2 weeks

Question for user: Do you have access to CRSP?