Intangible capital: depreciation rates and stocks

This repository contains the parameter estimates for intangible capital accumulation and the estimated knowledge and organization capital stocks from Ewens, Peters and Wang (2024), "Measuring Intangible Capital with Market Prices." It also contains the code that builds the stocks from Compustat, so the stocks can be refreshed as Compustat updates.

Update history

• April 2019: updated the goodwill adjustment and fixed a bug in the estimation code. The parameter estimates changed.
• October 2023: fixed a small data error in the estimation sample. Small changes to the parameter estimates and stocks; no meaningful change to the relative performance of the paper's intangible stocks.
• September 2026: the stocks are now rebuilt from current Compustat with the October 2023 parameter estimates (unchanged) and refreshed as new fiscal years close in Compustat (a few times a year, since fiscal year-ends differ across firms). The construction code is in pipeline/. Coverage extends from fiscal 1975 through the latest reported fiscal year.

Primer on capitalizing intangibles: the perpetual inventory model

See Sections 2 and 4 of Ewens, Peters and Wang (2024). Parameters of interest:

• : the depreciation rate of knowledge capital investment (i.e. R&D)
• : the depreciation rate of organization capital investment (i.e. SG&A)
• : the percent of SG&A spending that is treated as investment in organizational capital.

Each of these parameters is estimated for the five Fama-French industries.

Parameter estimates

The depreciation and investment parameters (csv) can be merged onto any dataset of firm-time R&D and SG&A flows using SIC codes. The variables are

• sic: SIC code
• knowDepr: the estimate for the knowledge capital depreciation
• organDepr: the estimate for the organizational capital depreciation
• gamma: the estimate for the percentage of SG&A spending that is considered investment in organizational capital
• industry5: the major Fama-French industry the SIC belongs to. Ewens, Peters and Wang put sic >= 8000 & sic <= 8099 in industry5 == 1 and move some "high-tech" TV/radio providers into consumer (classification scheme).

Estimates (Oct. 2023) with bootstrapped standard errors:

Stocks for Compustat firms

Current release: intangibleCapital_20260924.csv. One row per Compustat firm and fiscal year, 534,089 firm-years, 46,082 firms, fiscal years 1975 to 2026, built from Compustat as of June 30, 2026.

Columns:

• gvkey: the Compustat unique identifier
• fyear: the fiscal year
• orgCapital: organization capital (net) using SG&A
• knowCapital: knowledge capital (net) using R&D
• note: blank when the stocks are computed from reported data. Otherwise one of: total assets not reported; R&D and SG&A reported (Compustat has no total assets for the firm-year but the flows are reported and used as they are); total assets not reported; R&D and SG&A interpolated from adjacent years; total assets not reported; R&D and SG&A missing; no stock computed (missing flows are set to zero only when total assets are reported, so nothing can be filled here); no stock: an earlier year had missing R&D or SG&A (a gap earlier in the firm's history leaves every later stock missing).

All dollars are nominal, in Compustat units (millions). The stocks are net assets, not gross, so a year-on-year change is net investment. Every Compustat firm-year is kept; a missing stock is explained in note rather than dropped.

To load in Stata:

import delimited "https://github.com/michaelewens/Intangible-capital-stocks/raw/master/intangibleCapital_20260924.csv", clear

In Python:

pd.read_csv("https://github.com/michaelewens/Intangible-capital-stocks/raw/master/intangibleCapital_20260924.csv")

The previous release, intangibleCapital_090123.dta (fiscal 1975 to 2019, September 2023), is kept for replication of work that used it.

Historical stocks change between releases

Updates to Compustat (restated financials, revised header industry codes, added or re-dated firm-years), to the deflator, and to the founding-year data can change a firm's historical stocks, because each stock is built from the firm's whole history. Read STOCK_CHANGES.md for the size of the changes between the 2023 and 2026 releases, what drives them, and ten worked examples.

How the stocks are built

The industry-level parameter estimates are applied to each firm's full history of R&D and SG&A in Compustat with the perpetual inventory method. The initial stock at a firm's first Compustat year is imputed from its age (Ritter founding year where available) and the average growth of R&D and SG&A around the IPO. Missing R&D is set to zero from 1977 when assets are reported; missing SG&A is set to zero when assets are reported; remaining gaps are interpolated. Flows are deflated with the CPI for the accumulation and the stocks are converted back to nominal dollars. pipeline/README.md documents every step in order.

Code to construct stocks

pipeline/ holds a Python implementation of the construction, verified against the authors' Stata code on identical inputs (507,638 firm-years, correlation 1.000000). It requires WRDS access for Compustat and the CRSP/Compustat link. See pipeline/README.md for the run steps.

Dijun Liu's earlier Python script reproduces the 2019 stocks with correlations above .99.

Updates

The stocks are annual. New releases are posted a few times a year as fiscal years close in Compustat. RELEASES.md records, for each release, how much the stocks moved relative to the previous one; release.py builds and publishes a release in one command. To be notified, use GitHub's Watch → Custom → Releases on this repository.

Citation

@article{ewensPetersWang2024,
  title={Measuring Intangible Capital with Market Prices},
  author={Ewens, Michael and Peters, Ryan and Wang, Sean},
  journal={Management Science},
  volume={71},
  number={1},
  pages={407--427},
  year={2024},
  doi={10.1287/mnsc.2021.02058}
}

Ewens, Michael, Ryan Peters and Sean Wang. "Measuring Intangible Capital with Market Prices." Management Science 71.1 (2024): 407-427.