Current approaches to measurement

• Constrained number of layers (4 to 8 in most papers); no metric for U.S. firms
• # of CEO reports: view of the top, <1% of employees; data pre-2000
• Homogeneous structures imposed across firms; titles mapped to one occupational code system

Hierarchies: measurable and important

• Resumes of 18.1 million workers at 3,454 US public firms
• Firm-specific algorithm estimates the number of hierarchical layers
• Average firm has 6.4 layers (median 6), pyramidal shape
• More layers: longer tenure, more internal promotion, higher operating performance, lower volatility
• AI adoption predicts flattening within firm

Three classes of models

• Knowledge hierarchy: managers as problem solvers handling exceptions from below; layers add specialists (Garicano 2000; Garicano and Rossi-Hansberg 2006; Caliendo et al. 2015)
• Information hierarchy: CEO’s span limited by information processing → delegation via layers (Radner 1993; Bolton and Dewatripont 1994; Stein 2002)
• Incentive hierarchy: promotions provide incentives; layers create the career path; tournaments (Williamson 1967; Lazear and Rosen 1981; Malcomson 1984)
• The tradeoff: gains from specialization vs. costs of communication + fixed cost of training managers

Goal: worker histories with public firms

• Two vendors: Revelio (2013–2024, default) + CoreSignal (longer histories; used where Revelio inadequate)

In and out of sample differences

• In-sample firms are larger, more levered, higher-level profits (directions unchanged)

Summary statistics

Algorithm rules

Not all employees

Layers over time: flat to weakly increasing

• Mean 6.39 (2015) → 6.55 (2024); median 6 throughout
• 19% of firms ever change their layer count

Is LinkedIn representative? BLS comparison

• Covers 73% of reported occupations and 89% of BLS employment in them
• Occupations absent industry-wide: mid-pay nationally
• Occupations absent within a firm: lower-pay relative to that firm, sit mid-rank or off the main ILM
• Undercounted workers: the bottom of the firm

Worker experience in higher layers

• The clustering algorithm sorts workers by experience level (since first job or BA)

External gradients: wages, work styles, skill

• BLS wages: top rank quintile pays about twice the bottom
• O*NET work styles: leadership, innovation, initiative rise up the hierarchy
• O*NET Job Zone: top quintile sits more than a full zone above the bottom

Known job ladders: investment banks

• Jefferies: 7 layers; analyst → associate → VP → MD land on rising layers (modal layers 2/3/5/6)
• Evercore: 7 layers; same weakly increasing ladder (2/3/4/5)
• Lazard not in sample

Human capital: education

• High-layer firms naturally have more promotion opportunities; theory predicts wage increases with layer

Human capital: tenure and turnover

• Fewer new hires (−0.015***), more layer promotions (0.0053***), longer tenure (0.067***/0.094***)

Employee equity and the CEO

• ESOP more likely (0.095***); CEO pay ratio higher (35.3***)
• Internal CEO: positive but not significant

Productivity and costs

• Benefits: ROA 0.0376***, net margin 0.184** (gross margin n.s.)
• Costs: revenue per employee −110.4*** (“more managers not directly involved in production”); wages positive but n.s. (“more expensive human capital”, small Compustat wage sample)

Innovation: less R&D, more patents, same quality

• R&D/assets −0.021***; # patents Poisson +1.23***, IHS +0.25***; citations n.s.

More layers, lower stock volatility

• −0.044*** / −0.042*** (raw / winsorized); N 20,071; 3,146 firms
• About 3% lower volatility for a 1 s.d. increase in log layers

Business or financial risk?

• Lower operating-asset volatility (−0.033**); cash-flow vol n.s.; debt n.s.; less cash (−0.025***)

Hierarchy-sorted portfolio returns

• Each year sort firms into quartiles of lagged log layers; long top quartile, short bottom; value-weighted; 2016 to 2024 (108 months)
• Negative alphas regardless of the asset pricing model (CAPM through the 6-factor model); statistically significant in 5 of 6 specifications

Interpreting the negative alphas

• Explanation 1: investors overestimate high-layer firms’ future performance. Little support: IBES analyst EPS forecasts show no systematic over- or underestimation of high- vs. low-layer firms
• Explanation 2 (suggestive): layers may capture a priced risk factor the models miss. High-layer firms are significantly less risky across several metrics; low-layer firms may carry an “organizational risk” premium
• Factor loadings agree: profitability (RMW) loads positively, size (SMB) negatively; high-layer stocks look like large-cap quality firms

Putting it altogether

• Tradeoffs of a more hierarchical structure; theory only partial guidance
• Prior measures impose a common 4 to 8 layer grid on every firm
• Firm-specific measure: middle half of firms 5 to 8 layers in 2024; 18% outside the 4 to 8 band; up to 18 layers
• O*NET Job Zone clustering: every public firm 5 layers, no variation

Going public: more layers after the IPO

• Bernstein (2012); Bias et al. (JF 2026) German hierarchization

Introduction of AI: fewer layers within firm

• Garicano (2000): cheaper knowledge acquisition lowers the need for layers
• Babina et al. (2024) measures; all four AI proxies negative (−0.42*** … −0.072***); firm FE; N 6,386

Industry variation of hierarchical layers

• Few papers predict hierarchy differences by industry
• Stein (2002): banks with more soft information should have fewer layers (“decentralized”)