Corporate Debt Persistence Extends Across Product-Market Networks

Looking beyond a firm’s own history

Corporate leverage describes a company’s use of debt in its capital structure. In conventional partial-adjustment models, firms move gradually toward a target leverage ratio because changing their financing mix is costly. Persistence in debt ratios is therefore interpreted as evidence of slow adjustment.

In their November 2026 preprint, Stylianos Asimakopoulos and Vasilis Sarafidis, both affiliated with Brunel University London, argue that this interpretation can miss a network component. Their question is whether current leverage also depends on the previous year’s leverage among economically connected firms—not simply on what the company itself borrowed before.

Mapping economic connections

The authors combine accounting and market information from the CRSP–Compustat Merged database with annual Hoberg–Phillips text-based product-similarity measures. Their balanced panel contains 549 U.S. publicly traded firms observed throughout 1990–2023, providing 18,666 firm-year observations before the lags required for estimation.

The primary outcome is book leverage: total debt divided by total assets. Market leverage provides an alternative measure in robustness checks.

Rather than placing every company into a single industry peer group, the network weights firms according to similarity in their product descriptions. More similar firms receive greater relative weight. Connections are updated annually, and the previous year’s leverage is weighted using the connections prevailing in that same previous year.

Separating interaction from shared circumstances

Similar financing choices need not mean that firms influence one another. Connected companies may share economic conditions, or firms with similar financing behaviour may operate in similar product markets.

The authors address these concerns by allowing model coefficients to differ across firms, including their responses to past leverage and financial characteristics. They also allow unobserved common shocks to affect firms differently.

Their instrumental-variable method uses variation in observed characteristics after removing estimated common-factor components. Firm-specific estimates are then averaged, with uncertainty calculations allowing economically proximate firms to have correlated coefficients. Using past rather than simultaneous peer leverage establishes temporal ordering, but the authors explicitly acknowledge that timing alone does not eliminate endogeneity.

What the estimates show

In the preferred annual-network model, the mean coefficient on a firm’s own lagged book leverage is 0.346. The coefficient on network-weighted lagged leverage is 0.283. Both are statistically significant at the 1% level, according to the preprint.

The network coefficient represents approximately 45% of the sum of those two mean coefficients. This is a comparison of coefficient magnitudes, not a finding that networks explain 45% of leverage variation or total system persistence. The authors stress that system persistence also depends on network structure and propagation.

Including network dependence lowers estimated own-firm persistence relative to an otherwise heterogeneous model without network terms. Under the partial-adjustment interpretation, the implied average annual adjustment speed rises from 53.8% to 65.4%. This describes movement toward the model’s target leverage conditional on connected firms’ financing histories, not a directly observed target.

Why changing connections matter

Holding product-market weights fixed produces a network coefficient of 0.570, compared with 0.283 when weights change annually. The estimated own-lag coefficient changes much less.

The authors also report substantially larger indirect effects under fixed connections. With annual weights, network transmission reinforces the estimated leverage responses associated with cash, profitability and tangible assets. By contrast, the indirect effect associated with firm size is small and statistically insignificant. Network amplification is therefore not uniform across financial characteristics.

Firm-specific estimates show significant product-market clustering in own-leverage persistence and responses to tangibility, profitability and size. However, the response to connected firms’ past leverage does not show statistically significant clustering.

What the findings establish—and leave open

Positive lagged network dependence remains statistically significant with market leverage and several alternative network specifications. Industry results are less uniform: Transportation and Retail have statistically insignificant network coefficients.

The evidence concerns a balanced sample of continuously observed U.S. listed firms, rather than all businesses. Its interpretation also depends on the model’s identification assumptions, including valid instruments and restrictions on unobserved disturbances and network timing. The study does not distinguish whether benchmarking, learning or competitive responses generate the estimated dependence.

The authors’ central conclusion is that leverage persistence should not be read solely as a firm-level adjustment phenomenon. Their results attribute an important component to connected firms’ past financing choices, while demonstrating that its measured strength depends on how evolving economic relationships and firm differences are represented.