Corporate land-asset adjustment after the formal establishment of China’s national parks: evidence from boundary-distance exposure

1 Introduction

Protected areas now cover an expanding share of the globe, and land and economic activity around their statutory boundaries may adjust as conservation regimes change. The Kunming–Montreal Global Biodiversity Framework calls for effective conservation of at least 30 percent of terrestrial and inland water areas and of marine and coastal areas by 2030 (Convention on Biological Diversity, 2022). When large areas are placed under long-term protection, surrounding land use, development pressure, and other spatial linkages may change, generating effects beyond statutory boundaries (Ament and Cumming, 2016; Pfaff and Robalino, 2017; Boillat et al., 2022). Evaluations confined to ecological outcomes inside parks or aggregate regional indicators may therefore miss factor reallocation by firms in surrounding areas.

China’s national park system pilots provide a setting with a clear institutional transition in which to study such firm responses. China began piloting a national park system in 2015. The specific pilots for the first five parks were launched between 2016 and 2019, and all five were formally established in October 2021. Relative to the pilot stage, formal establishment clarified long-run conservation objectives, boundary governance, and planning coordination, giving firms a more stable basis for forming expectations about future development conditions around the parks. In China, construction land is supplied primarily through local governments, and firms’ access to land is jointly shaped by supply arrangements, planning constraints, and regulatory enforcement (Cai et al., 2013; Cai et al., 2017; Wang et al., 2020; Peng and Zhou, 2024). Land is also location-specific, costly to reallocate, and valuable as collateral (Tuzel, 2010; Chaney et al., 2012; Cvijanović, 2014). Land-use-right accounts may therefore register firms’ adjustment to changing long-term spatial conditions relatively early.

This decision margin remains largely unexplored. The protected-area literature has examined forest loss, biodiversity, household welfare, employment, and regional growth, and has begun to study corporate production and capital-market responses (Andam et al., 2008; Ferraro and Hanauer, 2014; Szabó and Ujhelyi, 2024; Garel et al., 2024; Akbari et al., 2026). The land-regulation literature, in turn, shows that planning and land-supply constraints can alter firm location, investment, and output (Cheshire et al., 2015; Zhang et al., 2024; Zhao et al., 2025). The former rarely observes firms’ long-lived land assets, whereas the latter seldom treats ecological protection as a source of spatial constraint. Together, these literatures leave a direct question: when national parks move from institutional pilots to formal establishment, how do firms at different distances from park boundaries adjust their holdings of land-use rights? Related questions concern when the adjustment occurs, what spatial gradient it follows, and how it relates in time to construction and operating outcomes.

We answer these questions using data for nonfinancial, non-real-estate firms listed on China’s Shanghai and Shenzhen A-share markets from 2015 to 2025. We construct the gross and net book values of land-use rights and related flows directly from the notes to firms’ annual reports. Spatial exposure is measured using each firm’s last valid pre-policy office location, which is fixed over the sample period, and its distance to the nearest national park boundary. The baseline design compares the relative change for firms 30–100 km and 100–200 km from the boundary as the nearest park moves from its pilot stage into the early formal-establishment period in 2022–2023. Because formal establishment occurred in October, 2021 is treated as an institutional transition year. The gross and net book values of land-use rights rise relative to the comparison group primarily in 2022–2023 and then weaken markedly in 2024–2025, while estimates at placebo formal-establishment dates are close to zero. The 30–50 km and 50–100 km bands both show sizable responses. Additions and purchases of land-use rights are concentrated in 2022, whereas the book value of buildings and structures declines and operating outcomes change little over the same period.

This paper makes three contributions. First, it extends the analysis of the economic effects of protected areas to a firm-level decision margin: long-lived corporate land assets. Existing work focuses mainly on ecological outcomes, household welfare, employment, and firm operating performance. Land-use-right accounts provide a distinct measure of long-term asset adjustment relative to fixed-asset construction and contemporaneous operations. Second, the paper exploits the institutional sequence of staggered park pilots followed by unified formal establishment to distinguish the exploratory pilot stage, the early formal-establishment period, and the later period. The concentration of land responses in the first two full fiscal years after formal establishment provides dynamic evidence on how the stabilization of a conservation regime is reflected in long-term corporate investment. Third, the paper links firms’ fixed pre-policy office locations to distance from national park boundaries to identify differences across spatial exposure groups. Both the 30–50 km and 50–100 km bands display sizable land-asset adjustments, while continuous-distance and alternative-window estimates show pronounced distance heterogeneity.

2 Literature review

2.1 Protected areas, national parks, and economic outcomes

A first challenge in evaluating protected areas is their nonrandom siting. Protected areas are often located where development pressure is lower, terrain is less favorable, or population density is lower, so simple comparisons across protected and unprotected areas can confound initial conditions with policy effects. Quasi-experimental studies using matching, boundary comparisons, and related designs generally find that protected areas reduce forest loss and habitat conversion and improve local biodiversity, although effects vary with governance, enforcement capacity, location, and external development pressure (Andam et al., 2008; Geldmann et al., 2013; Nolte et al., 2013; Blackman et al., 2015; Gray et al., 2016; Geldmann et al., 2019; Zang et al., 2022; Chen et al., 2023; Xu et al., 2025). This literature establishes protected areas as spatial governance regimes with meaningful constraints and also suggests that their effects may extend beyond statutory boundaries.

The economic consequences of protection depend on the channels through which those constraints operate. Restrictions on resource use may contract traditional production, while tourism, infrastructure, ecosystem services, and amenity values can generate new income and employment. Welfare and growth effects therefore vary substantially across settings (Sims, 2010; Andam et al., 2010; Ferraro and Hanauer, 2014; Sims and Alix-Garcia, 2017; Naidoo et al., 2019; Sims et al., 2019; Walls et al., 2020; Zhang et al., 2020; Ferris and Frank, 2021; Cheng et al., 2023; Balietti et al., 2024; Szabó and Ujhelyi, 2024). Recent work has begun to examine corporate production, employment, and market valuation (Garel et al., 2024; Akbari et al., 2026), but there is still little direct evidence on how firms adapt to conservation regimes through long-lived spatial inputs. Because land acquisition typically precedes plant construction and capacity formation, corporate land accounts provide a separate window onto these earlier asset adjustments.

2.2 Land-use regulation and corporate land-asset allocation

Land-use regulation affects corporate investment through the supply of developable land, land prices, and location-specific returns. Constraints on development rights change the use value of regulated parcels and can also affect land values and welfare through neighboring land supply and spatial externalities (Turner et al., 2014). Firm-level evidence further shows that planning and land-supply constraints can alter production locations, capital investment, and output, while land-investment rules in Chinese manufacturing can widen the gap between the marginal product and user cost of land (Cheshire et al., 2015; Zhang et al., 2024; Zhao et al., 2025). These findings imply that when ecological protection changes the long-run availability of developable space, land itself can become an early margin of corporate adjustment.

Land also has a dual role as an operating input and a balance-sheet asset. Once acquired, land is costly to relocate or reconfigure, and real-estate values can affect investment and capital structure through collateral channels (Tuzel, 2010; Chaney et al., 2012; Cvijanović, 2014). In China, construction-land supply is additionally shaped by local fiscal conditions, land-market institutions, and regulatory implementation (Cai et al., 2013; Cai et al., 2017; Wang et al., 2020; Peng and Zhou, 2024). Changes in long-run land supply and planning conditions around national parks can therefore influence the timing of land acquisition, the scale of land holdings, and the composition of land assets. Existing land-regulation research rarely examines large-scale ecological protection as a source of persistent spatial change. This paper links protected-area governance with corporate land-asset allocation.

2.3 Spatial scale and boundary effects

The economic effects of conservation policy may extend beyond statutory boundaries. Restrictions on land development can alter the supply of developable land and the location of projects in neighboring areas, while planning coordination, infrastructure, tourism, and ecological governance can reshape development conditions around a park (Pfaff and Robalino, 2017). Existing evidence shows that protected-area effects are sensitive to proximity and the radius of analysis and should therefore be examined at multiple spatial scales (Ament and Cumming, 2016; Herrera et al., 2019; Boillat et al., 2022). The ecological effects of China’s first national parks likewise exhibit spatial heterogeneity (Cao et al., 2025). Distance is thus both a measure of policy exposure and a substantive dimension of the geographic reach of economic responses.

Taken together, the protected-area literature shows that ecological governance can reshape surrounding economic activity. The land-regulation literature shows that firms adjust land inputs in response to persistent spatial constraints, and research on spatial scale shows that policy effects can vary sharply with distance. We bring these strands into a common firm-level framework. The transition from national park pilots to formal establishment provides the temporal variation, land-use-right accounts capture firms’ long-term spatial inputs, and distance to the park boundary measures spatial exposure. This framework allows us to examine how corporate land accounts change across institutional stages and where, and when, those changes are concentrated.

3 Theoretical framework and hypotheses

3.1 Formal national park establishment and corporate land-asset allocation

Relative to the pilot stage, formal national park establishment made the institutional regime more durable and predictable by clarifying establishment plans, boundary governance, and planning coordination. The pilot stage primarily tested institutional arrangements and management systems. After formal establishment, firms could form more stable expectations about future developable space, construction-land supply, and project approval around the parks. Because land-use regulation operates through the supply of developable land, prices, and location-specific returns (Turner et al., 2014; Cheshire et al., 2015; Zhang et al., 2024), the greater institutional certainty associated with formal establishment can affect when firms acquire land and how much land they choose to hold.

Land-use rights are especially likely to reflect these expectations early. Industrial and commercial land is location-specific, slow to acquire, infrequently traded, and costly to reallocate. Once a firm has established production, supply-chain, and market linkages in a location, relocation becomes expensive. When firms anticipate tighter future land supply or changing acquisition conditions, securing land in advance or retaining existing land reserves can preserve future investment options and reduce the cost of searching for land when projects are launched. Zhao et al. (2025) note that manufacturing firms may reserve land for future expansion. The collateral value of land and real estate can further increase the value of retaining such assets (Tuzel, 2010; Chaney et al., 2012; Cvijanović, 2014). Accordingly, the earliest response to formal establishment may appear in the gross and net book values of land-use rights, while fixed-asset construction and operating performance follow different adjustment horizons.

Hypothesis 1: Following the formal establishment of national parks, firms whose pre-policy offices are closer to park boundaries increase their holdings of land-use rights relative to firms located farther away.

3.2 Peripheral spatial exposure and distance heterogeneity

The effect of formal national park establishment on corporate land decisions may vary with distance from the boundary. Planning coordination, expected land supply, and links to transport and industrial networks all have spatial dimensions. Firms in the nearer periphery are more directly exposed to changes in surrounding land markets and planning conditions, while these linkages generally weaken with distance. At the same time, accessibility, industrial linkages, and land-market conditions differ across distance bands, making a nonlinear spatial gradient plausible. Protected-area spillover studies likewise document strong dependence on distance and analytical scale (Ament and Cumming, 2016; Pfaff and Robalino, 2017; Herrera et al., 2019; Boillat et al., 2022).

The baseline sample has estimable support beginning at 30 km from a park boundary. We use 100 km to distinguish the nearer and farther exposure groups and compare land-account changes across the 30–50 km, 50–100 km, and 100–200 km bands. Piecewise-linear distance functions, alternative treatment windows, and progressively more distant comparison groups then assess how sensitive the spatial pattern is to any single cutoff. Figure 1 summarizes the pilot launch dates for the first five national parks and their common formal-establishment date in 2021.

Hypothesis 2: Corporate land-asset adjustment following formal national park establishment is heterogeneous across distance. Within the range supported by the sample, firms whose pre-policy offices are closer to park boundaries respond more strongly on average, and the distance relationship may be nonlinear.

4 Research design

4.1 Data sources and sample construction

The study covers firms listed on China’s Shanghai and Shenzhen A-share markets from 2015 to 2025 and links them to the first five national parks formally established in 2021. China launched the national-level national park system pilot in 2015. The pilot timing for the five parks follows official documents: the Pilot Plan for the Sanjiangyuan National Park System, issued in March 2016 by the General Offices of the CPC Central Committee and the State Council; the Implementation Plan for the Wuyishan National Park System Pilot Area, approved by the National Development and Reform Commission in June 2016; the January 31, 2017 notice issuing the pilot plans for the Northeast China Tiger and Leopard and Giant Panda national park systems (Ting Zi [2017] No. 6); and the Pilot Plan for the Hainan Tropical Rainforest National Park System, issued by the National Park Administration in July 2019. The State Council approved the establishment of the five parks on September 30, 2021 (Guohan [2021] Nos. 101–105), and October 12, 2021, when the first group of national parks was formally announced, is used as the institutional transition date (National Forestry and Grassland Administration, 2021). National-park boundaries are represented by a research-compiled WGS84 (EPSG:4326) vector layer, and the national cartographic context uses a cartographic base derived from the standard China map associated with map approval No. GS(2023)2767. Firm office addresses, geographic coordinates, and standard financial and governance variables come from the CSMAR database. Provincial variables follow National Bureau of Statistics definitions. GDP per capita and the secondary-industry share are compiled from regional accounts in successive China Statistical Yearbooks and the National Data portal. The secondary-industry share equals secondary-industry value added divided by regional GDP. Fiscal self-sufficiency equals general public budget revenue divided by general public budget expenditure. Land-use rights, buildings and structures, and construction-project fields are collected from listed firms’ annual reports. All datasets are merged by stock code and fiscal year.

Land-use-right variables are drawn from the notes to the consolidated financial statements in listed firms’ annual reports. The notes on intangible assets provide the year-end gross and net book values of land-use rights, together with additions, purchases, reductions, amortization, impairment, and related accounting information. Notes on fixed assets and construction in progress are used to construct the buildings-and-structures and new/expanded-production-base outcomes. Monetary amounts are converted to RMB 10,000, and land stocks and flows are matched to the financial, governance, and spatial data using the same stock-code–fiscal-year key. Flow components are coded only when they are explicitly disclosed in the annual report. Undisclosed components remain missing. Where disclosures are comparable, gross land value, accumulated amortization, impairment, and net book value are reconciled through accounting identities, and year-end balances are checked against the following year’s opening balances.

Spatial exposure is measured using each firm’s last valid pre-policy office location and held fixed throughout the sample period. Land outcomes come from consolidated financial statements and therefore capture land-use-right assets at the corporate-group level. Specific projects, production bases, and major operating locations disclosed in annual reports provide supplementary information on firms’ spatial footprints. The baseline estimates accordingly compare changes in group-level land accounts across firms with different exposure to national parks. Figure 2 shows the national distribution of the baseline sample, park boundaries, and the baseline distance groups.

We construct the estimation sample in four steps. First, we retain nonfinancial, non-real-estate A-share firms. Second, we exclude ST, *ST, and PT firms. Third, to keep the analytical sample fixed across control specifications, we drop observations missing a valid pre-policy office address, geographic coordinates, distance to a national park, land-use-right data, or core controls. Fourth, we remove singleton observations that are not estimable under firm, national park × year, and industry × year fixed effects. The complete-case candidate sample contains 1,048 firm-year observations, and the final baseline sample contains 854. Because the data are annual, each park’s official pilot-launch year is treated as the launch year, and PilotStage begins with the first complete fiscal year thereafter and runs through 2020. Data from 2015 are used to construct lagged variables, so the baseline estimation years are 2016–2020 and 2022–2025. Formal establishment occurred in October 2021. We therefore treat 2021 as the institutional transition year, with full-year post-establishment periods beginning in 2022. 2022–2023 define the early formal-establishment period and 2024–2025 the later period. Continuous variables are winsorized at the 1st and 99th percentiles.

4.2 Spatial exposure and variable construction

4.2.1 Spatial exposure

The baseline spatial sample is restricted to firms located 30–200 km outside statutory national park boundaries. The nearest firm in the final estimation sample is 30.41 km from a boundary. Distances are calculated from each fixed pre-formal-establishment office coordinate to each national-park boundary using a park-centered local azimuthal-equidistant projection based on WGS84 coordinates; the nearest park is defined by the minimum boundary distance. Treat equals one for firms whose fixed pre-policy office is 30–100 km from the nearest national park boundary and zero for firms 100–200 km away. The 100 km threshold defines the baseline comparison. Subsequent analyses split the nearer group into 30–50 km and 50–100 km bands, estimate a piecewise-linear distance function, vary the treatment window, and progressively move the comparison group outward to assess sensitivity to the distance definition.

4.2.2 Land-asset outcomes

The primary outcomes are the gross and net book values of land-use rights, which capture cumulative acquisition cost and the book stock after amortization and impairment, respectively. Gross Land is defined as ln[year-end gross book value of land-use rights (RMB 10,000) + 1] and is less sensitive to differences in amortization schedules. Net Land is defined as ln[year-end net book value (RMB 10,000) + 1] and reflects the land stock after accumulated amortization and impairment. LandShare equals the gross book value of land-use rights divided by year-end total assets, and LandIntensity equals the gross book value of land-use rights divided by operating revenue. Together, the four measures capture the scale of land assets, their weight in the balance sheet, and their intensity relative to firm operations.

4.2.3 Controls and extended outcomes

Candidate firm-level controls capture lifecycle and size (LnAge, Size), capital structure and financing conditions (Lev, Cash, CFO), operating conditions (ROA, Growth), asset structure (Tang), and governance and ownership (Top1, SOE). All firm variables enter with a one-year lag. Regional controls include provincial GDP per capita, the secondary-industry share, and fiscal self-sufficiency at the fixed office location and enter contemporaneously. Because some lagged financial variables observed after pilot launch may themselves reflect the evolving policy regime, the preferred specification relies on firm, national park × year, and industry × year fixed effects without time-varying firm or regional controls. Restricted and Full specifications are reported as sensitivity checks. Restricted adds lagged LnAge, Size, Lev, Tang, Top1, and SOE together with provincial PGDP, Industry, and Fiscal. Full additionally includes lagged ROA, Growth, Cash, and CFO.

Extended outcomes are organized around three stages: land acquisition, construction, and operating performance. LandIncrease and LandPurchase record current-period additions to and purchases of land-use rights. Buildings measures buildings and structures within fixed assets, and NewBase identifies annual-report disclosures of newly built or expanded production bases, factories, production lines, or other major self-constructed projects. Asset turnover, sales growth, ROA, operating cash flow, and cash holdings capture contemporaneous operating performance. Examining these outcomes alongside land stocks allows us to compare the timing of changes in land accounts, construction investment, and operating performance after formal establishment. Table 1 summarizes the variable definitions.

Table 1. Variable definitions.

Monetary variables are expressed in RMB 10,000 unless otherwise stated. Continuous variables are winsorized at the 1st and 99th percentiles.

4.3 Empirical models and identification strategy

Identification uses two dimensions of relative variation: within-firm changes as the nearest park moves from the pilot stage to formal establishment, and cross-sectional differences between firms 30–100 km and 100–200 km from the boundary. Firm fixed effects absorb time-invariant firm characteristics. National park × year fixed effects compare firms linked to the same nearest park in the same year, and industry × year fixed effects absorb industry-wide shocks. For treated firms, observations before the first complete pilot fiscal year form the omitted stage category.

4.3.1 Staged baseline model

The staged baseline specification is presented in Equation 1.

Yit = β P PilotStageit + β E EarlyFormalit + β L LateFormalit + μ i + λ nt + δ jt + εit (1)

PilotStage equals one for treated firms during complete fiscal years after the official pilot launch of their nearest national park and through 2020. EarlyFormal covers treated firms in 2022–2023, and LateFormal covers treated firms in 2024–2025. The model includes firm, nearest-national-park × year, and industry × year fixed effects. The main estimand is the difference between the EarlyFormal and PilotStage coefficients, which captures the change for the nearer group relative to the farther group as the nearest park moves from complete pilot years into the early formal-establishment period. Contrasts involving LateFormal describe subsequent evolution. Standard errors are clustered by the pre-policy office city.

4.3.2 Pre-formal-establishment diagnostics and year-specific dynamics

Yit = β P PilotStageit + ∑ τ = 2022 2025 β τ ( FormalYea r τ × Trea t i ) + μ i + λ nt + δ jt + εit (2)

Equation 2 decomposes the formal-establishment period into the four complete fiscal years from 2022 through 2025. Each year-specific interaction directly estimates the group difference relative to PilotStage and therefore traces the post-establishment path. The pre-formal-establishment diagnostic uses firms whose nearest national park entered the pilot by 2017 and assigns placebo formal-establishment dates in 2018 and 2019. To preserve sufficient observations before each placebo date, this diagnostic window begins in the park’s official pilot-launch year. The exercise tests whether a same-direction relative jump for the nearer group was already present before actual formal establishment. Trend sensitivity is assessed by adding a differential linear trend, a differential quadratic trend, and park-specific differential linear trends between the treatment and comparison groups.

4.3.3 Distance gradients and spatial counterfactuals

The distance analysis proceeds in three steps. First, we split the 30–100 km treatment group into 30–50 km and 50–100 km bands and compare each with the 100–200 km group. Second, we estimate a piecewise-linear distance function with knots at 50, 100, and 150 km, allowing the slope to vary across distance intervals. Third, we vary the treatment window and the lower bound of the comparison group and estimate a far-ring placebo comparing 100–150 km with 150–200 km. For the translated-boundary counterfactual, each draw selects a translation distance uniformly from 250 to 500 km and a bearing uniformly from 0 to 360 degrees, applying the same translation vector to all five park polygons. Firm distances, nearest parks, exposure groups, and the estimation sample are reconstructed after every translation. Valid translations have same-park overlap of no more than 20% and retain at least 40 eligible firms, 8 pseudo-treated firms, 8 pseudo-control firms, and 10 city clusters. The algorithm retains 50 valid translations. 484 candidate translations were generated to obtain this set under the stated geometry and support criteria.

4.4 Inference and robustness checks

Baseline standard errors are clustered by the pre-policy office city. The main stage contrast is supplemented with 3,999 wild-cluster bootstrap replications, the year-specific analysis uses 1,999 wild-cluster replications, and several space–time HAC estimators provide additional inference checks. The preferred fixed-effect structure combines firm, national park × year, and industry × year effects, preserving the boundary-distance variation across cities linked to the same nearest park in a given year. City × year fixed effects are reported as a sensitivity specification and absorb all common annual variation within each city. In the baseline sample, 4 of 30 cities contain firms from both distance groups. 17 of 213 city-year cells contain both groups, drawn from 3 cities. A matched-sample estimate re-runs the preferred specification on the observations retained by the city × year model to separate fixed-effect structure from sample composition. Trend sensitivity is examined with differential linear, differential quadratic, and park-specific differential linear trends. Additional checks use a common 2019–2020 pilot-stage comparison window, sequential park omission, alternative distance windows, a far-ring placebo, and translated park boundaries.

5 Empirical results

5.1 Sample structure and descriptive statistics

Table 2 summarizes the distribution of the variables and the composition of the baseline sample. Panel A shows substantial dispersion in the land-asset measures across firms. Panel B shows that Giant Panda National Park accounts for 65.69% of baseline firm-year observations and Wuyishan National Park for 26.46%, or 92.15% combined. These two parks therefore contribute most of the observations underlying the pooled estimate. Panel C also shows concentration within the 30–100 km range, motivating the leave-one-park-out and alternative-distance-window exercises. Panel D shows relatively even coverage across the complete fiscal years on both sides of the institutional transition.

Table 2. Descriptive statistics and baseline sample structure.

Panel A reports descriptive statistics. Panels B–C report baseline composition by nearest national park and distance band. The entries are firm-year observations, firms, and (where shown) shares of the 854-observation baseline sample. Panel D reports the distribution of baseline firm-year observations across complete fiscal years used in estimation. Continuous variables are winsorized at the 1st and 99th percentiles.

Figure 3 provides a complementary view of the sample’s spatial and temporal structure. Panel A reports firm-year observations by nearest national park, while Panel B compares annual observations in the 30–100 km and 100–200 km groups. After excluding the 2021 transition year, both groups remain continuously represented before and after formal establishment. Together with Table 2, the subsequent leave-one-park-out and distance-window analyses assess whether the concentration of observations and differences in group size materially affect the main estimates.

5.2 Formal national park establishment and corporate land assets

5.2.1 Stage effects

Table 3 reports the staged baseline model. Relative to firms 100–200 km from a park boundary, firms 30–100 km away experience increases of 0.433 and 0.461 log points in Gross Land and Net Land, respectively, when moving from the pilot stage to 2022–2023. The city-clustered t-values are 3.63 and 3.50. LandIntensity moves in the same direction, while the LandShare estimate is positive and less precise under conventional city-clustered inference. The two land-stock estimates remain similar when lagged firm and regional controls are added in the Restricted and Full sensitivity specifications. From 2022–2023 to 2024–2025, Gross Land and Net Land decline by 0.363 and 0.440 log points, respectively, indicating that the land-account adjustment is concentrated in the first two complete fiscal years after formal establishment.

Table 3. Stage effects from the pilot period to formal establishment.

All regressions include firm, park × year, and industry × year fixed effects. Coefficient rows report city-clustered t-values in parentheses. *, , and * denote significance at the 10, 5, and 1% levels. The preferred specification contains no time-varying firm or regional controls. Restricted adds lagged LnAge, Size, Lev, Tang, Top1, and SOE together with provincial PGDP, Industry, and Fiscal. Full further adds lagged ROA, Growth, Cash, and CFO. The wild-cluster row reports named test p-values from 3,999 bootstrap replications.

5.2.2 Pre-formal-establishment diagnostics and year-specific dynamics

Table 4 reports the pre-formal-establishment diagnostics together with year-specific post-establishment estimates. Panel A shows that estimates for all four land measures are close to zero and statistically insignificant at placebo formal-establishment dates in 2018 and 2019. Panel B shows that the post-establishment changes in land stocks are concentrated in 2022 and 2023: the year-specific coefficients for Gross Land are 0.520 and 0.366, and those for Net Land are 0.570 and 0.383. The estimates weaken markedly in 2024 and 2025. Under wild-cluster inference in Panel C, Gross Land is significant at the 5 percent level in both 2022 and 2023; Net Land is significant at the 10 percent level in 2022 and at the 5 percent level in 2023. Figure 4A shows the same temporal concentration and subsequent narrowing.

Table 4. Pre-establishment placebos and year-specific effects.

Panel A uses artificial formal-establishment breaks within a broader pre-formal diagnostic window beginning in each park’s official pilot-launch year. This diagnostic is reported separately from the complete-fiscal-year PilotStage used in the main model. The placebo sample includes firms whose nearest park entered the pilot by 2017 (429 observations; 24 city clusters). Hainan is outside this sample because its pilot began in 2019. Panel B reports city-clustered t-values in parentheses. Panel C reports named wild-cluster p-values from 1,999 replications.

5.2.3 Distance heterogeneity

Table 5 examines the spatial range of the land-account response. Relative to firms 100–200 km from a park boundary, both the 30–50 km and 50–100 km groups show positive changes in Gross Land and Net Land. Joint tests for the piecewise-linear distance function yield p-values of 0.0772 for Gross Land and 0.0862 for Net Land, while LandShare and LandIntensity show stronger joint evidence of spatial variation. Figure 4B displays a segmented distance profile. The two core land-stock estimates also remain positive across the alternative treatment windows and comparison-group boundaries reported in Panels B and C.

Table 5. Distance heterogeneity and alternative distance windows.

Panel A reports the baseline 30–50 and 50–100 km bands and the joint test for the piecewise-linear distance function. Panels B–C vary the treatment and comparison windows. Regression cells report city-clustered t-values in parentheses. Named joint tests report p-values directly. All specifications include firm, park × year, and industry × year fixed effects. The joint p-values test whether the four EarlyFormal−PilotStage spline-basis contrast coefficients are jointly zero.

5.3 Identification robustness

Table 6 evaluates the estimates under alternative inference methods, regional annual variation, and differential-trend specifications. Across six space–time HAC estimators, the Gross Land and Net Land contrasts remain strongly significant. Province × year fixed effects yield coefficients of 0.5492 and 0.6109. Re-estimating the preferred model on the same 738 observations retained by the city × year specification produces Gross Land and Net Land contrasts of 0.4451 and 0.4878, both significant at conventional levels and close to the full-sample estimates. Under city × year fixed effects, Gross Land and Net Land are −0.0006 and 0.0351, while LandShare remains positive and significant and LandIntensity is positive at the 10 percent level. The three differential-trend specifications retain positive Gross Land and Net Land point estimates with wider confidence intervals. Restricting the pilot-stage comparison to 2019–2020 yields coefficients of 0.3158 for Gross Land and 0.3291 for Net Land. These results show strong support across alternative inference procedures and several sample and fixed-effect checks, alongside greater uncertainty in the most granular city-year and flexible-trend specifications.

Table 6. Robustness, influence diagnostics, and spatial falsification tests.

Panel A reports named bootstrap p-values; Panels B–G report coefficients with t-values in parentheses. Panel H reports the actual-boundary coefficient with its city-clustered t-value and the empirical placebo p-values. The province × year specification uses 826 observations and 28 clusters. The city × year specification uses 738 observations and 19 city clusters. The matched-sample baseline re-estimates the preferred fixed-effect structure on these same 738 observations. The common 2019–2020 pilot window uses 545 observations and 28 clusters. Panel G uses 244 observations and 15 city clusters. Panel H uses 50 valid boundary translations of 250–500 km; empirical p-values compare the absolute t-statistic at the actual boundary with the translated-boundary distribution.

Alternative spatial specifications provide a second set of checks on the baseline distance definition. Gross Land and Net Land remain positive across five alternative treatment windows and as the lower bound of the comparison group is moved outward. When the treatment window is narrowed to 50–100 km, the estimates are 0.4837 and 0.5096. When 150–200 km is used as the comparison group, they are 0.4384 and 0.4935. Sequential park omission also preserves positive Gross Land and Net Land coefficients, with wider uncertainty when Giant Panda National Park is omitted. The far-ring placebo comparing 100–150 km with 150–200 km is statistically insignificant for all four land measures. Across 50 valid translated boundaries, no placebo absolute t-statistic reaches the actual-boundary statistic for either Gross Land or Net Land. The two-sided plus-one empirical p-value is 0.0196 for both outcomes.

5.4 Spatial exposure and measurement consistency

Section 5.4 evaluates the office-location exposure measure using two complementary checks, as summarized in Table 7. Among observations with both fixed pre-policy office and registered addresses available, the two address measures identify the same nearest park in 87.7% of records and the same 100-km exposure class in 80.9%. We then reconstruct nearest park, pilot timing, the 30–100 km treatment group, the 100–200 km comparison group, park × year fixed effects, and city clusters from the fixed pre-policy registered address. After fixed-effect pruning, the registered-address sample contains 796 firm-year observations and 32 city clusters. The EarlyFormal − PilotStage coefficients are −0.0234 for Gross Land (t = −0.10), −0.0603 for Net Land (t = −0.23), −0.0113 for LandShare (t = −0.97), and −0.0063 for LandIntensity (t = −0.30), all estimated imprecisely around zero.

Table 7. Spatial exposure, alternative-address regressions, and location validation.

Annual-report location evidence refers to disclosures that identify a specific land parcel, project, production base, factory, or major operating location. Address-consistency percentages are calculated for observations with both fixed pre-policy office and registered addresses available. Registered-address regressions reconstruct the nearest park, pilot timing, treatment status, park × year fixed effects, and city clusters using the fixed pre-policy registered address. Coefficient rows report city-clustered t-values in parentheses. GIS reproduction uses the research-compiled WGS84 (EPSG:4326) national-park vector layer and park-centered local azimuthal-equidistant projections for distance calculation.

The registered-address specification produces estimates close to zero for all four land measures. On the same 796 observations, the office-based estimates remain positive: 0.4783 for Gross Land (t = 4.08), 0.5162 for Net Land (t = 3.92), 0.0113 for LandShare (t = 1.97), and 0.0474 for LandIntensity (t = 3.00). These results show sensitivity to the headquarters-location measure. The baseline uses each firm’s fixed pre-policy office location, which is observed before formal establishment, held fixed throughout the sample period, and applied consistently in the main spatial design. Annual-report disclosures of identifiable land parcels, projects, production bases, or major operating locations provide supplementary spatial information for 269 firm-years. An independent distance recalculation using the same park-centered local azimuthal-equidistant projection method reproduces the nearest national park for all 180 baseline firms, with distance differences attributable to numerical rounding.

5.5 Land-account adjustment, construction, and operating outcomes

Table 8 organizes the additional evidence around land flows, year-specific dynamics, land-account components, and operating outcomes. Panel A compares land additions, land purchases, buildings and structures, and new or expanded production bases in the common 336-observation sample. The preferred LandPurchase contrast is positive at the 10 percent level, while the average LandIncrease contrast is positive with wider uncertainty. Panel B shows that both flow measures rise sharply in 2022. Panel C uses the 172 observations for which all accounting components are estimable. DeltaGross is positive at the 5 percent level and LandReduction is negative at the 10 percent level. The remaining components are estimated less precisely. The accounting evidence therefore points most clearly to a higher change in gross land stock and lower land reductions in the common-component sample, with 2022 additions and purchases providing complementary flow evidence.

Table 8. Land-account adjustment, annual dynamics, account components, and operating outcomes.

Panel A uses the common 336-observation sample (94 firms; 23 city clusters). Panel B uses the same sample. Panel C uses 172 estimable observations and 17 city clusters; each component is estimated separately after asinh transformation and 1st/99th-percentile winsorization, so the component-specific coefficients are interpreted separately. In Panel C, DeltaGross denotes the year-to-year change in the gross book value of land-use rights; MergerIncrease denotes additions to land-use rights arising from business combinations; and BridgeResidual is the reconciliation residual defined as DeltaGross minus net disclosed land flows (LandIncrease − LandReduction). Panel D uses the 854-observation baseline sample and 30 city clusters. All regressions include firm, park × year, and industry × year fixed effects. Coefficient rows report city-clustered t-values in parentheses.

Panel D returns to the full baseline sample to examine contemporaneous operating outcomes. Asset turnover, sales growth, ROA, operating cash flow, and cash holdings show no statistically significant EarlyFormal − PilotStage differences. Combined with Panels A–C, the results indicate that land-account changes are concentrated earlier than observable operating changes in the sample. Buildings decline in 2022–2023, while reported land additions and purchases show their strongest increases in 2022.

6 Conclusion

6.1 Main findings

We examine how land-use-right accounts change across firms with different headquarters-based spatial exposure as China’s first five national parks move from institutional pilots to formal establishment. Under the preferred specification, firms whose fixed pre-policy offices are 30–100 km from a park boundary show higher gross and net book values of land-use rights in 2022–2023 relative to firms 100–200 km away, and the gaps narrow markedly in 2024–2025. Placebo formal-establishment dates in 2018 and 2019 yield estimates close to zero. The first two complete fiscal years after formal establishment are therefore the period in which the land-account differences are most concentrated in the sample.

The office-based spatial results also show systematic distance variation. Firms in both the 30–50 km and 50–100 km bands display sizable land-stock changes, the piecewise-distance tests reject the null that the spline-basis contrast coefficients are jointly zero at the 10 percent level for Gross Land and Net Land, and the alternative distance windows preserve positive land-stock estimates. Sequential park omission retains positive coefficients, with wider uncertainty after omitting Giant Panda National Park. The far-ring placebo shows no comparable change, and the actual-boundary statistics exceed those from all 50 valid translated-boundary draws. The pooled magnitude is informed mainly by firms associated with Giant Panda and Wuyishan National Parks, which together account for 92.15% of baseline observations.

The land-flow and operating analyses clarify the composition and timing of the accounting change. In the 336-observation flow sample, the strongest increases in reported land additions and purchases occur in 2022. In the 172-observation common-component sample, DeltaGross rises and LandReduction declines, while the individual acquisition components are estimated less precisely. Buildings decline in 2022–2023 and contemporaneous operating outcomes change little. The combined evidence is consistent with an early adjustment in land accounts preceding broader changes in fixed-asset construction and operations over the observed horizon.

The identification exercises provide additional context for the preferred estimates. Gross Land and Net Land remain positive under wild-cluster inference, space–time HAC estimators, province × year fixed effects, a common pilot-stage window, alternative distance definitions, and sequential park omission. City × year fixed effects and flexible differential trends produce wider uncertainty for the two land-stock measures, while the registered-address specification yields estimates close to zero. Taken together, the results document a temporally concentrated land-account adjustment under the fixed pre-policy office-location design and show how the estimated magnitude varies with alternative controls for local annual shocks, trends, and headquarters-location measurement.

6.2 Theoretical implications

These findings connect the protected-area literature with research on firms’ long-lived asset allocation. Tuzel (2010), Chaney et al. (2012), and Cvijanović (2014) emphasize the roles of asset values and collateral capacity in linking real estate to corporate investment and capital structure. In the 336-observation flow sample, the largest positive changes in reported land additions and purchases occur in 2022, while Buildings declines and contemporaneous operating outcomes change little. In the 172-observation common-component sample, DeltaGross rises and LandReduction falls. Together, the results show that land-use-right accounts contain information on long-lived asset adjustment that is distinct from fixed-asset construction and contemporaneous operating performance. The individual acquisition components provide supporting evidence with greater sampling uncertainty.

Research on land-use regulation typically explains firm responses through supply constraints, development intensity, and allocative efficiency (Turner et al., 2014; Cheshire et al., 2015; Zhao et al., 2025). Our results further indicate that, under the same institutional transition, changes in corporate land accounts vary substantially with distance from the protected boundary: both the 30–50 km and 50–100 km bands exhibit sizable adjustments. This pattern is consistent with the scale dependence emphasized in the protected-area literature (Ament and Cumming, 2016; Pfaff and Robalino, 2017; Boillat et al., 2022) and extends the spatial-scale question to firms’ long-lived asset decisions. In our sample, the land-account adjustment varies by distance band, and the piecewise distance profile is consistent with a nonlinear spatial pattern.

6.3 Policy implications

The early years following formal national park establishment are the period in which land-asset changes among the listed firms in our sample are most concentrated. National park authorities, territorial spatial planning agencies, and construction-land supply authorities can strengthen information coordination during the initial implementation of park plans by monitoring disclosed changes in corporate land-use rights alongside surrounding industrial and commercial land supply, project approvals, construction starts, and commissioning progress. Such coordination can help identify interactions between ecological protection and firms’ long-term land allocation at an earlier stage.

Monitoring can also be stratified by distance, with the relevant radius tailored to the land market, planning conditions, and industrial linkages around each national park. In our sample, listed firms in the 30–100 km group exhibit more pronounced changes in land accounts, providing an empirical reference for where closer monitoring may be most informative. For firms operating across regions, project locations, subsidiary addresses, and parcel coordinates can be combined with group-level land assets to link balance-sheet changes more closely to specific projects and construction progress and thereby support more granular land-use governance around national parks.

6.4 Scope and directions for future research

This study measures corporate spatial exposure using fixed pre-policy office locations and observes group-level land-asset changes through consolidated land-use-right accounts. The registered-address specification yields estimates close to zero, showing that the estimated land-stock relationship varies with the headquarters-location proxy. Sample composition also matters for external validity: Giant Panda and Wuyishan National Parks account for 92.15% of baseline firm-year observations. The pooled relationship is therefore most informative about listed firms associated with these two parks, while the smaller park samples provide limited precision for park-specific magnitudes.

Future research linking parcel transactions, subsidiary locations, project approvals, and construction-permit coordinates can connect group-level land-account changes with specific parcels and project stages. Larger samples around the less represented national parks would also support more precise park-specific estimates, and longer time series can examine how early land allocation relates to subsequent fixed assets, operating revenue, and productivity.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: He Linbo, Hunan Agricultural University, China

Reviewed by: Donghao Duan, Hunan University, China

Zeyu Cao, Beijing Forestry University, China