Environmental Code Reform, Enterprise Abatement Outlays, and Stationary Industrial Emissions: Subnational Evidence from Kazakhstan, 2010–2025

Staggered industrial environmental codes pair multi-year preparatory monitoring windows with scheduled transitions toward best-available-technique permitting and progressive fee multipliers, raising an open mechanism question: do historically emission-intensive industrial regions adjust during early preparatory compliance through enterprise current abatement outlays and end-of-pipe pollutant capture, or does emission-intensity compression emerge as integrated-permit and fee-escalation thresholds arrive? Assembling a boundary-harmonized panel across all 16 consolidated parent regions of Kazakhstan from 2010 to 2025 (N = 256) around Environmental Code No. 400-VI (enacted January 2021, entering into general force July 2021, with Category I integrated permitting and scheduled fee multipliers commencing January 2025), we estimate exposure-gradient regressions with 9,999 null-imposed wild cluster bootstrap replications. Across 2022–2025, one standard deviation higher predetermined 2017–2019 stationary emission exposure is associated with no statistically resolved change in log real enterprise current environmental protection outlays (+0.0947 log points; clustered SE = 0.0869; bootstrap p = 0.3303) and statistically unresolved shifts in stationary pollutant capture shares (-0.0167; 95% bootstrap set [-0.0773, 0.0293]; p = 0.4876) and log captured tonnage (-0.0284; p = 0.8925). Conversely, relative compression in stationary emission intensity concentrates in the 2025 commencement year (-32.85 tonnes per billion 2020 tenge of real industrial output; SE = 13.37; bootstrap p = 0.0080; inverted 95% bootstrap set [-70.60, -8.91]; six-outcome within-phase Holm p = 0.0480; 24-test Holm p = 0.1920). This 2025 gradient strengthens without contemporaneous output controls (-56.79; bootstrap p = 0.0007) and persists across parent-clustered 17-region panels (-30.47; p = 0.0014), 2014–2016 anchors (-34.35; p = 0.0064), and directionally across all 16 single-region deletions (15 significant at 5%; Pavlodar omission p = 0.0660), while coexisting with pre-2020 downward exposure drift that bounds interpretation to conditional subnational convergence.