PaGNet: A Panel-Aware GBDT--Neural Network for Multi-Target Corporate Tax Avoidance Proxy Forecasting

RESUMO

Forecasting corporate tax avoidance proxies from firm-year panel data is challenging because predictive signals are distributed across short firm histories and related targets, while screening-oriented use requires transparent model behavior. We propose PaGNet (Panel-Aware GBDT-Neural Network), a two-branch hybrid that combines a LightGBM branch using panel-temporal summaries with a Panel-MLP branch using attention-pooled temporal aggregation in a shared-trunk multi-task learning framework. A per-target validation-optimal blender produces both the final prediction and a compact branch-reliance diagnostic without trainable fusion parameters. On the KoTaP panel of 1,754 Korean listed firms from 2011-2024, evaluated under a leakage-free, shared-hyperparameter protocol across four feature regimes, accrual targets route stably to the LightGBM branch, where PaGNet raises explained variance over the strongest six baselines by roughly 0.08-0.11 on the primary split. GETR often leans toward the neural branch while CETR exposes a validation-test branch-selection mismatch rather than stable assignment. A panel-flatten control shows most accrual gains come from observed multi-year base-panel values, with PaGNet's panel-aware representation adding a smaller but directionally consistent refinement. Rolling-origin analysis confirms stable routing and bounds ETR diagnostics to split-specific behavior, identifying a far-horizon split where supervised models underperform naive persistence. PaGNet is therefore best viewed not as a universally superior tabular learner but as a proxy-aware panel model competitive for forecasting with explicit per-target branch reporting.