DEEP NEURAL NETWORK-BASED MODELLING OF INTERGOVERNMENTAL FISCAL TRANSFERS: AN EMPIRICAL INVESTIGATION INTO THE NATIONAL FINANCE COMMISSION AWARD AND THE CONSTITUTIONAL, FISCAL AND ADMINISTRATIVE CONTESTATIONS UNDERLYING FEDERAL DEVOLUTION IN PAKISTAN

Abstract

Pakistan's National Finance Commission (NFC) Award is the pivot of the federal bargain, yet more than fifteen years after the Seventh Award no successor has been agreed. This article assembles an official province-year dataset of NFC transfers (FY1997–FY2027) and combines compositional data analysis, small-cluster panel econometrics and deep learning to examine how the Award has operated in practice. Transfers tracked the formula almost exactly under the 1997 Award but have departed from it persistently since 2010, with a structural break in FY2020-21; the realized allocation is consistent with a revealed population weight of about 0.68 rather than the statutory 0.82. Per-capita transfers have dispersed and then partially converged, while the 1996 Award projections overstated realized transfers by 49 log points, a pure bias error. Regularized deep neural networks and an LSTM forecast transfer growth about as well as simple benchmarks, but none is significantly better. Counterfactuals show that the 2023 census would move 3.25 points of the pool away from Punjab and that no hold-harmless redesign below one point is feasible. The findings clarify the constitutional, fiscal and administrative stakes of the Tenth NFC.