BMS-Conference

This repository holds the code and data behind our CONFLUENCE-2026 conference paper (BMS College of Commerce and Management, Bengaluru, Theme 5: Digital Economy, Business Analytics & Strategic Management):

R Vinay Kumar and S. Dutta (2026). Fiscal Discipline and Foreign Direct Investment among India's Top 10 FDI-Receiving States: A Computational Panel Analysis, FY 2019-20 to FY 2023-24. Department of Advanced Computing, St Joseph's University, Bengaluru.

It contains the raw inputs, the processed panel, the official documents the figures were copied from, every script, and the outputs of the last full run. You can rebuild each table and figure and every number quoted in the paper from these files alone.

Reproduce

You need Python 3.11 or newer (tested on 3.13) and pdftotext from Poppler, which ships with Git for Windows and with most Linux distributions.

pip install -r requirements.txt
python code/run_all.py

run_all.py runs the scripts in order and saves one log per script in results/logs/. The last step, verify_paper_numbers.py, reads the regenerated outputs and compares them with each value printed in the paper (137 checks). It exits with an error if any of them differ. The full run took about 50 minutes on a laptop. Most of that time goes on the wild cluster bootstraps (1,999 draws per coefficient) and the random-forest grid search.

Layout

data/raw/           inputs as collected: FDI, GSDP, Census controls, FC-16 fiscal tables
data/processed/     MASTER_PANEL_DATASET.csv (10 states x 5 years) and its column dictionary
data/sources/       the FC-16 Volume II annexures and the DPIIT FDI factsheet, as published
code/               analysis scripts (listed below)
results/            regenerated tables, logs and diagnostic figures
figures/            Figures 1 and 2 of the paper

Which script produces what

In the paper	Script	Output
Source checks: 200 fiscal values vs FC-16; sample vs DPIIT ranking	verify_against_fc16.py, verify_against_dpiit.py	results/logs/
Table 1, variance inflation factors	analysis.py	results/econometrics/Table1_*.csv, Table5_VIF.csv
Table 2, Mundlak test, rank correlations, regional statistics, variance split, FY 2023-24 FDI shares, growth models	conference_estimates.py	results/conference/paper_numbers.csv
Wild cluster bootstrap procedure (shared by Table 2)	specification_tests.py	results/diagnostics/TableS4_wild_cluster_bootstrap.csv
Table 3, permutation test, leave-one-state-out ranks, out-of-sample R²	ml_validation_v2.py	results/ml/panel_aware_v2/
Urbanisation leakage in the first machine-learning run	ml_validation.py	results/ml/lasso_shap_original_7var/TableM3_shap_importance.csv
Regional confound, detailed	adversarial_checks.py, adversarial_checks_2.py	results/logs/
26-state extension	widen_panel.py	results/logs/widen_panel.log
Figures 1 and 2	make_figures.py	figures/
Every number in the paper, checked	verify_paper_numbers.py	results/logs/verify_paper_numbers.log

review3_reruns.py and ml_validation_corrected.py are supporting reruns kept for completeness.

Data sources

Series	File	Publisher
Fiscal deficit, revenue deficit, outstanding liabilities, own tax revenue (% of GSDP), FY 2019-20 to FY 2023-24	data/raw/fiscal_data_source_tables/, data/sources/FC16_Vol2_Annexures.pdf	Sixteenth Finance Commission, Report Volume II, Annexures 5.1-5.5 (Government of India, 2025)
State-wise FDI equity inflows, October 2019 to March 2024	data/raw/FDI_top_states_fiscalyear_INRcrore.csv	Department for Promotion of Industry and Internal Trade
National FDI equity inflow and later state totals	data/sources/DPIIT_FDI_Factsheet_June2025.pdf	DPIIT quarterly FDI fact sheet, data to June 2025
GSDP at current prices, 2011-12 base	data/raw/GSDP_statewise_2018-19_to_2023-24.csv	Ministry of Statistics and Programme Implementation
Literacy and urbanisation	data/raw/Control_Variables.xlsx	Census of India 2011

Things to know about the data

• DPIIT began its state-wise series in October 2019. The FY 2019-20 FDI column therefore covers six months (October to March).
• DPIIT shows no separate Uttar Pradesh figure before FY 2021-22, which leaves 48 FDI observations out of 50 state-years.
• FC-16 does not tabulate capital expenditure. The panel's capital-expenditure column is GFD minus RD (capital outlay plus net lending, net of miscellaneous capital receipts), which runs high for states with revenue surpluses. review3_reruns.py repeats the FDI model with reported capital outlay for the two years where it is available.
• The ten states are the largest FDI recipients among the states that FC-16 reports on. Delhi ranks fourth on DPIIT's list but has no FC-16 data, so West Bengal is included instead.

Contact

R Vinay Kumar, rvinaykumar6924@gmail.com