Practical research code for optimizing LBO covenants under IFRS-16 and frozen-GAAP conventions. Clean, installable Python package with reproducible analysis and a public benchmark.
Purpose: reproducible experiments, publishable manuscript, and an optimization library for covenant design.
- Clone and install (editable):
git clone https://github.com/Aniket2002/ifrs16-lbo-engine.git
cd ifrs16-lbo-engine
pip install -e .- Accor case study (small, reproducible):
python analysis/scripts/case_study_accor.pypytest -qsrc/lbo/— core library (installable)analysis/— experiments, scripts, andpaper/(manuscript + figures)data/— small CSV inputs and benchmarktests/— unit and integration tests
License: MIT
git clone https://github.com/Aniket2002/ifrs16-lbo-engine.git
cd ifrs16-lbo-engine
pip install -e . # Installs as package# 1. Reproduce Accor case study (real company analysis)
python analysis/scripts/case_study_accor.py
# 2. Generate theoretical guarantee proofs
python analysis/scripts/theoretical_guarantees.py
# 3. Run Bayesian calibration pipeline
python analysis/calibration/bayes_calibrate.py
# 4. Create all paper figures in one command
make figuresdocker build -t ifrs16-lbo .
docker run ifrs16-lbo python analysis/scripts/case_study_accor.py📦 Production-Ready Python Package
├── 🔧 src/lbo/ # Core library (pip installable)
│ ├── optimization/ # Bayesian covenant optimization
│ ├── workflows/ # LBO modeling pipelines
│ └── models/ # IFRS-16 compliant engines
│
├── 🔬 analysis/ # Research & experiments
│ ├── scripts/ # Executable analyses
│ ├── calibration/ # Bayesian parameter fitting
│ ├── paper/ # LaTeX manuscript + figures
│ └── figures/ # Generated visualizations
│
├── � data/ # Input datasets
├── 🧪 tests/ # Comprehensive test suite
└── 📋 benchmark_dataset_v1.0/ # Public research benchmark
- Complex derivative pricing: IFRS-16 lease liability valuation
- Risk management: Covenant breach probability modeling
- Regulatory compliance: Dual accounting standard handling
- Portfolio optimization: Multi-objective PE fund optimization
- Bayesian inference: PyMC hierarchical modeling at scale
- Uncertainty quantification: Posterior predictive distributions
- Model validation: Cross-validation with financial time series
- Feature engineering: Financial ratio transformation pipelines
- Clean architecture: Domain-driven design with clear interfaces
- Performance optimization: Closed-form solutions vs Monte Carlo
- Testing strategy: Property-based testing for financial invariants
- Documentation: Research-grade technical writing
- Fast screening for optimization algorithms (10x speedup)
- Validation framework against full simulation with error bounds
- Stochastic optimization of (ICR, Leverage, Sweep) thresholds
- Pareto frontiers mapping IRR vs breach risk trade-offs
- ε-constraint formulation with risk-first covenant design
- Policy maps showing optimal covenant levels vs risk tolerance
- Wilson confidence intervals for success rate estimation
- Bootstrap percentile CIs for robust IRR quantiles
- Multiple IRR calculation methods for cross-validation
- Sobol global sensitivity with first-order (S₁) and total-effect (Sₜ) indices
- F1-F6: Standard methods figures (Monte Carlo, S&U, Sobol, stress)
- F7: Prior vs posterior distributions (Bayesian shrinkage)
- F8: Analytic vs simulation validation (approximation quality)
- F9: First-order elasticities (parameter sensitivities)
- F10: Pareto frontiers (IRR vs breach risk) ← KEY NOVEL OUTPUT
- F11: Policy maps (optimal covenant levels vs risk tolerance)
- Optimized vs baseline covenants (ΔIRR, ΔP(breach), Δheadroom)
- Posterior parameter estimates by firm with shrinkage metrics
- Sobol sensitivity indices with confidence intervals
- Validation statistics (analytic vs simulation errors)
# Benchmark Results (vs Traditional Methods)
covenant_breach_auc: 0.76 ± 0.05 # +18% improvement
headroom_rmse: 0.28 # 46% reduction
computational_speedup: 12.3x # Analytic vs Monte Carlo
model_accuracy: ε ≤ 0.12 # Mathematical guarantee# Material impact quantification
ifrs16_leverage: 5.1x vs frozen_gaap_leverage: 12.6x
ifrs16_icr: 10.6x vs frozen_gaap_icr: 2.6x
covenant_sensitivity: "High - requires dual-convention analysis"- 📄 Full manuscript:
analysis/paper/main.tex - 🔢 Mathematical proofs: Propositions with deterministic error bounds
- 📈 Empirical validation: Multi-company benchmark testing
- 🏆 Publication-ready: Structured for top finance journals
Proposition 1: Analytic Screening Guarantee
├── ε ≤ 0.12 bounded approximation error
├── Computational complexity O(1) vs O(n³)
└── Formal proof in mathematical appendix
Proposition 2: Frontier Monotonicity
├── Pareto-efficiency under uncertainty
├── Bayesian posterior convergence
└── Risk-adjusted optimization guarantees
- 🏢 5 hotel operators with public financial data
- 📋 3 standardized tasks for method comparison
- ✅ Integrity verified with SHA256 checksums
- 🔓 Open access under CC-BY-4.0 license
This project demonstrates advanced capabilities across multiple domains:
| Skill Category | Specific Demonstrations |
|---|---|
| Quantitative Finance | IFRS-16 compliance, derivative valuation, risk modeling, portfolio optimization |
| Machine Learning | Bayesian inference, uncertainty quantification, hierarchical modeling, validation |
| Software Engineering | Clean architecture, performance optimization, comprehensive testing, CI/CD |
| Research Excellence | Mathematical rigor, reproducible science, academic writing, benchmark creation |
| Business Impact | Real company analysis, regulatory compliance, decision support systems |
Key Technical Differentiators:
- ✅ Production-ready code (not just research prototype)
- ✅ Mathematical guarantees (not just empirical results)
- ✅ End-to-end pipeline (data → model → optimization → deployment)
- ✅ Regulatory expertise (IFRS-16, dual accounting standards)
- ✅ Open source contribution (public benchmark for research community)
- Single source of truth: All optimization logic in clearly documented modules
- Deterministic pipeline:
make paper-optimizationreproduces all F1-F11 results - Complete provenance: Git tracking, parameter logging, computational environment capture
- Graceful degradation: Framework works with/without optional optimization dependencies
- Methods contribution: Novel algorithmic framework, not just empirical application
- Practical impact: Industry-applicable covenant optimization with measurable value
- Literature gap: First optimization approach to IFRS-16 covenant design under uncertainty
- Scalable framework: Extensible to other deal structures and accounting standards
When using this framework in academic research, please reference:
- Bayesian hierarchical calibration methodology with partial pooling
- Analytic headroom approximations for IFRS-16 covenant dynamics
- Stochastic optimization formulation for covenant package design
- Pareto frontier analysis for risk-return covenant trade-offs
- IFRS-16 lease treatment methodology with proper debt classification
- Monte Carlo simulation parameters (400+ scenarios, seed=42)
- Sobol sensitivity analysis with first-order and total-effect indices
- Statistical confidence interval methods (Wilson, Bootstrap percentile)
optimize_covenants.py- Covenant optimization enginelbo_model_analytic.py- Analytic approximation frameworkanalysis/calibration/bayes_calibrate.py- Bayesian calibration
docs/model_specification.md- Mathematical formulation
Aniket Bhardwaj | Quantitative Finance Researcher
📧 bhardwaj.aniket2002@gmail.com
🔗 LinkedIn | GitHub
Open to opportunities in:
- 🏦 Quantitative Finance (Buy-side, Sell-side, Fintech)
- 🤖 Machine Learning Engineering (Finance, Risk, Optimization)
- 📊 Data Science (Financial Services, Regulatory Technology)
- 🔬 Research Engineering (Academic-Industry Bridge Roles)
@article{bhardwaj2025ifrs16lbo,
title={Covenant Optimization in LBO Structures Under IFRS-16:
Fast Analytic Approximations with Deterministic Error Bounds},
author={Bhardwaj, Aniket},
journal={arXiv preprint arXiv:XXXX.XXXXX},
year={2025},
url={https://github.com/Aniket2002/ifrs16-lbo-engine}
}🌟 Advanced quantitative finance framework combining academic rigor with production engineering excellence.