How antares-acies/portfolio-valuation works
Project type and purpose
• This is a batch-oriented financial data processing tool for portfolio valuation, implemented primarily in Python with experimental C++ modules for performance-critical routines (found under EXP_CPP).
• It ingests a variety of CSV datasets (cashflows, product master, curve repositories, etc.), applies valuation models, and emits parquet/CSV outputs.
File structure
• Top level:
• main.py – entry point that orchestrates the valuation run
• generate_request.py – builds/serializes request payloads
• config.py – holds configuration (paths, constants)
• helper.py – utility functions used across modules
• Portfolio_Valuation.py – core orchestration of data read → compute → write
• Valuation_Models.py – financial model implementations
• Data directories (input):
• “Portfolio Valuation Data” and “Read Data Func Data” (and their sample-data counterparts) – grouped CSVs
• Output directory:
• output/Cashflow_Engine_Outputs – generated parquet and CSV results
• EXP_CPP folder:
• three sub-cases (1. POC success, 2. schedule, 3. pybind11) with mixed .cpp, .py, executables – indicates prototype C++ routines, some exposed via pybind11.
Components and interactions
a. Data ingestion layer
– File system sources (CSV directories)
– Data-reader functions (in helper.py or dedicated read functions)
b. Core valuation engine
– Portfolio_Valuation.py: orchestrates read → model apply → write
– Valuation_Models.py: contains pricing and cashflow models
c. C++ acceleration (optional/experimental)
– EXP_CPP modules for schedule computation
– Wrapped via build_ext.py & pybind11 for Python calls
d. Request/request-builder
– generate_request.py constructs request objects or payload definitions
e. Configuration/Utilities
– config.py for paths & constants
– helper.py for common utilities (date handling, I/O helpers)
f. Output
– Parquet/CSV writer routines embedded in Portfolio_Valuation or helper
– output folder hierarchy
Architectural patterns and design principles
• Modular layering: ingestion → business logic → output
• Separation of concerns: models isolated in Valuation_Models, orchestration in Portfolio_Valuation
• Extensibility point: C++ POC modules can be swapped in via pybind11
• File-based data flow: explicit CSV → in-memory objects → parquet/CSV