Portfolio Risk Automation

Overview

I build workflows for recurring portfolio-risk reports, using Python, Parquet, and DuckDB for data preparation and Excel/VBA for delivery. Prepared monthly data can be reused and reconciled without repeatedly reading the source workbooks.

Context

Monthly spreadsheets can mix source records, reference mappings, calculations, and manual corrections. As reporting history grows, that makes it harder to explain a changed number or reproduce an earlier report.

My Role

• Designed a staged workflow around preserved source workbooks and versioned outputs.
• Built transformations and monthly analytical feeds.
• Used Parquet and DuckDB to reduce repeated workbook processing.
• Connected processed feeds to Excel/VBA reporting workflows.
• Added reconciliation, mapping coverage, conflict flags, and contract-level traceability.
• Documented cases where definitions or source coverage were not yet comparable.

Approach

I keep source workbooks unchanged and write mapped, typed data to separate files before building reporting tables. Reconciliation runs separately from the transformation steps. Ambiguous mappings stay flagged for review.

Architecture

Preserved monthly sources
  -> schema and identifier validation
  -> controlled reference mapping
  -> partitioned Parquet staging
  -> DuckDB analytical layer
  -> dimensional portfolio feeds
  -> Excel/VBA dashboards
  -> reconciliation and exception reports

Validation and Evidence

The internal project includes dashboard feed documentation, reconciliation tests, mapping audits, exception outputs, and portable VBA modules. Public material will use generated contracts and simplified dimensions.

Results

The reporting layer can reuse prepared data. Reconciliation and exception outputs help trace monthly differences to source periods, identifiers, mappings, and business definitions. I do not claim a processing-time reduction without a measured comparison.

Limitations

• Business definitions must be aligned before numerical comparisons are treated as equivalent.
• Reference mappings can change over time and require dated provenance.
• Excel remains useful as a delivery interface but should not be the only source of transformation logic.

Technology Stack

Python, Polars, pandas, Parquet, DuckDB, Jupyter, Excel, VBA.

Publication Note

The public version will not contain employer workbooks, contracts, product mappings, branch references, internal dimensions, or reporting values. It will demonstrate the architecture with synthetic portfolio data.