frequency-conversion in macro-data initialization

[agurgone]

Summary

Depends on PR #83.

This PR fixes several frequency-conversion issues in macro-data initialization. The main problem was that some data sources were implicitly treated as quarterly even when the model was configured with a different time_unit. This could produce incorrectly scaled household cash flows, exogenous series, firm flows, and initial loan values.

Issue

The model configuration defines the number of model periods per year through time_unit, but several initialization paths still used hard-coded quarterly assumptions.

Examples included:

• HFCS income and monthly cash-flow variables being divided by hard-coded quarterly factors.
• HFCS monthly variables such as rent, mortgage payments, consumer loan payments, pension contributions, transfers, and insurance payments not being consistently converted into model-period values.
• Exogenous quarterly source data being consumed without conversion to the configured model frequency.
• Compustat quarterly firm flow variables not being adjusted for the configured period length.
• Loan maturity defaults in data configuration being duplicated rather than sourced from the model defaults.
• Initial credit-market totals summing the full loan stack rather than principal balances only.

These issues can materially distort initialization, especially when using a non-quarterly time_unit.

Changes

• Added explicit yearly_factor = 12 / time_unit plumbing through macro-data readers and synthetic population initialization.
• Rescaled HFCS monthly variables into model-period values.
• Replaced hard-coded quarterly income scaling with configuration-driven scaling.
• Added monthly HFCS variable mappings for housing, debt service, and other cash-flow variables.
• Converted exogenous quarterly growth rates and level series to the configured model period.
• Updated exogenous period lookup to use timestamps/model-period starts rather than quarter-string assumptions.
• Updated annual exchange-rate handling to repeat values according to the configured number of model periods per year.
• Updated Compustat firm/bank readers to respect the configured simulation quarter and time_unit.
• Scaled active quarterly Compustat flow variables according to the configured period.
• Added tests for Compustat period handling.
• Sourced data loan maturity defaults from BankParameters instead of duplicating hard-coded values.
• Clarified that configured maturities are model periods, not months.
• Fixed initial credit-market time-series totals to use principal balances only.

Notes

The FRA data config still stores loan maturities in model periods. This PR does not convert config values from months at runtime; instead, it makes the expectation explicit and aligns defaults with the model configuration.

Validation

Ran targeted validation before committing:

• ruff check on touched Python files
• git diff --check
• tests/test_macro_data/unit/test_data_wrapper.py
• targeted macro-data/macromodel pytest suite covering:
  • exogenous readers
  • Compustat readers
  • configuration
  • synthetic population
  • synthetic firms/banks
  • central government
  • credit market
  • country/exogenous simulation compatibility

Result: targeted tests passed.