Single-sector default, heterogeneous CES epsilon, and UK-calibrated structural parameters

[vahid-ahmadi]

Summary

• Default to M=1 (single sector): The 8-sector industry calibration is now opt-in via multi_sector=True on solve_steady_state() / run_transition_path(), or multi-sector CLI flag. This makes the default run faster and matches OG-Core's built-in M=1 defaults.
• Enable heterogeneous CES epsilon in multi-sector mode: Uses calibrated literature values (Chirinko 2008, Knoblach et al. 2020) ranging from 0.40 (Real Estate) to 1.30 (Business Services), instead of forcing all sectors to Cobb-Douglas (ε=1.0).
• Recalibrate TFP (Z) for CES: Solow residuals now use the CES production function when ε≠1, ensuring Z values are consistent with the production technology.
• UK-calibrated structural parameters with sources:

Parameter	Old	New	Source
frisch	0.4	0.35	Blundell, MaCurdy & Meghir (1999)
g_y_annual	0.01	0.011	OBR EFO Nov 2025, Table 1.1
delta_annual	0.05	0.065	ONS Capital stocks & fixed capital consumption (CAPSTK)
beta_annual	0.96	0.965	ONS Households saving ratio (NRJS)
world_int_rate_annual	0.0175	0.02	Bank of England yield curve data

• Sector-specific depreciation rates (multi-sector mode): Tax depreciation (delta_tau_annual) now varies by sector using ONS CAPSTK data, from 3% (Real Estate — long-lived structures) to 10% (Info & Finance — IT equipment).

Usage

# Single sector (default, fast)
uv run python examples/run_oguk.py ss pooled
uv run python examples/run_oguk.py tpi

# 8-sector with heterogeneous CES
uv run python examples/run_oguk.py ss pooled multi-sector
uv run python examples/run_oguk.py tpi multi-sector

Dependencies

Requires PSLmodels/OG-Core#1096 (numerical guards for CES production functions) to prevent NaN when ε<1 and inputs are near zero.

Test plan

• 8-sector SS with heterogeneous epsilon converges (resource constraint ~1e-6)
• Unit tests pass with updated parameter values
• Single-sector SS/TPI produces same results as before this PR
• 8-sector TPI with heterogeneous epsilon converges (requires OG-Core#1096)

🤖 Generated with Claude Code

[vahid-ahmadi]

Future improvement suggestions

Calibration

1. Open-economy parameters (zeta_D, zeta_K)

Currently at defaults (zeta_D=0.4, zeta_K=0.1). The UK runs a persistent current account deficit (~3-4% of GDP) and foreign investors hold ~30% of gilts. These should be calibrated from:

• ONS Balance of Payments (Pink Book)
• DMO gilt holder data

2. Government spending composition (alpha_G)

Currently auto-calibrated as a fiscal residual in SS. OBR publishes detailed departmental spending breakdowns that could anchor this, especially in multi-sector mode where government demand varies across sectors:

• OBR Public finances databank
• HM Treasury PESA

3. Immigration age profile

We use UN World Population Prospects data, which is smoothed. ONS publishes UK-specific immigration data by single year of age that captures the sharp peak at ages 18-25:

• ONS Long-term international migration

4. UK wealth distribution (lambdas and e)

The current J=7 ability types use US-derived earnings-by-age profiles. Re-estimating from UK data would improve distributional results:

• HMRC Survey of Personal Incomes (SPI)
• ONS Family Resources Survey (FRS)
• ONS Wealth and Assets Survey

5. Sector-specific effective corporate tax rates

Currently a flat 27% CIT across all 8 sectors. Effective rates vary in the UK (e.g. North Sea oil has a supplementary charge, R&D-intensive sectors get enhanced capital allowances):

• HMRC Corporation Tax statistics
• HMRC effective tax rates by sector
• OBR policy costings for capital allowances impact

Performance

6. Parallel calibration

The 3 budget-window years in calibrate() are estimated sequentially but are independent. Could run in parallel with concurrent.futures.ProcessPoolExecutor to cut calibration time by ~60%.