UK CCL Replication

Replication of Chien, Cole, Lustig (2023, "What about Japan?") — Sections 3 to 5 — applied to UK data, 2009-Q1 to 2024-Q4.

What this is

A pipeline to construct the UK consolidated public sector balance sheet (Bank of England + central government + LGPS) and compute the realized return wedge between assets and liabilities, plus the implied wedge from CCL's asset-pricing equation. Counterfactuals follow the same structure as CCL Tables 4 panels B/C.

Quick start

cd ccl_uk
python3 -m scripts.run_tests        # confirm pipeline mechanics
python3 -m scripts.consolidate      # see consolidated balance sheet (stub data)
python3 -m scripts.wedge            # see realized/implied wedge (stub data)

All scripts run on stub data by default until real data is downloaded into data/raw/. Each loader has a real implementation (load_*) and a stub (load_*_stub); the real ones currently raise NotImplementedError and need finishing once data is in place.

Architecture

ccl_uk/
├── data/
│   ├── raw/                # downloaded files go here (see docs/data_manifest.md)
│   └── processed/
├── docs/
│   └── data_manifest.md    # spec: every dataset, source URL, format
├── output/                 # final tables/figures land here
└── scripts/
    ├── config.py           # parameters, paths, sub-period definitions
    ├── schemas.py          # standard balance-sheet/returns DataFrame shapes
    ├── load_gdp.py         # ONS YBHA loader
    ├── load_boe.py         # BoE Bankstats + APF
    ├── load_central_government.py  # DMO + ONS PSA + UKGI
    ├── load_lgps.py        # LGPS aggregate
    ├── consolidate.py      # combine entities, net cross-holdings
    ├── wedge.py            # realized returns, net-return-as-%-GDP, implied omega
    └── run_tests.py        # sanity checks

The pipeline is deliberately layered so each loader is independent: when you finish the BoE loader you can verify it on its own without breaking anything else.

Data work — priority order

The data manifest in docs/data_manifest.md lists every series. To get a first headline number, you need:

1. GDP (ONS YBHA) — quarterly, instant download
2. Gilts outstanding + APF holdings (DMO + BoE APF reports) — for the gilt netting that drives the QE story
3. BoE balance sheet quarterly (BoE Bankstats Tables A1.1, A1.2) — bank reserves, currency, FX reserves
4. LGPS aggregate (Scheme Advisory Board annual reports) — assets by class

Items 1-4 produce the consolidated balance sheet. Then for the wedge:

5. Returns: gilts (already in Bloomberg pull as G0L0), FTSE All-Share, MSCI World ex-UK, Bank Rate
6. UK equity beta and ERP (Damodaran is fine for first cut)

This gives the headline omega. Counterfactuals (no-QE, currency-hedged) come last.

Methodology notes specific to UK

LGPS consolidation is debatable. CCL consolidates GPIF for Japan because Japanese public pension reserves were historically used to fund FILP. The UK case is weaker: LGPS is funded, with pension liabilities owed to specific local authority workers. The pipeline runs both ways. Headline reports include LGPS; robustness reports without.

The asset-to-debt ratio will be much lower than Japan's. Japan's was 0.66 in 2023; the UK's is likely ~0.20-0.30. This means the wedge mechanism is going to operate primarily through the QE-induced liability transformation (long-duration gilts replaced by floating-rate reserves), not through leveraged risky-asset holdings. The story to tell is about duration matching, not carry-trade returns.

Counterfactual sizing for UK QE is smaller than for Japan. UK QE literature (Joyce, Tong, Woods 2011; D'Amico-King 2013) suggests UK QE compressed gilt yields by roughly 50-100 bps, vs CCL's 200-300 bps for Japan. This means the UK no-QE counterfactual should be more conservative.

The 2022 LDI episode is a natural lab. During September 2022 the gilt market briefly crashed and the BoE intervened by purchasing long-dated gilts on financial-stability grounds (i.e., to halt forced selling by pension funds with leveraged liability-driven investment portfolios). This is plausibly the cleanest UK episode of forced bond-holding-as-financial-repression in the sample. Worth a dedicated paragraph in the discussion.

What this does NOT do

• No duration analysis (CCL Section 6). Out of scope per project decision.
• No household welfare analysis (CCL Section 7). Out of scope.
• No CIP-deviation as repression evidence (CCL Section 5.1). UK is not Japan in this respect; sterling is not a carry-funded currency.
• No structural model. Pure descriptive empirical replication.

Known issues / things to handle when real data arrives

• All loaders currently raise NotImplementedError for the real path; need to finish per actual file structure when data is in data/raw/.
• Stub data is rough; the wedge values from stubs are not meaningful.
• The asset-class-to-return-series mapping in wedge.py (asset_return_map) assumes asset_foreign_securities returns the equity index. If LGPS data lets you split foreign equity from foreign bonds, refine the mapping.
• The liab_pension_funded row gets no return assignment in the current code (treated as accrual-basis). This is defensible but worth documenting.
• "Annualized GDP × % of GDP" ratios assume quarterly GDP × 4 ≈ annualized rate. Fine for sample averages, but for a single quarter you may want trailing-4Q sum instead. Refine if precision matters.