SocialMurderInstitute.org

Researching and publishing data and analysis attributing harms in society to the organizations and policies directly responsible for them.

A program of CastroMedia.org

Complex Futures: Forecasting Federal Policy Outcomes

Complex Futures is a living research specification for evaluating how alternative federal policy packages may reshape economic outcomes over time. It presents conditional futures rather than certain predictions and makes its assumptions, evidence requirements, and limits explicit.

Contents

• Project design
• Policy comparison and methods
• System and dashboard architecture
• Evidence and identification
• Forecasting and futures cones
• Validation and data engineering
• MVP, governance, and development

Working title: Policy-Regime Futures Model (PRFM)
Version: 0.1
Purpose: Research and engineering specification for deep-research agents, data agents, econometric agents, simulation agents, and application-development agents
Primary use case: Estimate and communicate how Fed-style interest-rate policy and alternative federal monetary-fiscal-credit regimes reshape future distributions of unemployment, inflation, real wages, output, financial stability, and sectoral credit allocation.

1. Executive Summary

This project will build a transparent, reproducible forecasting and counterfactual simulation system capable of comparing:

1. Interest-rate-centered monetary policy;
2. Direct and indirect credit-control regimes;
3. Hybrid interest-rate and credit-control regimes;
4. A national job guarantee and related fiscal stabilizers;
5. Alternative tax, spending, wage, industrial, housing, and financial-regulation settings.

The system is not intended to produce a single deterministic prediction. Its central output is a set of evolving conditional distributions: quantified ranges of economic outcomes under specified policy choices and assumptions. These distributions will be presented as nested futures cones:

• Probable: outcomes with substantial model-supported probability under the estimated data-generating process;
• Plausible: outcomes consistent with empirical evidence, causal mechanisms, and defensible structural assumptions, including low-frequency regimes not well represented in recent data;
• Possible: outcomes not ruled out by physical, accounting, institutional, or logical constraints, including stress cases and model uncertainty.

The model will combine panel econometrics, causal inference from historical policy episodes, dynamic forecasting, regime-dependent parameters, scenario simulation, and uncertainty decomposition. Multivariate regression remains a core component, but a simple static regression is insufficient because credit controls are heterogeneous treatments, policy adoption is endogenous, effects are lagged, policy instruments interact, transmission mechanisms change across institutional regimes, and the desired output includes counterfactual distributions rather than fitted historical associations.

The initial minimum viable product should answer questions such as:

Holding the fiscal stance and external environment constant, what is the estimated distribution of unemployment, inflation, and real-wage outcomes over 1, 2, 5, and 10 years if federal policymakers substitute targeted credit controls for some portion of interest-rate tightening?

How does a national job guarantee alter the inflation-unemployment-real-wage distribution under alternative credit-allocation rules?

Which outcomes are robust across models, and which depend heavily on assumptions about credit leakage, bank behavior, import constraints, productivity, wage bargaining, or political durability?

Policy Regime Comparison and Methodological Framework

Fundamental question

The project’s central question is:

Compared with Fed-style interest-rate management, how do MMT-informed fiscal and monetary coordination, a job guarantee, and targeted credit controls affect inflation, employment, real wages, investment, and financial stability?

The project treats this as a comparison of explicit policy regimes, not a contest between economic schools. MMT-informed coordination is not a single intervention: it describes a fiscal-monetary operating framework whose effects depend on the concrete fiscal, employment, credit, and institutional choices within it.

Comparable policy regimes

Each scenario must state its fiscal stance, external environment, supply assumptions, phase-in, enforcement, and central-bank response. The initial comparison set is:

1. Rate-hike stabilization: interest-rate-centered management of inflation, without a job guarantee or targeted credit controls.
2. Job guarantee stabilization: stable or low policy rates combined with a federally funded job guarantee and stated fiscal rules.
3. Targeted credit stabilization: stable or low policy rates combined with credit controls that restrain specified inflationary or speculative lending while protecting productive investment.
4. Coordinated job guarantee and credit controls: a job guarantee and targeted credit controls combined with explicit fiscal and central-bank coordination.

The project may later evaluate hybrid regimes, but it must not compare incoherent bundles. A scenario must specify the policy instruments, legal authority, implementation capacity, timing, and response to inflation, unemployment, credit growth, and supply constraints.

Shared accounting core

All model families must respect common accounting relationships and use the same observed state variables. At minimum:

[ Y_t = C_t + I_t + G_t + NX_t ]

[ Y_t = W_t + \Pi_t + \text{taxes less subsidies}_t ]

[ \text{assets}_t = \text{liabilities}_t ]

The shared outcome set includes inflation, employment and unemployment, real wages, output, sectoral investment, sectoral credit allocation, housing and asset prices, debt service, financial stress, and distributional outcomes. These accounting and measurement constraints make disagreements between models testable rather than rhetorical.

Complementary explanatory frameworks

The project uses neoclassical, Marxian, MMT/post-Keynesian, institutional, and empirical methods as complementary sources of mechanisms and hypotheses. No framework is assumed to exhaustively explain policy transmission, and no framework-specific claim is treated as established without evidence.

These are complementary contributions, not fixed labels for every model or researcher. The project will use the mechanisms that are conceptually coherent, empirically measurable, and useful for the specific policy and historical setting under study.

Mechanism registry

For every causal claim used in estimation or simulation, the project must maintain a mechanism record containing:

• The policy lever and precise treatment definition;
• The outcome and affected population, sector, or institution;
• The proposed causal mechanism and its timing;
• Whether the claim is a shared proposition, a framework-specific proposition, or an open empirical question;
• The observable variables, source data, transformations, and identification strategy;
• Competing explanations, boundary conditions, and expected failure modes;
• Historical evidence and out-of-sample validation results.

For example, a rate increase may be modeled as affecting debt service, investment, consumption, inflation, unemployment, real wages, and profitability. The shared proposition is that it changes credit conditions and spending; the empirical analysis must then test distributional exposure, sectoral investment, capacity effects, and the competing explanations for any observed inflation response.

2. Project Objectives

2.1 Primary objective

Construct a dynamic policy-regime model that estimates how policy levers alter:

• Expected macroeconomic trajectories;
• Forecast uncertainty;
• Downside and upside tail risks;
• Feasible combinations of inflation, unemployment, real wages, and output;
• Sectoral allocation of credit;
• Distributional outcomes;
• Financial-instability risks;
• Political and administrative feasibility.

2.2 Secondary objectives

The system should:

• Reconstruct historical counterfactuals;
• Distinguish credit-supply effects from credit-demand effects;
• Estimate heterogeneous treatment effects across countries, sectors, institutions, and business-cycle states;
• Quantify the degree to which policy changes narrow, widen, shift, skew, or bifurcate the futures cone;
• Expose uncertainty rather than hide it;
• Permit analysts to inspect assumptions, source data, transformations, and model disagreements;
• Support both academic analysis and a public-facing policy simulator.

2.3 Non-objectives

The project should not claim:

• Exact prediction of political-economic history;
• A universal effect of “credit controls” independent of instrument design;
• That historical natural experiments perfectly identify modern policy effects;
• That a confidence interval from one statistical model exhausts genuine uncertainty;
• That observational equivalence constitutes causal proof;
• That all future possibilities can be assigned meaningful frequentist probabilities.

3. Conceptual Model

3.1 State-transition formulation

Let the economy at time (t) be represented by a state vector:

[ S_t = [ Y_t,\ \pi_t,\ U_t,\ E_t,\ RW_t,\ W_t,\ P_t,\ C_t,\ C^s_t,\ I_t,\ H_t,\ B_t,\ F_t,\ X_t,\ M_t,\ ER_t,\ D_t,\ Q_t,\ Z_t ] ]

where, illustratively:

• (Y): real output or output gap;
• (\pi): inflation;
• (U): unemployment;
• (E): employment and participation;
• (RW): real wages;
• (W): nominal wage growth;
• (P): price-level and sectoral price indices;
• (C): aggregate credit;
• (C^s): sectoral credit vector;
• (I): private investment;
• (H): housing and land-market state;
• (B): bank balance-sheet state;
• (F): fiscal stance;
• (X, M): exports and imports;
• (ER): exchange-rate state;
• (D): household, firm, financial, and public debt;
• (Q): productive capacity, inventories, and bottlenecks;
• (Z): exogenous and latent conditions.

Policy is represented by a vector:

[ A_t = [ r_t,\ T_t,\ G_t,\ JG_t,\ CC_t,\ MPP_t,\ FXC_t,\ WP_t,\ IP_t,\ SP_t ] ]

where:

• (r): policy interest rate and rate corridor;
• (T): taxes and transfers;
• (G): government spending;
• (JG): job-guarantee parameters;
• (CC): credit-control instruments;
• (MPP): other macroprudential policies;
• (FXC): exchange and capital controls;
• (WP): wage and price policies;
• (IP): industrial policy;
• (SP): social-policy variables.

The state evolves according to:

[ S_{t+1} = f(S_t, A_t, R_t, Z_{t+1}; \theta_t) + \epsilon_{t+1} ]

where:

• (R_t) describes the institutional or policy regime;
• (\theta_t) contains estimated transmission parameters;
• (\epsilon) contains innovations and unobserved shocks;
• (\theta_t) may vary by time, regime, country, sector, and economic state.

The core causal question is not merely whether a credit-control indicator predicts unemployment. It is:

[ P(S_{t+h}\mid do(A_t=a), S_t, R_t) ]

compared with:

[ P(S_{t+h}\mid do(A_t=a’), S_t, R_t) ]

for horizons (h) and alternative policy packages (a) and (a’).

System Architecture and Dashboard Contract

Complex Futures is a policy-scenario and forecasting system. A multivariate regression is an important component, but it cannot by itself establish causal effects, capture long-horizon cohort pathways, or provide a safe interface for policy experimentation. The system therefore combines shared data infrastructure, linked model families, explicit scenario definitions, and uncertainty-aware public outputs.

Variable roles and dynamic policy relationships

Every modeled variable must be classified as an outcome, a policy lever, or a contextual constraint. Some variables may occupy more than one role at different points in time; the model must state the role used in each estimation or scenario.

The system’s dynamic state relationship is:

[ S_{t+1} = f(S_t, A_t, X_t; \theta) + \epsilon_t ]

where (S_t) is the economic and social state, (A_t) is the policy package, (X_t) is structural and external context, (\theta) contains estimated parameters, and (\epsilon_t) captures shocks and irreducible uncertainty.

Observed policy is endogenous: policymakers react to economic conditions, political pressures, and institutions. The system must therefore distinguish an observed policy association from an intervention effect:

[ A_t = g(S_t, \text{political conditions}_t, \text{institutions}_t) + \eta_t ]

For example, a rate increase may follow high inflation. A regression that does not account for this reaction can mistake the conditions that prompted a rate increase for its effects. Causal policy designs, event studies, policy-shock identification, and explicit scenario interventions are required alongside dynamic forecasting.

Data and provenance pipeline

Source data + policy-event records + historical documents
                    ↓
Versioned provenance, cleaning, harmonization, and data vintages
                    ↓
Common state variables, policy ontology, and sectoral balance sheets
                    ↓
Dynamic forecasting + causal-effect + cohort/structural model families
                    ↓
Baseline forecast and policy counterfactuals with uncertainty cones
                    ↓
Interactive dashboard with documented policy-package controls

Each source and derived series must retain its identifier, provider, retrieval time, license, vintage, geography, coverage, units, transformation history, missing-data treatment, and source citation. Policy events must additionally retain their primary evidence, coding decision, effective date, announcement date, target population, intensity, enforcement, and known avoidance channels.

Linked model families

The system is modular. Modules must share definitions, time conventions, accounting constraints, and uncertainty interfaces; they may not silently overwrite each other’s outputs.

1. Macro-financial model: inflation, employment, output, real wages, interest rates, credit, assets, debt, and financial stress at monthly or quarterly frequency.
2. Policy-effect model: causal estimates from historical rate changes, fiscal changes, credit controls, job programs, and comparable policy events.
3. Sectoral model: housing, education, health, energy, finance, infrastructure, and productive investment, including sectoral credit allocation and capacity effects.
4. Cohort and distribution model: long-run outcomes by age, income, race, geography, education, wealth, and other legally and ethically appropriate population categories.
5. Accounting and feasibility model: sectoral balances, public budget constraints, financial balance sheets, labor, productive capacity, and physical limits.

Scenario contract

Users and analysts do not set isolated knobs without context. Every scenario must be a complete policy package that specifies:

• The policy change, legal authority, implementation date, phase-in, duration, and reversal rule;
• The affected population, sector, institution, or credit category;
• Fiscal financing, use of revenue, and central-bank operating assumptions;
• Enforcement, administrative capacity, compliance, and leakage assumptions;
• The business-cycle state, external conditions, supply constraints, and baseline assumptions;
• The expected causal channels, relevant historical evidence, and extrapolation warnings.

The scenario engine compares each package with a documented baseline trajectory. It must report changes at immediate, medium-term, and long-run horizons for inflation, employment, real wages, output, sectoral investment, housing and asset prices, debt service, financial stress, public finance, and distributional outcomes.

Dashboard contract

The public dashboard must make policy assumptions inspectable. It should allow users to compare a baseline with one or more permitted scenarios, review the selected inputs and assumptions, inspect the mechanism and evidence behind each output, and view disagreement across model families.

The dashboard must present distributions and uncertainty cones rather than a single authoritative number. It must clearly distinguish estimated causal effects, forecast correlations, structural assumptions, expert scenarios, and outcomes outside historical support. Controls must enforce policy-package constraints rather than imply that a tax rate, education appropriation, or interest rate can be changed independently of financing, timing, institutional capacity, and other policy choices.

Long-horizon cohort pathways

Some policy effects operate through cohorts and cannot be inferred reliably from a single aggregate macroeconomic regression. An education-funding reduction, for example, may have near-term fiscal effects, medium-term effects on staffing and attainment, and long-run effects on earnings, productivity, health, crime exposure, tax revenue, and inequality.

The cohort model must represent those pathways explicitly and report the assumptions, empirical support, timing, distributional exposure, and uncertainty at each link. It must feed its estimated labor-force skill, earnings, public-service, and social-outcome effects back into the macro-financial and public-finance modules without double counting.

4. Credit Controls: Treatment Taxonomy and Encoding

“Credit controls” must never be represented only by a single binary variable. The treatment ontology should distinguish instruments, targets, intensity, enforcement, and leakage.

4.1 Instrument families

A. Quantitative aggregate controls

• Aggregate bank-lending ceilings;
• Bank-specific loan-growth quotas;
• Credit-growth reference paths;
• Supplementary reserve or deposit requirements triggered by excessive balance-sheet growth;
• Funding-growth limits;
• Central-bank refinancing quotas.

B. Sectoral allocation controls

• Required lending shares for housing, agriculture, manufacturing, exports, small business, green investment, or public priorities;
• Prohibitions or caps on lending to land speculation, securities purchases, luxury consumption, mergers, or other designated uses;
• Sector-specific reserve requirements;
• Differential capital requirements;
• Differential risk weights;
• Directed refinancing facilities;
• Preferential rediscounting;
• Public development-bank lending targets.

C. Borrower-based controls

• Loan-to-value limits;
• Debt-to-income limits;
• Debt-service-to-income limits;
• Interest-coverage rules;
• Maturity limits;
• Amortization requirements;
• Margin requirements;
• Consumer installment-credit terms;
• Down-payment requirements.

D. Price-based credit controls

• Interest-rate ceilings or floors;
• Usury limits;
• Deposit-rate controls;
• Differential central-bank funding rates;
• Credit taxes or levies;
• Fees on particular loan categories;
• Subsidized rates for targeted sectors.

E. Institutional and balance-sheet controls

• Reserve requirements;
• Liquidity requirements;
• Capital buffers;
• Countercyclical capital buffers;
• Asset concentration limits;
• Foreign-currency lending restrictions;
• Open-position limits;
• Loan-loss provisioning rules;
• Restrictions on securitization or off-balance-sheet migration.

F. Administrative and qualitative controls

• Window guidance;
• Moral suasion;
• Lending “recommendations” backed by supervisory authority;
• Credit plans;
• Case-by-case approval;
• Public-bank mandates;
• Licensing and charter restrictions;
• Exchange controls that reduce circumvention.

4.2 Required treatment fields

Each policy action should be stored as an event with:

• Country;
• Jurisdictional level;
• Announcement date;
• Legal effective date;
• Enforcement date;
• Relaxation date;
• Repeal date;
• Instrument family and subtype;
• Target lender class;
• Target borrower class;
• Target sector;
• Target loan purpose;
• Target currency;
• Quantitative threshold;
• Direction: tightening, easing, or reallocation;
• Intended objective;
• Stated motivation;
• Enforcement mechanism;
• Penalty;
• Exemptions;
• Coverage share;
• Estimated bindingness;
• Anticipation period;
• Expected duration;
• Actual duration;
• Complementary policies;
• Concurrent interest-rate action;
• Concurrent fiscal action;
• Capital-account regime;
• Financial openness;
• Evidence source;
• Coding confidence;
• Evidence conflicts.

4.3 Recommended numerical representation

For instrument (j), country (i), and time (t):

[ CC_{ijt} = Direction_{ijt} \times Intensity_{ijt} \times Coverage_{ijt} \times Bindingness_{ijt} \times Enforcement_{ijt} ]

All terms should be retained separately in the data, not only multiplied.

Recommended normalized scales:

• Direction: (-1) easing, (0) no action, (+1) tightening;
• Intensity: 0–1 or standardized instrument-specific change;
• Coverage: estimated share of relevant lending affected, 0–1;
• Bindingness: estimated probability or degree that the rule constrains behavior, 0–1;
• Enforcement: administrative capacity and expected compliance, 0–1;
• Leakage: expected displaced share through nonbanks, foreign lending, offshore markets, trade credit, securitization, or reclassification, 0–1.

A net effective control measure may be:

[ EffectiveCC_{ijt} = CC_{ijt}(1-Leakage_{ijt}) ]

This is an estimated latent construct and must carry uncertainty.

4.4 Sectoral credit-allocation vector

The model should explicitly track shares and growth rates of credit to:

• Residential mortgages;
• Commercial real estate;
• Construction and land development;
• Consumer revolving credit;
• Consumer durable credit;
• Financial-sector lending;
• Large nonfinancial corporations;
• Small and medium enterprises;
• Manufacturing;
• Agriculture;
• Energy;
• Transportation;
• Public infrastructure;
• Export industries;
• Green and decarbonization projects;
• Health, education, and care sectors;
• State and local governments.

The effect of reallocating credit may be more important than changing aggregate credit volume.

5. Outcome Variables

5.1 Primary outcomes

Inflation

Use multiple measures:

• Headline CPI;
• Core CPI;
• PCE or national equivalent;
• GDP deflator;
• Producer-price inflation;
• Wage inflation;
• Housing-cost inflation;
• Food and energy inflation;
• Tradable versus nontradable inflation;
• Sectoral price indices;
• Inflation dispersion;
• Inflation persistence;
• Inflation expectations where available.

Forecast:

• Level;
• Change;
• Cumulative price-level effect;
• Probability of exceeding policy thresholds;
• Duration above threshold;
• Tail risk.

Unemployment and employment

Use:

• Harmonized unemployment rate;
• Employment-to-population ratio;
• Labor-force participation;
• Underemployment;
• Involuntary part-time employment;
• Long-term unemployment;
• Job-finding and separation rates;
• Vacancy rate;
• Hours worked;
• Labor-market tightness;
• Informal employment where relevant;
• Job-guarantee enrollment.

Unemployment is not sufficient as the sole labor-market dependent variable because definitions, participation changes, and discouraged-worker effects can mask deterioration.

Real wages

Use:

• Real average hourly earnings;
• Real median wages;
• Real compensation per employee;
• Labor share;
• Wage share by income decile;
• Real disposable household income;
• Unit labor costs;
• Productivity-adjusted wages;
• Real wage by sector, occupation, gender, race/ethnicity, region, and educational level where permitted by data.

Real wage outcomes should be deflated using both aggregate CPI and household-group-specific cost indices where possible.

5.2 Secondary macroeconomic outcomes

• Real GDP growth;
• Output gap;
• Industrial production;
• Productivity;
• Capacity utilization;
• Fixed investment;
• Business formation;
• Firm exits;
• Consumption;
• Household saving;
• Trade balance;
• Current account;
• Exchange rate;
• Import prices;
• Fiscal balance;
• Public debt ratios;
• Tax receipts;
• Job-guarantee spending;
• Automatic-stabilizer spending.

5.3 Financial outcomes

• Aggregate private credit growth;
• Credit-to-GDP ratio and gap;
• Credit spreads;
• Lending standards;
• Approval rates;
• Bank profitability;
• Net interest margin;
• Delinquency and default;
• Nonperforming loans;
• Bank capital;
• Liquidity;
• Asset prices;
• Housing prices and rents;
• Equity prices;
• Financial stress;
• Crisis probability;
• Shadow-bank growth;
• Foreign borrowing;
• Credit leakage and disintermediation.

5.4 Distributional and social outcomes

• Income and wealth shares;
• Poverty;
• Material hardship;
• Housing cost burden;
• Household debt-service burden;
• Regional employment;
• Racial, gender, age, and disability employment gaps;
• Union density;
• Strike activity;
• Wage bargaining coverage;
• Public-service access;
• Economic-security indices.

6. Explanatory Variables and Controls

6.1 Monetary variables

• Policy rate;
• Real policy rate;
• Short-term market rates;
• Long-term sovereign yields;
• Yield curve;
• Central-bank balance sheet;
• Reserve remuneration;
• Reserve requirements;
• Monetary aggregates;
• Exchange-rate policy;
• Forward guidance;
• Quantitative easing or tightening;
• Monetary-policy shock series.

6.2 Fiscal variables

• Government consumption;
• Public investment;
• Transfers;
• Tax rates by type;
• Effective tax rates;
• Cyclically adjusted fiscal balance;
• Fiscal impulse;
• Government employment;
• Social-insurance replacement rates;
• Job-guarantee wage, eligibility, hours, geographic coverage, and spending;
• Public procurement;
• Automatic stabilizers.

6.3 Labor institutions

• Union density;
• Collective bargaining coverage;
• Minimum wage;
• Employment protection;
• Unemployment benefits;
• Payroll taxes;
• Strike restrictions;
• Public-sector bargaining;
• Wage-indexation rules;
• Job-guarantee presence and design;
• Immigration and labor-supply conditions.

6.4 Supply and capacity variables

• Productivity;
• Capacity utilization;
• Energy supply and prices;
• Food prices;
• Import dependency;
• Inventory-to-sales ratios;
• Logistics pressure;
• Housing supply;
• Vacancy rates;
• Demographics;
• Labor-force growth;
• Capital stock;
• Climate and disaster shocks;
• War and sanctions;
• Pandemic indicators;
• Commodity terms of trade.

6.5 Political and institutional variables

• Electoral cycle;
• Government ideology;
• Legislative control;
• Central-bank independence;
• Financial-regulatory authority;
• State capacity;
• Rule-of-law and enforcement measures;
• Capital-account openness;
• Exchange controls;
• Bank concentration;
• Public-bank share;
• Financialization;
• International agreements;
• Policy durability;
• credibility and expectations.

Kingdon’s problem, policy, and politics streams can be added as a separate adoption-and-durability module rather than mixed directly into the macro outcome equation.

7. Data Architecture and Sources

7.1 Core international macro-financial datasets

BIS Credit to the Non-Financial Sector

Use for quarterly credit aggregates, borrower sectors, and long historical coverage for many economies.

Source:

• BIS Data Portal, Credit to the non-financial sector
  https://www.bis.org/statistics/totcredit.htm
• BIS data interface
  https://data.bis.org/topics/TOTAL_CREDIT/data

BIS Credit-to-GDP Gaps

Use as an indicator of excessive credit and financial-cycle position.

Source:

• https://data.bis.org/topics/CREDIT_GAPS

IMF Integrated Macroprudential Policy Database (iMaPP)

Use as the starting event database for monthly tightening and loosening actions across multiple macroprudential instruments. It includes instrument indicators and policy-action descriptions, but must be augmented for older direct controls and qualitative instruments.

Source:

• https://www.elibrary-areaer.imf.org/Macroprudential/Pages/iMaPPDatabase.aspx
• https://www.elibrary-areaer.imf.org/Macroprudential/Pages/AboutUs.aspx

IMF International Financial Statistics

Use for monetary aggregates, credit, interest rates, exchange rates, prices, and balance-of-payments variables. Access conditions and licensing must be checked.

World Bank Global Financial Development Database

Use for annual cross-country characteristics of financial institutions and markets. The currently described release covers long-run annual indicators but is less current than some operational sources, so it should be treated as a historical structural dataset.

Source:

• https://www.worldbank.org/en/publication/gfdr/data/global-financial-development-database

World Development Indicators

Use for macroeconomic, demographic, fiscal, trade, and development controls. The Indicators API offers programmatic access to thousands of time series.

Source:

• https://datahelpdesk.worldbank.org/knowledgebase/articles/889392-about-the-indicators-api-documentation

Jordà-Schularick-Taylor Macrohistory Database

Use for long-run annual data across advanced economies, including credit, interest rates, prices, output, public finances, housing, and crisis chronologies. This is especially valuable for regime history and long-period validation.

Source:

• https://www.macrohistory.net/database/

7.2 Labor and wage data

ILOSTAT

Use for harmonized unemployment, employment, participation, hours, informal employment, earnings, labor underutilization, and demographic breakdowns.

Sources:

• https://ilostat.ilo.org/data/
• https://ilostat.ilo.org/data/bulk/

OECD Data Explorer

Use for harmonized monthly and quarterly unemployment, CPI, wages, productivity, labor institutions, national accounts, and sectoral indicators for OECD countries.

Sources:

• https://data-explorer.oecd.org/
• https://www.oecd.org/en/data/indicators/unemployment-rate.html
• https://www.oecd.org/en/data/indicators/inflation-cpi.html

National statistical offices

Examples:

• U.S. Bureau of Labor Statistics;
• U.S. Bureau of Economic Analysis;
• U.K. Office for National Statistics;
• Statistics Canada;
• Eurostat;
• Japan Statistics Bureau;
• Statistics Korea;
• national labor-force surveys and establishment surveys.

Use national data for higher frequency, subgroup detail, and historical documentation, but maintain harmonized international series separately.

7.3 U.S.-specific sources

FRED and ALFRED

Use FRED for programmatic retrieval and ALFRED for real-time vintages. Vintage data are essential to prevent look-ahead bias and to recreate what policymakers could have known at the time.

Source:

• https://fred.stlouisfed.org/docs/api/fred/
• https://fred.stlouisfed.org/docs/api/fred/overview.html

Federal Reserve

• Financial Accounts of the United States;
• Senior Loan Officer Opinion Survey;
• Call Reports;
• Y-14 data where accessible;
• Survey of Consumer Finances;
• Consumer Credit;
• Flow of Funds;
• historical FOMC materials;
• regulations and policy announcements.

FDIC

• BankFind Suite;
• Historical Statistics on Banking;
• Summary of Deposits;
• failed-bank data.

Census and IPUMS

• Current Population Survey;
• American Community Survey;
• County Business Patterns;
• Business Dynamics Statistics;
• Longitudinal Employer-Household Dynamics;
• historical census microdata via IPUMS.

7.4 Credit-control event-source archive

Deep-research agents must build a versioned source archive from:

• Central-bank annual reports;
• Central-bank bulletins;
• Government gazettes;
• Banking statutes;
• Regulatory circulars;
• legislative histories;
• archival policy minutes;
• IMF Article IV reports;
• OECD country surveys;
• BIS papers;
• World Bank financial-sector reports;
• academic monographs;
• contemporary newspapers;
• bank association records;
• historical statistical yearbooks.

Every coded policy event must cite a primary source when one is obtainable.

8. Candidate Historical Natural and Quasi-Natural Experiments

The following are candidate episodes, not automatically valid experiments. Each requires a dedicated identification memo.

8.1 United States: 1980 Credit Controls

President Carter authorized Federal Reserve use of the Credit Control Act in March 1980. Measures included reserve requirements on certain consumer credit and voluntary credit restraint. The episode was short, highly salient, and accompanied by major monetary tightening and recession.

Research value:

• Sharp announcement and implementation dates;
• Differential exposure by credit category;
• Consumer-credit and retail-spending effects;
• Strong anticipation and behavioral response;
• Useful for studying abrupt aggregate and consumer-credit restraint.

Identification risks:

• Simultaneous Volcker monetary tightening;
• Recession and oil-price effects;
• policy reversal;
• public signaling;
• anticipatory borrowing;
• endogenous adoption during severe inflation.

Primary starting source:

• Federal Reserve history of cyclical macroprudential policy:
  https://www.federalreserve.gov/pubs/feds/2013/201329/index.html

Candidate designs:

• Interrupted time series;
• category-level difference-in-differences;
• bank-level exposure design;
• local projections using exposure intensity;
• synthetic control using comparable economies;
• event study around announcement and suspension.

8.2 United Kingdom: Supplementary Special Deposits Scheme (“the Corset”)

The U.K. imposed penalties on excessive growth of banks’ interest-bearing eligible liabilities. The scheme operated in variants during the 1970s and was suspended in 1980. It restrained measured bank funding but also induced avoidance and migration to alternative channels.

Research value:

• Clear instrument mechanics;
• repeated activation and modification;
• documented leakage;
• bank and nonbank substitution;
• useful for estimating how institutional openness affects control effectiveness.

Primary sources:

• Bank of England review:
  https://www.bankofengland.co.uk/quarterly-bulletin/1982/q1/the-supplementary-special-deposits-scheme
• Bank of England causal historical research:
  https://www.bankofengland.co.uk/-/media/boe/files/working-paper/2016/monetary-versus-macroprudential-policies-causal-impacts-of-interest-rates.pdf

Candidate designs:

• Local projections using historical instrument indices;
• bank balance-sheet panel;
• difference between covered banks and substitute intermediaries;
• policy-on/policy-off episodes;
• leakage modeling;
• comparison with interest-rate shocks.

8.3 Japan: Bank of Japan Window Guidance

The Bank of Japan used quantitative and qualitative guidance to influence bank loan growth for decades, beginning in the postwar period and ending in 1991. Its effectiveness varied with financial liberalization and market structure.

Research value:

• Long duration;
• bank-specific guidance;
• interaction with administered rates and central-bank lending;
• gradual erosion through liberalization;
• strong case for regime-dependent treatment effects.

Primary sources:

• Bank of Japan, effectiveness and historical transition:
  https://www.boj.or.jp/en/research/wps_rev/rev_2010/data/rev10e04.pdf
• Bank of Japan, policy in the 1980s:
  https://www.imes.boj.or.jp/research/papers/english/me33-5.pdf

Candidate designs:

• Bank-level quota exposure;
• sectoral credit allocation;
• threshold or formula-based assignment if archival rules permit;
• state-dependent local projections;
• interaction with liberalization indices;
• structural break and varying-coefficient models.

8.4 United Kingdom: Hire-Purchase and Consumer-Credit Controls

The U.K. repeatedly varied down-payment and maturity requirements for installment purchases.

Research value:

• Product-level exposure;
• discrete policy changes;
• durable-goods purchases as proximate outcomes;
• potential high-frequency identification;
• direct analogue for borrower-based controls.

Candidate designs:

• Difference-in-differences across controlled and uncontrolled goods;
• event studies;
• distributed-lag models;
• household expenditure microdata;
• industry production and sales comparisons.

8.5 United States: Regulation W and Consumer Installment Credit

Federal Reserve Regulation W restricted terms of consumer installment credit in several periods, including wartime and postwar use.

Research value:

• Product and loan-purpose targeting;
• repeated changes;
• interaction with supply constraints and demobilization;
• potential monthly retail and production data.

Identification risks:

• wartime rationing;
• price controls;
• material shortages;
• demobilization;
• major structural breaks.

8.6 France: Encadrement du Crédit

France used quantitative bank-credit ceilings and selective credit policies during the postwar era, with progressive liberalization and eventual removal.

Research value:

• explicit quantitative ceilings;
• preferential treatment for designated activities;
• long policy history;
• useful comparison between administrative allocation and market-based policy.

Required research:

• Banque de France archives;
• French government economic plans;
• OECD financial-market surveys;
• bank-level and sectoral loan series;
• liberalization chronology.

8.7 South Korea: Directed Credit and Heavy and Chemical Industry Drive

Korea used state influence over banks, directed credit, preferential rates, and sectoral allocation during industrialization, followed by liberalization.

Research value:

• strong sectoral treatment;
• policy-directed investment;
• potential firm- and industry-level exposure;
• interaction with export policy and public guarantees.

Identification risks:

• simultaneous industrial policy;
• trade strategy;
• repression of labor;
• foreign aid and geopolitical support;
• rapid structural transformation.

Required sources:

• Bank of Korea ECOS;
• Statistics Korea KOSIS;
• Korean development-bank archives;
• World Bank historical reports;
• firm and industry panels.

8.8 China: Window Guidance, Loan Quotas, and Sectoral Restrictions

China has used bank loan quotas, window guidance, reserve requirements, property-sector controls, and targeted relending.

Research value:

• contemporary high-frequency episodes;
• differentiated treatment across bank types, cities, sectors, and borrower classes;
• property and local-government financing channels;
• interaction between public and private banks.

Identification risks:

• data revisions and transparency;
• concurrent administrative policies;
• local implementation heterogeneity;
• state-owned enterprise preference;
• shadow banking and reclassification.

8.9 Loan-to-Value and Debt-to-Income Changes Across Countries

Frequent changes in LTV and DTI rules provide a larger panel of modern macroprudential actions.

Research value:

• extensive event databases;
• targeted housing-credit channel;
• multiple treatment intensities;
• suitable for panel local projections and difference-in-differences.

Limit:

These are primarily housing and financial-stability controls, not a complete substitute for interest-rate policy.

8.10 Reserve-Requirement Differentiation

Countries including Brazil, China, Turkey, India, and others have varied reserve requirements by bank, liability, maturity, currency, or credit type.

Research value:

• measurable quantitative changes;
• bank-level exposure;
• differential transmission;
• interaction with capital flows.

8.11 Removal or Liberalization Episodes

Abolition may provide cleaner discontinuities than adoption:

• U.K. Corset suspension;
• Japanese window-guidance termination;
• Scandinavian lending-control liberalizations;
• New Zealand credit-allocation and rate deregulation;
• French financial liberalization;
• capital-account opening that increases leakage.

These episodes can reveal both the direct effect of controls and the degree to which markets circumvent them.

9. Identification Strategy

No single identification method should determine the headline result. The system should use a hierarchy of evidence.

9.1 Event studies

Estimate dynamic responses before and after policy announcements and implementation.

Requirements:

• Test pre-trends;
• distinguish announcement from effective date;
• include anticipation windows;
• account for reversals;
• cluster uncertainty appropriately;
• report sensitivity to event-window choice.

9.2 Difference-in-differences

Compare more-exposed and less-exposed units:

• Banks;
• firms;
• sectors;
• households;
• products;
• regions;
• loan categories.

Use modern estimators robust to staggered treatment and heterogeneous effects. Avoid naive two-way fixed-effects estimates where treatment timing varies.

9.3 Synthetic controls

Construct a weighted counterfactual for a treated country, sector, or region using unaffected units.

Use:

• Classical synthetic control;
• augmented synthetic control;
• synthetic difference-in-differences;
• Bayesian structural time series.

9.4 Instrumental variables

Potential instruments may arise from:

• Formula-based quotas;
• bank exposure determined before the policy;
• regulatory thresholds;
• institutional eligibility;
• foreign monetary shocks under fixed exchange regimes;
• political timing;
• administrative boundaries.

Instrument validity must be defended explicitly. Do not use instruments merely because they are correlated with treatment.

9.5 Regression discontinuity

Possible where controls depend on:

• Loan size;
• LTV threshold;
• bank size;
• sector classification;
• borrower income;
• geographic eligibility;
• regulatory capital threshold.

9.6 Local projections

Estimate horizon-specific impulse responses:

[ Y_{i,t+h}-Y_{i,t-1} = \alpha_{i,h} + \lambda_{t,h} + \beta_h PolicyShock_{it} + \Gamma_h X_{it} + \epsilon_{i,t+h} ]

Extend with interactions:

[ \beta_h = \beta_{0h} + \beta_{1h} FinancialOpenness + \beta_{2h} Slack + \beta_{3h} PublicBankShare + \beta_{4h} Enforcement + \beta_{5h} JobGuarantee ]

Local projections are suitable for transparent dynamic response estimation and nonlinear state dependence.

9.7 Panel vector autoregression and Bayesian VAR

Use to model dynamic interdependence among:

• Credit;
• output;
• unemployment;
• prices;
• wages;
• interest rates;
• investment;
• exchange rates.

Identification options:

• Recursive restrictions;
• sign restrictions;
• narrative shocks;
• external instruments;
• heteroskedasticity;
• policy-announcement surprises.

9.8 Time-varying parameter and regime-switching models

Use because transmission parameters vary with:

• Financial liberalization;
• institutional structure;
• inflation regime;
• labor bargaining power;
• global capital mobility;
• crisis state;
• public-bank share;
• supply constraints.

Candidates:

• Time-varying parameter VAR;
• Markov-switching VAR;
• Bayesian dynamic linear model;
• hidden Markov model;
• varying-coefficient panel model;
• Gaussian process coefficient functions;
• dynamic factor model.

9.9 Structural causal models

Develop a directed acyclic graph and structural causal model for each policy family. Explicitly distinguish:

• Confounders;
• mediators;
• colliders;
• treatment selection;
• measurement error;
• feedback over time.

A static DAG is insufficient for the complete system, but dynamic causal graphs help prevent invalid controls.

9.10 Structural macroeconomic model

A later phase may include a stock-flow-consistent, heterogeneous-agent, or sectoral balance-sheet model. Its role is not to replace empirical estimates but to:

• Enforce accounting consistency;
• represent mechanisms not observable in historical data;
• explore unprecedented policy packages;
• model distributional and balance-sheet channels;
• generate plausible but low-frequency scenarios.

9.11 Ensemble approach

The final forecast should combine:

• Reduced-form dynamic regression;
• Causal local projections;
• Bayesian VAR or state-space model;
• panel treatment-effect model;
• synthetic-control estimates for key episodes;
• structural simulation;
• expert-elicited stress scenarios.

Weights should depend on out-of-sample performance, calibration, domain relevance, and model dependence. Do not average mechanically when models encode incompatible assumptions; preserve model disagreement as uncertainty.

10. From Historical Effects to Modern Counterfactuals

The core transfer problem is:

How can an effect estimated under one historical regime be applied to a different present or future regime?

10.1 Conditional transportability

Estimate treatment effects as functions of context:

[ \tau(x) = E[Y(1)-Y(0)\mid X=x] ]

Context features should include:

• Financial openness;
• bank concentration;
• public-bank share;
• nonbank share;
• capital controls;
• digital-finance capacity;
• household leverage;
• corporate leverage;
• mortgage-market structure;
• union density;
• inflation regime;
• unemployment gap;
• capacity utilization;
• import dependence;
• exchange-rate regime;
• reserve-currency status;
• fiscal stance;
• administrative capacity;
• policy credibility;
• enforcement and leakage.

10.2 Hierarchical partial pooling

Use multilevel Bayesian models so that:

• Each episode receives an episode-specific estimate;
• related instruments share information;
• countries share information conditionally;
• estimates shrink when data are sparse;
• truly different regimes may remain different.

Example:

[ \tau_{episode} \sim N( \mu + \gamma^\top Context_{episode}, \sigma^2_{\tau} ) ]

10.3 Similarity-weighted evidence

For a target scenario, assign higher weight to historical episodes with similar:

• Instrument;
• target;
• financial structure;
• macro state;
• enforcement;
• leakage;
• external environment.

The similarity function must be documented and sensitivity-tested.

10.4 Mechanism decomposition

Decompose total effect into:

1. Credit-volume channel;
2. Sectoral-reallocation channel;
3. Asset-price channel;
4. Interest-income and debt-service channel;
5. Exchange-rate channel;
6. Investment and capacity channel;
7. Consumption channel;
8. Wage-bargaining and labor-demand channel;
9. Expectations channel;
10. Financial-instability channel;
11. Leakage channel.

Historical transfer is more defensible when the mechanism exists in both source and target settings.

10.5 Extrapolation flags

The simulator must warn when a scenario is outside historical support. Compute:

• Mahalanobis distance;
• propensity-score overlap;
• convex-hull membership;
• density ratio;
• nearest-neighbor distance;
• posterior predictive support.

Do not label an extrapolated estimate “high confidence” merely because the model’s conditional variance is small.

11. Forecasting Architecture

11.1 Baseline dynamic regression

For each outcome (y):

[ y_{i,t+h} = \alpha_i + \lambda_t + \sum_{k=1}^{K}\phi_k y_{i,t-k} + \sum_{m}\sum_{k=0}^{K}\beta_{m,k} A_{m,i,t-k} + \gamma^\top X_{it} + \delta^\top(A_{it}\times X_{it}) + \epsilon_{i,t+h} ]

This captures lagged outcomes, policy lags, controls, and interactions.

11.2 Joint state-space model

Use a Bayesian state-space model to estimate latent:

• Potential output;
• natural or noninflationary labor-market state;
• financial-cycle state;
• supply constraint;
• inflation trend;
• policy-regime state;
• credit-control effective intensity.

Parameters may evolve:

[ \theta_t = \theta_{t-1} + \nu_t ]

with shrinkage to prevent spurious drift.

11.3 Policy scenario engine

Each scenario should specify:

• Start date;
• phase-in;
• instrument values;
• compliance;
• enforcement;
• duration;
• feedback rules;
• triggers;
• automatic adjustment;
• political-reversal risk;
• external assumptions;
• job-guarantee design;
• fiscal accommodation;
• central-bank response rule.

The engine runs Monte Carlo trajectories through the posterior predictive distribution.

11.4 Endogenous policy reaction

Policies respond to outcomes. Ignoring this creates bias.

Model:

[ A_t = g(S_t, Political_t, Institutions_t, Expectations_t) + \eta_t ]

For historical estimation, identify exogenous policy innovations where possible. For simulation, allow user-selected reaction rules.

Examples:

• Raise sectoral reserve requirements if housing-credit growth exceeds a threshold;
• Expand job-guarantee positions if private unemployment rises;
• relax productive-credit quotas when capacity utilization falls;
• tighten speculative-credit controls when land-price growth exceeds a band.

11.5 Model frequency

Recommended architecture:

• Monthly operational model for inflation, credit, employment, rates, and financial conditions;
• Quarterly core macro model for GDP, wages, investment, and sectoral accounts;
• Annual long-run model for structural variables and historical episodes.

Use mixed-frequency state-space methods rather than forcing all data into annual frequency.

12. Jobs Guarantee Module

The national job guarantee should be modeled as both a policy lever and an automatic stabilizer.

12.1 Parameters

• Eligibility;
• take-up;
• offered wage;
• benefit package;
• hours;
• geographic availability;
• administrative lag;
• job matching;
• training component;
• public procurement;
• material and capital requirements;
• local implementation;
• transition rate to private employment;
• wage floor effect;
• productivity of output;
• import content;
• funding mechanism;
• tax offsets;
• program capacity constraints.

12.2 State variables

• Job-guarantee enrollment;
• vacancies;
• program wage bill;
• nonwage cost;
• produced public services;
• local labor tightness;
• private-sector displacement;
• private wage response;
• skill accumulation;
• transition rates;
• program bottlenecks.

12.3 Interaction with credit controls

Test scenarios where:

• Productive credit expands alongside job-guarantee employment;
• Speculative mortgage and asset credit is restrained;
• credit is directed toward capacity needed to absorb increased demand;
• small-business working capital is protected;
• public and cooperative development banks finance complementary investment.

This interaction may determine whether employment gains create capacity expansion or merely encounter supply bottlenecks.

13. Confidence, Prediction, and Futures Cones

13.1 Statistical confidence interval

A confidence interval concerns repeated-sampling uncertainty about an estimated parameter under a specified model. It is not the correct label for every future range.

Use for:

• Estimated treatment coefficients;
• impulse responses;
• average treatment effects;
• regression parameters.

13.2 Credible interval

A Bayesian credible interval summarizes posterior uncertainty conditional on the model, priors, and data.

Use for:

• Time-varying parameters;
• latent states;
• posterior treatment effects;
• posterior predictive simulations.

13.3 Prediction interval

A prediction interval includes both parameter uncertainty and future shocks under the assumed model.

Use for:

• Future inflation;
• unemployment;
• real wages;
• output;
• credit growth.

13.4 Model uncertainty interval

Construct across plausible models, specifications, identification assumptions, and data vintages.

Possible methods:

• Bayesian model averaging;
• stacking;
• bootstrap across specifications;
• multiverse analysis;
• robust Bayesian analysis;
• interval union or weighted mixture;
• model-confidence sets.

13.5 Scenario uncertainty

External assumptions are not always probabilistically estimable. Represent scenarios for:

• Oil shock;
• war;
• pandemic;
• climate disaster;
• productivity boom;
• financial crisis;
• capital flight;
• political reversal;
• rapid credit leakage;
• housing supply response.

13.6 Nested cone definitions

Probable cone

Operational definition:

The central posterior predictive mass under the reference ensemble and explicitly stated policy scenario.

Recommended display:

• 50%, 67%, 80%, and 90% predictive bands;
• median trajectory;
• probability of policy-relevant thresholds;
• model-calibration score.

This cone is probabilistic only conditional on the included models and assumptions.

Plausible cone

Operational definition:

The union or weighted envelope of trajectories generated by empirically defensible models, alternative parameterizations, historical analogues, structural mechanisms, and external scenarios that pass plausibility criteria.

A scenario is plausible when:

• It respects accounting identities;
• its causal mechanism is documented;
• parameters fall within defensible empirical or expert ranges;
• it does not require internally contradictory institutions;
• it survives review by economic, historical, and domain agents.

The plausible cone may include outcomes with no stable probability estimate.

Possible cone

Operational definition:

The broadest set of trajectories not excluded by hard constraints.

Hard constraints include:

• National accounting identities;
• sectoral balance sheets;
• labor-force bounds;
• production and capacity constraints;
• physical resource limits;
• legal and institutional rules specified in the scenario;
• nonnegative quantities where required;
• debt and flow consistency.

The possible cone includes rare shocks, structural breaks, and model failure. It should not be marketed as a probabilistic forecast.

13.7 Cone geometry metrics

For each policy scenario calculate:

• Center shift;
• width at each horizon;
• skewness;
• kurtosis;
• upper and lower tail movement;
• probability of threshold breaches;
• expected shortfall;
• downside semivariance;
• multimodality;
• overlap with baseline;
• Wasserstein distance between distributions;
• Kullback-Leibler divergence where appropriate;
• robustness across models;
• proportion of trajectories satisfying multiple goals.

13.8 Feasibility cones

Separate economic outcomes from implementation feasibility.

Estimate:

• Administrative feasibility;
• legal feasibility;
• political adoption probability;
• durability;
• compliance;
• evasion;
• international constraints;
• required data and supervisory capacity.

Kingdon’s three streams can inform an adoption-window model:

[ P(Adoption_t) = logit^{-1}( \alpha + \beta_P Problem_t + \beta_L PolicyReadiness_t + \beta_G Politics_t + \beta_W WindowAlignment_t ) ]

This should be reported separately from the economic effect conditional on adoption.

14. Validation and Falsification

14.1 Backtesting

Use rolling and expanding windows:

• Train through year (t);
• forecast (t+h);
• compare with observed outcomes;
• repeat across countries and regimes.

14.2 Real-time vintage validation

Use ALFRED and national vintage datasets to avoid revised-data hindsight.

14.3 Historical holdout episodes

Reserve entire policy episodes from training. Ask whether the system predicts their outcome distribution without seeing the episode.

14.4 Placebo tests

• False policy dates;
• unaffected sectors;
• unaffected loan types;
• pre-treatment outcomes;
• shuffled treatments;
• geographic placebo units.

14.5 Negative controls

Use outcomes or exposures that should not respond under the proposed mechanism.

14.6 Calibration

For predicted intervals:

• 50% intervals should contain roughly 50% of outcomes;
• 80% intervals roughly 80%;
• test conditional calibration by regime and horizon.

Metrics:

• Coverage;
• interval score;
• continuous ranked probability score;
• log predictive density;
• Brier score for events;
• probability integral transform;
• calibration slope and intercept.

14.7 Comparative baselines

Compare against:

• Random walk;
• autoregression;
• Phillips-curve baseline;
• standard VAR;
• professional forecasts;
• central-bank projections;
• IMF/OECD forecasts;
• no-policy-change model.

14.8 Sensitivity

Test:

• Alternative policy codings;
• treatment dates;
• lag lengths;
• outcome definitions;
• deflators;
• samples;
• priors;
• fixed versus time-varying effects;
• exclusion of major crises;
• financial-openness assumptions;
• leakage assumptions;
• policy endogeneity controls.

14.9 Falsification conditions

The project must predefine what findings would weaken major claims. Examples:

• No reproducible effect on targeted credit categories;
• no stronger response in more-exposed units;
• effects vanish when monetary and fiscal co-movements are controlled;
• estimated benefits occur only under historically unsupported enforcement assumptions;
• inflation improvements are offset by output and employment losses comparable to rate tightening;
• cross-model results are too unstable for a useful policy distinction.

15. Data Engineering Specification

15.1 Repository structure

prfm/
  README.md
  pyproject.toml
  configs/
  data_catalog/
  data_raw/
  data_external/
  data_interim/
  data_processed/
  event_database/
  metadata/
  notebooks/
  src/
    ingest/
    harmonize/
    policy_coding/
    features/
    causal/
    forecasting/
    simulation/
    uncertainty/
    validation/
    visualization/
    api/
  tests/
  reports/
  dashboards/
  model_cards/
  provenance/

15.2 Data standards

Every series should include:

• Canonical variable ID;
• source ID;
• source URL;
• country;
• region;
• sector;
• unit;
• frequency;
• seasonal adjustment;
• nominal/real status;
• deflator;
• vintage date;
• observation date;
• release date;
• transformation;
• missingness code;
• quality flag;
• license;
• retrieval timestamp;
• checksum.

15.3 Revisions

Never overwrite raw data. Store:

• Retrieval snapshot;
• source vintage;
• revision history;
• transformation code;
• lineage manifest.

15.4 Harmonization

Create separate layers:

1. Raw national concepts;
2. internationally harmonized concepts;
3. model-ready transformed variables.

Do not silently splice series with definitional breaks.

15.5 Missing data

Use:

• Explicit missingness;
• interpolation only where defensible;
• Kalman smoothing for latent-state models;
• multiple imputation;
• uncertainty propagation.

Never treat imputed policy events as observed facts.

15.6 Policy event knowledge graph

Entities:

• Policy action;
• instrument;
• authority;
• legal document;
• lender;
• borrower;
• sector;
• objective;
• country;
• economic episode;
• evidence source.

Relations:

• Introduced;
• amended;
• tightened;
• loosened;
• repealed;
• targeted;
• exempted;
• enforced by;
• circumvented through;
• accompanied by;
• cited by.

16. Agentic Work Packages

Agent A: Policy Ontology Agent

Deliverables:

• Credit-control taxonomy;
• machine-readable schema;
• controlled vocabulary;
• coding manual;
• edge-case guidance;
• example records.

Acceptance test:

Two independent coding agents should reach high agreement on a common sample.

Agent B: Historical Episode Research Agents

One agent per country or episode.

Deliverables:

• Chronology;
• primary-source archive;
• treatment coding;
• concurrent-policy table;
• identification opportunities;
• threats to validity;
• candidate datasets;
• confidence grading.

Agent C: Data Acquisition Agent

Deliverables:

• API and bulk-download connectors;
• license registry;
• raw snapshots;
• automated update jobs;
• checksums;
• source availability tests.

Agent D: Harmonization Agent

Deliverables:

• crosswalks;
• frequency conversion;
• currency and price transformations;
• seasonal-adjustment handling;
• metadata and break detection;
• harmonized panel.

Agent E: Causal Graph Agent

Deliverables:

• dynamic causal diagrams by instrument family;
• confounder and mediator lists;
• forbidden-control warnings;
• identification assumptions;
• negative controls.

Agent F: Econometric Agent

Deliverables:

• event studies;
• difference-in-differences;
• local projections;
• synthetic controls;
• IV/RD where justified;
• treatment-effect heterogeneity;
• specification multiverse.

Agent G: Forecasting Agent

Deliverables:

• dynamic regression;
• Bayesian VAR;
• state-space model;
• time-varying parameters;
• probabilistic forecasts;
• backtests.

Agent H: Structural Simulation Agent

Deliverables:

• stock-flow-consistent sectoral model;
• job-guarantee module;
• credit-allocation mechanisms;
• accounting checks;
• stress scenarios.

Agent I: Uncertainty and Futures Agent

Deliverables:

• probable/plausible/possible cone engine;
• model disagreement metrics;
• scenario envelopes;
• extrapolation warnings;
• feasibility cone;
• visualization specification.

Agent J: Validation Agent

Deliverables:

• holdout plan;
• vintage backtests;
• placebo tests;
• calibration reports;
• model comparison;
• falsification report.

Agent K: Application Agent

Deliverables:

• scenario API;
• interactive dashboard;
• parameter controls;
• provenance display;
• downloadable scenario manifest;
• charting of distributions and cones;
• accessible explanations.

Agent L: Reproducibility and Audit Agent

Deliverables:

• environment lockfiles;
• deterministic builds where possible;
• model cards;
• data cards;
• lineage reports;
• replication instructions;
• audit logs.

17. Minimum Viable Product

17.1 Scope

Countries:

• United States;
• United Kingdom;
• Japan;
• France;
• South Korea.

Period:

• Long-run annual panel where available;
• post-1960 quarterly model;
• post-1990 monthly model for modern validation.

Policies:

• Policy interest rates;
• aggregate credit controls;
• borrower-based housing controls;
• sectoral credit allocation;
• reserve requirements;
• job-guarantee scenarios as a counterfactual module.

Outcomes:

• CPI inflation;
• unemployment;
• employment-to-population ratio;
• real wages;
• real GDP;
• aggregate and sectoral credit;
• house prices;
• investment.

17.2 MVP model stack

1. Policy event database;
2. Dynamic panel local projections;
3. Bayesian time-varying parameter model;
4. synthetic-control case studies;
5. posterior scenario simulator;
6. cone visualization;
7. backtesting and calibration report.

17.3 MVP user workflow

1. Select country and start date;
2. select baseline;
3. alter interest-rate path;
4. add credit controls by instrument and sector;
5. set enforcement and leakage;
6. set fiscal and job-guarantee parameters;
7. choose external scenarios;
8. run simulation;
9. inspect median, quantiles, tails, and model disagreement;
10. inspect causal evidence and historical analogues;
11. download scenario and provenance.

18. Recommended Technical Stack

• Python;
• Polars or pandas;
• DuckDB;
• PostgreSQL;
• Parquet;
• dbt or equivalent transformation framework;
• PyMC, Stan, or NumPyro;
• statsmodels;
• linearmodels;
• EconML or DoubleML for selected heterogeneous-effect workflows;
• scikit-learn;
• ArviZ;
• Prefect, Dagster, or Airflow;
• FastAPI;
• Plotly or Observable for interactive uncertainty graphics;
• Docker;
• Git LFS or object storage for large immutable raw files;
• DVC or lakeFS for data versioning;
• Quarto for reports.

All core estimates should be reproducible without proprietary software. Proprietary datasets may be optional extensions, not dependencies of the public baseline.

19. Output and Visualization Requirements

19.1 Core plots

• Fan charts;
• nested futures cones;
• scenario distribution overlays;
• impulse-response functions;
• treatment-effect forest plots;
• sectoral credit Sankey diagrams;
• historical analogue maps;
• model-disagreement plots;
• threshold probability charts;
• policy frontier plots.

19.2 Policy frontier

Display combinations such as:

• Inflation versus unemployment;
• real-wage growth versus inflation;
• employment versus financial-crisis risk;
• productive investment versus housing inflation.

Rather than claiming one optimal policy, show the estimated frontier and which policy packages dominate others under defined criteria.

19.3 Cone-change visualization

For baseline (B) and policy (P), show:

• Shift in median;
• narrowing or widening;
• tail-risk change;
• overlap;
• probability of achieving a target bundle.

Example target bundle:

[ U < 4\%,\quad 1\% < \pi < 4\%,\quad \Delta RW > 1.5\% ]

Report:

[ P(TargetBundle\mid Policy) - P(TargetBundle\mid Baseline) ]

20. Governance, Transparency, and Interpretation

20.1 Model cards

Each model card must state:

• Intended use;
• data;
• identification assumptions;
• training period;
• policy support;
• known failure modes;
• calibration;
• uncertainty included;
• uncertainty omitted;
• extrapolation behavior.

20.2 Normative assumptions

The model should separate estimated facts from normative choices. Users may choose objective weights, but default reports should show outcomes separately rather than collapsing them into one welfare score.

20.3 Political economy

Where the project evaluates the claim that interest-rate policy disciplines labor by weakening employment and wages, it should operationalize and test the mechanisms rather than encode the conclusion as an assumption.

Relevant measurable channels include:

• Unemployment response;
• vacancy response;
• wage growth;
• labor share;
• union activity;
• strike outcomes;
• household debt service;
• firm financing;
• distribution of interest income;
• asset-price effects;
• sectoral employment.

20.4 Communication rule

Every chart and headline estimate must answer:

• Compared with what?
• Conditional on what?
• Over what horizon?
• Under which models?
• With what uncertainty?
• Is the scenario interpolated or extrapolated?
• Which historical episodes provide support?
• What would falsify the result?

21. Development Phases

Phase 0: Concept and ontology

• Finalize policy taxonomy;
• define outcomes;
• create causal diagrams;
• define cone semantics;
• establish governance.

Phase 1: Data foundation

• Build source registry;
• ingest core datasets;
• create policy-event schema;
• code priority episodes;
• produce harmonized panel.

Phase 2: Causal case studies

• U.S. 1980;
• U.K. Corset;
• Japanese window guidance;
• one borrower-based international panel;
• one directed-credit development case.

Phase 3: Forecasting baseline

• Dynamic regressions;
• Bayesian VAR;
• state-space model;
• backtesting;
• vintage validation.

Phase 4: Scenario and cone engine

• Monte Carlo simulation;
• policy packages;
• external shocks;
• probable/plausible/possible cones;
• extrapolation alerts.

Phase 5: Jobs guarantee and structural layer

• automatic stabilizer;
• wage floor;
• capacity response;
• sectoral credit interaction;
• stock-flow consistency.

Phase 6: Public application

• API;
• dashboard;
• documentation;
• downloadable reports;
• open replication package.

22. Acceptance Criteria

The first production candidate is acceptable only when:

1. Every policy event has traceable evidence;
2. credit controls are represented by type, target, intensity, coverage, bindingness, enforcement, and leakage;
3. at least three independent model families produce forecasts;
4. historical holdout performance is reported;
5. predictive intervals are calibrated or explicitly corrected;
6. model uncertainty is visible;
7. probable, plausible, and possible cones are not conflated;
8. extrapolation is flagged;
9. baseline and policy scenarios are reproducible from a manifest;
10. the system can estimate differences in inflation, unemployment, and real wages over multiple horizons;
11. the job-guarantee module interacts with credit allocation and capacity;
12. all charts expose assumptions and provenance;
13. the public interface never presents a single path as “the forecast.”

23. Initial Research Questions for Deep-Research Agents

1. Which historical credit-control episodes offer the strongest within-country variation and least-confounded assignment?
2. For each instrument, what is the best measure of effective rather than announced intensity?
3. How much lending shifted to uncontrolled institutions or offshore channels?
4. Which controls changed aggregate credit, and which merely reallocated it?
5. What were the effects on investment, employment, prices, wages, housing, and financial stability?
6. Did controls work differently under capital controls, public banking, or concentrated banking systems?
7. How did policy effects change with slack, inflation, and external constraints?
8. Which episodes combine credit controls with high employment or job-guarantee-like policies?
9. Are borrower-based controls informative for productive-credit controls, or should they remain separate treatment classes?
10. Can historical bank-level assignment rules support credible instruments or discontinuities?
11. How should enforcement and leakage be measured from archival evidence?
12. Which modern datasets provide sectoral credit at monthly or quarterly frequency?
13. How can real-wage distributions be harmonized across historical regimes?
14. Which aspects of a modern national job guarantee lie outside historical support?
15. What range of outcomes remains plausible after accounting for structural model disagreement?

24. Initial Coding Tasks for Agentic Coding Agents

1. Create the repository and environment;
2. implement the data catalog schema;
3. write BIS, World Bank, FRED/ALFRED, OECD, and ILOSTAT connectors;
4. build immutable raw-data snapshots;
5. implement country and variable crosswalks;
6. create the policy-event relational schema;
7. create validation rules for policy records;
8. implement panel-building and transformation pipelines;
9. implement baseline local projections;
10. implement Bayesian dynamic regression;
11. implement forecast simulation;
12. implement calibration metrics;
13. implement distribution-distance and cone-width metrics;
14. implement scenario manifests;
15. implement provenance endpoints;
16. build prototype fan-chart and nested-cone views;
17. write unit, integration, and data-quality tests;
18. create model and data card templates.

25. Final Methodological Position

The project can accurately be described as beginning with multivariate dynamic regression. However, its substantive goal requires more than fitting outcome variables on policy variables.

The additional requirements arise because:

• The treatment is multidimensional;
• policy is endogenous;
• effects are dynamic;
• coefficients vary by regime and state;
• historical evidence must be transported to new contexts;
• future shocks matter;
• models disagree;
• unprecedented policy packages require structural constraints;
• statistical confidence is narrower than futures uncertainty;
• “possible,” “plausible,” and “probable” refer to different epistemic sets.

The project’s research contribution is therefore not a novel regression equation by itself. It is a disciplined architecture for turning fragmented historical evidence about credit governance into transparent, conditional distributions of alternative economic futures.

Selected Starting References and Data Portals

• Bank for International Settlements, Credit to the non-financial sector:
  https://www.bis.org/statistics/totcredit.htm
• BIS Data Portal, Total Credit:
  https://data.bis.org/topics/TOTAL_CREDIT/data
• BIS Credit-to-GDP gaps:
  https://data.bis.org/topics/CREDIT_GAPS
• IMF Integrated Macroprudential Policy Database:
  https://www.elibrary-areaer.imf.org/Macroprudential/Pages/iMaPPDatabase.aspx
• World Bank Global Financial Development Database:
  https://www.worldbank.org/en/publication/gfdr/data/global-financial-development-database
• World Bank Indicators API:
  https://datahelpdesk.worldbank.org/knowledgebase/articles/889392-about-the-indicators-api-documentation
• ILOSTAT bulk data:
  https://ilostat.ilo.org/data/bulk/
• OECD Data Explorer:
  https://data-explorer.oecd.org/
• FRED API:
  https://fred.stlouisfed.org/docs/api/fred/
• Jordà-Schularick-Taylor Macrohistory Database:
  https://www.macrohistory.net/database/
• Federal Reserve, History of Cyclical Macroprudential Policy in the United States:
  https://www.federalreserve.gov/pubs/feds/2013/201329/index.html
• Bank of England, Supplementary Special Deposits Scheme:
  https://www.bankofengland.co.uk/quarterly-bulletin/1982/q1/the-supplementary-special-deposits-scheme
• Bank of Japan, Effectiveness of Window Guidance:
  https://www.boj.or.jp/en/research/wps_rev/rev_2010/data/rev10e04.pdf