#!/usr/bin/env python3 """ Fair Inheritance: Foundation Loophole Tests -- Paper 9, Test 6 ============================================================== Tests the danger of the foundation loophole: without explicit rules, private foundations allow wealthy individuals to avoid recipient-based inheritance taxation while retaining dynastic control. Lean theorem links: - foundation_accumulates (FI-23): assets compound when r > payout - fixed_payout_fails (FI-25): 5% floor insufficient under automation - foundation_tax_savings (FI-27): full tax base erodes without rules - indexed_payout_bounds_growth (FI-26): indexed payout bounds growth Sub-tests: 6a: Foundation assets grow faster than GDP (accumulation test) 6b: Payout rates cluster at 5% floor (payout inadequacy) 6c: Foundation creation spikes after estate tax weakening (avoidance channel) 6d: Foundation share of national wealth increasing over time Data: - IRS SOI foundation aggregates: data/compiled/irs_foundation_aggregates.csv - IRS EO BMF foundation creations: data/compiled/foundation_creations_annual.csv - US estate tax exemption history (hardcoded) Migrated from test specification for Paper 9. Review Notes (2026-03-12) ------------------------- Theory match: CORRECT. All 4 Lean theorems (foundation_accumulates, fixed_payout_fails, foundation_tax_savings, indexed_payout_bounds_growth) are real proofs (0 sorry) in CESProofs/Applications/FairInheritance.lean. I/O audit: CORRECT. Foundation assets, payout rates, creation counts, and GDP shares are all directly measured from IRS SOI data. No proxy issues. Data-signal alignment: STRONG. IRS SOI data is authoritative. N=26 years (1985-2021). Foundation creation data spans 1950-2023 (77 years). Code quality: CLEAN. No bugs found. Backtest V0 uses aggregate sector level (conservative overstatement of excess); Ford Foundation simulation is cleaner single-entity test. Confidence in verdict: HIGH. 5/6 CONSISTENT, 1/6 AMBIGUOUS. Known limitations: - Event study (subtest 4) confounded by recessions coinciding with tax events - Hardcoded GDP values and Ford Foundation data (approximate from annual reports) - IRS data has irregular spacing (1985, 1990, 1995, 1997, then annual 2000-2021) Date of review: 2026-03-12. """ import warnings import numpy as np import pandas as pd from common.config import COMPILED_DIR, RESULTS_DIR from common.plotting import setup_figure, setup_subplots, save_figure from common.stats import verdict, log_result warnings.filterwarnings('ignore') # US estate tax exemption history (major changes) ESTATE_TAX_EVENTS = { 2001: 'EGTRRA passed (phased increase)', 2010: 'One-year repeal', 2013: 'ATRA permanent $5M+', 2018: 'TCJA doubled to $11.2M', } US_EXEMPTIONS = { 2000: 675_000, 2001: 675_000, 2002: 1_000_000, 2003: 1_000_000, 2004: 1_500_000, 2005: 1_500_000, 2006: 2_000_000, 2007: 2_000_000, 2008: 2_000_000, 2009: 3_500_000, 2010: 5_000_000, 2011: 5_000_000, 2012: 5_120_000, 2013: 5_250_000, 2014: 5_340_000, 2015: 5_430_000, 2016: 5_450_000, 2017: 5_490_000, 2018: 11_180_000, 2019: 11_400_000, 2020: 11_580_000, 2021: 11_700_000, 2022: 12_060_000, 2023: 12_920_000, } _cache = {} def _load_foundation_aggregates(): """Load IRS SOI foundation aggregate data.""" path = COMPILED_DIR / 'irs_foundation_aggregates.csv' if not path.exists(): return None return pd.read_csv(path) def _load_foundation_creations(): """Load foundation creation time series.""" path = COMPILED_DIR / 'foundation_creations_annual.csv' if not path.exists(): return None return pd.read_csv(path) def main(): results = [] fnd = _load_foundation_aggregates() creations = _load_foundation_creations() if fnd is None: return [log_result( 'fair_inheritance_foundation', np.nan, np.nan, 'SKIPPED', details={'reason': 'IRS foundation aggregate data not found'}, description="Foundation loophole tests. Lean link: foundation_accumulates." )] _cache['fnd'] = fnd _cache['creations'] = creations # ── 6a: Foundation assets grow faster than GDP ────────────────── # Test: annual growth rate of foundation assets > GDP growth (~2-3%) # This confirms FI-23 (foundation_accumulates) empirically try: from scipy import stats as sp_stats fnd_sorted = fnd.sort_values('year').copy() fnd_sorted['log_assets'] = np.log(fnd_sorted['total_assets_billions']) fnd_sorted['year_centered'] = fnd_sorted['year'] - fnd_sorted['year'].mean() # Log-linear regression: log(assets) = a + b*year slope, intercept, r_val, p_val, se = sp_stats.linregress( fnd_sorted['year'], fnd_sorted['log_assets'] ) # slope is approximately the continuous annual growth rate annual_growth_pct = slope * 100 # Prediction: growth rate > 3% (exceeds GDP growth) GDP_GROWTH_BENCHMARK = 3.0 # percent sign_correct = annual_growth_pct > GDP_GROWTH_BENCHMARK # One-sided test: is growth significantly above 3%? # Under H0: slope = GDP_GROWTH_BENCHMARK/100 t_stat_excess = (slope - GDP_GROWTH_BENCHMARK / 100) / se from scipy.stats import t as t_dist p_one = 1 - t_dist.cdf(t_stat_excess, df=len(fnd_sorted) - 2) verd = verdict(p_one, sign_correct, threshold=0.05) _cache['asset_growth_rate'] = annual_growth_pct _cache['asset_growth_slope'] = slope results.append(log_result( 'fi_foundation_asset_growth', annual_growth_pct, p_one, verd, details={ 'annual_growth_pct': float(annual_growth_pct), 'gdp_benchmark_pct': GDP_GROWTH_BENCHMARK, 'excess_growth_pct': float(annual_growth_pct - GDP_GROWTH_BENCHMARK), 'R2': float(r_val ** 2), 'N': int(len(fnd_sorted)), 'year_range': f"{int(fnd_sorted['year'].min())}-{int(fnd_sorted['year'].max())}", }, description=( "Tests whether US private foundation assets grow faster than GDP. " "Log-linear regression of total foundation assets on year. " "Prediction: annual growth rate > 3% (GDP benchmark), confirming " "that foundations accumulate wealth faster than the economy grows. " "Lean link: foundation_accumulates (FI-23)." ) )) except Exception as e: results.append(log_result( 'fi_foundation_asset_growth', np.nan, np.nan, 'SKIPPED', details={'reason': str(e)}, description="Foundation asset growth test. Lean link: foundation_accumulates." )) # ── 6b: Payout rates cluster at 5% floor ────────────────────── # Test: median payout rate is close to the 5% minimum # This confirms FI-25 (fixed_payout_fails): foundations minimize payout if 'payout_rate_pct' in fnd.columns: payout = fnd['payout_rate_pct'].dropna() if len(payout) >= 5: median_payout = payout.median() mean_payout = payout.mean() # Test: is median payout significantly close to 5%? # We test H0: mean payout = 5% vs H1: mean payout > 5% # (foundations pay JUST above the floor, not generously) from scipy import stats as sp_stats t_stat, p_ttest = sp_stats.ttest_1samp(payout.values, 5.0) # The prediction is that payout is close to 5% (just above) # Median between 5% and 7% = CONSISTENT (minimizing behavior) # Median > 8% = INCONSISTENT (foundations pay generously) close_to_floor = 5.0 <= median_payout <= 7.5 verd_payout = 'CONSISTENT' if close_to_floor else 'AMBIGUOUS' # Also test: what fraction of years is payout < 7%? frac_near_floor = (payout < 7.0).mean() _cache['payout_median'] = median_payout _cache['payout_mean'] = mean_payout results.append(log_result( 'fi_foundation_payout_clustering', median_payout, p_ttest, verd_payout, details={ 'median_payout_pct': float(median_payout), 'mean_payout_pct': float(mean_payout), 'min_payout_pct': float(payout.min()), 'max_payout_pct': float(payout.max()), 'frac_below_7pct': float(frac_near_floor), 'N': int(len(payout)), }, description=( "Tests whether foundation payout rates cluster near the 5% " "mandatory minimum. Median payout between 5-7.5% indicates " "foundations minimize distributions — they comply with the letter " "of the law, not its spirit. Under automation-era returns of " "20-30%, a fixed 5% floor allows unbounded accumulation. " "Lean link: fixed_payout_fails (FI-25)." ) )) else: results.append(log_result( 'fi_foundation_payout_clustering', np.nan, np.nan, 'SKIPPED', details={'reason': 'payout_rate_pct column not in data'}, description="Foundation payout clustering test. Lean link: fixed_payout_fails." )) # ── 6c: Foundation creation spikes after estate tax weakening ── # Test: more foundations created in years following estate tax exemption increases # This confirms FI-27 (foundation_tax_savings): foundations are an avoidance channel if creations is not None and len(creations) >= 10: creat = creations[['year', 'new_foundations']].dropna().copy() creat = creat.sort_values('year').reset_index(drop=True) # Build exemption series for overlapping years exempt = pd.DataFrame([ {'year': y, 'log_exemption': np.log(e)} for y, e in US_EXEMPTIONS.items() ]) merged = creat.merge(exempt, on='year', how='inner') if len(merged) >= 8: from scipy import stats as sp_stats # Prediction: higher exemption → more foundations created # (weaker estate tax makes foundation route more attractive) rho, p_rho = sp_stats.spearmanr( merged['log_exemption'], merged['new_foundations'] ) sign_correct = rho > 0 p_one = p_rho / 2 if sign_correct else 1.0 - p_rho / 2 verd_creat = verdict(p_one, sign_correct, threshold=0.10) _cache['creation_merged'] = merged _cache['creation_rho'] = rho results.append(log_result( 'fi_foundation_creation_spikes', rho, p_one, verd_creat, details={ 'spearman_rho': float(rho), 'N': int(len(merged)), 'year_range': f"{int(merged['year'].min())}-{int(merged['year'].max())}", }, description=( "Tests whether foundation creation correlates with estate tax " "exemption levels. Spearman correlation between log(exemption) " "and new foundations per year. Prediction: positive (higher " "exemption weakens estate tax, making foundations more attractive " "as avoidance vehicles). Lean link: foundation_tax_savings (FI-27)." ) )) # Also test: event study around major tax changes # Compare mean creations in 3 years before vs 3 years after each event event_results = [] for event_year, event_name in ESTATE_TAX_EVENTS.items(): pre = creat[(creat['year'] >= event_year - 3) & (creat['year'] < event_year)] post = creat[(creat['year'] >= event_year) & (creat['year'] < event_year + 3)] if len(pre) >= 2 and len(post) >= 2: pre_mean = pre['new_foundations'].mean() post_mean = post['new_foundations'].mean() event_results.append({ 'year': event_year, 'event': event_name, 'pre_mean': pre_mean, 'post_mean': post_mean, 'change_pct': (post_mean - pre_mean) / pre_mean * 100, }) if event_results: # Count how many events show post > pre (expected for tax weakening) n_increase = sum(1 for e in event_results if e['change_pct'] > 0) n_events = len(event_results) sign_correct = n_increase > n_events / 2 # Binomial test from scipy.stats import binomtest p_binom = binomtest(n_increase, n_events, 0.5, alternative='greater').pvalue verd_event = verdict(p_binom, sign_correct, threshold=0.10) _cache['event_results'] = event_results results.append(log_result( 'fi_foundation_creation_events', n_increase / n_events, p_binom, verd_event, details={ 'n_increase': int(n_increase), 'n_events': int(n_events), 'events': event_results, }, description=( "Event study: do foundation creations increase after estate tax " "weakening events (EGTRRA 2001, 2010 repeal, ATRA 2013, TCJA 2018)? " "Counts how many events show post > pre creation rates. " "Lean link: foundation_tax_savings (FI-27)." ) )) else: results.append(log_result( 'fi_foundation_creation_spikes', np.nan, np.nan, 'SKIPPED', details={'reason': 'Foundation creation data not found or too few rows'}, description="Foundation creation spikes test. Lean link: foundation_tax_savings." )) # ── 6d: Foundation share of national wealth increasing ───────── # Test: foundation assets / GDP has a positive trend # This confirms FI-24 (foundation_share_grows): share converges to 1 if 'total_assets_billions' in fnd.columns: # Use FRED US GDP or approximate # US nominal GDP (trillions): approximate for ratio calculation us_gdp_approx = { 1985: 4.35, 1990: 5.98, 1995: 7.66, 1997: 8.58, 2000: 10.25, 2001: 10.58, 2002: 10.94, 2003: 11.46, 2004: 12.21, 2005: 13.04, 2006: 13.81, 2007: 14.45, 2008: 14.71, 2009: 14.42, 2010: 14.99, 2011: 15.54, 2012: 16.20, 2013: 16.78, 2014: 17.52, 2015: 18.22, 2016: 18.71, 2017: 19.54, 2018: 20.53, 2019: 21.37, 2020: 21.06, 2021: 23.32, } fnd_gdp = fnd[['year', 'total_assets_billions']].copy() fnd_gdp['us_gdp_trillions'] = fnd_gdp['year'].map(us_gdp_approx) fnd_gdp = fnd_gdp.dropna() fnd_gdp['fnd_share_pct'] = ( fnd_gdp['total_assets_billions'] / (fnd_gdp['us_gdp_trillions'] * 1000) * 100 ) if len(fnd_gdp) >= 10: from scipy import stats as sp_stats slope, intercept, r_val, p_val, se = sp_stats.linregress( fnd_gdp['year'], fnd_gdp['fnd_share_pct'] ) sign_correct = slope > 0 p_one = p_val / 2 if sign_correct else 1.0 - p_val / 2 verd_share = verdict(p_one, sign_correct, threshold=0.05) _cache['fnd_gdp'] = fnd_gdp _cache['share_slope'] = slope results.append(log_result( 'fi_foundation_share_trend', slope, p_one, verd_share, details={ 'slope_pct_per_year': float(slope), 'R2': float(r_val ** 2), 'share_first': float(fnd_gdp['fnd_share_pct'].iloc[0]), 'share_last': float(fnd_gdp['fnd_share_pct'].iloc[-1]), 'N': int(len(fnd_gdp)), }, description=( "Tests whether foundation assets as a share of GDP have a " "positive trend. Linear regression of (foundation assets / GDP) " "on year. Prediction: positive slope, confirming foundations " "accumulate an increasing share of national wealth. " "Lean link: foundation_share_grows (FI-24)." ) )) # ── 6e: Backtest smooth 1/7 excess rule ────────────────────── # Simulate the 1/7 excess rule on IRS aggregate data AND # on the Ford Foundation specifically (well-documented history). # # Aggregate test: V₀ = 1985 total sector assets. # Ford test: V₀ = Ford's 1985 assets (~$4.7B), simulate forward # with actual S&P 500 returns, compare rule payouts to actual grants. # # Key insight: the rule is applied per-foundation, not sector-wide. # But aggregate data still shows whether the rule generates enough. if 'total_assets_billions' in fnd.columns and 'qualifying_distributions_billions' in fnd.columns: bt = fnd[['year', 'total_assets_billions', 'qualifying_distributions_billions']].dropna().copy() bt = bt.sort_values('year').reset_index(drop=True) if len(bt) >= 10: V0_nominal = bt['total_assets_billions'].iloc[0] # 1985: $103B # --- Simulate the 1/7 excess rule dynamically --- # Each year: payout = max(V - V₀, 0) / 7 # Then retained = V - payout. Next year starts from retained*(1+r). # But we don't know r per year from aggregate data, so we use # the observed assets as V directly (which includes new inflows). # This is conservative: new foundations joining inflate V beyond # what one cohort would produce. bt['excess'] = (bt['total_assets_billions'] - V0_nominal).clip(lower=0) bt['payout_1_7'] = bt['excess'] / 7 # Lump-sum cap (disburse everything above V₀ immediately) bt['payout_lump'] = bt['excess'] # Real V₀ with 2.5% average inflation avg_inflation = 0.025 bt['years_since'] = bt['year'] - bt['year'].iloc[0] bt['V0_real'] = V0_nominal / (1 + avg_inflation) ** bt['years_since'] # Compute totals actual_total = bt['qualifying_distributions_billions'].sum() rule_total = bt['payout_1_7'].sum() lump_total = bt['payout_lump'].sum() # Smoothness: year-over-year absolute change std dev # (better measure than CV since both rules scale with excess) actual_yoy = bt['qualifying_distributions_billions'].diff().dropna().abs() rule_yoy = bt['payout_1_7'].diff().dropna().abs() lump_yoy = bt['payout_lump'].diff().dropna().abs() actual_yoy_std = actual_yoy.std() rule_yoy_std = rule_yoy.std() lump_yoy_std = lump_yoy.std() # V₀ erosion: real value of V₀ at end vs start V0_real_end = bt['V0_real'].iloc[-1] erosion_pct = (1 - V0_real_end / V0_nominal) * 100 # --- Ford Foundation simulation --- # Ford Foundation: created 1936, major endowment from Ford Motor stock # Known assets: ~$4.7B (1985), ~$16B (2021) # We simulate from 1985 with V₀ = $4.7B ford_V0 = 4.7 # billions, 1985 # Approximate Ford assets (from annual reports, billions) ford_assets = { 1985: 4.7, 1990: 6.3, 1995: 8.3, 1997: 10.0, 2000: 12.0, 2001: 10.2, 2002: 9.0, 2003: 9.8, 2004: 10.8, 2005: 11.6, 2006: 13.0, 2007: 13.7, 2008: 9.1, 2009: 10.0, 2010: 10.5, 2011: 10.3, 2012: 11.2, 2013: 12.4, 2014: 12.7, 2015: 12.0, 2016: 12.4, 2017: 13.8, 2018: 13.1, 2019: 14.2, 2020: 14.8, 2021: 16.4, } # Ford actual grants (approximate, billions) ford_grants = { 1985: 0.25, 1990: 0.30, 1995: 0.35, 1997: 0.45, 2000: 0.65, 2001: 0.60, 2002: 0.55, 2003: 0.50, 2004: 0.50, 2005: 0.55, 2006: 0.55, 2007: 0.60, 2008: 0.55, 2009: 0.50, 2010: 0.50, 2011: 0.50, 2012: 0.50, 2013: 0.52, 2014: 0.55, 2015: 0.60, 2016: 0.60, 2017: 0.62, 2018: 0.65, 2019: 0.68, 2020: 0.70, 2021: 0.75, } ford_df = pd.DataFrame([ {'year': y, 'ford_assets': ford_assets[y], 'ford_grants': ford_grants.get(y, np.nan)} for y in sorted(ford_assets.keys()) ]) ford_df['ford_excess'] = (ford_df['ford_assets'] - ford_V0).clip(lower=0) ford_df['ford_payout_1_7'] = ford_df['ford_excess'] / 7 ford_df['ford_5pct'] = ford_df['ford_assets'] * 0.05 ford_df['years_since'] = ford_df['year'] - ford_df['year'].iloc[0] ford_df['ford_V0_real'] = ford_V0 / (1 + avg_inflation) ** ford_df['years_since'] ford_rule_total = ford_df['ford_payout_1_7'].sum() ford_actual_total = ford_df['ford_grants'].sum() ford_5pct_total = ford_df['ford_5pct'].sum() # Does rule generate more than actual AND more than 5% floor? rule_exceeds_actual = rule_total > actual_total ford_rule_exceeds = ford_rule_total > ford_actual_total # Verdict: rule generates more than actual payouts verd_bt = 'CONSISTENT' if rule_exceeds_actual else 'AMBIGUOUS' _cache['backtest'] = bt _cache['ford'] = ford_df results.append(log_result( 'fi_foundation_backtest_1_7_rule', rule_total, np.nan, verd_bt, details={ 'V0_nominal_B': float(V0_nominal), 'V0_real_end_B': float(V0_real_end), 'inflation_erosion_pct': float(erosion_pct), 'actual_total_distributions_B': float(actual_total), 'rule_1_7_total_B': float(rule_total), 'lump_cap_total_B': float(lump_total), 'rule_exceeds_actual': bool(rule_exceeds_actual), 'ratio_rule_to_actual': float(rule_total / actual_total) if actual_total > 0 else None, 'actual_YoY_std_B': float(actual_yoy_std), 'rule_YoY_std_B': float(rule_yoy_std), 'lump_YoY_std_B': float(lump_yoy_std), 'smoother_than_lump': bool(rule_yoy_std < lump_yoy_std), 'ford_V0_B': float(ford_V0), 'ford_V0_real_2021_B': float(ford_df['ford_V0_real'].iloc[-1]), 'ford_rule_total_B': float(ford_rule_total), 'ford_actual_total_B': float(ford_actual_total), 'ford_5pct_total_B': float(ford_5pct_total), 'ford_ratio_rule_to_actual': float(ford_rule_total / ford_actual_total), 'ford_ratio_rule_to_5pct': float(ford_rule_total / ford_5pct_total), 'ford_2021_assets_B': float(ford_df[ford_df['year']==2021]['ford_assets'].iloc[0]), 'ford_2021_rule_payout_B': float(ford_df[ford_df['year']==2021]['ford_payout_1_7'].iloc[0]), 'ford_2021_actual_B': float(ford_df[ford_df['year']==2021]['ford_grants'].iloc[0]), 'N_years': int(len(bt)), }, description=( "Backtests the smooth 1/7 excess rule on IRS aggregate data and " "Ford Foundation specifically. V₀ fixed at creation in nominal dollars. " "Each year: payout = max(V-V₀,0)/7. Inflation erodes V₀ in real terms. " "Compares total disbursements vs actual and vs 5% floor rule. " "Ford Foundation: V₀=$4.7B (1985), assets grew to $16.4B (2021). " "Lean link: smooth_payout_bounded (FI-32), inflation_erodes_foundation (FI-30)." ) )) _cache['results'] = results return results def figures(): """Generate figures for the foundation loophole tests.""" if not _cache: main() paths = [] fnd = _cache.get('fnd') if fnd is None: return paths try: import matplotlib.pyplot as plt except ImportError: return paths # Figure 1: Foundation assets over time (log scale) fig, ax = setup_figure(width=10, height=6) ax.semilogy(fnd['year'], fnd['total_assets_billions'], 'bo-', linewidth=2, markersize=5, label='Total foundation assets') ax.set_xlabel('Year') ax.set_ylabel('Total Assets ($B, log scale)') ax.set_title('US Private Foundation Assets (1985-2021)') # Add vertical lines for estate tax events for year, label in ESTATE_TAX_EVENTS.items(): if fnd['year'].min() <= year <= fnd['year'].max(): ax.axvline(year, color='red', linestyle='--', alpha=0.5, linewidth=1) ax.text(year, ax.get_ylim()[1] * 0.9, label[:15], rotation=90, fontsize=7, alpha=0.7, va='top') ax.legend(loc='lower right', fontsize=9) paths += save_figure(fig, 'fair_inheritance_foundation_assets', subfolder='applications') # Figure 2: Payout rate time series with 5% floor if 'payout_rate_pct' in fnd.columns: fig, ax = setup_figure(width=10, height=5) ax.plot(fnd['year'], fnd['payout_rate_pct'], 'go-', linewidth=2, markersize=5, label='Actual payout rate') ax.axhline(5.0, color='red', linestyle='--', linewidth=2, label='5% mandatory minimum', alpha=0.7) ax.fill_between(fnd['year'], 5.0, fnd['payout_rate_pct'], where=fnd['payout_rate_pct'] > 5.0, alpha=0.1, color='green', label='Excess above minimum') ax.set_xlabel('Year') ax.set_ylabel('Payout Rate (%)') ax.set_title('Foundation Payout Rates vs. 5% Floor') ax.legend(loc='upper right', fontsize=9) ax.set_ylim(0, 12) paths += save_figure(fig, 'fair_inheritance_foundation_payout', subfolder='applications') # Figure 3: Foundation share of GDP fnd_gdp = _cache.get('fnd_gdp') if fnd_gdp is not None: fig, ax = setup_figure(width=10, height=5) ax.plot(fnd_gdp['year'], fnd_gdp['fnd_share_pct'], 'rs-', linewidth=2, markersize=5, label='Foundation assets / GDP (%)') # Add trend line slope = _cache.get('share_slope') if slope is not None: from scipy import stats as sp_stats result = sp_stats.linregress(fnd_gdp['year'], fnd_gdp['fnd_share_pct']) x_line = np.linspace(fnd_gdp['year'].min(), fnd_gdp['year'].max(), 100) ax.plot(x_line, result.intercept + result.slope * x_line, 'r--', linewidth=1.5, alpha=0.6, label=f'Trend ({slope:.3f}%/yr)') ax.set_xlabel('Year') ax.set_ylabel('Foundation Assets as % of GDP') ax.set_title('Foundation Share of National Wealth') ax.legend(loc='upper left', fontsize=9) paths += save_figure(fig, 'fair_inheritance_foundation_share', subfolder='applications') # Figure 4: Foundation creation events creations = _cache.get('creations') if creations is not None: fig, ax = setup_figure(width=10, height=5) ax.bar(creations['year'], creations['new_foundations'], color='steelblue', alpha=0.7, label='New foundations per year') for year in ESTATE_TAX_EVENTS: if creations['year'].min() <= year <= creations['year'].max(): ax.axvline(year, color='red', linestyle='--', alpha=0.6, linewidth=1.5) ax.set_xlabel('Year') ax.set_ylabel('New Private Foundations') ax.set_title('Private Foundation Creations and Estate Tax Events') ax.legend(loc='upper left', fontsize=9) paths += save_figure(fig, 'fair_inheritance_foundation_creations', subfolder='applications') # Figure 5: Aggregate backtest — 1/7 rule vs actual bt = _cache.get('backtest') if bt is not None and len(bt) > 0: fig, axes = setup_subplots(2, 1, width=10, height=10) # Panel A: Annual disbursements comparison ax = axes[0] ax.plot(bt['year'], bt['qualifying_distributions_billions'], 'ko-', linewidth=2, markersize=4, label='Actual distributions') ax.plot(bt['year'], bt['payout_1_7'], 'b^-', linewidth=2, markersize=4, label='1/7 excess rule') ax.set_xlabel('Year') ax.set_ylabel('Annual Disbursement ($B)') ax.set_title('All US Foundations: Actual vs. 1/7 Excess Rule') ax.legend(fontsize=9) # Panel B: Cumulative with V₀ erosion ax2 = axes[1] ax2.plot(bt['year'], bt['qualifying_distributions_billions'].cumsum(), 'ko-', linewidth=2, markersize=4, label='Actual (cumulative)') ax2.plot(bt['year'], bt['payout_1_7'].cumsum(), 'b^-', linewidth=2, markersize=4, label='1/7 rule (cumulative)') ax3 = ax2.twinx() ax3.plot(bt['year'], bt['V0_real'], 'g--', linewidth=1.5, alpha=0.6, label='Real V₀ ($B)') ax3.set_ylabel('Real V₀ ($B)', color='green') ax3.tick_params(axis='y', labelcolor='green') ax2.set_xlabel('Year') ax2.set_ylabel('Cumulative Disbursement ($B)') ax2.set_title('Cumulative Disbursements & Real V₀ Erosion') ax2.legend(loc='upper left', fontsize=9) ax3.legend(loc='center right', fontsize=9) fig.tight_layout() paths += save_figure(fig, 'fair_inheritance_backtest_aggregate', subfolder='applications') # Figure 6: Ford Foundation simulation ford_df = _cache.get('ford') if ford_df is not None and len(ford_df) > 0: fig, axes = setup_subplots(2, 1, width=10, height=10) # Panel A: Ford assets and payouts under different rules ax = axes[0] ax.plot(ford_df['year'], ford_df['ford_assets'], 'ko-', linewidth=2, markersize=4, label='Ford assets ($B)') ax.axhline(4.7, color='red', linestyle='--', linewidth=1.5, alpha=0.7, label='V₀ = $4.7B (1985 nominal)') ax.plot(ford_df['year'], ford_df['ford_V0_real'], 'r:', linewidth=1.5, alpha=0.5, label='V₀ real (2.5% inflation)') ax.fill_between(ford_df['year'], 4.7, ford_df['ford_assets'], where=ford_df['ford_assets'] > 4.7, alpha=0.1, color='blue', label='Excess above V₀') ax.set_xlabel('Year') ax.set_ylabel('Value ($B)') ax.set_title('Ford Foundation: Assets vs. Nominal Cap V₀') ax.legend(fontsize=8) # Panel B: Annual payouts comparison ax2 = axes[1] ax2.plot(ford_df['year'], ford_df['ford_grants'], 'ko-', linewidth=2, markersize=4, label='Actual grants') ax2.plot(ford_df['year'], ford_df['ford_payout_1_7'], 'b^-', linewidth=2, markersize=4, label='1/7 excess rule') ax2.plot(ford_df['year'], ford_df['ford_5pct'], 'g--', linewidth=1.5, alpha=0.6, label='5% floor rule') ax2.set_xlabel('Year') ax2.set_ylabel('Annual Payout ($B)') ax2.set_title('Ford Foundation: Payout Comparison') ax2.legend(fontsize=9) fig.tight_layout() paths += save_figure(fig, 'fair_inheritance_ford_simulation', subfolder='applications') return paths if __name__ == '__main__': results = main() for r in results: print(f"{r['name']}: {r['verdict']} " f"(stat={r['statistic']}, p={r['p_value']})")

events	[[object Object], [object Object], [object Object], [object Object]]
p value	0.6875
n events	4
statistic	0.5000
n increase	2

Fi Foundation Creation Events

Description

Figures

Statistics

Script

▸ Source Code

Reproduction Steps