Update all Quantlets for R3 revision (30 Quantlets)

CFP_Calibration_Efficiency_Frontier/Metainfo.txt

@@ -1,25 +1,17 @@
 Name of Quantlet: CFP_Calibration_Efficiency_Frontier
 
-Published in: 'Distribution-Free Recalibration of Tail Quantile Forecasts under Temporal Dependence'
+Published in: Distribution-Free Recalibration of Tail Quantile Forecasts under Temporal Dependence
 
-Description: Generates the two-panel calibration-efficiency frontier (Figure 3)
-             showing the trade-off between empirical coverage and VaR width for
-             four representative forecasters (Chronos-Small, TimesFM 2.5, Lag-Llama,
-             GJR-GARCH) before and after conformal correction. Panel (a) shows the
-             zoomed view near the 99% target; panel (b) shows the full scale
-             including Chronos-Small at raw coverage ~61%.
+Description: Calibration-efficiency frontier showing raw-to-corrected trajectory for four representative forecasters on the coverage-width plane at alpha = 0.01 (Figure 3). Hollow markers indicate raw forecasts, filled markers indicate corrected, with arrows showing the conformal correction path.
 
-Keywords: calibration efficiency, VaR width, coverage, conformal prediction, frontier, time series foundation models
+Keywords: calibration efficiency, VaR width, coverage, conformal prediction, frontier, TSFM
 
-See also:
-- CO_full_evaluation
+See also: CO_full_evaluation, CO_violation_rates
 
-Author: Daniel Traian Pele, Stefan Lessmann, Wolfgang Karl Haerdle
+Author: Daniel Traian Pele
 
-Submitted: 2026-04-09
+Submitted: 2026-04-25
 
-Input: cfp_ijf_data/paper_outputs/tables/all_results.csv
+Datafile: cfp_ijf_data/paper_outputs/tables/all_results.csv
 
-Output: calibration_efficiency_frontier_v2.pdf, calibration_efficiency_frontier_v2.png
-
-Example: Chronos-Small moves from raw coverage 61.2% (width 0.003) to corrected coverage 98.9% (width 0.041), while TimesFM requires only a modest shift from 98.6% to 99.0%.
+Output: fig_frontier_killer.pdf, fig_frontier_killer.png

CFP_Calibration_Efficiency_Frontier/run_frontier.py

@@ -0,0 +1,142 @@
+"""
+CFP_Calibration_Efficiency_Frontier — run_frontier.py
+=====================================================
+Two-panel calibration-efficiency frontier (Figure 3).
+Left: raw forecasts (full y-range). Right: after conformal correction
+(zoomed). Legend identifies models by unique color+marker.
+
+Input:  cfp_ijf_data/paper_outputs/tables/all_results.csv
+        cfp_ijf_data/paper_outputs/tables/moirai11_full_results.csv
+Output: fig_frontier_killer.pdf/.png
+"""
+
+import pandas as pd
+import numpy as np
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+from pathlib import Path
+import warnings
+warnings.filterwarnings('ignore')
+
+plt.rcParams.update({
+    'font.family': 'sans-serif',
+    'axes.grid': False,
+    'savefig.transparent': True,
+    'axes.spines.top': False,
+    'axes.spines.right': False,
+    'font.size': 12,
+})
+
+SCRIPT_DIR = Path(__file__).resolve().parent
+BASE = SCRIPT_DIR.parent.parent
+DATA = BASE / 'cfp_ijf_data' / 'paper_outputs' / 'tables'
+FIG_DIR = BASE / 'figures'
+SLIDE_DIR = BASE / 'ICFS 2026'
+OUT = SCRIPT_DIR
+
+TSFMS = ['Chronos-Small', 'Chronos-Mini', 'TimesFM-2.5',
+         'Moirai-1.1', 'Moirai-2.0', 'Lag-Llama']
+BENCHMARKS = ['GJR-GARCH', 'GARCH-N', 'Hist-Sim', 'EWMA']
+SHOW = TSFMS + BENCHMARKS
+
+COLORS = {
+    'Chronos-Small': '#8B0000',
+    'Chronos-Mini':  '#C8102E',
+    'TimesFM-2.5':   '#DC143C',
+    'Moirai-1.1':    '#228B22',
+    'Moirai-2.0':    '#E06030',
+    'Lag-Llama':     '#7B2FBE',
+    'GJR-GARCH':     '#003DA5',
+    'GARCH-N':       '#4A90D9',
+    'Hist-Sim':      '#2CA02C',
+    'EWMA':          '#555555',
+}
+
+MARKERS = {
+    'Chronos-Small': 'o',
+    'Chronos-Mini':  's',
+    'TimesFM-2.5':   'D',
+    'Moirai-1.1':    '^',
+    'Moirai-2.0':    'v',
+    'Lag-Llama':     'P',
+    'GJR-GARCH':     'X',
+    'GARCH-N':       'p',
+    'Hist-Sim':      'h',
+    'EWMA':          '*',
+}
+
+DISPLAY = {
+    'Chronos-Small': 'Chronos-S',   'Chronos-Mini': 'Chronos-M',
+    'TimesFM-2.5':   'TimesFM 2.5', 'Moirai-1.1':   'Moirai 1.1',
+    'Moirai-2.0':    'Moirai 2.0',  'Lag-Llama':     'Lag-Llama',
+    'GJR-GARCH':     'GJR-GARCH',   'GARCH-N':       'GARCH-N',
+    'Hist-Sim':      'Hist-Sim',    'EWMA':          'EWMA',
+}
+
+df = pd.read_csv(DATA / 'all_results.csv')
+m11 = pd.read_csv(DATA / 'moirai11_full_results.csv')
+df = pd.concat([df, m11], ignore_index=True)
+d01 = df[df['alpha'] == 0.01].copy()
+
+data = {}
+for model in SHOW:
+    mdf = d01[d01['model'] == model]
+    data[model] = {
+        'raw_width':  abs(mdf['raw_width'].mean()),
+        'raw_cov':    1 - mdf['pihat_raw'].mean(),
+        'corr_width': abs(mdf['VaR_width'].mean()),
+        'corr_cov':   1 - mdf['pihat_cp'].mean(),
+    }
+
+fig, (ax_raw, ax_corr) = plt.subplots(1, 2, figsize=(13, 5.5), sharey=True,
+                                       gridspec_kw={'wspace': 0.08})
+
+handles = []
+for name in SHOW:
+    d = data[name]
+    c = COLORS[name]
+    m = MARKERS[name]
+    ms = 12 if m == '*' else 10
+    ax_raw.plot(d['raw_width'], d['raw_cov'], m, color=c, ms=ms,
+                mew=0.5, mec='black', zorder=5, clip_on=False)
+    ax_corr.plot(d['corr_width'], d['corr_cov'], m, color=c, ms=ms,
+                 mew=0.5, mec='black', zorder=5, clip_on=False)
+    h = plt.Line2D([0], [0], marker=m, color=c, ms=ms, mew=0.5,
+                   mec='black', ls='', label=DISPLAY[name])
+    handles.append(h)
+
+for ax in [ax_raw, ax_corr]:
+    ax.axhline(y=0.99, color='grey', ls='--', lw=1.5, zorder=1)
+    ax.set_xlabel(r'Mean VaR width', fontsize=13)
+    ax.tick_params(axis='both', labelsize=12)
+
+ax_raw.set_title('Raw Forecasts', fontsize=14, fontweight='bold', pad=10)
+ax_raw.set_ylabel(r'Empirical coverage $1 - \hat{\pi}$', fontsize=13)
+ax_raw.set_ylim(0.0, 1.08)
+ax_raw.set_xlim(-0.002, 0.052)
+ax_raw.text(0.048, 0.93, '99% target', fontsize=10, color='grey',
+            style='italic', ha='right', va='top')
+
+ax_corr.set_title('After Conformal Correction', fontsize=14,
+                   fontweight='bold', pad=10)
+ax_corr.set_xlim(0.033, 0.074)
+ax_corr.text(0.072, 0.93, '99% target', fontsize=10, color='grey',
+             style='italic', ha='right', va='top')
+
+plt.tight_layout()
+fig.subplots_adjust(bottom=0.22)
+fig.legend(handles=handles, loc='lower center',
+           bbox_to_anchor=(0.5, 0.0), ncol=5, fontsize=11,
+           frameon=False, handletextpad=0.3, columnspacing=1.2)
+FIG_DIR.mkdir(exist_ok=True)
+for ext in ['pdf', 'png']:
+    fig.savefig(OUT / f'fig_frontier_killer.{ext}',
+                dpi=300, bbox_inches='tight', pad_inches=0.05)
+    fig.savefig(FIG_DIR / f'fig_frontier_killer.{ext}',
+                dpi=300, bbox_inches='tight', pad_inches=0.05)
+    fig.savefig(SLIDE_DIR / f'fig_frontier_killer.{ext}',
+                dpi=300, bbox_inches='tight', pad_inches=0.05)
+
+plt.close(fig)
+print("Saved: fig_frontier_killer.pdf/.png")