From 2b94cc802d059320147f2c34da48f64bbc5b317b Mon Sep 17 00:00:00 2001
From: Variet Agent <agent@variet.net>
Date: Wed, 11 Mar 2026 16:00:07 +0900
Subject: [PATCH] feat(pipeline): update transition_matrices and config for 7x7
 Zt estimation

---
 config.yaml                 |  4 ++--
 data/transition_matrices.py | 25 +++++++++++++++++++------
 2 files changed, 21 insertions(+), 8 deletions(-)

diff --git a/config.yaml b/config.yaml
index 125a330..843c981 100644
--- a/config.yaml
+++ b/config.yaml
@@ -18,14 +18,14 @@ ecos:
 # 전이행렬 데이터 소스
 data:
   transition_source: "real"   # "real" (3사 실제) | "builtin" (내장 샘플)
-  transition_dir: null        # null이면 기본 data/real/
+  transition_dir: null        # null이면 기본 data/real_v2/
 
 # 모형 파라미터
 model:
   # 자산상관계수 (Basel IRB 기준 0.12~0.24, 기업 평균 ~0.20)
   rho: 0.20
   # 신용등급 체계 (한국 3사 공통)
-  rating_grades: ["AAA", "AA", "A", "BBB", "BB", "B", "CCC", "D"]
+  rating_grades: ["AAA", "AA", "A", "BBB", "BB", "B", "D"]  # 7x7 (CCC제외, Zt추정용)
 
 # 시나리오 설정
 scenarios:
diff --git a/data/transition_matrices.py b/data/transition_matrices.py
index ab1f255..c6a7b3e 100644
--- a/data/transition_matrices.py
+++ b/data/transition_matrices.py
@@ -15,8 +15,12 @@ from pathlib import Path
 from typing import Dict, List, Optional, Tuple
 
 
-# 등급 레이블
-RATING_GRADES = ["AAA", "AA", "A", "BBB", "BB", "B", "CCC", "D"]
+# Rating grade labels
+RATING_GRADES_8 = ["AAA", "AA", "A", "BBB", "BB", "B", "CCC", "D"]
+RATING_GRADES_7 = ["AAA", "AA", "A", "BBB", "BB", "B", "D"]
+
+# Default: 7x7 (CCC excluded for Zt estimation)
+RATING_GRADES = RATING_GRADES_7
 N_GRADES = len(RATING_GRADES)
 
 
@@ -205,7 +209,7 @@ def _load_real_matrices(data_dir: Optional[str] = None) -> Dict[int, np.ndarray]
     parse_pdf_matrices.py 로 생성된 3사 평균 CSV 사용.
     """
     if data_dir is None:
-        data_dir = str(Path(__file__).parent / "real")
+        data_dir = str(Path(__file__).parent / "real_v2")
     real_dir = Path(data_dir)
     
     if not real_dir.exists():
@@ -320,7 +324,11 @@ def get_default_rates(matrices: Dict[int, np.ndarray]) -> pd.DataFrame:
         index=연도, columns=등급, values=연간 PD
     """
     years = sorted(matrices.keys())
-    grades = RATING_GRADES[:-1]  # D 제외
+    n = list(matrices.values())[0].shape[0]
+    if n == 7:
+        grades = RATING_GRADES_7[:-1]  # AAA~B (D 제외)
+    else:
+        grades = RATING_GRADES_8[:-1]  # AAA~CCC (D 제외)
     
     data = {}
     for year in years:
@@ -332,10 +340,15 @@ def get_default_rates(matrices: Dict[int, np.ndarray]) -> pd.DataFrame:
 
 def display_matrix(tm: np.ndarray, title: str = "전이행렬") -> str:
     """전이행렬을 보기 좋게 포매팅"""
+    n = tm.shape[0]
+    if n == 7:
+        grades = RATING_GRADES_7
+    else:
+        grades = RATING_GRADES_8
     df = pd.DataFrame(
         tm,
-        index=RATING_GRADES,
-        columns=RATING_GRADES
+        index=grades,
+        columns=grades
     )
     # 백분율 표시
     df_pct = df * 100