수정본

backend/app/services/strategy_optimizer.py

@@ -47,6 +47,11 @@ class OptimizationResult:
     applied: bool
     apply_reason: str
     total_combinations: int
+    # IS/OOS 검증 결과 (v2 추가)
+    oos_sharpe: float = 0.0
+    oos_win_rate: float = 0.0
+    oos_mdd: float = 0.0
+    is_oos_sharpe_ratio: float = 0.0  # OOS Sharpe / IS Sharpe (1.0에 가까울수록 건전)
 
 
 # ─────────────────────────────────────────────────────────────────────────────
@@ -282,7 +287,7 @@ async def run_optimization(
     Returns:
         OptimizationResult
     """
-    logger.info(f"🔧 전략 최적화 시작 (trigger={trigger}, period={period_days}일)")
+    logger.info(f"전략 최적화 시작 (trigger={trigger}, period={period_days}일)")
 
     # 1. 현재 파라미터 저장 (롤백용)
     prev_params = await _get_current_params()
@@ -310,16 +315,26 @@ async def run_optimization(
             total_combinations=0,
         )
 
+    # ── IS/OOS 기간 분할 (70% IS / 30% OOS, 날짜 기준) ──
     end_date = date.today()
     start_date = end_date - timedelta(days=period_days)
+    is_days = int(period_days * 0.7)
+    is_end_date = start_date + timedelta(days=is_days)
+    oos_start_date = is_end_date + timedelta(days=1)
+    oos_end_date = end_date
 
-    # 4. 현재 파라미터 기준 Sharpe 계산
-    cur_score, cur_sharpe, cur_wr, cur_mdd = await _evaluate_params(
-        prev_params, stock_codes, start_date, end_date, initial_capital
+    logger.info(
+        f"   IS 기간: {start_date} ~ {is_end_date} ({is_days}일), "
+        f"OOS 기간: {oos_start_date} ~ {oos_end_date} ({period_days - is_days}일)"
     )
-    logger.info(f"   현재 성과 → Sharpe: {cur_sharpe:.3f}, 승률: {cur_wr:.1f}%, MDD: {cur_mdd:.1f}%")
 
-    # 5. 그리드 서치 (배치 5개씩 병렬)
+    # 4. 현재 파라미터 기준 Sharpe 계산 (IS 기간)
+    cur_score, cur_sharpe, cur_wr, cur_mdd = await _evaluate_params(
+        prev_params, stock_codes, start_date, is_end_date, initial_capital
+    )
+    logger.info(f"   현재 IS 성과 -> Sharpe: {cur_sharpe:.3f}, 승률: {cur_wr:.1f}%, MDD: {cur_mdd:.1f}%")
+
+    # 5. 그리드 서치 — IS 기간에서 최적 파라미터 탐색 (배치 5개씩 병렬)
     best_score = cur_score
     best_params = dict(prev_params)
     best_sharpe = cur_sharpe
@@ -330,7 +345,7 @@ async def run_optimization(
     for i in range(0, len(combinations), batch_size):
         batch = combinations[i:i + batch_size]
         tasks = [
-            _evaluate_params(p, stock_codes, start_date, end_date, initial_capital)
+            _evaluate_params(p, stock_codes, start_date, is_end_date, initial_capital)
             for p in batch
         ]
         results = await asyncio.gather(*tasks, return_exceptions=True)
@@ -345,39 +360,65 @@ async def run_optimization(
                 best_mdd = mdd
                 best_params = {**prev_params, **combo}
 
-    logger.info(f"   최적 성과 → Sharpe: {best_sharpe:.3f}, 승률: {best_win_rate:.1f}%, MDD: {best_mdd:.1f}%")
+    logger.info(f"   IS 최적 성과 -> Sharpe: {best_sharpe:.3f}, 승률: {best_win_rate:.1f}%, MDD: {best_mdd:.1f}%")
 
-    # 6. 자동 적용 여부 결정 (Sharpe 10% 이상 개선)
+    # 6. OOS 기간에서 최적 파라미터 검증 (과적합 탐지)
+    oos_score, oos_sharpe, oos_wr, oos_mdd = await _evaluate_params(
+        best_params, stock_codes, oos_start_date, oos_end_date, initial_capital
+    )
+    is_oos_ratio = (oos_sharpe / best_sharpe) if best_sharpe > 0 else 0.0
+    logger.info(
+        f"   OOS 검증 -> Sharpe: {oos_sharpe:.3f}, 승률: {oos_wr:.1f}%, MDD: {oos_mdd:.1f}%, "
+        f"IS/OOS Sharpe 비율: {is_oos_ratio:.2f}"
+    )
+
+    if is_oos_ratio < 0.5 and best_sharpe > 0:
+        logger.warning(
+            f"   [과적합 경고] OOS Sharpe / IS Sharpe = {is_oos_ratio:.2f} < 0.5 "
+            "-- 파라미터가 IS 기간에 과적합되었을 가능성 높음"
+        )
+
+    # 7. 자동 적용 여부 결정
+    # 조건: IS Sharpe 10% 이상 개선 AND OOS Sharpe > 0 (OOS에서도 양의 성과)
     improvement_threshold = 1.1
-    should_apply = (
+    _is_improved = (
         cur_sharpe > 0
         and best_sharpe >= cur_sharpe * improvement_threshold
         and best_params != prev_params
     )
     if cur_sharpe <= 0 and best_sharpe > 0:
-        should_apply = True
+        _is_improved = True
+
+    # OOS 검증 게이트: OOS Sharpe > 0 이어야 실전 적용 허가
+    _oos_passed = oos_sharpe > 0
+    _should_apply = _is_improved and _oos_passed
 
     applied = False
     apply_reason = ""
 
-    if should_apply:
+    if _should_apply:
         improvement_pct = ((best_sharpe - cur_sharpe) / max(abs(cur_sharpe), 0.001)) * 100
-        apply_reason = f"Sharpe {improvement_pct:.1f}% 향상 ({cur_sharpe:.3f} → {best_sharpe:.3f})"
+        apply_reason = (
+            f"IS Sharpe {improvement_pct:.1f}% 향상 ({cur_sharpe:.3f} -> {best_sharpe:.3f}), "
+            f"OOS Sharpe: {oos_sharpe:.3f}, IS/OOS 비율: {is_oos_ratio:.2f}"
+        )
         # 최적화된 파라미터만 적용 (prev와 다른 키만)
         changed = {k: v for k, v in best_params.items() if str(v) != str(prev_params.get(k, ""))}
         if changed:
             applied = await apply_params(changed, reason="auto_optimizer")
             if applied:
-                logger.info(f"✅ 자동 적용 완료: {apply_reason}")
+                logger.info(f"자동 적용 완료: {apply_reason}")
     else:
         if cur_sharpe <= 0 and best_sharpe <= 0:
             apply_reason = "insufficient_data"
+        elif not _oos_passed and _is_improved:
+            apply_reason = f"OOS 검증 실패 (OOS Sharpe: {oos_sharpe:.3f} <= 0, 과적합 의심)"
         elif best_sharpe < cur_sharpe * improvement_threshold:
             pct = ((best_sharpe - cur_sharpe) / max(abs(cur_sharpe), 0.001)) * 100
-            apply_reason = f"개선 미달 ({pct:+.1f}% < 10%)"
+            apply_reason = f"IS 개선 미달 ({pct:+.1f}% < 10%)"
         else:
             apply_reason = "no_improvement"
-        logger.info(f"⏭️  자동 적용 건너뜀: {apply_reason}")
+        logger.info(f"자동 적용 건너뜀: {apply_reason}")
 
     return OptimizationResult(
         best_params=best_params,
@@ -390,6 +431,10 @@ async def run_optimization(
         applied=applied,
         apply_reason=apply_reason,
         total_combinations=total_combos,
+        oos_sharpe=oos_sharpe,
+        oos_win_rate=oos_wr,
+        oos_mdd=oos_mdd,
+        is_oos_sharpe_ratio=round(is_oos_ratio, 4),
     )