fix: generate_report now uses actual position_weights from daily_records
Previously hardcoded equal weight (1/select_num), ignoring config weight type. Now reads position_weights from last daily_record, correctly showing rank-based weights.
This commit is contained in:
128
rotation/experiments/debug_clf_2022.py
Normal file
128
rotation/experiments/debug_clf_2022.py
Normal file
@@ -0,0 +1,128 @@
|
||||
"""分析 2022年4月底~5月初 CL=F 入选原因"""
|
||||
import os, sys, math
|
||||
from pathlib import Path
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
PROJECT_ROOT = Path(__file__).parent.parent.parent
|
||||
sys.path.insert(0, str(PROJECT_ROOT))
|
||||
|
||||
from rotation.simple_rotation import SimpleRotationStrategy, slope_r2_score
|
||||
|
||||
if 'FLASK_API_URL' not in os.environ:
|
||||
os.environ['FLASK_API_URL'] = 'https://k3s.tokenpluse.xyz'
|
||||
|
||||
strategy = SimpleRotationStrategy()
|
||||
strategy._preload_data()
|
||||
|
||||
# 分析日期范围
|
||||
start = pd.Timestamp('2022-04-15')
|
||||
end = pd.Timestamp('2022-05-10')
|
||||
n_days = strategy.config.factor.n_days # 25
|
||||
|
||||
print("=" * 80)
|
||||
print(f" 分析 CL=F 动量信号 ({start.date()} ~ {end.date()})")
|
||||
print(f" 窗口长度: {n_days} 天")
|
||||
print("=" * 80)
|
||||
|
||||
# 获取所有 signal_codes 的 score
|
||||
signal_codes = strategy.signal_codes
|
||||
date_range = pd.bdate_range(start, end)
|
||||
|
||||
for date in date_range:
|
||||
scores = {}
|
||||
for code in signal_codes:
|
||||
if code not in strategy.index_data:
|
||||
continue
|
||||
df = strategy.index_data[code]
|
||||
mask = df.index <= date
|
||||
recent = df.loc[mask]
|
||||
if len(recent) < n_days:
|
||||
continue
|
||||
prices = recent['close'].values[-n_days:]
|
||||
score = slope_r2_score(prices)
|
||||
scores[code] = (score, prices[-1])
|
||||
|
||||
if not scores:
|
||||
continue
|
||||
|
||||
# 排序
|
||||
ranked = sorted(scores.items(), key=lambda x: x[1][0], reverse=True)
|
||||
cl_rank = None
|
||||
for i, (code, (score, price)) in enumerate(ranked):
|
||||
if code == 'CL=F':
|
||||
cl_rank = i + 1
|
||||
break
|
||||
|
||||
cl_score = scores.get('CL=F', (None, None))[0]
|
||||
cl_price = scores.get('CL=F', (None, None))[1]
|
||||
|
||||
print(f"\n{date.strftime('%Y-%m-%d')} | CL=F score={cl_score:.4f}, price={cl_price:.2f}, rank={cl_rank}/{len(ranked)}")
|
||||
print(f" Top 5:")
|
||||
for i, (code, (score, price)) in enumerate(ranked[:5]):
|
||||
marker = " <<<" if code == 'CL=F' else ""
|
||||
print(f" #{i+1} {code:<15} score={score:>10.4f} price={price:.2f}{marker}")
|
||||
|
||||
# 详细分析 CL=F 价格走势
|
||||
print(f"\n{'='*80}")
|
||||
print(f" CL=F 价格走势 (2022年3月~5月)")
|
||||
print(f"{'='*80}")
|
||||
|
||||
df_cl = strategy.index_data['CL=F']
|
||||
mask = (df_cl.index >= '2022-03-01') & (df_cl.index <= '2022-05-15')
|
||||
cl_prices = df_cl.loc[mask, 'close']
|
||||
|
||||
for date, price in cl_prices.items():
|
||||
# 计算25天窗口的score
|
||||
mask2 = df_cl.index <= date
|
||||
recent = df_cl.loc[mask2]
|
||||
if len(recent) < n_days:
|
||||
continue
|
||||
prices = recent['close'].values[-n_days:]
|
||||
score = slope_r2_score(prices)
|
||||
normalized = prices / prices[0]
|
||||
slope, intercept = np.polyfit(np.arange(len(normalized)), normalized, 1)
|
||||
y_pred = slope * np.arange(len(normalized)) + intercept
|
||||
ss_res = np.sum((normalized - y_pred) ** 2)
|
||||
ss_tot = np.sum((normalized - np.mean(normalized)) ** 2)
|
||||
r2 = 1 - ss_res / ss_tot if ss_tot > 0 else 0
|
||||
|
||||
flag = ""
|
||||
if date.strftime('%Y-%m-%d') in ('2022-04-29', '2022-05-05'):
|
||||
flag = " <<< 入选日"
|
||||
|
||||
print(f" {date.strftime('%Y-%m-%d')} price={price:>8.2f} score={score:>10.4f} "
|
||||
f"slope={slope:>8.5f} R²={r2:.4f}{flag}")
|
||||
|
||||
# 分析 CL=F 的组内竞争
|
||||
print(f"\n{'='*80}")
|
||||
print(f" CL=F 所在组: 查看组内竞争")
|
||||
print(f"{'='*80}")
|
||||
|
||||
groups = strategy.config.asset_pools.by_group
|
||||
for group_name, assets in groups.items():
|
||||
group_codes = [a.signal_source for a in assets.values()]
|
||||
if 'CL=F' in group_codes:
|
||||
print(f" 组名: {group_name}")
|
||||
print(f" 组成员: {group_codes}")
|
||||
|
||||
# 4/29 和 5/5 的组内得分
|
||||
for target_date_str in ['2022-04-29', '2022-05-05']:
|
||||
target_date = pd.Timestamp(target_date_str)
|
||||
print(f"\n {target_date_str} 组内得分:")
|
||||
for code in group_codes:
|
||||
if code not in strategy.index_data:
|
||||
continue
|
||||
df = strategy.index_data[code]
|
||||
mask = df.index <= target_date
|
||||
recent = df.loc[mask]
|
||||
if len(recent) < n_days:
|
||||
print(f" {code:<15} 数据不足")
|
||||
continue
|
||||
prices = recent['close'].values[-n_days:]
|
||||
score = slope_r2_score(prices)
|
||||
marker = " <<< TOP1" if score == max(
|
||||
slope_r2_score(strategy.index_data[c].loc[strategy.index_data[c].index <= target_date]['close'].values[-n_days:])
|
||||
for c in group_codes if c in strategy.index_data and len(strategy.index_data[c].loc[strategy.index_data[c].index <= target_date]) >= n_days
|
||||
) else ""
|
||||
print(f" {code:<15} score={score:>10.4f}{marker}")
|
||||
175
rotation/experiments/r2_distribution.py
Normal file
175
rotation/experiments/r2_distribution.py
Normal file
@@ -0,0 +1,175 @@
|
||||
"""分析短债(931862.CSI)和CL=F的R²分布"""
|
||||
import os, sys
|
||||
from pathlib import Path
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
PROJECT_ROOT = Path(__file__).parent.parent.parent
|
||||
sys.path.insert(0, str(PROJECT_ROOT))
|
||||
|
||||
from rotation.simple_rotation import SimpleRotationStrategy
|
||||
|
||||
if 'FLASK_API_URL' not in os.environ:
|
||||
os.environ['FLASK_API_URL'] = 'https://k3s.tokenpluse.xyz'
|
||||
|
||||
strategy = SimpleRotationStrategy()
|
||||
strategy._preload_data()
|
||||
|
||||
n_days = strategy.config.factor.n_days # 25
|
||||
|
||||
def compute_r2(prices):
|
||||
prices = np.clip(prices, 0.01, None)
|
||||
y = prices / prices[0]
|
||||
x = np.arange(len(y))
|
||||
slope, intercept = np.polyfit(x, y, 1)
|
||||
y_pred = slope * x + intercept
|
||||
ss_res = np.sum((y - y_pred) ** 2)
|
||||
ss_tot = np.sum((y - np.mean(y)) ** 2)
|
||||
r2 = 1 - ss_res / ss_tot if ss_tot > 0 else 0
|
||||
return r2, slope
|
||||
|
||||
# 分析所有 signal_codes
|
||||
print("=" * 80)
|
||||
print(f" 各资产 R² 分布统计 (2020-01-10 ~ 2026-06-05, 窗口={n_days}天)")
|
||||
print("=" * 80)
|
||||
|
||||
all_r2_data = {}
|
||||
|
||||
for code in strategy.signal_codes:
|
||||
if code not in strategy.index_data:
|
||||
continue
|
||||
df = strategy.index_data[code]
|
||||
r2_list = []
|
||||
slope_list = []
|
||||
for i in range(n_days, len(df)):
|
||||
prices = df['close'].values[i-n_days:i]
|
||||
r2, slope = compute_r2(prices)
|
||||
r2_list.append(r2)
|
||||
slope_list.append(slope)
|
||||
|
||||
r2_arr = np.array(r2_list)
|
||||
slope_arr = np.array(slope_list)
|
||||
all_r2_data[code] = (r2_arr, slope_arr)
|
||||
|
||||
# 区分正slope和负slope时的R²
|
||||
pos_mask = slope_arr > 0
|
||||
neg_mask = slope_arr <= 0
|
||||
|
||||
print(f"\n {code} (n={len(r2_arr)})")
|
||||
print(f" 全部 R²: mean={r2_arr.mean():.4f} median={np.median(r2_arr):.4f} "
|
||||
f"p10={np.percentile(r2_arr, 10):.4f} p25={np.percentile(r2_arr, 25):.4f} "
|
||||
f"p75={np.percentile(r2_arr, 75):.4f} p90={np.percentile(r2_arr, 90):.4f}")
|
||||
print(f" slope>0: mean={r2_arr[pos_mask].mean():.4f} "
|
||||
f"p10={np.percentile(r2_arr[pos_mask], 10):.4f} "
|
||||
f"p25={np.percentile(r2_arr[pos_mask], 25):.4f} "
|
||||
f"count={pos_mask.sum()}")
|
||||
print(f" slope<=0: mean={r2_arr[neg_mask].mean():.4f} "
|
||||
f"p10={np.percentile(r2_arr[neg_mask], 10):.4f} "
|
||||
f"p25={np.percentile(r2_arr[neg_mask], 25):.4f} "
|
||||
f"count={neg_mask.sum()}")
|
||||
|
||||
# 重点对比：短债 vs CL=F
|
||||
print(f"\n{'=' * 80}")
|
||||
print(f" 重点对比: 931862.CSI (短债) vs CL=F (原油)")
|
||||
print(f"{'=' * 80}")
|
||||
|
||||
for code in ['931862.CSI', 'CL=F']:
|
||||
r2_arr, slope_arr = all_r2_data[code]
|
||||
pos_mask = slope_arr > 0
|
||||
r2_pos = r2_arr[pos_mask]
|
||||
|
||||
print(f"\n {code} (正趋势 R², n={len(r2_pos)})")
|
||||
# 分位数
|
||||
for p in [1, 5, 10, 15, 20, 25, 30, 50, 75, 90, 95, 99]:
|
||||
val = np.percentile(r2_pos, p)
|
||||
print(f" p{p:>2}: R²={val:.4f}")
|
||||
|
||||
# CL=F 入选时的 R² 统计
|
||||
print(f"\n{'=' * 80}")
|
||||
print(f" CL=F 入选持仓时的 R² (从 detail JSON 中提取)")
|
||||
print(f"{'=' * 80}")
|
||||
|
||||
import json
|
||||
detail_path = PROJECT_ROOT / "rotation" / "results" / "simple_rotation_detail.json"
|
||||
with open(detail_path) as f:
|
||||
detail = json.load(f)
|
||||
|
||||
clf_held_r2 = []
|
||||
for day in detail['days']:
|
||||
assets = day.get('assets', {})
|
||||
clf = assets.get('CL=F', {})
|
||||
if clf.get('is_held'):
|
||||
date = day['date']
|
||||
mom = clf.get('momentum', 0)
|
||||
rank = clf.get('rank')
|
||||
# 需要重新算这个日期的R² (用T-1数据)
|
||||
clf_held_r2.append((date, mom, rank))
|
||||
|
||||
# 取前20个和后20个持仓日
|
||||
print(f" CL=F 总持仓天数: {len(clf_held_r2)}")
|
||||
print(f"\n 入场日(added)的动量和R²:")
|
||||
|
||||
entry_days = []
|
||||
for day in detail['days']:
|
||||
if 'CL=F' in day.get('added', []):
|
||||
date = day['date']
|
||||
clf = day['assets']['CL=F']
|
||||
entry_days.append((date, clf['momentum'], clf.get('rank')))
|
||||
# 计算信号日(T-1)的R²
|
||||
sig_date = pd.Timestamp(date) - pd.Timedelta(days=1)
|
||||
df = strategy.index_data['CL=F']
|
||||
mask = df.index <= sig_date
|
||||
recent = df.loc[mask]
|
||||
if len(recent) >= n_days:
|
||||
prices = recent['close'].values[-n_days:]
|
||||
r2, slope = compute_r2(prices)
|
||||
print(f" {date}: momentum={clf['momentum']:>10.4f} R²={r2:.4f} slope={slope:.5f} "
|
||||
f"entry_price={clf.get('entry_price_idx', '?')}")
|
||||
|
||||
print(f"\n 离场日(removed)的前一天动量和R²:")
|
||||
exit_days = []
|
||||
for day in detail['days']:
|
||||
if 'CL=F' in day.get('removed', []):
|
||||
date = day['date']
|
||||
# 信号日是T-1
|
||||
sig_date = pd.Timestamp(date) - pd.Timedelta(days=1)
|
||||
df = strategy.index_data['CL=F']
|
||||
mask = df.index <= sig_date
|
||||
recent = df.loc[mask]
|
||||
if len(recent) >= n_days:
|
||||
prices = recent['close'].values[-n_days:]
|
||||
r2, slope = compute_r2(prices)
|
||||
else:
|
||||
r2, slope = 0, 0
|
||||
# 找前一天的momentum
|
||||
clf = day['assets'].get('CL=F', {})
|
||||
print(f" {date}: 信号日R²={r2:.4f} slope={slope:.5f} "
|
||||
f"momentum(当日)={clf.get('momentum', '?')} "
|
||||
f"price={clf.get('index_close', '?')}")
|
||||
|
||||
# 不同R²阈值下的回测影响估算
|
||||
print(f"\n{'=' * 80}")
|
||||
print(f" R² 阈值对 CL=F 信号的过滤效果")
|
||||
print(f"{'=' * 80}")
|
||||
|
||||
for threshold in [0.05, 0.10, 0.15, 0.20, 0.25, 0.30]:
|
||||
filtered_entries = []
|
||||
for day in detail['days']:
|
||||
if 'CL=F' in day.get('added', []):
|
||||
date = day['date']
|
||||
sig_date = pd.Timestamp(date) - pd.Timedelta(days=1)
|
||||
df = strategy.index_data['CL=F']
|
||||
mask = df.index <= sig_date
|
||||
recent = df.loc[mask]
|
||||
if len(recent) >= n_days:
|
||||
prices = recent['close'].values[-n_days:]
|
||||
r2, slope = compute_r2(prices)
|
||||
if r2 < threshold:
|
||||
clf = day['assets']['CL=F']
|
||||
# 查这个入场最终盈亏
|
||||
entry_price = clf.get('entry_price_etf')
|
||||
filtered_entries.append((date, r2, slope, clf['momentum']))
|
||||
|
||||
print(f" R²<{threshold:.2f} 过滤掉的入场: {len(filtered_entries)} 次")
|
||||
for date, r2, slope, mom in filtered_entries:
|
||||
print(f" {date}: R²={r2:.4f} slope={slope:.5f} momentum={mom:.4f}")
|
||||
@@ -1203,7 +1203,7 @@ class SimpleRotationStrategy:
|
||||
|
||||
# Build positions info for table
|
||||
# Sort holdings by momentum score descending
|
||||
weight = 1.0 / self.select_num if self.select_num > 0 else 1.0
|
||||
position_weights = last_rec.get('position_weights', {})
|
||||
sorted_holdings = sorted(holdings, key=lambda c: factors.get(c, 0) or 0, reverse=True)
|
||||
|
||||
# Determine previous holdings to distinguish "调入" vs "维持"
|
||||
@@ -1249,7 +1249,7 @@ class SimpleRotationStrategy:
|
||||
|
||||
positions_info.append({
|
||||
'name': name, 'code': code, 'etf': etf_code,
|
||||
'weight': weight, 'score': score,
|
||||
'weight': position_weights.get(code, 1.0 / len(holdings)), 'score': score,
|
||||
'idx_close': idx_close, 'etf_close': etf_close,
|
||||
'premium': premium, 'action': action,
|
||||
'entry_date': entry_date, 'entry_price': entry_price,
|
||||
@@ -1276,7 +1276,7 @@ class SimpleRotationStrategy:
|
||||
premium = self._get_latest_premium(trade_code, last_date)
|
||||
exit_positions.append({
|
||||
'name': name, 'code': code, 'etf': etf_code,
|
||||
'weight': weight, 'score': None,
|
||||
'weight': 0, 'score': None,
|
||||
'idx_close': idx_close, 'etf_close': etf_close,
|
||||
'premium': premium, 'action': '调出',
|
||||
'entry_date': None, 'entry_price': None,
|
||||
|
||||
162
scripts/diag_ic_calibration.py
Normal file
162
scripts/diag_ic_calibration.py
Normal file
@@ -0,0 +1,162 @@
|
||||
"""IC Calibration diagnostic analysis script"""
|
||||
import os, sys, numpy as np, pandas as pd
|
||||
from pathlib import Path
|
||||
from collections import Counter
|
||||
os.environ['FLASK_API_URL'] = 'https://k3s.tokenpluse.xyz'
|
||||
sys.path.insert(0, str(Path.cwd()))
|
||||
from rotation.simple_rotation import SimpleRotationStrategy
|
||||
from rotation.ic_calibration import ICCalibrator
|
||||
|
||||
def run_and_analyze(select_num, ic_enabled):
|
||||
s = SimpleRotationStrategy()
|
||||
s.config.rotation.select_num = select_num
|
||||
s.select_num = select_num
|
||||
if ic_enabled:
|
||||
s.config.rotation.ic_calibration.enabled = True
|
||||
s.ic_calibrator = ICCalibrator(fwd_days=5, min_samples=120, ic_method='pearson')
|
||||
r = s.run()
|
||||
return s, s.daily_records
|
||||
|
||||
print("=" * 90)
|
||||
print(" Running 4 experiments...")
|
||||
print("=" * 90)
|
||||
|
||||
s1, r1 = run_and_analyze(3, False)
|
||||
s2, r2 = run_and_analyze(3, True)
|
||||
s3, r3 = run_and_analyze(1, False)
|
||||
s4, r4 = run_and_analyze(1, True)
|
||||
|
||||
# === 1. 持仓频率分析 ===
|
||||
print("\n\n" + "=" * 90)
|
||||
print(" 1. 持仓频率对比")
|
||||
print("=" * 90)
|
||||
|
||||
for label, recs, s_obj in [
|
||||
("原始 select=3", r1, s1),
|
||||
("IC校准 select=3", r2, s2),
|
||||
("原始 select=1", r3, s3),
|
||||
("IC校准 select=1", r4, s4),
|
||||
]:
|
||||
hold_counter = Counter()
|
||||
for rec in recs:
|
||||
for h in rec['holdings']:
|
||||
nm = s_obj.config.asset_pools.assets.get(h)
|
||||
name = nm.name if nm else h
|
||||
hold_counter[name] += 1
|
||||
|
||||
print(f"\n [{label}] 持仓天数分布:")
|
||||
for name, cnt in hold_counter.most_common():
|
||||
pct = cnt / len(recs) * 100
|
||||
print(f" {name:<12}: {cnt:>5} 天 ({pct:>5.1f}%)")
|
||||
|
||||
# === 2. 关键资产选中变化 ===
|
||||
print("\n\n" + "=" * 90)
|
||||
print(" 2. IC校准 select=3: 关键资产被选中变化")
|
||||
print("=" * 90)
|
||||
|
||||
targets = ['日经225', '德国DAX', '纳指100', '恒生指数', '黄金', '有色金属', '中证红利低波', '原油', '创业板指']
|
||||
for name_key in targets:
|
||||
orig = sum(1 for r in r1 if any(
|
||||
name_key == (s1.config.asset_pools.assets.get(h).name if s1.config.asset_pools.assets.get(h) else h)
|
||||
for h in r['holdings']))
|
||||
calib = sum(1 for r in r2 if any(
|
||||
name_key == (s2.config.asset_pools.assets.get(h).name if s2.config.asset_pools.assets.get(h) else h)
|
||||
for h in r['holdings']))
|
||||
print(f" {name_key:<12}: 原始={orig:>5}天 -> IC校准={calib:>5}天 (差={calib-orig:+d})")
|
||||
|
||||
# === 3. 收益质量分析 ===
|
||||
print("\n\n" + "=" * 90)
|
||||
print(" 3. 持有期收益质量")
|
||||
print("=" * 90)
|
||||
|
||||
for label, recs in [
|
||||
("原始 select=3", r1), ("IC校准 select=3", r2),
|
||||
("原始 select=1", r3), ("IC校准 select=1", r4),
|
||||
]:
|
||||
arr = np.array([r['daily_return'] for r in recs])
|
||||
win = arr[arr > 0]
|
||||
lose = arr[arr < 0]
|
||||
print(f"\n [{label}]:")
|
||||
print(f" 日均: {arr.mean()*100:+.4f}% 胜率: {(arr>0).mean()*100:.1f}%")
|
||||
print(f" 盈利均: {win.mean()*100:+.4f}% 亏损均: {lose.mean()*100:+.4f}% 盈亏比: {abs(win.mean()/lose.mean()):.3f}")
|
||||
|
||||
# Worst 5 days
|
||||
sorted_rets = sorted(recs, key=lambda x: x['daily_return'])
|
||||
worst5 = [r['daily_return'] * 100 for r in sorted_rets[:5]]
|
||||
best5 = [r['daily_return'] * 100 for r in sorted_rets[-5:]]
|
||||
print(f" 最差5天: {['%.2f%%' % v for v in worst5]}")
|
||||
print(f" 最好5天: {['%.2f%%' % v for v in best5]}")
|
||||
|
||||
# === 4. 分段绩效 ===
|
||||
print("\n\n" + "=" * 90)
|
||||
print(" 4. 分段绩效: 冷启动期(前170天) vs 校准生效期")
|
||||
print("=" * 90)
|
||||
|
||||
for label, recs in [
|
||||
("原始 select=1", r3), ("IC校准 select=1", r4),
|
||||
("原始 select=3", r1), ("IC校准 select=3", r2),
|
||||
]:
|
||||
cold = recs[:170]
|
||||
warm = recs[170:]
|
||||
|
||||
cold_rets = np.array([r['daily_return'] for r in cold])
|
||||
warm_rets = np.array([r['daily_return'] for r in warm])
|
||||
|
||||
cold_cum = np.prod(1 + cold_rets) - 1
|
||||
warm_cum = np.prod(1 + warm_rets) - 1
|
||||
cold_ann = (1 + cold_cum) ** (252 / len(cold_rets)) - 1
|
||||
warm_ann = (1 + warm_cum) ** (252 / len(warm_rets)) - 1
|
||||
|
||||
warm_nav = np.cumprod(1 + warm_rets)
|
||||
warm_peak = np.maximum.accumulate(warm_nav)
|
||||
warm_dd = (warm_nav - warm_peak) / warm_peak
|
||||
warm_maxdd = warm_dd.min()
|
||||
|
||||
cold_nav = np.cumprod(1 + cold_rets)
|
||||
cold_peak = np.maximum.accumulate(cold_nav)
|
||||
cold_dd = (cold_nav - cold_peak) / cold_peak
|
||||
cold_maxdd = cold_dd.min()
|
||||
|
||||
print(f"\n [{label}]:")
|
||||
print(f" 冷启动(1-170天): 累计={cold_cum*100:+.2f}% 年化={cold_ann*100:.2f}% 胜率={(cold_rets>0).mean()*100:.1f}% 最大回撤={cold_maxdd*100:.2f}%")
|
||||
print(f" 校准后(170天+): 累计={warm_cum*100:+.2f}% 年化={warm_ann*100:.2f}% 胜率={(warm_rets>0).mean()*100:.1f}% 最大回撤={warm_maxdd*100:.2f}%")
|
||||
|
||||
# === 5. IC校准 select=3 退化的根因：新增持仓的收益质量 ===
|
||||
print("\n\n" + "=" * 90)
|
||||
print(" 5. IC校准 select=3: 新增持仓收益分析")
|
||||
print("=" * 90)
|
||||
|
||||
# For each day, check if holdings changed between original and calibrated
|
||||
changed_days = 0
|
||||
new_asset_rets = []
|
||||
removed_asset_rets = []
|
||||
for orig_rec, calib_rec in zip(r1, r2):
|
||||
if orig_rec['date'] != calib_rec['date']:
|
||||
continue
|
||||
orig_set = set(orig_rec['holdings'])
|
||||
calib_set = set(calib_rec['holdings'])
|
||||
if orig_set != calib_set:
|
||||
changed_days += 1
|
||||
# The return of the calibrated portfolio on that day
|
||||
new_asset_rets.append(calib_rec['daily_return'])
|
||||
removed_asset_rets.append(orig_rec['daily_return'])
|
||||
|
||||
print(f"\n 持仓变化天数: {changed_days} / {len(r1)}")
|
||||
if new_asset_rets:
|
||||
new_arr = np.array(new_asset_rets)
|
||||
old_arr = np.array(removed_asset_rets)
|
||||
diff = new_arr - old_arr
|
||||
print(f" 变化日 - IC校准收益: {new_arr.mean()*100:+.4f}% 原始收益: {old_arr.mean()*100:+.4f}% 差: {diff.mean()*100:+.4f}%")
|
||||
print(f" 变化日胜率: IC校准={((new_arr>0).mean()*100):.1f}% 原始={((old_arr>0).mean()*100):.1f}%")
|
||||
print(f" 变化日累计: IC校准={new_arr.sum()*100:+.2f}% 原始={old_arr.sum()*100:+.2f}% 差={diff.sum()*100:+.2f}%")
|
||||
|
||||
# === 6. IC stats ===
|
||||
print("\n\n" + "=" * 90)
|
||||
print(" 6. IC校准最终状态")
|
||||
print("=" * 90)
|
||||
calibrator = s2.ic_calibrator
|
||||
for code in sorted(calibrator._history.keys()):
|
||||
stats = calibrator.get_stats(code)
|
||||
nm = s2.config.asset_pools.assets.get(code)
|
||||
name = nm.name if nm else code
|
||||
print(f" {name:<12} ({code:<14}): n={stats.n_samples:>5} IC={stats.ic_value:+.4f} sign={stats.ic_sign:+d}")
|
||||
Reference in New Issue
Block a user