179 lines
9.7 KiB
Python
179 lines
9.7 KiB
Python
"""Test configuration for factor module tests."""
|
||
import sys
|
||
import os
|
||
|
||
# Add vnpy source to path
|
||
_VNPY_SRC = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "vnpy_v4.4.0"))
|
||
if _VNPY_SRC not in sys.path:
|
||
sys.path.insert(0, _VNPY_SRC)
|
||
|
||
|
||
# ==================== 合成财务静态域(财务因子批测试共用) ====================
|
||
# 6 只股 × 24 报告期(2018Q1~2023Q4;2018-2019 为 16 季滚动窗/5 年 CAGR 预热史),
|
||
# NOTICE_DATE 错位: Q1→当年4-28 / H1→当年8-29 / Q3→当年10-27 / 年报→次年4-25
|
||
# 数值全部手算可验证(断言用),公式见各 build 函数内注释。
|
||
from datetime import date
|
||
|
||
import polars as pl
|
||
|
||
REPORT_DATES = [
|
||
f"{y}-{m}" for y in (2018, 2019, 2020, 2021, 2022, 2023) for m in ("03-31", "06-30", "09-30", "12-31")
|
||
]
|
||
|
||
# 单季 REV/NP 序列(累计=年内前缀和): 报告期索引 i=0..23(2018Q1..2023Q4)
|
||
# 2020 起的段与 P0 期完全一致(旧断言零回归);2018-2019 为新增预热史
|
||
REV_Q = ([70, 100, 110, 120] + [80, 110, 120, 130] # 2018 / 2019
|
||
+ [90, 130, 140, 150] + [100, 120, 150, 160] # 2020 / 2021
|
||
+ [110, 140, 150, 180] + [130, 140, 170, 180]) # 2022 / 2023
|
||
NP_Q = ([7, 10, 11, 12] + [8, 11, 12, 13] # 2018 / 2019
|
||
+ [9, 13, 14, 15] + [10, 12, 15, 16] # 2020 / 2021
|
||
+ [11, 14, 15, 18] + [13, 14, 17, 18]) # 2022 / 2023
|
||
|
||
# 资产负债表(时点存量,随报告期线性演化;2021Q1 起算,idx=i-12)
|
||
TA_OF = lambda i: 1000 + 50 * max(i - 12, 0)
|
||
EQ_OF = lambda i: 500 + 20 * max(i - 12, 0)
|
||
AR_OF = lambda i: 100 + 10 * max(i - 12, 0)
|
||
SC_OF = lambda i: 100 if i < 20 else 110 # 2023 起股本扩张 10%(NSI=0.1)
|
||
|
||
SYN_STOCKS = {
|
||
"600000.SH": {"vt": "600000.SSE", "scale": 1.0,
|
||
"drop_periods": [], "null_notice_periods": []},
|
||
"000001.SZ": {"vt": "000001.SZSE", "scale": 1.0,
|
||
"drop_periods": ["2022-03-31"], # 缺 2022Q1 → 单季差分/TTM 链 NaN
|
||
"null_notice_periods": ["2022-09-30"]}, # NOTICE_DATE 缺失 → 该报告期跳过
|
||
"300001.SZ": {"vt": "300001.SZSE", "scale": 2.0,
|
||
"drop_periods": [], "null_notice_periods": []},
|
||
# 三只分块等值测试扩容股(全史无残缺,不同 scale 增截面多样性)
|
||
"600004.SH": {"vt": "600004.SSE", "scale": 0.5,
|
||
"drop_periods": [], "null_notice_periods": []},
|
||
"000333.SZ": {"vt": "000333.SZSE", "scale": 1.7,
|
||
"drop_periods": [], "null_notice_periods": []},
|
||
"300124.SZ": {"vt": "300124.SZSE", "scale": 3.0,
|
||
"drop_periods": [], "null_notice_periods": []},
|
||
# 银行模板股(OPERATE_COST 缺失 → _is_fin;INVENTORY 缺失测 null 传播):
|
||
# 盈利质量/成长/费用类特征应置 NaN,盈利能力/估值族保留(§7 红线 5)
|
||
"601398.SH": {"vt": "601398.SSE", "scale": 4.0, "bank": True,
|
||
"drop_periods": [], "null_notice_periods": []},
|
||
}
|
||
|
||
|
||
def _notice_date(report_date: str) -> str:
|
||
"""A 股典型披露节奏(年报次年 4-25 / 一季报 4-28 / 中报 8-29 / 三季报 10-27)."""
|
||
y, m = int(report_date[:4]), int(report_date[5:7])
|
||
if m == 3:
|
||
return f"{y}-04-28"
|
||
if m == 6:
|
||
return f"{y}-08-29"
|
||
if m == 9:
|
||
return f"{y}-10-27"
|
||
return f"{y + 1}-04-25"
|
||
|
||
|
||
def _cum_in_year(q_values: list[float], i: int) -> float:
|
||
"""报告期 i 的年内累计值 = 当年前几季单季之和."""
|
||
year_start = (i // 4) * 4
|
||
return float(sum(q_values[year_start:i + 1]))
|
||
|
||
|
||
def build_synthetic_static(root: str) -> str:
|
||
"""写合成静态域 parquet 树(data_dir),返回 static 根目录路径.
|
||
|
||
目录结构与 NAS 一致: static/{income,balance,cashflow}/{code}.{SH|SZ}_{table}.parquet
|
||
forecast 按报告期全市场文件: static/forecast/{YYYYMMDD}_forecast.parquet
|
||
"""
|
||
static_dir = os.path.join(str(root), "static")
|
||
for table in ("income", "balance", "cashflow"):
|
||
os.makedirs(os.path.join(static_dir, table), exist_ok=True)
|
||
|
||
for file_code, spec in SYN_STOCKS.items():
|
||
s, drop, null_notice = spec["scale"], set(spec["drop_periods"]), set(spec["null_notice_periods"])
|
||
is_bank = spec.get("bank", False)
|
||
income_rows, balance_rows, cashflow_rows = [], [], []
|
||
for i, rd in enumerate(REPORT_DATES):
|
||
if rd in drop:
|
||
continue
|
||
notice = None if rd in null_notice else _notice_date(rd)
|
||
rev_c, np_c = s * _cum_in_year(REV_Q, i), s * _cum_in_year(NP_Q, i)
|
||
common = {
|
||
"REPORT_DATE": f"{rd} 00:00:00",
|
||
"NOTICE_DATE": (f"{notice} 00:00:00" if notice else None),
|
||
"UPDATE_DATE": f"{notice} 00:00:00" if notice else None,
|
||
}
|
||
income_rows.append({**common,
|
||
"TOTAL_OPERATE_INCOME": rev_c,
|
||
# 银行模板无营业成本(OPERATE_COST 缺失 → 金融股判定)
|
||
"OPERATE_COST": None if is_bank else 0.6 * rev_c,
|
||
"PARENT_NETPROFIT": np_c, "DEDUCT_PARENT_NETPROFIT": 0.9 * np_c,
|
||
"TOTAL_PROFIT": 1.1 * np_c, "INVEST_INCOME": 0.05 * np_c,
|
||
"FAIRVALUE_CHANGE_INCOME": 0.01 * np_c,
|
||
"ASSET_IMPAIRMENT_LOSS": 0.02 * np_c, "CREDIT_IMPAIRMENT_LOSS": 0.01 * np_c,
|
||
# P1 新原料(比率设计成可手算: τ=0.25 / INT=0.2·NP / 研发5% / 销售8%)
|
||
"RESEARCH_EXPENSE": 0.05 * rev_c, "SALE_EXPENSE": 0.08 * rev_c,
|
||
"FE_INTEREST_EXPENSE": 0.2 * np_c, "INCOME_TAX": 0.275 * np_c,
|
||
"BASIC_EPS": _cum_in_year(NP_Q, i) / SC_OF(i)}) # EPS 与 scale 无关
|
||
# IBD 只给 SHORT_LOAN 一列(其余组件列缺失,测 schema 缺列容错)
|
||
balance_rows.append({**common,
|
||
"TOTAL_ASSETS": s * TA_OF(i), "TOTAL_PARENT_EQUITY": s * EQ_OF(i),
|
||
"ACCOUNTS_RECE": s * AR_OF(i), "OTHER_RECE": s * (5 + max(i - 12, 0)),
|
||
"GOODWILL": 50.0, "SHARE_CAPITAL": float(SC_OF(i)),
|
||
"SHORT_LOAN": s * (100 + max(i - 12, 0)),
|
||
# P1 新原料: 存货/货币资金(银行无存货 → null 传播)
|
||
"INVENTORY": None if is_bank else s * (200 + 5 * max(i - 12, 0)),
|
||
"MONETARYFUNDS": s * (150 + 10 * max(i - 12, 0))})
|
||
cashflow_rows.append({**common,
|
||
"NETCASH_OPERATE": 1.2 * np_c, "SALES_SERVICES": 1.05 * rev_c,
|
||
"ACCEPT_INVEST_CASH": 0.1 * rev_c,
|
||
# P1 新原料(DA 合计 = 0.10·REV / capex 15% / 债务净发行 = +0.10·REV)
|
||
"FA_IR_DEPR": 0.06 * rev_c, "IA_AMORTIZE": 0.02 * rev_c,
|
||
"LPE_AMORTIZE": 0.01 * rev_c, "USERIGHT_ASSET_AMORTIZE": 0.01 * rev_c,
|
||
"CONSTRUCT_LONG_ASSET": 0.15 * rev_c,
|
||
"RECEIVE_LOAN_CASH": 0.3 * rev_c, "ISSUE_BOND": 0.05 * rev_c,
|
||
"PAY_DEBT_CASH": 0.25 * rev_c})
|
||
|
||
for table, rows in (("income", income_rows), ("balance", balance_rows), ("cashflow", cashflow_rows)):
|
||
pl.DataFrame(rows).write_parquet(
|
||
os.path.join(static_dir, table, f"{file_code}_{table}.parquet"))
|
||
|
||
# forecast: 按报告期全市场文件(中文列,归母净利润行优先 + 无净利润行 fallback)
|
||
# 预测数值 = 预告中值(元): A=30(实际 27 → beat=-0.1) / B=40(beat=-0.325)
|
||
os.makedirs(os.path.join(static_dir, "forecast"), exist_ok=True)
|
||
pl.DataFrame([
|
||
{"股票代码": "600000", "预测指标": "归属于上市公司股东的净利润",
|
||
"预测数值": 30.0, "业绩变动幅度": 56.79, "预告类型": "预增", "公告日期": date(2023, 7, 15)},
|
||
{"股票代码": "000001", "预测指标": "归属于上市公司股东的净利润",
|
||
"预测数值": 40.0, "业绩变动幅度": -30.0, "预告类型": "预减", "公告日期": date(2023, 7, 20)},
|
||
{"股票代码": "300001", "预测指标": "营业收入",
|
||
"预测数值": 999.0, "业绩变动幅度": 5.0, "预告类型": "略增", "公告日期": date(2023, 7, 10)},
|
||
# 北交所两段前缀(92x/43x)→ .BJSE 映射用例
|
||
{"股票代码": "920001", "预测指标": "归属于上市公司股东的净利润",
|
||
"预测数值": 50.0, "业绩变动幅度": 20.0, "预告类型": "预增", "公告日期": date(2023, 7, 18)},
|
||
{"股票代码": "430047", "预测指标": "净利润",
|
||
"预测数值": 20.0, "业绩变动幅度": 10.0, "预告类型": "预增", "公告日期": date(2023, 7, 19)},
|
||
]).write_parquet(os.path.join(static_dir, "forecast", "20230630_forecast.parquet"))
|
||
pl.DataFrame([
|
||
{"股票代码": "600000", "预测指标": "净利润",
|
||
"预测数值": 65.0, "业绩变动幅度": 100.0, "预告类型": "扭亏", "公告日期": date(2023, 10, 15)},
|
||
]).write_parquet(os.path.join(static_dir, "forecast", "20230930_forecast.parquet"))
|
||
# 迟到预告用例(F07 锚语义): 2022 年报披露 2023-04-25,预告公告 2023-05-10
|
||
# 晚于披露日 → 兑现差锚 = max(披露日, 公告日) = 05-10(实际 NP 58,中值 55)
|
||
pl.DataFrame([
|
||
{"股票代码": "600000", "预测指标": "归属于上市公司股东的净利润",
|
||
"预测数值": 55.0, "业绩变动幅度": 15.0, "预告类型": "预增", "公告日期": date(2023, 5, 10)},
|
||
]).write_parquet(os.path.join(static_dir, "forecast", "20221231_forecast.parquet"))
|
||
return static_dir
|
||
|
||
|
||
import pytest
|
||
|
||
|
||
@pytest.fixture(scope="session")
|
||
def synthetic_static(tmp_path_factory) -> str:
|
||
"""session 级合成静态域根目录(test_fundamental_* 共用)."""
|
||
return build_synthetic_static(tmp_path_factory.mktemp("fund_static"))
|
||
|
||
|
||
@pytest.fixture(scope="session")
|
||
def np_q_series() -> list[float]:
|
||
"""合成归母净利单季序列(SUE 期望值独立重算用)."""
|
||
return list(NP_Q)
|