EDINETのCSVから財務数値を取り出す｜XBRLをPythonで扱う

import zipfile
from pathlib import Path

import pandas as pd


def read_edinet_csv(zip_path: Path) -> pd.DataFrame:
    """EDINETのCSV(ZIP)から提出本文書のデータを読む"""
    with zipfile.ZipFile(zip_path) as zf:
        targets = [n for n in zf.namelist() if "jpcrp" in n and n.endswith(".csv")]
        if not targets:
            raise ValueError(f"対象CSVが見つかりません: {zip_path}")

        with zf.open(targets[0]) as f:
            # UTF-16LE / タブ区切り
            return pd.read_csv(f, encoding="utf-16le", sep="\t")


df = read_edinet_csv(Path("edinet/S1000001.zip"))
print(df.columns.tolist())
print(len(df), "行")

COL_ITEM = "項目名"
COL_VALUE = "値"
COL_CONSOLIDATED = "連結・個別"
COL_PERIOD = "相対年度"


def pick(df: pd.DataFrame, item: str) -> str | None:
    """連結・当期の値を1つ取り出す"""
    hit = df[
        (df[COL_ITEM] == item)
        & (df[COL_CONSOLIDATED] == "連結")
        & (df[COL_PERIOD] == "当期")
    ]
    if hit.empty:
        return None
    return hit.iloc[0][COL_VALUE]


for item in ["売上高", "営業利益", "経常利益", "当期純利益"]:
    print(item, pick(df, item))

def to_number(value) -> float | None:
    """文字列の値を数値に変換する。変換できなければ None"""
    return pd.to_numeric(value, errors="coerce")

ALIASES = {
    "売上高": ["売上高", "営業収益", "売上収益"],
    "営業利益": ["営業利益", "営業利益又は営業損失（△）"],
    "経常利益": ["経常利益", "経常利益又は経常損失（△）"],
    "当期純利益": [
        "親会社株主に帰属する当期純利益",
        "当期純利益",
        "親会社の所有者に帰属する当期利益",
    ],
}


def pick_any(df: pd.DataFrame, candidates: list[str]) -> float | None:
    """候補の項目名を順に試して、最初に見つかった値を返す"""
    for name in candidates:
        value = pick(df, name)
        if value is not None:
            return to_number(value)
    return None

def extract_summary(zip_path: Path, meta: dict) -> dict:
    """1社分のZIPから主要数値を取り出す"""
    df = read_edinet_csv(zip_path)
    row = {
        "docID": meta["docID"],
        "証券コード": meta["secCode"],
        "会社名": meta["filerName"],
        "決算期": meta["periodEnd"],
    }
    for label, candidates in ALIASES.items():
        row[label] = pick_any(df, candidates)
    return row


rows = []
for meta in docs:  # 書類一覧APIで絞り込んだリスト
    zip_path = Path("edinet") / f"{meta['docID']}.zip"
    if not zip_path.exists():
        continue
    try:
        rows.append(extract_summary(zip_path, meta))
    except Exception as e:  # 1社の失敗で全体を止めない
        print("skip:", meta["docID"], e)

summary = pd.DataFrame(rows)
summary.to_csv("summary.csv", index=False, encoding="utf-8-sig")