217 lines
9.3 KiB
Python
217 lines
9.3 KiB
Python
"""
|
|
Script Name:
|
|
Description: 获取沪深300成分股的最新股价, 并计算年内涨幅, 924以来的涨幅, 市盈率, 股息率等。
|
|
需要调用futu的获取快照接口。
|
|
https://openapi.futunn.com/futu-api-doc/quote/get-market-snapshot.html
|
|
|
|
Author: [Your Name]
|
|
Created Date: YYYY-MM-DD
|
|
Last Modified: YYYY-MM-DD
|
|
Version: 1.0
|
|
|
|
Modification History:
|
|
- YYYY-MM-DD [Your Name]:
|
|
- YYYY-MM-DD [Your Name]:
|
|
- YYYY-MM-DD [Your Name]:
|
|
"""
|
|
|
|
import pymysql
|
|
import logging
|
|
import csv
|
|
import os
|
|
import config
|
|
import time
|
|
from datetime import datetime
|
|
from futu import OpenQuoteContext, RET_OK # Futu API client
|
|
|
|
# 配置日志
|
|
config.setup_logging()
|
|
|
|
# 1. 获取 hs300 表数据
|
|
def get_hs300_data():
|
|
conn = pymysql.connect(**config.db_config)
|
|
cursor = conn.cursor(pymysql.cursors.DictCursor)
|
|
cursor.execute("SELECT code, code_name FROM hs300")
|
|
hs300_data = cursor.fetchall()
|
|
cursor.close()
|
|
conn.close()
|
|
return hs300_data
|
|
|
|
# 2. 批量获取市场快照
|
|
def get_market_snapshots(codes):
|
|
quote_ctx = OpenQuoteContext(host='127.0.0.1', port=11111) # 替换为实际Futu API地址和端口
|
|
snapshot_results = []
|
|
batch_size = 350
|
|
|
|
for i in range(0, len(codes), batch_size):
|
|
code_batch = codes[i:i+batch_size]
|
|
ret, data = quote_ctx.get_market_snapshot(code_batch)
|
|
if ret == RET_OK:
|
|
snapshot_results.extend(data.to_dict('records'))
|
|
else:
|
|
logging.error(f"获取市场快照失败: {data}")
|
|
|
|
quote_ctx.close()
|
|
return snapshot_results
|
|
|
|
# 3. 插入或更新 futu_market_snapshot 表
|
|
def insert_or_update_snapshot(snapshot_data):
|
|
conn = pymysql.connect(**config.db_config)
|
|
cursor = conn.cursor()
|
|
query = """
|
|
INSERT INTO futu_market_snapshot (
|
|
code, name, update_time, last_price, open_price, high_price, low_price, prev_close_price,
|
|
volume, turnover, turnover_rate, suspension, listing_date, equity_valid, issued_shares,
|
|
total_market_val, net_asset, net_profit, earning_per_share, outstanding_shares, net_asset_per_share,
|
|
circular_market_val, ey_ratio, pe_ratio, pb_ratio, pe_ttm_ratio, dividend_ttm, dividend_ratio_ttm,
|
|
dividend_lfy, dividend_lfy_ratio
|
|
) VALUES (%s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s)
|
|
ON DUPLICATE KEY UPDATE
|
|
last_price = VALUES(last_price), open_price = VALUES(open_price), high_price = VALUES(high_price),
|
|
low_price = VALUES(low_price), prev_close_price = VALUES(prev_close_price), volume = VALUES(volume),
|
|
turnover = VALUES(turnover), turnover_rate = VALUES(turnover_rate), suspension = VALUES(suspension),
|
|
issued_shares = VALUES(issued_shares), total_market_val = VALUES(total_market_val), net_asset = VALUES(net_asset),
|
|
net_profit = VALUES(net_profit), earning_per_share = VALUES(earning_per_share), outstanding_shares = VALUES(outstanding_shares),
|
|
net_asset_per_share = VALUES(net_asset_per_share), circular_market_val = VALUES(circular_market_val),
|
|
ey_ratio = VALUES(ey_ratio), pe_ratio = VALUES(pe_ratio), pb_ratio = VALUES(pb_ratio),
|
|
pe_ttm_ratio = VALUES(pe_ttm_ratio), dividend_ttm = VALUES(dividend_ttm), dividend_ratio_ttm = VALUES(dividend_ratio_ttm),
|
|
dividend_lfy = VALUES(dividend_lfy), dividend_lfy_ratio = VALUES(dividend_lfy_ratio)
|
|
"""
|
|
|
|
try:
|
|
for record in snapshot_data:
|
|
cursor.execute(query, (
|
|
record['code'], record['name'], record['update_time'], record['last_price'], record['open_price'],
|
|
record['high_price'], record['low_price'], record['prev_close_price'], record['volume'], record['turnover'],
|
|
record['turnover_rate'], record['suspension'], record['listing_date'], record['equity_valid'], record['issued_shares'],
|
|
record['total_market_val'], record['net_asset'], record['net_profit'], record['earning_per_share'], record['outstanding_shares'],
|
|
record['net_asset_per_share'], record['circular_market_val'], record['ey_ratio'], record['pe_ratio'], record['pb_ratio'],
|
|
record['pe_ttm_ratio'], record['dividend_ttm'], record['dividend_ratio_ttm'], record['dividend_lfy'], record['dividend_lfy_ratio']
|
|
))
|
|
conn.commit()
|
|
except Exception as e:
|
|
logging.error(f"插入或更新快照数据时出错: {e}")
|
|
conn.rollback()
|
|
finally:
|
|
cursor.close()
|
|
conn.close()
|
|
|
|
def process_snapshot_data():
|
|
hs300_data = get_hs300_data()
|
|
codes = [item['code'] for item in hs300_data]
|
|
|
|
# 分批拉取市场快照数据
|
|
snapshot_data = get_market_snapshots(codes)
|
|
|
|
# 插入或更新数据到数据库
|
|
insert_or_update_snapshot(snapshot_data)
|
|
logging.info(f"Successfully get market snapshots and write to db.")
|
|
|
|
|
|
def write_to_csv(file_path, fieldnames, data):
|
|
"""
|
|
Writes data to a CSV file.
|
|
|
|
:param file_path: Path to the CSV file.
|
|
:param fieldnames: A list of field names (headers) for the CSV file.
|
|
:param data: A list of dictionaries where each dictionary represents a row.
|
|
"""
|
|
try:
|
|
# Check if file exists to determine if we need to write headers
|
|
file_exists = os.path.isfile(file_path)
|
|
|
|
# Open file in append mode ('a' means append, newline='' is needed for correct CSV formatting)
|
|
with open(file_path, mode='w+', newline='', encoding='utf-8') as csv_file:
|
|
writer = csv.DictWriter(csv_file, fieldnames=fieldnames)
|
|
|
|
# Write headers
|
|
writer.writeheader()
|
|
|
|
# Write the data rows
|
|
writer.writerows(data)
|
|
|
|
logging.info(f"Successfully wrote data to {file_path}")
|
|
except Exception as e:
|
|
logging.error(f"Error while writing to CSV file {file_path}: {str(e)}")
|
|
raise
|
|
|
|
def calculate_yield():
|
|
conn = pymysql.connect(**config.db_config)
|
|
cursor = conn.cursor(pymysql.cursors.DictCursor)
|
|
|
|
try:
|
|
hs300_data = get_hs300_data()
|
|
results = []
|
|
for item in hs300_data:
|
|
code = item['code']
|
|
name = item['code_name']
|
|
|
|
# 1. 获取最近快照数据
|
|
cursor.execute("SELECT * FROM futu_market_snapshot WHERE code=%s ORDER BY update_time DESC LIMIT 1", (code,))
|
|
row1 = cursor.fetchone()
|
|
row1['close'] = row1['last_price']
|
|
row1['time_key'] = row1['update_time']
|
|
|
|
# 2. 获取 2024-01-01 之前的历史数据
|
|
cursor.execute("SELECT * FROM hs300_his_kline_none WHERE code=%s AND time_key<'2024-01-01' ORDER BY time_key DESC LIMIT 1", (code,))
|
|
row2 = cursor.fetchone()
|
|
|
|
# 3. 获取 2024-09-24 之前的历史数据
|
|
cursor.execute("SELECT * FROM hs300_his_kline_none WHERE code=%s AND time_key<'2024-09-24' ORDER BY time_key DESC LIMIT 1", (code,))
|
|
row3 = cursor.fetchone()
|
|
|
|
# 4. 读取复权数据
|
|
cursor.execute("SELECT * FROM futu_rehab WHERE code=%s ORDER BY ex_div_date ASC", (code,))
|
|
rehab_res = cursor.fetchall()
|
|
|
|
# 5. 计算复权价格
|
|
for row in [row1, row2, row3]:
|
|
qfq_close = row['close']
|
|
time_key_date = row['time_key'].date() # 将 datetime 转换为 date
|
|
for rehab in rehab_res:
|
|
if rehab['ex_div_date'] >= time_key_date:
|
|
qfq_close = (qfq_close * rehab['forward_adj_factorA']) + rehab['forward_adj_factorB']
|
|
row['qfq_close'] = qfq_close
|
|
|
|
# 6. 计算收益率
|
|
year_yield = row1['qfq_close'] / row2['qfq_close'] - 1 if row1 and row2 else None
|
|
yield_0924 = row1['qfq_close'] / row3['qfq_close'] - 1 if row3 and row1 else None
|
|
|
|
# 7. 收集结果
|
|
result = {
|
|
'code': code,
|
|
'name': name,
|
|
'year_begin_date': row2['time_key'].date(),
|
|
'year_begin_close': row2['close'],
|
|
'0924_date': row3['time_key'].date(),
|
|
'0924_close': row3['close'],
|
|
'latest_date': row1['time_key'].date(),
|
|
'latest_close': row1['close'],
|
|
'year_yield': year_yield,
|
|
'yield_0924': yield_0924,
|
|
'total_market_val': row1.get('total_market_val', None),
|
|
'pe_ttm_ratio': row1.get('pe_ttm_ratio', None),
|
|
'dividend_ratio_ttm': row1.get('dividend_ratio_ttm', None),
|
|
'dividend_lfy_ratio': row1.get('dividend_lfy_ratio', None),
|
|
}
|
|
results.append(result)
|
|
logging.info(f"{result}")
|
|
time.sleep(0.1)
|
|
|
|
# 写入CSV
|
|
# 指定字段名称
|
|
# 获取当前日期 格式化为 yyyymmdd
|
|
current_date = datetime.now()
|
|
date_string = current_date.strftime('%Y%m%d')
|
|
fieldnames = ['code', 'name', 'year_begin_date', 'year_begin_close', '0924_date', '0924_close', 'latest_date', 'latest_close', 'year_yield', 'yield_0924', 'total_market_val', 'pe_ttm_ratio', 'dividend_ratio_ttm', 'dividend_lfy_ratio']
|
|
write_to_csv(f'./result/stat_growth{date_string}.csv', fieldnames, results)
|
|
|
|
except Exception as e:
|
|
logging.error(f"计算收益率时出错: {e}")
|
|
finally:
|
|
cursor.close()
|
|
conn.close()
|
|
|
|
if __name__ == "__main__":
|
|
process_snapshot_data()
|
|
calculate_yield() |