baseline / Kronos /examples /run_backtest_kronos.py
humblman's picture
Upload folder using huggingface_hub
ccd4d5a verified
Raw
History Blame Contribute Delete
16.8 kB
# run_backtest.py
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import os
from datetime import datetime, timedelta
import warnings
warnings.filterwarnings('ignore')
# 设置中文字体
plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False
class KronosBacktester:
"""
Kronos模型回测类
"""
def __init__(self, data_dir, model_dir, initial_capital=100000):
"""
初始化回测器
参数:
data_dir: 数据目录
model_dir: 模型预测结果目录
initial_capital: 初始资金
"""
self.data_dir = data_dir
self.model_dir = model_dir
self.initial_capital = initial_capital
self.results = {}
def load_historical_data(self, stock_code):
"""
加载历史数据
"""
csv_file = os.path.join(self.data_dir, f"{stock_code}_stock_data.csv")
if not os.path.exists(csv_file):
raise FileNotFoundError(f"数据文件不存在: {csv_file}")
df = pd.read_csv(csv_file, encoding='utf-8-sig')
# 检查列名并标准化
column_mapping = {
'日期': 'date',
'开盘价': 'open',
'最高价': 'high',
'最低价': 'low',
'收盘价': 'close',
'成交量': 'volume',
'成交额': 'amount'
}
# 重命名列
for old_col, new_col in column_mapping.items():
if old_col in df.columns:
df = df.rename(columns={old_col: new_col})
df['date'] = pd.to_datetime(df['date'])
df.set_index('date', inplace=True)
df = df.sort_index()
print(f"✅ 加载历史数据: {len(df)} 条记录")
print(f"时间范围: {df.index.min()}{df.index.max()}")
return df
def load_predictions(self, stock_code):
"""
加载模型预测结果
"""
# 尝试不同的预测文件命名
pred_files = [
os.path.join(self.model_dir, f"{stock_code}_kronos_predictions.csv"),
os.path.join(self.model_dir, f"{stock_code}_detailed_predictions.csv"),
os.path.join(self.model_dir, f"{stock_code}_predictions.csv")
]
pred_df = None
for pred_file in pred_files:
if os.path.exists(pred_file):
pred_df = pd.read_csv(pred_file, encoding='utf-8-sig')
print(f"✅ 找到预测文件: {pred_file}")
break
if pred_df is None:
raise FileNotFoundError(f"未找到预测文件,请检查目录: {self.model_dir}")
# 标准化列名
column_mapping = {
'日期': 'date',
'预测收盘价': 'predicted_close',
'收盘价': 'predicted_close',
'预测成交量': 'predicted_volume',
'成交量': 'predicted_volume'
}
for old_col, new_col in column_mapping.items():
if old_col in pred_df.columns:
pred_df = pred_df.rename(columns={old_col: new_col})
pred_df['date'] = pd.to_datetime(pred_df['date'])
pred_df.set_index('date', inplace=True)
pred_df = pred_df.sort_index()
print(f"✅ 加载预测数据: {len(pred_df)} 条记录")
print(f"预测时间范围: {pred_df.index.min()}{pred_df.index.max()}")
return pred_df
def align_data(self, hist_df, pred_df):
"""
对齐历史数据和预测数据的时间范围
"""
# 找到历史数据的最后日期
last_hist_date = hist_df.index.max()
# 筛选预测数据,从历史数据结束后开始
pred_df_aligned = pred_df[pred_df.index > last_hist_date]
if len(pred_df_aligned) == 0:
# 如果没有未来的预测数据,使用所有预测数据
pred_df_aligned = pred_df.copy()
print("⚠️ 警告:预测数据没有未来的日期,使用所有预测数据")
print(f"✅ 数据对齐: 历史数据结束于 {last_hist_date}, 预测数据从 {pred_df_aligned.index.min()} 开始")
return pred_df_aligned
def calculate_trading_signals(self, hist_df, pred_df, threshold=0.02):
"""
计算交易信号
"""
# 对齐数据
pred_df = self.align_data(hist_df, pred_df)
# 合并历史数据和预测数据
combined = pd.concat([
hist_df[['close']].rename(columns={'close': 'actual'}),
pred_df[['predicted_close']].rename(columns={'predicted_close': 'predicted'})
], axis=1)
# 计算预测收益率
combined['pred_return'] = combined['predicted'].pct_change()
# 生成交易信号
combined['signal'] = 0
combined['signal'] = np.where(combined['pred_return'] > threshold, 1, # 买入信号
np.where(combined['pred_return'] < -threshold, -1, 0)) # 卖出信号
# 过滤信号:避免频繁交易
combined['position'] = combined['signal'].replace(to_replace=0, method='ffill').fillna(0)
return combined
def run_backtest(self, combined_df):
"""
运行回测
"""
# 初始化资金和持仓
capital = self.initial_capital
position = 0
trades = []
# 回测记录
backtest_results = pd.DataFrame(index=combined_df.index)
backtest_results['capital'] = capital
backtest_results['position'] = 0
backtest_results['returns'] = 0.0
backtest_results['price'] = combined_df['actual'].combine_first(combined_df['predicted'])
for i, (date, row) in enumerate(combined_df.iterrows()):
current_price = row['actual'] if not pd.isna(row['actual']) else row['predicted']
signal = row['position']
# 跳过无效价格
if pd.isna(current_price):
continue
# 执行交易
if i > 0: # 从第二天开始
prev_position = backtest_results['position'].iloc[i - 1] if i > 0 else 0
# 平仓信号
if prev_position != 0 and signal == 0:
# 平仓
capital = position * current_price
position = 0
trades.append({
'date': date,
'action': 'SELL',
'price': current_price,
'shares': prev_position,
'capital': capital
})
# 开仓信号
elif prev_position == 0 and signal != 0:
# 计算可买股数(假设全仓交易)
shares = int(capital / current_price)
if shares > 0:
position = shares * signal
capital -= shares * current_price
trades.append({
'date': date,
'action': 'BUY',
'price': current_price,
'shares': shares * signal,
'capital': capital
})
# 更新持仓市值
portfolio_value = capital + position * current_price
# 记录结果
backtest_results.loc[date, 'capital'] = portfolio_value
backtest_results.loc[date, 'position'] = position
backtest_results.loc[date, 'price'] = current_price
# 计算日收益率
if i > 0:
prev_value = backtest_results['capital'].iloc[i - 1]
if prev_value > 0:
backtest_results.loc[date, 'returns'] = (portfolio_value - prev_value) / prev_value
return backtest_results, trades
def calculate_metrics(self, backtest_results, trades):
"""
计算回测指标
"""
returns = backtest_results['returns'].replace([np.inf, -np.inf], np.nan).dropna()
if len(returns) == 0:
return {
'总收益率': 0,
'年化收益率': 0,
'波动率': 0,
'夏普比率': 0,
'最大回撤': 0,
'胜率': 0,
'平均交易收益': 0,
'交易次数': 0,
'最终资金': self.initial_capital
}
total_return = (backtest_results['capital'].iloc[-1] - self.initial_capital) / self.initial_capital
annual_return = (1 + total_return) ** (252 / len(returns)) - 1
# 波动率
volatility = returns.std() * np.sqrt(252)
# 夏普比率(假设无风险利率为3%)
risk_free_rate = 0.03
sharpe_ratio = (annual_return - risk_free_rate) / volatility if volatility > 0 else 0
# 最大回撤
cumulative_returns = (1 + returns).cumprod()
peak = cumulative_returns.expanding().max()
drawdown = (cumulative_returns - peak) / peak
max_drawdown = drawdown.min()
# 交易统计
trade_returns = []
buy_trades = [t for t in trades if t['action'] == 'BUY']
sell_trades = [t for t in trades if t['action'] == 'SELL']
for i in range(min(len(buy_trades), len(sell_trades))):
buy = buy_trades[i]
sell = sell_trades[i]
trade_return = (sell['price'] - buy['price']) / buy['price']
trade_returns.append(trade_return)
win_rate = len([r for r in trade_returns if r > 0]) / len(trade_returns) if trade_returns else 0
avg_trade_return = np.mean(trade_returns) if trade_returns else 0
metrics = {
'总收益率': total_return,
'年化收益率': annual_return,
'波动率': volatility,
'夏普比率': sharpe_ratio,
'最大回撤': max_drawdown,
'胜率': win_rate,
'平均交易收益': avg_trade_return,
'交易次数': len(trades),
'最终资金': backtest_results['capital'].iloc[-1]
}
return metrics
def plot_backtest_results(self, backtest_results, metrics, stock_code, output_dir):
"""
绘制回测结果图表
"""
fig, (ax1, ax2, ax3) = plt.subplots(3, 1, figsize=(15, 12))
# 1. 资金曲线
ax1.plot(backtest_results.index, backtest_results['capital'],
linewidth=2, label='策略资金曲线', color='#1f77b4')
ax1.axhline(y=self.initial_capital, color='red', linestyle='--',
label=f'初始资金 ({self.initial_capital:,.0f}元)')
ax1.set_ylabel('资金 (元)', fontsize=12)
ax1.legend()
ax1.grid(True, alpha=0.3)
ax1.set_title(f'{stock_code} Kronos模型回测结果', fontsize=14, fontweight='bold')
# 2. 收益率曲线
cumulative_returns = (1 + backtest_results['returns'].fillna(0)).cumprod()
ax2.plot(backtest_results.index, cumulative_returns,
linewidth=2, label='策略累计收益', color='#2ca02c')
# 基准收益(买入持有)
price_returns = backtest_results['price'].pct_change().fillna(0)
benchmark_returns = (1 + price_returns).cumprod()
ax2.plot(backtest_results.index, benchmark_returns,
linewidth=2, label='基准收益(买入持有)', color='#ff7f0e', alpha=0.7)
ax2.set_ylabel('累计收益', fontsize=12)
ax2.legend()
ax2.grid(True, alpha=0.3)
# 3. 回撤曲线
peak = cumulative_returns.expanding().max()
drawdown = (cumulative_returns - peak) / peak
ax3.fill_between(backtest_results.index, drawdown, 0,
alpha=0.3, color='red', label='回撤')
ax3.set_ylabel('回撤', fontsize=12)
ax3.set_xlabel('日期', fontsize=12)
ax3.legend()
ax3.grid(True, alpha=0.3)
# 添加指标文本
metrics_text = (
f"总收益率: {metrics['总收益率']:.2%}\n"
f"年化收益率: {metrics['年化收益率']:.2%}\n"
f"夏普比率: {metrics['夏普比率']:.2f}\n"
f"最大回撤: {metrics['最大回撤']:.2%}\n"
f"胜率: {metrics['胜率']:.2%}\n"
f"交易次数: {metrics['交易次数']}\n"
f"最终资金: {metrics['最终资金']:,.0f}元"
)
ax1.text(0.02, 0.98, metrics_text, transform=ax1.transAxes, fontsize=10,
verticalalignment='top', bbox=dict(boxstyle="round,pad=0.3",
facecolor="lightyellow", alpha=0.8))
plt.tight_layout()
# 保存图表
os.makedirs(output_dir, exist_ok=True)
chart_file = os.path.join(output_dir, f'{stock_code}_backtest_results.png')
plt.savefig(chart_file, dpi=300, bbox_inches='tight')
print(f"📊 回测图表已保存: {chart_file}")
plt.show()
def run_complete_backtest(self, stock_code, output_dir, threshold=0.02):
"""
运行完整的回测流程
"""
print(f"🎯 开始 {stock_code} 回测分析")
print("=" * 50)
try:
# 1. 加载数据
print("步骤1: 加载历史数据和预测数据...")
hist_df = self.load_historical_data(stock_code)
pred_df = self.load_predictions(stock_code)
# 2. 计算交易信号
print("步骤2: 计算交易信号...")
combined_df = self.calculate_trading_signals(hist_df, pred_df, threshold)
# 3. 运行回测
print("步骤3: 运行回测...")
backtest_results, trades = self.run_backtest(combined_df)
# 4. 计算指标
print("步骤4: 计算回测指标...")
metrics = self.calculate_metrics(backtest_results, trades)
# 5. 绘制结果
print("步骤5: 生成回测图表...")
self.plot_backtest_results(backtest_results, metrics, stock_code, output_dir)
# 6. 打印详细报告
print("\n" + "=" * 70)
print(f"📊 {stock_code} 回测报告")
print("=" * 70)
for key, value in metrics.items():
if isinstance(value, float):
if '率' in key or '收益' in key or '回撤' in key:
print(f" {key}: {value:.2%}")
else:
print(f" {key}: {value:.2f}")
else:
print(f" {key}: {value}")
print(f"\n交易记录 (共{len(trades)}次交易):")
for i, trade in enumerate(trades[-10:], 1): # 显示最后10次交易
print(f" 交易{i}: {trade['date'].strftime('%Y-%m-%d')} "
f"{trade['action']} {abs(trade['shares'])}股 @ {trade['price']:.2f}元")
return metrics, backtest_results, trades
except Exception as e:
print(f"❌ 回测过程中出现错误: {e}")
import traceback
traceback.print_exc()
return None, None, None
def main():
"""
主函数:运行Kronos模型回测
"""
# 配置参数
BACKTEST_CONFIG = {
"stock_code": "000831", # 要回测的股票代码
"data_dir": r"D:\lianghuajiaoyi\Kronos\examples\data", # 历史数据目录
"model_dir": r"D:\lianghuajiaoyi\Kronos\examples\yuce", # 模型预测结果目录
"output_dir": r"D:\lianghuajiaoyi\Kronos\examples\backtest", # 回测结果输出目录
"initial_capital": 100000, # 初始资金
"threshold": 0.02 # 交易阈值(2%)
}
print("🤖 Kronos模型回测系统")
print("=" * 50)
print(f"回测股票: {BACKTEST_CONFIG['stock_code']}")
print(f"初始资金: {BACKTEST_CONFIG['initial_capital']:,.0f}元")
print(f"交易阈值: {BACKTEST_CONFIG['threshold']:.1%}")
print()
# 创建回测器并运行
backtester = KronosBacktester(
data_dir=BACKTEST_CONFIG["data_dir"],
model_dir=BACKTEST_CONFIG["model_dir"],
initial_capital=BACKTEST_CONFIG["initial_capital"]
)
metrics, results, trades = backtester.run_complete_backtest(
stock_code=BACKTEST_CONFIG["stock_code"],
output_dir=BACKTEST_CONFIG["output_dir"],
threshold=BACKTEST_CONFIG["threshold"]
)
if metrics:
print(f"\n✅ {BACKTEST_CONFIG['stock_code']} 回测完成!")
print(f"📁 结果保存在: {BACKTEST_CONFIG['output_dir']}")
if __name__ == "__main__":
main()