“投资范围包括所有中国股市的股票。复合反转策略将等权分配给短期反转和长期反转两个子策略，形成期分别为3-1个月和36-13个月，短期反转部分不包括冷却期。根据滞后动量的平均值将投资组合分为十分位，做多最低十分位并做空最高十分位。该策略每月再平衡，且每个反转策略按市值加权。”
资产类别：股票 | 地区：中国 | 频率：每月 | 市场：股票 | 关键词：长期，短期，反转，中国
策略概述
投资范围包括所有中国股市股票。复合反转策略将等权分配给两个子策略：一半为短期反转，一半为长期反转。形成期分别为3-1个月和36-13个月。这意味着在短期反转部分不包括冷却期。计算每只股票在上述期间的滞后动量。根据形成期动量的平均值将投资组合分为十分位。做多最低的十分位，做空最高的十分位。该策略每月再平衡，每个反转策略按市值加权。
策略合理性
中国股市的行为与大多数其他市场不同。研究表明，在2008年金融危机之前，存在动量效应，但在危机之后不再明显，动量效应似乎已经转变为今天观察到的反转效应。
该策略的功能性也得到了功能数据分析（FDA）的支持，正是通过FDA发现了这一效应。FDA允许捕捉线性和非线性的横截面模式以及动态时间序列的演变。这一过程能够将收益分解为经验功能组成部分，从而以一种全新的方式捕捉反常收益。
论文来源
The Evolvement of Momentum Effects in China: Evidence from Functional Data Analysis [点击浏览原文]
- 刘振亚、李博、王世轩
<摘要>
与发达证券市场相比，中国股市的动量或反转效应存在不一致的现象。我们使用基于功能数据分析（FDA）的新范式来解决这一争议，目的是调和先前的不一致。与传统方法相比，基于FDA的范式能够识别非线性横截面模式和动态时间序列演变。我们的实证结果提供了有力证据，表明在2008年全球金融危机后，中期动量效应消失，市场以反转效应为主。此外，我们在各种设定中没有发现永久动量效应的证据，但确实发现了中国市场中的短期（1-6个月）和长期（3年）反转效应的显著证据。
回测表现
| 年化收益率 | 27.88% |
| 波动率 | 19.95% |
| Beta | -0.078 |
| 夏普比率 | 1.4 |
| 索提诺比率 | N/A |
| 最大回撤 | N/A |
| 胜率 | 53% |
完整python代码
from AlgorithmImports import *
#endregion
class CombinationoftheLongtermandtheShorttermReversalinChina(QCAlgorithm):
def Initialize(self):
self.SetStartDate(2000, 1, 1)
self.SetCash(100000)
self.data:dict = {}
self.quantity:dict[Symbol, int] = {}
# https://www.tradingview.com/markets/stocks-hong-kong/market-movers-large-cap/
self.tickers:list[str] = [
'0700','1299','3690','9618','0883','0388','9633','1810','2388','1876',
'0011','0016','1024','0066','0267','1109','0688','2020','0669','0981',
'0003','0020','0001','0960','0002','2269','1113','2015','0027','2319',
'2328','0012','0291','0316','0788','2313','1929','2057','2331','2382',
'1928','1038','0968','0762','6618','2618','0881','1972','0006','2688',
'1997','0175','1821','1093','6098',
# NOTE price data error exclusion
# '1913','2007','6969','0151','9961',
# '1177','0004','0017','0083','1308',
# '0992','1378','0322','3692','6823',
'0868','6186','3323','1209','0101','2638','0586','3800','0836','0270',
'1179','6862','0288','1193','0019','0656','0135','2066','1099','0384',
'3799','9889','0916','0241','1359','0144','0489','3311','1044','0268'
]
# long and short term period
self.st_period:int = 3 * 21
self.st_period_skip:int = 1 * 21
self.lt_period:int = 36 * 21
self.lt_period_skip:int = 13 * 21
self.quantile:int = 10
self.max_missing_days = 5
self.SetWarmup(self.lt_period, Resolution.Daily)
for ticker in self.tickers:
# price data
data = self.AddData(ChineseStock, ticker, Resolution.Daily)
data.SetLeverage(10)
data.SetFeeModel(CustomFeeModel())
self.data[ticker] = SymbolData(self.lt_period)
self.recent_month:int = -1
def OnData(self, data:Slice):
# store daily prices
for ticker in self.tickers:
if ticker in data and data[ticker]:
self.data[ticker].update_close(data[ticker].Value)
if self.Time.month == self.recent_month:
return
self.recent_month = self.Time.month
lt_momentum:dict = { ticker : self.data[ticker].momentum(self.lt_period, self.lt_period_skip) for ticker in self.tickers if self.data[ticker].closes_are_ready() and (self.Time.date() - self.Securities[ticker].GetLastData().Time.date()).days < self.max_missing_days}
st_momentum:dict = { ticker : self.data[ticker].momentum(self.st_period, self.st_period_skip) for ticker in self.tickers if self.data[ticker].closes_are_ready() and (self.Time.date() - self.Securities[ticker].GetLastData().Time.date()).days < self.max_missing_days}
if len(lt_momentum) >= self.quantile:
# both lt and st reversal strategy sorting
sorted_by_lt_momentum:list = sorted(lt_momentum.items(), key = lambda x: x[1], reverse=True)
self.vw_reversal_strategy_quantity(sorted_by_lt_momentum)
sorted_by_st_momentum:list = sorted(st_momentum.items(), key = lambda x: x[1], reverse=True)
self.vw_reversal_strategy_quantity(sorted_by_st_momentum)
# trade execution
invested:list = [x.Key for x in self.Portfolio if x.Value.Invested]
for symbol in invested:
if symbol not in self.quantity:
self.Liquidate(symbol)
for symbol, q in self.quantity.items():
self.MarketOrder(symbol, q)
self.quantity.clear()
def vw_reversal_strategy_quantity(self, sorted_by_momentum:dict) -> None:
quantile:int = int(len(sorted_by_momentum) / self.quantile)
# long and short leg
short:list = [x[0] for x in sorted_by_momentum[:quantile]]
long:list = [x[0] for x in sorted_by_momentum[-quantile:]]
# calculate weights and quantities
w:float = 0.5 / len(short)
for ticker in short:
q:int = int(np.floor(-(self.Portfolio.TotalPortfolioValue * w) / self.data[ticker].closes[0]))
if ticker not in self.quantity:
self.quantity[ticker] = q
else:
self.quantity[ticker] += q
w:float = 0.5 / len(long)
for ticker in long:
q:int = int(np.floor((self.Portfolio.TotalPortfolioValue * w) / self.data[ticker].closes[0]))
if ticker not in self.quantity:
self.quantity[ticker] = q
else:
self.quantity[ticker] += q
class SymbolData():
def __init__(self, max_momentum_period:int) -> None:
self.closes = RollingWindow[float](max_momentum_period)
def update_close(self, close:float) -> None:
self.closes.Add(close)
def closes_are_ready(self) -> bool:
return self.closes.IsReady
def momentum(self, momentum_period:int, skip_period:int) -> float:
performance:float = self.closes[skip_period - 1] / self.closes[momentum_period - 1] - 1
return performance
# Custom fee model
class CustomFeeModel():
def GetOrderFee(self, parameters):
fee = parameters.Security.Price * parameters.Order.AbsoluteQuantity * 0.00005
return OrderFee(CashAmount(fee, "USD"))
class ChineseStock(PythonData):
''' https://finance.yahoo.com/ '''
def GetSource(self, config, date, isLiveMode):
return SubscriptionDataSource("data.quantpedia.com/backtesting_data/equity/hong_kong_stocks/{0}.csv".format(config.Symbol.Value), SubscriptionTransportMedium.RemoteFile, FileFormat.Csv)
def Reader(self, config, line, date, isLiveMode):
# Example Line Format:
# 2003-03-12;590.3811645507812
data = ChineseStock()
data.Symbol = config.Symbol
if not line[0].isdigit(): return None
split = line.split(';')
data.Time = datetime.strptime(split[0], "%Y-%m-%d") + timedelta(days=1)
data.Value = float(split[1])
return data
