Overall Statistics |
Total Orders
17
Average Win
11.88%
Average Loss
-3.08%
Compounding Annual Return
3.606%
Drawdown
28.900%
Expectancy
2.642
Start Equity
100000
End Equity
181637.66
Net Profit
81.638%
Sharpe Ratio
0.133
Sortino Ratio
0.069
Probabilistic Sharpe Ratio
0.019%
Loss Rate
25%
Win Rate
75%
Profit-Loss Ratio
3.86
Alpha
-0.001
Beta
0.217
Annual Standard Deviation
0.105
Annual Variance
0.011
Information Ratio
-0.343
Tracking Error
0.162
Treynor Ratio
0.064
Total Fees
$45.15
Estimated Strategy Capacity
$930000.00
Lowest Capacity Asset
ERUS URI1LRYQ5ISL
Portfolio Turnover
0.09%
|
# https://quantpedia.com/strategies/value-factor-effect-within-countries/ # # The investment universe consists of 32 countries with easily accessible equity markets (via ETFs, for example). At the end of every year, # the investor calculates Shiller’s “CAPE” Cyclically Adjusted PE) ratio, for each country in his investment universe. CAPE is the ratio of # the real price of the equity market (adjusted for inflation) to the 10-year average of the country’s equity index (again adjusted for inflation). # The whole methodology is explained well on Shiller’s home page (http://www.econ.yale.edu/~shiller/data.htm) or # http://turnkeyanalyst.com/2011/10/the-shiller-pe-ratio/). The investor then invests in the cheapest 33% of countries from his sample if those # countries have a CAPE below 15. The portfolio is equally weighted (the investor holds 0% cash instead of countries with a CAPE higher than 15) # and rebalanced yearly. #region imports from AlgorithmImports import * #endregion class ValueFactorCAPEEffectwithinCountries(QCAlgorithm): def Initialize(self): self.SetStartDate(2008, 1, 1) self.SetCash(100000) self.symbols = { "Australia" : "EWA", # iShares MSCI Australia Index ETF "Brazil" : "EWZ", # iShares MSCI Brazil Index ETF "Canada" : "EWC", # iShares MSCI Canada Index ETF "Switzerland" : "EWL", # iShares MSCI Switzerland Index ETF "China" : "FXI", # iShares China Large-Cap ETF "France" : "EWQ", # iShares MSCI France Index ETF "Germany" : "EWG", # iShares MSCI Germany ETF "Hong Kong" : "EWH", # iShares MSCI Hong Kong Index ETF "Italy" : "EWI", # iShares MSCI Italy Index ETF "Japan" : "EWJ", # iShares MSCI Japan Index ETF "Korea" : "EWY", # iShares MSCI South Korea ETF "Mexico" : "EWW", # iShares MSCI Mexico Inv. Mt. Idx "Netherlands" : "EWN", # iShares MSCI Netherlands Index ETF "South Africa" : "EZA", # iShares MSCI South Africe Index ETF "Singapore" : "EWS", # iShares MSCI Singapore Index ETF "Spain" : "EWP", # iShares MSCI Spain Index ETF "Sweden" : "EWD", # iShares MSCI Sweden Index ETF "Taiwan" : "EWT", # iShares MSCI Taiwan Index ETF "UK" : "EWU", # iShares MSCI United Kingdom Index ETF "USA" : "SPY", # SPDR S&P 500 ETF "Russia" : "ERUS", # iShares MSCI Russia ETF "Israel" : "EIS", # iShares MSCI Israel ETF "India" : "INDA", # iShares MSCI India ETF "Poland" : "EPOL", # iShares MSCI Poland ETF "Turkey" : "TUR" # iShares MSCI Turkey ETF } self.quantile:int = 3 self.max_missing_days:int = 31 self.leverage:int = 2 for country, etf_symbol in self.symbols.items(): data = self.AddEquity(etf_symbol, Resolution.Daily) data.SetLeverage(self.leverage) data.SetFeeModel(CustomFeeModel()) # CAPE data import. self.cape_data = self.AddData(CAPE, 'CAPE', Resolution.Daily).Symbol self.recent_month:int = -1 def OnData(self, data:Slice) -> None: if self.Time.month == self.recent_month: return self.recent_month = self.Time.month if self.recent_month != 12: return price = {} for country, etf_symbol in self.symbols.items(): if etf_symbol in data and data[etf_symbol]: # cape data is still comming in if self.Securities[self.cape_data].GetLastData() and (self.Time.date() - self.Securities[self.cape_data].GetLastData().Time.date()).days <= self.max_missing_days: country_cape = self.Securities['CAPE'].GetLastData().GetProperty(country) if country_cape < 15. and country_cape != 0.: price[etf_symbol] = data[etf_symbol].Value long = [] # Cape and price sorting. if len(price) >= self.quantile: sorted_by_price = sorted(price.items(), key = lambda x: x[1], reverse = True) tercile = int(len(sorted_by_price) / self.quantile) long = [x[0] for x in sorted_by_price[-tercile:]] # Trade execution. invested = [x.Key for x in self.Portfolio if x.Value.Invested] for symbol in invested: if symbol not in long: self.Liquidate(symbol) for symbol in long: if symbol in data and data[symbol]: self.SetHoldings(symbol, 1 / len(long)) # NOTE: IMPORTANT: Data order must be ascending (datewise) # Data source: https://indices.barclays/IM/21/en/indices/static/historic-cape.app class CAPE(PythonData): def GetSource(self, config, date, isLiveMode): return SubscriptionDataSource("data.quantpedia.com/backtesting_data/economic/cape_by_country.csv", SubscriptionTransportMedium.RemoteFile, FileFormat.Csv) _header_columns:List[str] = [] def Reader(self, config, line, date, isLiveMode): data = CAPE() data.Symbol = config.Symbol if not line[0].isdigit(): CAPE._header_columns = line.split(',')[1:] return None split = line.split(',') data.Time = datetime.strptime(split[0], "%Y-%m-%d") + timedelta(days=1) for i, col in enumerate(CAPE._header_columns): if split[i+1] != '': data[col] = float(split[i+1]) else: data[col] = 0. data.Value = float(split[1]) return data # Custom fee model. class CustomFeeModel(FeeModel): def GetOrderFee(self, parameters): fee = parameters.Security.Price * parameters.Order.AbsoluteQuantity * 0.00005 return OrderFee(CashAmount(fee, "USD"))