使用 pandas 在 Python 中研究回测环境

# backtest.pyclass Portfolio(object):    """An abstract base class representing a portfolio of     positions (including both instruments and cash), determined    on the basis of a set of signals provided by a Strategy."""    __metaclass__ = ABCMeta    @abstractmethod    def generate_positions(self):        """Provides the logic to determine how the portfolio         positions are allocated on the basis of forecasting        signals and available cash."""        raise NotImplementedError("Should implement generate_positions()!")    @abstractmethod    def backtest_portfolio(self):        """Provides the logic to generate the trading orders        and subsequent equity curve (i.e. growth of total equity),        as a sum of holdings and cash, and the bar-period returns        associated with this curve based on the 'positions' DataFrame.        Produces a portfolio object that can be examined by         other classes/functions."""        raise NotImplementedError("Should implement backtest_portfolio()!")

# random_forecast.pyimport numpy as npimport pandas as pdimport Quandl   # Necessary for obtaining financial data easilyfrom backtest import Strategy, Portfolioclass RandomForecastingStrategy(Strategy):    """Derives from Strategy to produce a set of signals that    are randomly generated long/shorts. Clearly a nonsensical    strategy, but perfectly acceptable for demonstrating the    backtesting infrastructure!"""        def __init__(self, symbol, bars):        """Requires the symbol ticker and the pandas DataFrame of bars"""        self.symbol = symbol        self.bars = bars    def generate_signals(self):        """Creates a pandas DataFrame of random signals."""        signals = pd.DataFrame(index=self.bars.index)        signals['signal'] = np.sign(np.random.randn(len(signals)))        # The first five elements are set to zero in order to minimise        # upstream NaN errors in the forecaster.        signals['signal'][0:5] = 0.0        return signals

# random_forecast.pyclass MarketOnOpenPortfolio(Portfolio):    """Inherits Portfolio to create a system that purchases 100 units of     a particular symbol upon a long/short signal, assuming the market     open price of a bar.    In addition, there are zero transaction costs and cash can be immediately     borrowed for shorting (no margin posting or interest requirements).     Requires:    symbol - A stock symbol which forms the basis of the portfolio.    bars - A DataFrame of bars for a symbol set.    signals - A pandas DataFrame of signals (1, 0, -1) for each symbol.    initial_capital - The amount in cash at the start of the portfolio."""    def __init__(self, symbol, bars, signals, initial_capital=100000.0):        self.symbol = symbol                self.bars = bars        self.signals = signals        self.initial_capital = float(initial_capital)        self.positions = self.generate_positions()    def generate_positions(self):        """Creates a 'positions' DataFrame that simply longs or shorts        100 of the particular symbol based on the forecast signals of        {1, 0, -1} from the signals DataFrame."""        positions = pd.DataFrame(index=signals.index).fillna(0.0)        positions[self.symbol] = 100*signals['signal']        return positions    def backtest_portfolio(self):        """Constructs a portfolio from the positions DataFrame by         assuming the ability to trade at the precise market open price        of each bar (an unrealistic assumption!).         Calculates the total of cash and the holdings (market price of        each position per bar), in order to generate an equity curve        ('total') and a set of bar-based returns ('returns').        Returns the portfolio object to be used elsewhere."""        # Construct the portfolio DataFrame to use the same index        # as 'positions' and with a set of 'trading orders' in the        # 'pos_diff' object, assuming market open prices.        portfolio = self.positions*self.bars['Open']        pos_diff = self.positions.diff()        # Create the 'holdings' and 'cash' series by running through        # the trades and adding/subtracting the relevant quantity from        # each column        portfolio['holdings'] = (self.positions*self.bars['Open']).sum(axis=1)        portfolio['cash'] = self.initial_capital - (pos_diff*self.bars['Open']).sum(axis=1).cumsum()        # Finalise the total and bar-based returns based on the 'cash'        # and 'holdings' figures for the portfolio        portfolio['total'] = portfolio['cash'] + portfolio['holdings']        portfolio['returns'] = portfolio['total'].pct_change()        return portfolio

if __name__ == "__main__":    # Obtain daily bars of SPY (ETF that generally     # follows the S&P500) from Quandl (requires 'pip install Quandl'    # on the command line)    symbol = 'SPY'    bars = Quandl.get("GOOG/NYSE_%s" % symbol, collapse="daily")    # Create a set of random forecasting signals for SPY    rfs = RandomForecastingStrategy(symbol, bars)    signals = rfs.generate_signals()    # Create a portfolio of SPY    portfolio = MarketOnOpenPortfolio(symbol, bars, signals, initial_capital=100000.0)    returns = portfolio.backtest_portfolio()    print returns.tail(10)

              SPY  holdings    cash  total   returnsDate                                                2014-01-02 -18398    -18398  111486  93088  0.0000972014-01-03  18321     18321   74844  93165  0.0008272014-01-06  18347     18347   74844  93191  0.0002792014-01-07  18309     18309   74844  93153 -0.0004082014-01-08 -18345    -18345  111534  93189  0.0003862014-01-09 -18410    -18410  111534  93124 -0.0006982014-01-10 -18395    -18395  111534  93139  0.0001612014-01-13 -18371    -18371  111534  93163  0.0002582014-01-14 -18228    -18228  111534  93306  0.0015352014-01-15  18410     18410   74714  93124 -0.001951