TimeReturn doesn't seem to work

adamb032

I am trying to get a basic buy and hold strategy working, but TimeReturn does not seem to be working... my code:

start_date = datetime(2019,9,30)
end_date = datetime(2019,10,5)
data_df = get_eod_data('SPY', start_date, end_date)
data_df['close_change'] = data_df['close'].pct_change()
print(data_df[['close', 'close_change']])


class BuyAndHold(bt.Strategy):

    def __init__(self):
        self.logger = logging.getLogger(__name__)
        self.logger.info(f'Initializing strategy: BuyAndHold')
        
    def nextstart(self):
        self.logger.info(f"BUY {self.data.datetime.date(0)}: {self.data.close[0]:.2f}")
        self.order_target_value(target=self.broker.get_cash())


    def stop(self):
        self.logger.info(f"starting value: {self.broker.startingcash}")
        self.logger.info(f"ending value  : {self.broker.getvalue()}")
        roi = (self.broker.get_value() / self.broker.startingcash) - 1.0
        self.logger.info(f"roi: {roi}")


data = bt.feeds.PandasData(dataname=data_df, openinterest=None)

cerebro = bt.Cerebro(stdstats=False)
cerebro.adddata(data)

cerebro.addanalyzer(btanalyzers.TimeReturn, _name='returns')
cerebro.addobserver(bt.observers.Broker)
cerebro.addobserver(bt.observers.DrawDown)
cerebro.addsizer(bt.sizers.PercentSizer, percents=100)
cerebro.addstrategy(BuyAndHold)

results = cerebro.run()  # run it all

#cerebro.plot(style='bar')  # and plot it with a single command

returns_in_ordered_dict = results[0].analyzers.getbyname('returns').get_analysis()
returns_df = pd.DataFrame([returns_in_ordered_dict]).T.rename(columns={0: 'return'})
print(returns_df)
print(286.601471 /288.957764-1)

the output:

                 close  close_change
trade_date                          
2019-09-30  288.957764           NaN
2019-10-01  285.520630     -0.011895
2019-10-02  280.477051     -0.017664
2019-10-03  282.774933      0.008193
2019-10-04  286.601471      0.013532
13:41:53 [                      buy_hold]:INFO: Initializing strategy: BuyAndHold
13:41:53 [                      buy_hold]:INFO: BUY 2019-09-30: 288.96
13:41:53 [                      buy_hold]:INFO: starting value: 10000.0
13:41:53 [                      buy_hold]:INFO: ending value  : 9887.776226027483
13:41:53 [                      buy_hold]:INFO: roi: -0.011222377397251737
              return
2019-09-30  0.000000
2019-10-01 -0.014897
2019-10-02 -0.017407
2019-10-03  0.008071
2019-10-04  0.013333

-0.008154454711242876

since this is a buy and hold, and it buys on the first bar (288.96) (the close on 2019-9-30), over these five days, the buy and hold should return -81.544bps, however using hte TimeReturn, it says my return is -112.22bps. also the daily returns are wrong. is there something I am not understanding here ?

run-out

@adamb032 I believe what is happening, is in your dataframe, you are calculating the pct_change of the price. TimeReturn in backtrader is working off the value of the account.

Try adding this into your analyzer.

cerebro.addanalyzer(bt.analyzers.TimeReturn, **dict(_name='returns', data=data))

By telling the analyzer to use data, you will get identical outputs to your dataframe.

Here are the outputs on my end:

Original data with percent change. 
                           Close      close_change
Date                                
2019-09-30  296.769989           NaN
2019-10-01  293.239990     -0.011895
2019-10-02  288.059998     -0.017665
2019-10-03  290.420013      0.008193
2019-10-04  294.350006      0.013532

TimeReturn using the data as parameter in the analyzer: 
                      return
2019-09-30  0.002703
2019-10-01 -0.011895
2019-10-02 -0.017665
2019-10-03  0.008193
2019-10-04  0.013532

Date/Close/Cash/Value of the account: 
2019-10-01 close: 293.24, cash: 174.58, value: 9851.50
2019-10-02 close: 288.06, cash: 174.58, value: 9680.56
2019-10-03 close: 290.42, cash: 174.58, value: 9758.44
2019-10-04 close: 294.35, cash: 174.58, value: 9888.13

adamb032

@run-out ah i see, i have another question, it looks like the first bar's returns are close/open-1 (ie, the open->close) returns, is there a way to force TimeReturn to only use close->close (so the first value is nan/0) or would that require basically writing my own wrapper around TimeReturn ?

run-out

@adamb032 When I was reading the docs today it says that the first bar will be open to close if open is available, otherwise not data exist. Not sure if this works but try setting firstopen to False.

- ``firstopen`` (default: ``True``)
        When tracking the returns of a ``data`` the following is done when
        crossing a timeframe boundary, for example ``Years``:
          - Last ``close`` of previous year is used as the reference price to
            see the return in the current year
        The problem is the 1st calculation, because the data has** no
        previous** closing price. As such and when this parameter is ``True``
        the *opening* price will be used for the 1st calculation.
        This requires the data feed to have an ``open`` price (for ``close``
        the standard [0] notation will be used without reference to a field
        price)
        Else the initial close will be used.