Some questions about the backtest module

Alphathon
#1

Hi,Mr.

Under the guidence of the Tutorial and your answers in the forums,we successfully created a payout function,but we met some new problems about the backtest function,as is shown below:

port = alphien.portfolio.Portfolio(tkr)

port.payout(payout,data)

port.evaluate()

port.backtest()

The payout funtion has a dataframe input which contains the ohlc data,and a weightmatrix output,just like the example you showed yesterday in other questions in the forums.But we cannot get the backtest function worked,the error is as follows:

---------------------------------------------------------------------------

AttributeError                            Traceback (most recent call last)

<ipython-input-8-4df52dd42d2e> in <module>

1 port = alphien.portfolio.Portfolio(tkr)

2 port.payout(payout,data)

----> 3 port.evaluate()

4 port.backtest()


/mnt/public/IT/Libs/Python/alphien/portfolio/portfolio.py in evaluate(self, startDate, endDate, zoom, period, lastN, computationDataPoints, rollEval, alignAndFillDates)

374         if rollEval is None:

375             self.singleEval(startDate=startDate, endDate=endDate, zoom=zoom, period=period, lastN=lastN, computationDataPoints=computationDataPoints,

–> 376                  rollEval=False, alignAndFillDates=alignAndFillDates)

377         else:

378             self.rollEval(startDate=startDate, endDate=endDate, zoom=zoom, period=period, rollingWindow=rollEval, computationDataPoints=computationDataPoints)


/mnt/public/IT/Libs/Python/alphien/portfolio/portfolio.py in singleEval(self, startDate, endDate, zoom, period, lastN, computationDataPoints, rollEval, alignAndFillDates, composeLogic)

456         #align indicator with the underlying asset

457         if alignAndFillDates:

–> 458             weights = mergeAndFill(self.features.pxs, weights)

459

460         #remove leading NAs


AttributeError: ‘NoneType’ object has no attribute ‘pxs’

Can you tell us what goes wrong and give us some more precise details of this Portfolio module?

#2

Could you give me some more details please, especially the size of “data”, as well as the size of the output from your function “payout”. If you give me the code that generated “data”, even better. Thanks

#3

The input data is a dataframe that contains six columns which are date,ticker and the four ohlc data,the output dataframe is the weightmatrix just like other payout in the Tutorial.

#4

Hi Andrew,

The input is not clear to me. How can a column be ticker? Let’s take an example and say we’re only looking at data for 2016, so around 253 days. Let not include dates in the dimensions as they are the index.

Your input dataframe is 253 x 5 (provided ticker is something like ‘bb_live’). Or it is 253 x (5*number of tickers)?

If you create a portfolio like:

pf = alphien.portfolio.Portfolio([‘AMZN.US Equity’, ‘AAPL.US equity’])

The portfolio pf has 2 assets. So your payout has to have 2 columns. If we backtest for 2016, the payout should be number_of_rebalancing_dates x 2, each column being the allocation for AMZN and AAPL respectively. Your number of rebalancing dates have to be less than 253 (you could rebalance only once a month for e.g., and you’d get 12 x 2 for 2016.


To summarise, your portfolio function can take an input data frame of any dimension (columns) but should output a dataframe of exactly “number_of_assets_in_the_portfolio” columns. The number of row is capped by the number of dates in your backtest.


Hope this help.

#5

My input data has the form like this:

date	asset	open_price	close_price	high_price	low_price

0	2007-01-03	A.US Equity	34.9792	34.30	35.480	34.05

1	2007-01-04	A.US Equity	34.2894	34.41	34.600	33.46

2	2007-01-05	A.US Equity	34.2894	34.09	34.400	34.00

3	2007-01-08	A.US Equity	33.9696	33.97	34.080	33.68

4	2007-01-09	A.US Equity	34.0695	34.01	34.320	33.63

…	…	…	…	…	…	…

1491069	2016-12-23	ZTS.US Equity	53.2000	53.78	53.780	53.18

1491070	2016-12-27	ZTS.US Equity	53.7800	53.72	54.150	53.65

1491071	2016-12-28	ZTS.US Equity	53.7600	53.44	53.850	53.35

1491072	2016-12-29	ZTS.US Equity	53.4747	53.62	53.779	53.40

1491073	2016-12-30	ZTS.US Equity	53.6400	53.53	53.740	53.27

And my output weightmatrix is:

asset	A.US Equity	AA.US Equity	AAL.US Equity	AAP.US Equity	AAPL.US Equity	ABBV.US Equity	ABC.US Equity	ABK.US Equity	ABT.US Equity	ACAS.US Equity	…	XOM.US Equity	XRAY.US Equity	XRX.US Equity	XTO.US Equity	XYL.US Equity	YHOO.US Equity	YUM.US Equity	ZBH.US Equity	ZION.US Equity	ZTS.US Equity

date																					

2007-01-03	0.0	0.0	0.0	0.00	0.00	0.0	0.0	0.00	0.0	0.0	…	0.00	0.0	0.00	0.00	0.0	0.00	0.0	0.00	0.00	0.0

2007-01-04	0.0	0.0	0.0	0.00	0.00	0.0	0.0	0.00	0.0	0.0	…	0.00	0.0	0.00	0.00	0.0	0.00	0.0	0.02	0.00	0.0

2007-01-05	0.0	0.0	0.0	0.00	0.00	0.0	0.0	0.00	0.0	0.0	…	0.00	0.0	0.00	0.02	0.0	0.00	0.0	0.00	0.00	0.0

2007-01-08	0.0	0.0	0.0	0.00	0.02	0.0	0.0	0.00	0.0	0.0	…	0.02	0.0	0.00	0.02	0.0	0.00	0.0	0.00	0.00	0.0

2007-01-09	0.0	0.0	0.0	0.00	0.02	0.0	0.0	0.02	0.0	0.0	…	0.00	0.0	0.00	0.00	0.0	0.00	0.0	0.00	0.00	0.0

…	…	…	…	…	…	…	…	…	…	…	…	…	…	…	…	…	…	…	…	…	…

I am sure the number of columns and the number of assets are matched,if not,it will raise another error different from the Traceback.

#6

I see. So the input format is different from the expected format.

The input dataframe should have dates as indices; here your index is not a date. This is also why I was confused with the fact that there is a ticker column. With a date as index, you won’t be able to have duplicated dates as row: a row is a single date. This is quite important as our backtesting engine will align all dates and handle non trading days.

Instead of having input dimensions = ( nb_days * nb_tickers ) x ( 6 ), try to have ( nb_days ) x ( number_of_tickers * 4 ) or 5 as you may want to include adjusted price ‘bb_live’ in here. This is natively returned by getHistoryData. If the variable tkr is a list of tickers, this call returns the expected format:

px = alphien.data.getHistoryData(tkr, field=[‘ohlc’, ‘bb_live’])

With this as input and the weight matrix as you generate it, dates will align well and the backtest will work.