Can someone help me?

@Vladimir, @Dan Whitnable, @Grant Kiehne

Thank you for your help in advance.

The behavior you are seeing stems from the fact that resample by more than a 1W increment is ambiguous. Resampling 'W-FRI' is easy. It will fetch every Friday. However, resampling '2W-FRI' actually needs one more piece of information. It needs to know which Friday to start counting. Pandas doesn't have an explicit way of setting the 'starting Friday'. So, if one starts resampling with '2W-FRI' this week it will choose one set of Fridays. However, if one starts resampling next week it will choose a different set of Fridays. Unfortunately, one cannot even use 'even' or 'odd' Fridays. Because the last Friday of the year can fall in either the 51, 52, or 53 week, every other Friday is not always even or odd.

So, a workaround is to first define all the possible Fridays one wants to resample on. Use the date_range method something like this

import pandas as pd  
context.resample_dates = pd.date_range(start='1/1/2001',  
                                           end='1/1/2021',  
                                           freq='2W-FRI',  
                                           tz='UTC')

The 'starting Friday' can be set by choosing a different 'start' date. Ensure the 'end' date includes all the resample dates in the backtest. (The attached algo uses the get_environment method to determine this`) .

Now, rather than resample simply select just those dates from close which are in the 'resample_list'. Maybe like this

    close = data.history(context.signal, 'close', 30, '1d')  
    # Use `dropna` to narrow the dates to just those in 'close'.  
    prices_biweek_close = close[context.resample_dates].dropna()

That 'workaround' should resolve the dates abruptly changing issue. However, if markets weren't open on one of the desired Fridays, there won't be any data. This is different from the behavior of the resample method which would fetch the last known price.

There are ways to get prices if markets weren't open but they introduce some other anomalies. Is there a particular reason you wish to check specific 'every other Fridays'?

@Dan Whitnable

Thank you very much for your help. I don't have a reason for checking specific 'every other Fridays'. My aim is to get the ATR data in a biweekly fashion, which would include biweekly close, high and low prices from the past. Say past 20 biweekly HLC price to calculate the past 20 biweekly ATR.

I am actually OK with the data change abruptly on Monday or Friday. It is just weird to me, with the reshape function the biweekly data changes during the week (e.g. Thursday).

Meanwhile, I also tried to use weekly resample for close price. This also generate issues.

For example, if you run the below test from 01/01/2020 to 04/29/2020, in the log you will get the log on 2020-01-14:

2020-01-14 15:59 myfunc:24 INFO SPY weekly close = 2019-08-23 00:00:00+00:00 282.114
2019-08-30 00:00:00+00:00 289.692
2019-09-06 00:00:00+00:00 295.249
2019-09-13 00:00:00+00:00 298.200
2019-09-20 00:00:00+00:00 296.671
2019-09-27 00:00:00+00:00 293.915
2019-10-04 00:00:00+00:00 292.930
2019-10-11 00:00:00+00:00 294.811
2019-10-18 00:00:00+00:00 296.522
2019-10-25 00:00:00+00:00 300.164
2019-11-01 00:00:00+00:00 304.682
2019-11-08 00:00:00+00:00 307.448
2019-11-15 00:00:00+00:00 310.304
2019-11-22 00:00:00+00:00 309.478
2019-11-29 00:00:00+00:00 312.742
2019-12-06 00:00:00+00:00 313.280
2019-12-13 00:00:00+00:00 315.728
2019-12-20 00:00:00+00:00 321.120
2019-12-27 00:00:00+00:00 322.860
2020-01-03 00:00:00+00:00 322.440
2020-01-10 00:00:00+00:00 325.700
2020-01-17 00:00:00+00:00 327.350

However, the log you will get on 2020-03-25 will be:
2020-03-25 15:59 myfunc:24 INFO SPY weekly close = 2019-11-01 00:00:00+00:00 302.906
2019-11-08 00:00:00+00:00 305.657
2019-11-15 00:00:00+00:00 308.496
2019-11-22 00:00:00+00:00 307.675
2019-11-29 00:00:00+00:00 310.920
2019-12-06 00:00:00+00:00 311.454
2019-12-13 00:00:00+00:00 313.888
2019-12-20 00:00:00+00:00 319.248
2019-12-27 00:00:00+00:00 320.978
2020-01-03 00:00:00+00:00 320.561
2020-01-10 00:00:00+00:00 323.802
2020-01-17 00:00:00+00:00 330.075
2020-01-24 00:00:00+00:00 326.754
2020-01-31 00:00:00+00:00 319.875
2020-02-07 00:00:00+00:00 330.264
2020-02-14 00:00:00+00:00 335.632
2020-02-21 00:00:00+00:00 331.576
2020-02-28 00:00:00+00:00 292.157
2020-03-06 00:00:00+00:00 295.677
2020-03-13 00:00:00+00:00 269.351
2020-03-20 00:00:00+00:00 228.950
2020-03-27 00:00:00+00:00 246.220

Please note the week close data for the week of 2020-01-10 (marked as bold and italic). Somehow it changes during weeks even in the same test. This also is very confusing, which makes neither weekly nor biweekly resample function work for me.

All prices fetched via the data.history method are adjusted as of the current simulated day in the backtest. So, the adjusted price for 2020-01-10, fetched on 2020-03-25, will include the $1.4056 dividend (ex date 2020-01-10). The adjusted price for 2020-01-10, fetched on 2020-01-14 is not adjusted for this (future) dividend. Generally, one cannot save and compare price/volume data fetched on different days because of splits and dividends.

Comparisons will be correct if you do not compare to saved values.

@Dan Whitnable

Yes Yes, you are right. I did not notice the price adjustment. Thank you for mentioning that.

To follow up on the biweekly close price method. I guess the method you mentioned above is fetching the biweekly date and use those dates to get the biweekly close price, right? Is there any way to get a biweekly high/low price data?

Thank you for your time :-)

Probably the best method to get high/low data is use the resample method but, instead of last, use max or min. Also fetch the associated 'high' and 'low' data. That will get you the highest high and the lowest low for the 2 weeks.

One thing to remember is the latest period (ie the latest 2 weeks or portion of 2 weeks) will be resampled over a lesser period (less than 2 weeks). The value for that last period my also change every day. For example, if every day there is a new high, then that last 2W value will change every day.

Hi Dan

Thank you very much for the help. I'll try this method.

Thank you again.

This is my strategy to solve the problem. Hope it can help others who have the same goal. :-)