Why use natural date instead of bar date?

[none2003]

Hi

I'm testing WeighInvVol algo，and I found it uses natural date instead of bar date, for example, I have following data

2023-12-25 1.042 1.117 1.064 4.634 2.299 106.870
2023-12-26 1.041 1.121 1.105 4.647 2.308 106.873
2023-12-27 1.038 1.128 1.072 4.660 2.335 106.806
2023-12-28 1.034 1.128 1.058 4.662 2.310 106.627
2023-12-29 1.030 1.130 1.051 4.648 2.268 106.562
2024-01-02 1.030 1.136 1.064 4.666 2.299 106.678
2024-01-03 1.023 1.146 1.047 4.663 2.286 106.835

the strategy is

s1 = bt.Strategy('Inv Vol',
                 [bt.algos.RunMonthly(run_on_first_date=False, run_on_end_of_period=False, run_on_last_date=False),
                  bt.algos.SelectAll(),
                  bt.algos.WeighInvVol(lookback=pd.DateOffset(days=5), lag=pd.DateOffset(days=1)),
                  bt.algos.Rebalance()])

according to code of WeighInvVol:

prc = target.universe.loc[t0 - self.lookback : t0, selected]

When t0 is '2024-01-01', the data range will be '2023-12-27':'2024-01-01', so the prc will be:

2023-12-27 1.038 1.128 1.072 4.660 2.335 106.806
2023-12-28 1.034 1.128 1.058 4.662 2.310 106.627
2023-12-29 1.030 1.130 1.051 4.648 2.268 106.562

clearly, it uses natural date.

If use bar date, since '2024-01-02' is the first available date of January, so 't0' should be '2024-01-02', and the prc calculation could be:

prc = target.universe[target.universe.index < t0][selected].tail(self.lookback)

then prc will be:

2023-12-25 1.042 1.117 1.064 4.634 2.299 106.870
2023-12-26 1.041 1.121 1.105 4.647 2.308 106.873
2023-12-27 1.038 1.128 1.072 4.660 2.335 106.806
2023-12-28 1.034 1.128 1.058 4.662 2.310 106.627
2023-12-29 1.030 1.130 1.051 4.648 2.268 106.562

I think this make more sense.

Another problem of using natural date is, the data range of every period varies, "t0 - self.lookback : t0" could give range from 3 to 5, this inconsistency is detrimental to subsequent calculations of inverse volatility.