importing orders from backtesting.py strategy to create a vbt.Portfolio

[eervin123]

I've read through the example to port strategies from backtrader and was attempting to do the same for backtesting.py however, I'm a bit stumped on the vbt.Portfolio.from_orders() class.

I feel this might be simple but I couldn't quite understand why in your example @polakowo you were using from_signals() instead of from_orders() my hunch was that you were either long or in cash, so it was easier to code using from_signals().

example below:

bt_entries_mask = bt_transactions[bt_transactions.amount > 0]
bt_entries_mask.index = bt_entries_mask.index
bt_exits_mask = bt_transactions[bt_transactions.amount < 0]
bt_exits_mask.index = bt_exits_mask.index

bt_entries = pd.Series.vbt.signals.empty_like(ohlcv['Close'])
bt_entries.loc[bt_entries_mask.index] = True
bt_exits = pd.Series.vbt.signals.empty_like(ohlcv['Close'])
bt_exits.loc[bt_exits_mask.index] = True

vbt.settings.portfolio['fees'] = 0.075 / 100 #0.0025 # in %
bt_pf = vbt.Portfolio.from_signals(ohlcv['Close'], bt_entries, bt_exits, price=ohlcv['Close'].vbt.fshift(1))

The reason I am asking is because the documentation around long entries and long exits versus short entries and short exits is a bit unclear to me.

For example, backtesting.py will produce the following trades dataframe in a long/short strategy. (note, the sample dataset below, appears to enter short upon exiting long, but the full series, has partial entries and exits as well. This is why I was attempting to use the from_orders() which would be more explicit about the size.
Also, of note, is that this is on a series with 4hr candles. I am assuming I could apply the original dataframe index to have a common index.

Size	EntryBar	ExitBar	EntryPrice	ExitPrice	PnL	ReturnPct	EntryTime	ExitTime	Duration
7724	147	949	12.94584	283.98	2.093468e+06	20.936004	2017-02-19 00:00:00+00:00	2017-07-03 04:00:00+00:00	134 days 04:00:00
-7739	949	1044	283.41204	231.95	3.982647e+05	0.181580	2017-07-03 04:00:00+00:00	2017-07-19 00:00:00+00:00	15 days 20:00:00
11151	1044	1045	232.41390	225.06	-8.200334e+04	-0.031641	2017-07-19 00:00:00+00:00	2017-07-19 04:00:00+00:00	0 days 04:00:00
-11173	1045	1047	224.60988	243.50	-2.110593e+05	-0.084102	2017-07-19 04:00:00+00:00	2017-07-19 12:00:00+00:00	0 days 08:00:00
9421	1047	1048	243.98700	228.71	-1.439246e+05	-0.062614	2017-07-19 12:00:00+00:00	2017-07-19 16:00:00+00:00	0 days 04:00:00
...	...	...	...	...	...	...	...	...	...
-305381	10647	10729	4132.86770	4141.81	-2.730809e+06	-0.002164	2021-12-10 08:00:00+00:00	2021-12-24 00:00:00+00:00	13 days 16:00:00
303455	10729	10738	4150.09362	4053.42	-2.933609e+07	-0.023294	2021-12-24 00:00:00+00:00	2021-12-25 12:00:00+00:00	1 days 12:00:00
-304063	10738	10740	4045.31316	4073.17	-8.470234e+06	-0.006886	2021-12-25 12:00:00+00:00	2021-12-25 20:00:00+00:00	0 days 08:00:00
299305	10740	10756	4081.31634	3918.68	-4.867787e+07	-0.039849	2021-12-25 20:00:00+00:00	2021-12-28 12:00:00+00:00	2 days 16:00:00
-299905	10756	10835	3910.84264	3098.00	2.437756e+08	0.207843	2021-12-28 12:00:00+00:00	2022-01-10 16:00:00+00:00	13 days 04:00:00

here is a picture incase the markdown of that table didn't format correctly.

The question is how to convert this dataframe into a vbt.Portfolio.from_orders() or from_signals()

I don't really care which one to use, I just wanted to be able to understand it and explain it to someone else.

The Direction variable appears to be for a strategy not for an order.

{
    "LongOnly": 0,
    "ShortOnly": 1,
    "Both": 2
}

any help would be much appreciated.

I'd be happy to share the notebook for the documentation if it is at all helpful.

Though, I would suspect that you would do a much better job of that.

[polakowo]

@eervin123 the notebook was created by one of the users, I can't really answer the design decisions that were made. Regarding your dataframe with trades, there is an issue you need to solve. If you want to use either from_orders or from_signals, you need to make sure that there is at most one order per bar. At bar 949 you exited at 283.98 and entered at 283.41204, which isn't possible to do with one order. But you can go another way - without simulation at all. Portfolio constructor takes close and order records that were generated during the simulation. Since you already have this info, you need to convert your dataframe into an array of order records and just pass to the Portfolio class. For instance, on order of 7724 for 12.94584 (buy side), then an order of 7724 for 283.98 (sell side), then 7739 for 283.41204 (sell side), and so on.

[eervin123]

Okay I'll play around with this. It does bring up another question. During backtesting is there a way to close a long position on a signal and open a short position on that same signal both on the closing price of the bar? (This is a fairly common approach for simple first pass testing) Or would you need to close the long position at the end of the bar and then open a short position on the open of the next bar?

[polakowo]

You can reverse the position using a single order, yes. Just print the following vbt.Portfolio.from_signals([1, 2], [True, False], [False, True], direction='both').positions.records_readable for illustration.

[eervin123]

Thank you, helpful. I'll dive in a bit before I pepper you with questions.