Multiple assets, multiple trade signals per asset

[Sanjan611]

With my understanding of vectorbt and group_by so far, I've worked out how to build a Portfolio for

(i) single asset, single trading signal
(ii) single asset, multiple trading signals
(iii) multiple assets, single trading signal

I'm now trying to work out how to have multiple assets and multiple trading signals with a Portfolio (I'm not sure if this is within or outside the functionalities of vectorbt or if it's really pushing it!).

Setup
I've got two assets AAPL and AMZN. I've defined a few trading rules (eg. based on golden cross, RSI, etc), and using those I've got a total of n trading signals (entries and exits) for two assets. So in my case, each asset has n/2 trading signals associated with it.

assets_close     # A DataFrame with two columns with close price. Keys = ['AAPL', 'AMZN']
entries          # A DataFrame with n columns. 
exits            # A DataFrame with n columns. 

(I'm not sure what the keys for my entries and exits should be)

When I build the Portfolio, how can I get the backtesting stats per asset (and so across all trading rules) or per trading rule (and so across all assets)?

What I've noticed
If I use a single trading rule and so my entries parameter has two columns (one for each asset) -

pf = vbt.Portfolio.from_signals(
    close = assets_close,
    entries = entries, # two columns, keys = [1, 2]
    exits = exits,     # two columns, keys = [1, 2]
    group_by = ['AAPL', 'AMZN']
) 

I'm able to access pf['AAPL'] and pf['AMZN], but I don't know which of the two columns from entries and exits it used for the backtesting!

[polakowo]

@Sanjan611 here's an example:

import pandas as pd
import vectorbt as vbt

price = pd.DataFrame({
    'p1': [1, 2, 3, 4],
    'p2': [5, 6, 7, 8]
})
price.columns.name = 'asset'
entries = pd.DataFrame({
    'en1': [True, False, False, False],
    'en2': [False, True, False, False],
    'en3': [False, False, True, False],
    'en4': [False, False, False, True]
})
entries.columns.name = 'entries'
exits = pd.DataFrame({
    'ex1': [False, False, False, True],
    'ex2': [False, False, False, True],
    'ex3': [False, False, False, True],
    'ex4': [False, False, False, True]
})
exits.columns.name = 'exits'
entries = entries.vbt.stack_index(pd.Index(['p1', 'p1', 'p2', 'p2'], name='asset'))
exits = exits.vbt.stack_index(pd.Index(['p1', 'p1', 'p2', 'p2'], name='asset'))
portfolio = vbt.Portfolio.from_signals(price, entries, exits)  # not grouped portfolio
print(portfolio.total_return())
entries  asset  exits
en1      p1     ex1      3.000000
en2      p1     ex2      1.000000
en3      p2     ex3      0.142857
en4      p2     ex4      0.000000
Name: total_return, dtype: float64
print(portfolio.total_return(group_by='asset'))  # group not grouped portfolio
asset
p1    2.000000
p2    0.071429
Name: total_return, dtype: float64
# or
portfolio = vbt.Portfolio.from_signals(price, entries, exits, group_by='asset')  # grouped portfolio
print(portfolio.total_return())
asset
p1    2.000000
p2    0.071429
Name: total_return, dtype: float64
print(portfolio.total_return(group_by=False))  # ungroup grouped portfolio
entries  asset  exits
en1      p1     ex1      3.000000
en2      p1     ex2      1.000000
en3      p2     ex3      0.142857
en4      p2     ex4      0.000000
Name: total_return, dtype: float64

Broadcasting in Portfolio.from_signals is internally done using https://polakowo.io/vectorbt/docs/base/reshape_fns.html#vectorbt.base.reshape_fns.broadcast, which can broadcast just like NumPy + it can broadcast if some inputs contain a subset of column levels of other inputs (as in the example above).

[polakowo]

@bibinvargheset It’s all explained in the documentation.

[forLinDre]

@polakowo, I am also attempting to group by portfolio by asset, or in my case the column heading 'symbol'. I have tried specifying group_by in the portfolio like so:

`RMA_strat_pf = vbt.Portfolio.from_signals(
open = Open_wobuf,
high = High_wobuf,
low = Low_wobuf,
close = Close_wobuf,
group_by = 'symbol',

entries = RMA_long_strat_entries,

...
`

I also tried to specify the group in a statistic like total_return with RMA_strat_pf.total_return(group_by = 'symbol'). Both methods results in an error shown below. Could you please hint to what I might be doing wrong? My portfolio wrapper columns are show below the error.

`

ValueError Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_17932/2931782157.py in
----> 1 RMA_strat_pf.total_return(group_by = 'symbol')

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\utils\decorators.py in wrapper(instance, *args, **kwargs)
442 # If not, do not invoke lru_cache
443 return func(instance, *args, **kwargs)
--> 444 return cached_func(*args, **kwargs)
445
446 def clear_cache(instance):

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\utils\decorators.py in partial_func(*args, **kwargs)
412 def partial_func(*args, **kwargs) -> tp.Any:
413 # Ignores non-hashable instances
--> 414 return func(instance, *args, **kwargs)
415
416 _func = None

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\portfolio\base.py in total_return(self, group_by, wrap_kwargs)
4384 wrap_kwargs: tp.KwargsLike = None) -> tp.MaybeSeries:
4385 """Get total profit per column/group."""
-> 4386 init_cash = to_1d_array(self.get_init_cash(group_by=group_by))
4387 total_profit = to_1d_array(self.total_profit(group_by=group_by))
4388 total_return = nb.total_return_nb(total_profit, init_cash)

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\utils\decorators.py in wrapper(instance, *args, **kwargs)
442 # If not, do not invoke lru_cache
443 return func(instance, *args, **kwargs)
--> 444 return cached_func(*args, **kwargs)
445
446 def clear_cache(instance):

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\utils\decorators.py in partial_func(*args, **kwargs)
412 def partial_func(*args, **kwargs) -> tp.Any:
413 # Ignores non-hashable instances
--> 414 return func(instance, *args, **kwargs)
415
416 _func = None

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\portfolio\base.py in get_init_cash(self, group_by, wrap_kwargs)
4233 init_cash = to_1d_array(self._init_cash)
4234 if self.wrapper.grouper.is_grouped(group_by=group_by):
-> 4235 group_lens = self.wrapper.grouper.get_group_lens(group_by=group_by)
4236 init_cash = nb.init_cash_grouped_nb(init_cash, group_lens, self.cash_sharing)
4237 else:

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\utils\decorators.py in wrapper(instance, *args, **kwargs)
442 # If not, do not invoke lru_cache
443 return func(instance, *args, **kwargs)
--> 444 return cached_func(*args, **kwargs)
445
446 def clear_cache(instance):

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\utils\decorators.py in partial_func(*args, **kwargs)
412 def partial_func(*args, **kwargs) -> tp.Any:
413 # Ignores non-hashable instances
--> 414 return func(instance, *args, **kwargs)
415
416 _func = None

~\Anaconda3\envs\strategy_lib_vectorbt\lib\site-packages\vectorbt\base\column_grouper.py in get_group_lens(self, group_by, **kwargs)
272 """See get_group_lens_nb."""
273 if not self.is_sorted(group_by=group_by):
--> 274 raise ValueError("group_by must lead to groups that are coherent and sorted "
275 "(such as [0, 0, 1, 2, 2, ...])")
276 group_by = self.resolve_group_by(group_by=group_by, **kwargs)

ValueError: group_by must lead to groups that are coherent and sorted (such as [0, 0, 1, 2, 2, ...])
`

MultiIndex([(8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), (8, 14, 64, 198, True, True, True, True, 26, -0.02, -0.02, ...), ... (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...), (8, 14, 64, 198, True, True, True, True, 28, 0.0, 0.0, ...)], names=['custom_EWM_window1', 'custom_EWM_window2', 'custom_EWM_window3', 'custom_EWM_window4', 'custom_ewm1', 'custom_ewm2', 'custom_ewm3', 'custom_ewm4', 'custom_change_window', 'custom_RMA1_change_thres', 'custom_RMA2_change_thres', 'custom_entry_st', 'custom_entry_et', 'custom_exit_st', 'custom_sl_stop', 'custom_sl_trail', 'custom_tp_stop', 'custom_RMA1_entry_EMA1_trending_EMA2', 'custom_RMA1_entry_EMA1_trending_RMA1', 'custom_RMA1_entry_EMA2_trending_RMA1', 'custom_RMA2_entry_EMA1_trending_EMA2', 'custom_RMA2_entry_EMA1_trending_RMA2', 'custom_RMA2_entry_EMA2_trending_RMA2', 'symbol'], length=216)

[polakowo]

@forLinDre as the error says, you need to sort your column levels by symbol in order to proceed (['BTC-USDT', 'BTC-USDT', 'ETH-USDT', 'ETH-USDT'] instead of ['BTC-USDT', 'ETH-USDT', 'BTC-USDT', 'ETH-USDT']).

[forLinDre]

@polakowo, kind of a weird requirement for grouping columns in my opinion but I'm sure there is a simple explanation... most importantly, thank you for your quick help resolving the matter!

[polakowo]

You're using grouping in the opposite way than it was designed. Usually, users are grouping by top column levels. But you're grouping by the lowest column level, meaning that each of your groups will not contain different assets but different hyperparameter combinations. I don't know about your intentions much, but yeah, sorting will solve the error.