How to set a benchmark?

[shiosai]

When having a portfolio with some trades I can easily calculate stats and some metrics like alpha. Both are using the underlying data as default benchmark but how if I want to set my own - e.g. like the SPY?
I see benchmark_rets but what would I supply exactly? Lets assume I have the SPY close series with exactly the same length as the portfolio data - would I supply it as is? Do I have to do any normalization or similar?

[polakowo]

@shiosai, yes you can pass SPY returns as benchmark_rets. You can use spy.vbt.to_returns() to convert price series to return series before passing. There is no normalization needed. Although vectorbt additionally forward and backward fills NaNs before calculating returns, so you can do the same using price.ffill().bfill() with pandas.

[shiosai]

Perfect - at one point I tried spy.vbt.returns() but it didn't come to me to use spy.vbt.to_returns()

[polakowo]

vbt.returns returns an accessor for working with returns and expects the data to be already a returns series, so beware of this. It's similar to vbt.signals.

[syedshahab698]

where exactly do we need to pass this parameter benchmark_rets? sorry I am new to vector bt.

[syedshahab698]

Thank you. This is helpful.

Also if I am using multiple assets which starts at different dates. how is the default benchmark calculated?

[polakowo]

@syedshahab698 multiple starting dates mean that some assets have starting price of nan. By default, portfolio forward and then backward fills nan values. Backward filling nans in this case is safe and doesn't expose you to look-ahead bias because it's applied only during the analysis stage. Once asset price is set, portfolio takes your initial capital and applies asset price to it, which becomes benchmark value (pf.benchmark_value()). Benchmark returns (pf.benchmark_returns) are then calculated based on benchmark value.

df = pd.DataFrame({'a': [10, 11, 12, 13], 'b': [np.nan, np.nan, 12, 13]})
print(vbt.Portfolio.from_orders(df).market_returns())
          a         b
0  0.000000  0.000000
1  0.100000  0.000000
2  0.090909  0.000000
3  0.083333  0.083333

[eervin123]

@polakowo thank you for this wonderful project. I'm just beginning to dig in and work with the examples in the documentation. I have a quick question regarding the above thread:

Is it correct to assume that there is no way to set a default custom benchmark in the beginning when I establish a portfolio or strategy?

I see and understand that we are required to pass benchmark_rets into the various methods every time we want to compare to our custom benchmark?

Please forgive me if I'm overlooking something. I have figured out how to evaluate stats vs. my custom benchmark, but I have to type in (settings=dict(benchmark_rets=...)) every time. It would be nice if I could set this up early on as a default benchmark for the portfolio or strategy.

[polakowo]

@eervin123 that's a great idea. Let's make a benchmark argument in Portfolio.

Edit: for stats you can also set benchmark_rets globally:

vbt.settings['portfolio']['stats']['settings']['benchmark_rets'] = my_benchmark_rets

But you would need to take care of grouping if the portfolio is grouped. Having a benchmark argument of the same shape as close that does this automatically would be probably more convenient.

[eervin123]

awesome, thank you @polakowo . I wish I could contribute a PR but my coding sucks. I will promise to help with documentation and other areas that you might need help. Like I said, I just jumped into it but as I go deeper I'm sure I'll find other areas where I can lend a hand.