Hi Maksim,

You may know that detailed analysis of full backtests is possible in Research through Pyfolio. Here is a post introducing the feature. Note that the result of get_backtest is assigned to a variable bt; this object has some attributes that you may find useful (check them out by running dir(bt)).

Specifically, bt.positions is a DataFrame containing data on all the positions the algorithm held during the backtest. The cost_basis column tells you the average price paid for the position, and the last_sale_price column tells you the price at which the last share of that position is sold/covered. By taking the difference of these two columns you can get the returns of every position. Let me know if this works for you!

Hi Nathan,

thanks for the replay. I tried it for the simplest algorithm and I'm confused.
bt.positions - is empty for my backtest but bt.transactions is valid.
I tried to use bt. transactions to compute return. But I got a return that different from the result of standard qbacktest.

My notebook.

Hi Maksim,

bt.positions is empty because your algorithm is only day trading; the positions are only recorded at the end of each day.

Your returns computation using bt.transactions has a couple problems with how the commissions are applied. Here's a corrected version:

def set_commistion(x):  
    return max(abs(x), 1.)  
#because int bt.transactions commision is None, I have to compute it by myself  
tr['_commis']  = tr['amount']*0.0075  
tr['_commis'] = tr['_commis'].apply(set_commistion)  

Basically, we need to put abs(x) instead of x because commissions are positive whether you're buying or selling. And .apply() doesn't work inplace; you have to reassign the result.

So once you make that change the returns calculation should be correct.

Thank you Nathan! It's bug in my computation, I will fix it and try it again. I'll share the results in this thread.