Pandas/Python 中的矢量化回测器：循环遍历每只股票作为一个新的数据帧，还是将它们全部放在一个数据帧中？

Question

I've been trying to build my own simple vectorized backtester in Pandas/Python to create a simple way to test some trading strategies. I have been using this article as a guide and it has been pretty helpful.

I want to perform a simple portfolio backtest of say 10 stocks/ETFs. For each stock I will have a dataframe which will have a date as a row index and the columns will be the Open, High, Low, Close prices for that date (financial time series data). So I will have say 10 of these dataframes that will 4 columns each. What would be the most pythonic and efficient way to do the backtest:

1. Work on each dataframe separately, by looping through and carrying out my calculations on each dataframe then summing the profits at the end.

OR

2. Concatenating all the dataframes together and just working on the one dataframe

In the example article I have been using, he works with just one dataframe, but he just uses Close price, so when he does this he doesn't need a column multi-index. I would need a column multi-index (level 0 is the stock name, level 1 is the Close, Open, High, Low, etc) and given my beginner pandas status, that's making things complicated for me. I've been thinking it would be easier for me to create a loop and work with 10 separate dataframes, but I'm wondering if this is just lazy and will hinder my development in the long run.

Answer 1

A df of closes is the simplest. You need a multiindex https://pandas.pydata.org/pandas-docs/stable/user_guide/advanced.html to use the other fields. The issue I found with multiindex is that adding columns to it requires some hacking of the df every change.