Commodity Futures Hierarchical Data Structure

Question

I for the life of me cant seem to get the structure I want and have it function properly, so in a fit of rage I come to you guys.

Setup: I have a Directory called Futures_Contracts and inside is about 30 folders all named with the underlying asset, and finally inside the 6 nearest expiration contracts in csv format. Each csv is identical in format and contains Date,O,H,L,C,V,OI,Expiration Month.

Note: O H L C V OI is open, high, low, close, volume, open interest (for those not familiar) also assume close is synonymous with settlement below

Task: From here the goal is to load in the futures data into a multi-index pandas dataframe in such a way that the top-level index is the underlying commodity symbol, the mid-level index is the expiration Month-Year, and finally the OHLC data. The end goal is to have something that I can start hacking at the zipline module to get it running on futures. So visually:

My Feeble attempt:

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from pandas import DataFrame, Series
import datetime
plt.figsize(16,8)

deliveries = {}
commoidities = {}
columns = 'open', 'high', 'low', 'settle', 'volume', 'interest', 'delivery' #Contract fields
path = os.getcwdu()+'/Futures_Contracts/' #Futures Path
for sym in os.listdir(path):
    if sym[0] != '.': #Weed out hidden files
        deliveries[sym] = []
        i = 0
        for contract in os.listdir(path + sym):
            temp = pd.io.parsers.read_csv(path + sym + '/' + contract, index_col=0, parse_dates = True, names = columns)#pull in the csv
            deliveries[sym].append(str(contract[:-4][-1] + contract[:-4][:-1][-2:])) #add contract to dict in form of MonthCode-YY
            commodities[sym] = deliveries[sym]
            commodities[sym][i] = temp
            i += 1

This somewhat works, however this is really a nested dict that holds a dataframe at the end. Therefore slicing is extremely clunky:

commodities['SB2'][0]['settle'].plot()
commodities['SB2'][3]['settle'].plot()
commodities['SB2'][4]['settle'].plot()
commodities['SB2'][3]['settle'].plot()
commodities['SB2'][4]['settle'].plot()
commodities['SB2'][5]['settle'].plot()

and yields

Optimally I will be able to slice across each of the indexes so that I can compare data across assets, expiration, date and value. Furthermore label what I am looking at, as you can see in the matplotlib chart everything is simply named 'settle'

There is surely a way to do this, but I am just not smart enough to figure it out.

Answer 1

I think you're going to be much better off getting this into one DataFrame, so consider using a MultiIndex. Here's a toy example, which I think will translate well to your code:

In [11]: dfN13 = pd.DataFrame([[1, 2]], columns=[['N13', 'N13'], ['a', 'b']])

In [12]: dfM13 = pd.DataFrame([[3, 4]], columns=[['M13', 'M13'], ['a', 'b']])

These are the DataFrames in your example, but the column's first level it just the asset name.

In [13]: df = pd.concat([dfN13, dfM13], axis=1)

In [14]: df
Out[14]:
   N13     M13
     a  b    a  b
0    1  2    3  4

For convenience we can label the columns-levels and index.

In [15]: df.columns.names = ['asset', 'chart']

In [16]: df.index.names = ['date']  # well, not in this toy example

In [17]: df
Out[17]:
asset  N13     M13
chart    a  b    a  b
date
0        1  2    3  4

Note: This looks quite like your spreadsheet.

And we can grab out a specific chart (e.g. ohlc) using xs:

In [18]: df.xs('a', level='chart', axis=1)
Out[18]:
asset  N13  M13
date
0        1    3

In [19]: df.xs('a', level='chart', axis=1).plot()  # win