Submit Alpha

When an alpha is scrutinized by various performance and correlation criteria, it is ready to be submitted into the Alpha DB. The author is required to provide some descriptional information about this alpha; for example, its fitting universe, its category etc. Most importantly, the source code must be re-constructed to facilitate daily update.

Some guidelines to code re-construction:

• Replace orca.alpha.base.BacktestingAlpha by orca.alpha.base.ProductionAlpha
• Replace any data cache dependency by MongoDB APIs
• Remove any redundant or un-used data
• Check for forward-looking bias

Usually data items used in an alpha have considerable overlap with existing alphas of similar categories, Orca will provide a mechanism to group these alpha codes in order and schedule update flexible enought to reduce database pressure.

Example

Research code. File cache.py:

from orca import DATES
from orca.mongo import QuoteFetcher

quote = QuoteFetcher()
returns5 = quote.fetch('returns5', 20090101, DATES[-5])

from orca.data import HDFSaver
hdf = HDFSaver('./.cache')
hdf['returns5'] = returns5

File alpha.py:

from orca.data import HDFLoader

hdf = HDFLoader('./.cache')
returns5 = hdf['returns5']

from orca.alpha import BacktestingAlpha

class AlphaReturns5(BacktestingAlpha):

    def __init__(self):
        super(AlphaReturns5, self).__init__()
        self.returns5 = returns5.shift(1)

    def generate(self, date):
        self.alphas[date] = -self.returns5.ix[date]

To submit, these two files should be merged and re-constructed as:

from orca.mongo import QuoteFetcher
from orca.alpha import ProdcutionAlpha

class AlphaReturns5(ProdcutionAlpha):

    def __init__(self):
        super(AlphaReturns5, self).__ini__()
        self.quote = QuoteFetcher(datetime_index=True, reindex=True)

    def generate(self, date):
        returns5 = self.quote.fetch_daily('returns5', date, offset=1)
        return -returns5