Volatility isn't all that predictive apparently.

https://www.quantopian.com/posts/2016-7-30-nyc-workshop#

no factor tear sheet it didnt work

Workflow Checklist for Factor Modeling

• Come up with hypothesis for whether a factor might be predictive of
  returns. Obtain and clean data for that factor. Run single-factor
  analysis using alphalens and determine whether it’s a viable factor.
  
• Once viability has been determined, check if factor is correlated
  with a list of known factors and factors already present in your
  model. If your new factor can be completely described as a weighted
  sum of existing factors, then it’s not that interesting. If your
  factor has new information, then add it to your set of existing
  factors.
  
• Develop a model for weighting each factor in your final
  ranking scheme. Weightings could be equal, volatility weighted, or
  based on something more sophisticated like a adaboost or other signal
  boosting/expert weighting algorithms.
  
• Put your new and improved
  multi-factor model in an algorithm along with the weighting scheme if
  it’s more than a fixed or equal weighting.
  
• Use portfolio construction
  techniques to ensure that the trades produced by your multi-factor
  ranking scheme do not contain excess risk. Things to consider
  include: CVaR, sector exposure, risk factor exposure. Generally you
  will want to create a list of longs and shorts based on your
  multi-factor ranking scheme, and then set the weights based on
  portfolio optimization techniques as described above to ensure you
  don’t have excess exposure to any given factor.
  
• Put the algorithm up
  to trade on paper for a little while. A higher frequency rebalancing
  algorithm will need less time to accumulate the same confidence in
  future performance as an algorithm that rarely trades. On the other
  hand, be careful not to paper trade it until the alpha you originally
  found is old news and exhausted.
  
• Do constant risk monitoring of the
  algorithm to determine if alpha is decaying or risk exposures and
  returns are outside of forecasted limits.
  
• Put algorithm to trade
  live.
  
• Continue risk monitoring.
  

Delaney,

Thank you for the great workshop today.

Attached is a backtest I finally got to work after getting the Long/Short algo to filter only on the Basic Materials Sector. The factor research I quickly did during the workshop showed that 'Retained Earnings to Assets' and 'Dividend Growth' had some promise in picking longs and shorts in Basic Materials over the 2014 calendar year. It's a very small portfolio of stocks in this sample portfolio, but I thought I would share with the class anyway. It was good to see how factor research matched expected portfolio results in a backtest.

Thanks again.

Hugh

Factor Tear Sheet Attached...

Backtest results...