Use Student's T to totally overfit expected stock returns

[willb]

This commit changes the simulation to use Student's T instead of a normal distribution. Specifically:

1. instead of calculating mean and variance, we fit a Student's T distribution to observed stock returns,
2. each simulation step samples from the appropriate T distribution to determine predicted returns

Note that the worker image must have scipy version 0.17.1 installed.