This post is meant to just outline my attempt at trying to 'rescue' my strategy 'Warren Buffett on the Move' from too much overfitting. Or reduce the likelihood of overfitting at least hopefully. I've 'walked back' some of the factors that I believe may be overfit, and retrained the model on the original training period.

Model Setup:
$10MM starting capital with Quantopian's default slippage and commission cost model.

Economic Hypothesis:
Buying 'great companies at reasonable prices' once they are trending higher. Shorting the opposite.

Training and OOS testing periods w/ average Sharpe:

Training: 01-01-2011 -- 01-01-2017 | Sharpe: 3.36

1st OOS test: 02-01-2017 -- 11-01-2018 | Sharpe: 1.98
2nd OOS test: 02-02-2005 -- 31-12-2010 | Sharpe: 3.04
3rd OOS test: 12-01-2018 -- 15-01-2019 | Sharpe: 2.44

The first OOS test period should be taken with a big grain of salt, as I had previously used this data to train the previous model on. The second and third OOS periods however I believe can be considered more reliable as I had only used them once each to test on previously, so I consider them still reasonably ok to test on again one more time.

In short, it's most likely still a bit overfit, but I think (hope) the second and third OOS tests indicate that the strategy is now at least less overfit than the original one. My plan is to leave this strategy running in the contest for at least 3 months, as the final (true) OOS test. Time will tell.

Attached is the tearsheet with the training period and 1st OOS test. 2nd and 3rd OOS test tearsheets to follow.