I got the idea for this trading strategy from an academic paper (http://www.ljmu.ac.uk/Images_Everyone/2nd_revised.pdf). I didn't go through the paper that closely so our implementations are likely very different.
The basic idea is that the strategy would buy the n-worst performers in the market. The worst performers can be found by comparing open price versus prior day's close (overnight performance) or comparing close and open on the same day (intraday performance). The strategy then sells the purchased stocks the next day hoping that the stocks experienced a mean-reversion.
My strategy is slightly different because I'm not sure if it's possible to review and trade overnight performance due to the fact that I can't buy a stock at its open price. So I can only look at intraday stock performance and buy the biggest losers. Also, my universe was the S&P 100 due to limitations in Quantopian.
I bought the 5 biggest losers. One (BIG) caveat is that I set the commission to be 0. I originally tried the strategy with the standard commission of 3 cents a share but the performance was consistently negative. Then, I had switched the strategy to do the opposite (short the biggest losers) and I was seeing similarly consistent negative performance.
I think with such high turnover and working off small corrections on average, any reasonable commission will eat up what little profits there are.
If someone wants to build on this, maybe a longer holding period would yield a better return. Or you can try messing with how many losers you trade or how you allocate capital to each loser.