Here's a fun experiment I tried using news sentiment, earnings avoidance, and a long-short equity strategy. For those who don't know what long-short strategies are, long-short equity strategies typically contain two main components: a ranking methodology and a long/short side.
“Long-short equity strategies are an incredibly robust family of
strategies. They depend on a methodology to rank equities, and perform
proportionally to how well the ranking system differentiates high and
low future returns. They avoid many forms of statistical bias and
noise, and are an excellent way to make money off a model that
predicts future returns for any given asset.” – Delaney, Quantopian
Engineer
Using James Christopher's news sentiment based long-short equity strategy I added an earnings call avoidance as well as a faster average sentiment calculation. I compared the performance of the algorithm with and without the earnings call avoidance and here are the results:
alpha increased by 0.0043625093271, a 17.06% improvement
returns increased by 0.0101466014443, a 17.47% improvement
sharpe increased by 0.0958536519917, a 49.00% improvement
volatility decreased by 0.00216994577249, a 2.46% improvement
max drawdown increased by 0.0159861238474, a 15.13% decrease Stock returns around earnings announcements are much more volatile than the daily average, by removing them from the strategy I was able to reduce volatility. The volatility reductions and returns improvments are greater if you use a 15 day before/after earnings avoidance, but I stuck with 7 days because this is a weekly strategy.
Full List of Changes:
- Rebalance period was changed from 1 month to 1 week
- Screened out all securities with a 7-day before/after earnings announcement period.
- Used a 7 day look-back window for Accern's News Sentiment in order to have a more up-to-date news sentiment average.
You can test out this strategy yourself by first heading to Quantopian Data at the links below to get the free sample versions of each dataset:
