Another day, and another batch of algorithms added to the Contextuall system. Our analysts upgraded our S&P500 signals, building 40 different algorithms to predict the direction of the stock market over the next 10 trading sessions.
We used a supervised learning process to train our machine learning algorithms to predict the direction of the S&P500 index.
The training sets were made up of closing prices between April 2008 and the current day, adjusted for dividends and splits.
Each historical observation was either labeled "Up" (if the index increased in value over the next 10 trading session), or "Down" (if the index decreased in value over the next 15 trading session).
In other words:
1. If SP500(n+10) > SP500(n) --> Label "Up"
2. If SP500(n+10) < SP500(n) --> Label "Down"
To backtest the accuracy of each model, we used an in-sample vs. out-sample testing procedure. This meant that we only selected algorithms that remained stable and accurate when presented with fresh, previously unseen data.
Historical Accuracy: Majority Vote (All Algorithms)
By following the majority vote of all the algorithms described above, Contextuall's system of algorithms would have achieved an accuracy rate of 77.66% since 6/22/2011.
In other words, by simply following the directional signal of the majority of our algorithms since 6/22/2011, Contextuall would have correctly predicted 77.66% of the up/down directions of the S&P500 index over the next 10 trading sessions. (More detailed backtesting is included below in an embedded spreadsheet)
Current Signal: Up (Majority Vote, All Algorithms)
Based on closing values from the August 28 close, 30 out of our 40 algorithms (75%) project gains for the S&P500 index over the next 10 trading sessions.
Here is a graphic representation of the majority vote trends over the last year.