How We're Using AI to Enhance ProfitSource (Part 1)
In this article, the first of a several part series, we’ll share an overview of our Artificial Intelligence (AI) project that is arguably the biggest enhancement for ProfitSource ever
We‘ll be explaining how we’ve been integrating AI into the ProfitSource trade selection mechanism… ultimately resulting in each Elliott Wave trade setup being accompanied by a probability score which denotes the likelihood of the pattern hitting is target based on the AI’s analysis!
ProfitSource has been generating Elliott Wave signals since 2005, in fact over 71,000 so far! We provided this information to our AI algorithm, plus a range of other information about each stock and the market, to see what it could teach us about trade selection based upon these signals.
Clearly no approach to trading will generate 100% winning trades, far from it in fact, however some traders seem to have a sense for which trade set-ups to enter and which of those to avoid.
Our AI project was aimed at trying to uncover which are the factors in any Elliott Wave trade set-up that are most likely to result in the markets hitting the projected target range generated by ProfitSource.
So what were the “Headline” Results?
In simple terms, the AI was able to identify trading setups that have a greater than 65% probability of hitting the forecast Elliott Wave level. That level of probability applies to about 6% of Elliott Wave trade setups on average.
In trading terms, this means that you should be able to use these high-probability trade setups so that 2 out of every 3 trades is a winner on average… potentially even better, depend on how you managed the trades. That’s pretty exciting!
Looking at US Markets over the past 15 years, ProfitSource has generated approximately 4,700 potential trade setups each year on average. If 6% can now be identified as high probability, that’s approximately 270 a year or roughly 1 high-probability opportunity on average per day. For traders this is a great outcome which gives a very manageable number of trades to work with.
A Range of Interesting Outcomes and Learnings
As you might imagine, there was a huge amount of information that came out of this project and some very interesting learnings. At a high level some of the things we found really insightful were:
A) Bullish and Bearish Elliott Wave Patterns are Influenced by Different Factors – Most people would typically think of Elliott Wave analysis as being the same in both Long and Short directions. However, while we still count an Elliott Wave Impulse or Corrective pattern using the same rules and guidelines, only reversed in terms of directions, for both Bullish and Bearish patterns, our AI found that the factors that were most aligned to a high-probability outcome were different for Bullish and Bearish patterns.
We’ll discuss the most important factors for various Elliott Patterns in the next few articles in this series.
B) Bearish Patterns are Harder to Predict than Bullish Patterns – Although this probably isn’t surprising to traders who’ve spent time in the market, it was interesting to note that our AI found the number of Bearish Patterns that would reliably hit their forecast targets was a substantially smaller proportion than Bullish Patterns. This doesn’t mean that overall Bearish Patterns weren’t likely to hit their projected targets, but rather that finding metrics that allowed the AI to identify high probability candidates was more difficult. Real world experience tells us that when markets fall, they fall sharply and quickly, often in relation to unexpected news events (take the recent corona virus as an example). In contrast to this, we believe the fact that Bullish patterns develop over longer periods of more orderly market activity provides more opportunity for the AI to identify factors commonly associated with patterns having a high probability of hitting their targets.
C) Common Beliefs about Elliott and Fibonacci Were Confirmed – It’s important to realize that our AI wasn’t a trading expert or familiar with technical analysis. Our AI was merely given examples of Elliott Wave patterns and a range of other data about the markets and indicators at the time of those patterns. Therefore it was very interesting that the AI identified Elliott Wave length ratios of 38% and 61% as being optimal in the case of various patterns.
Those initiated with the subject will immediately recognize these as Fibonacci Ratios!
Whether these ratios are important because of the power of the Golden ratio in nature, or whether it’s just a self-fulfilling prophecy because many people use these ratios in their analysis… well, your guess is as good as ours. However, certainly, our AI found them to be important in predicting whether certain patterns would hit their target.
D) A Stock’s Performance Relative to its Peers and the Overall Market is Important – Something that the AI demonstrated was that a stock’s rank, in certain metrics, like price movement or indicator values, during the Elliott Wave pattern was also an important predictor in identifying higher-probability of a pattern hitting it’s target.
So What's Next?
What we’ve discussed so far is just the tip of the iceberg… there is a wealth of insights into not just Elliott Wave, but also the markets and how we trade them.
Over the next few months we’ll be sharing detail about the model the AI developed for selecting high probability patterns. We’ll cover Wave 4 and Wave 5, both long and short patterns, individually… and will tell you what counts most when you are selecting these trades.
Until next time.