[originally published on medium]
I’m a huge fan of Van Tharp and his books. I’ve read all (most?) of them. I also subscribe to his email newsletter.
In a recent newsletter, Mr. Tharp revealed a small investment he made in $AIEQ, an ETF recently launched which utilizes Artificial Intelligence to make trades. Van clearly recognizes the potential in this new paradigm of trading driven by machine learning, but he — like most of us — has some reservations and probably has more questions than answers.
I especially enjoyed this riff:
“Understanding how little is truly important for successful investing would be a lot for a machine to unlearn as it moves along in its journey.” ~ Van Tharp
We at Trade Ideas, LLC began our journey of learning (and UNlearning) what is and is not possible when applying Artificial Intelligence to the process of extracting Alpha from the stock market a few years ago. The end result of tireless brainstorming and tweaking allowed us to deliver Trade Ideas’ Investment Discovery Engine powered by Machine Learning A.I. We named her ‘HOLLY’ and she launched January 4, 2016. And she has been undergoing continuous innovative improvement in performance and experience since.
Given our experiences so far, we thought we might like to tackle some of the questions Mr. Tharp listed in his newsletter and hope this helps him and you down the road as we all look to integrate new edges into our daily trading workflow:
Does AI understand reward to risk ratios in its decisions?
The backtesting and optimization of every algorithm evaluated by our Machine Learning A.I. (‘HOLLY’) focuses on two elements: the pattern/event and the trading plan. The trading plan optimization considers hundreds of thousands of reward and risk ratios and scenarios in order to determine the most ideal entry price, profit target, stop loss, and the amount of time in the trade. We call these overnight preparations by the A.I. the “Quantitative Combine.”
Does AI know when it’s wrong about a position and then exit? Or instead, does it close a position when it finds a better trade or when it thinks the information about the current position as a top candidate is no longer useful?
When the market opens each day, the Machine Learning A.I. performs a real-time impact assessment of the current market conditions on any open positions. The A.I. makes a determination as to whether a Risk-On or Risk-Off approach must be taken. If market conditions are volatile without direction, HOLLY indicates a Risk-Off approach and suggests capturing profits even before predetermined exit conditions. If the market is clearly trending, HOLLY indicates a Risk-On approach and suggests that holds may exceed pre-determined limits absent other exit criteria. In other words, traders are getting paid to hold on to winners.
Does AI understand the power of exits and the value of “not being in the market” under certain conditions?
The value of ‘not being in the market’ was vividly on display by HOLLY’s activity during Brexit. Heading into the historic vote, markets were exhibiting elevated levels of volatility with no clear sense of direction. Only anxiety.
Meanwhile, HOLLY unemotionally went about her nightly “Quantitative Combine,” crunching the facts only told by price and volume actions, and set the game plan for the next day. As trading began to unfold at the open that Friday, HOLLY strictly adhered to the 3 Rules of Machine Learning as applied to the equity markets:
- Understand when an event in the market occurs for which it (HOLLY) has not prepared.
- Do no harm to the P&L. Take no action until more data collected.
- Once additional data is received and a Quantitative Combine is completed, boldly return to the markets — often going where few have the wherewithal to go.
What ended up happening that first full day of trading after the Brexit vote? HOLLY took a look at the volatile landscape, recognized the volatility, and decided no possible setups had acceptable risk-return odds. HOLLY did not enter into a single trade.
What will AI’s performance be like during huge down markets?
The Machine Learning A.I. (HOLLY) will identify high probabilities of return from short-based algorithms in the nightly Quantitative Combine that will likely perform well when markets plunge. Of course, we cannot guarantee profits, only that you’ll likely be positioned best to take statistical advantage of downward price action.
What will AIs performance be like during sideways markets? Or any of the various market types?
While HOLLY takes market direction and volatilities into consideration when selecting strategies to activate each trading day, the stocks chosen for trade entries on sideways market days often have little to no correlation with the major indices. These stocks usually have their own individual catalysts determining price direction. Here’s a recent example.
Is AI biased toward asset allocation? (Because very few humans really understand what’s important about asset allocation). For example, if it has a bias toward some sort of asset allocation built into its programming, it might be very hard to overcome that bias.
HOLLY does not take portfolio allocation or optimization into consideration when selecting trades. Each strategy and therefore each trade stand on their own merits and are entered when a statistical edge for each stock is deemed present. There is certainly room for improvement in this area as we continue to compile more data every day.
Does AI know the power of position sizing strategies?
Position sizing is indeed a most important consideration for every trader and every trade. And with our AI, we recommend all traders size their trades in accordance with prudent risk management principles and would do best to keep risk as close to constant across their entire spectrum of trades.
Our AI leaves position sizing to the determination of the user. This is mostly due to the fact that Trade Ideas, LLC is not a registered broker and does not execute trades for customers and therefore has no visibility into the trader’s real-time capital levels.
Is AI biased toward being 100% invested at all time? If so, can it go short when the market is going down?
I mostly address the answer to this question above in the discussion about Brexit. But no, HOLLY is not biased toward being fully invested. In fact, at many points during the trading day there may not be any trades currently active. And most certainly, HOLLY has algorithms that look for short trade opportunities when conditions warrant.
I hope these questions and our answers lead both Mr. Tharp and you, Dear Reader, down a path towards greater discovery of how AI can augment your current trading routine and further examination of ways to further leverage the power of machines to the (still) unmatched power of the human mind.
Sean McLaughlin, @chicagosean