Timing The 3x Bulls, Theory Into Practice
Summary
- The structural deficiencies of 3x Bull ETFs make them the obvious choice for market-timing strategies.
- Various technical issues must be resolved when transitioning from analysis to live trading.
- The algorithm gives advice during live tests; do not put it in control at first (or second).
- The focus on real-time action should lead to material improvements in the strategy.
The Legend
After my Market Timing article, I decided to start trading the signals. Like the vast majority of people, trading is difficult for me. No doubt, some people can do it quite well.
A trader will have money in a trading account, but a serious person will also have money in longer-term equity investments that are important sources of wealth. That situation can quickly become confusing, as successfully combining the two styles is problematic. In addition, non-trivial accounting issues make it difficult to get a clear, understandable handle on performance.
My previous articles show that Specter/FF5 signal returns beat long-term buy-and-hold returns for almost all US equities. However, the 3x Bulls profit from the signals orders of magnitude better without a lot of extra risk.
I'm neutral on the question of whether these unsound products were approved for use because of regulatory and industry incompetence, nefariousness, or some combination of 50 shades of gray. I'm willing to give the decision makers the benefit of the doubt.
The Trade Plan
Everyone has a plan till they get punched in the mouth. - Mike Tyson
The sages tell us that a trade plan is critical for success. If you were considering trading as a pastime and this didn't occur to you independently, relatively early in the process, maybe you should look for a different pastime.
The algorithm doesn't have a stock selection function and it has no understanding of risk. Those issues can be easily mitigated by trading 100 shares at a time with one or a few relatively decent generic 3x bulls like SPXL or TQQQ. However, it is important for the human partner to give these important strategy elements due consideration.
The algorithm continuously reinvests profits. This is an excellent technique for modeling human performance. As a human gets positive return feedback, he will prudently increase his investment. There probably are human investors that go through reinvestment calculations prior to taking positions; that type of behavior is eccentric, but probably harmless.
Trading at the exact close is also an artificial construct. It makes a lot of sense for the algorithm to buy and sell there because it is a known, verifiable point. There is no practical advantage for a human to trade there.
It is tough to make a living by trading, but even a mediocre human trader can plausibly try to work within the general parameters of the computer's purely mechanical performance without messing things up too badly. The results of using your judgment are an objective measure of skill.
It is tempting to give the algorithm control; after all, it seems to trade like Rambo and never wets its pants. Giving in to the temptation is wrong because it violates the BFF AI principle mentioned in Market Timing. The human has to be a contributing member of the human computer partnership.
The Aggregated Matrix
The matrix was briefly introduced in the Understanding Complex Strategies article. It took several days to code and I thought it was a nice technical achievement. The matrix is used to determine the historical profitability of FF5 market states. The investor wants to go long when the market state is clearly profitable and flat when it isn't.
Specter (as opposed to vanilla) performance results can be aggregated in a logically powerful manner because each security is long or flat at exactly the same times. Natural logs are the obvious numbers to use for aggregation. Of the many compelling advantages of natural logarithms, the possibility of using arithmetic instead of math to get good numbers is really useful.
Ironically, after rewriting the logic engine, I realized that the aggregate and non-aggregate matrices can be built with pivot tables. That makes them much better than my laboriously coded rendition. Pivot tables do a nice job with date handling, but more importantly are fast and efficient for dynamic indexing. There might have been a coding error in my interpretation of the matrix, the pivot table virtually eliminates that possibility of human error.
The aggregated matrix pivot table is shown below:
Created by the author with data from Norgate
The lookback period is about 4.5 years. I spend a lot of time looking at different length periods. This is a decent length for pivot tables because you get to see a bunch of years at once. The practical reason for this specific period is to pick up LABU.
The columns in green show the states that Classical Specter/FF5 is long. The five repeating character signals were selected as those consistently show positive returns for over 28 years. There is also a certain aesthetic elegance in the repeating pattern.
Note that the 3xs got slapped around a little on x00 in 2020, losing .32. That breaks down to a -.48 in Qtr 1 and a .16 gain in Qtr 2. -.32 would be the investor's return in that state, if he begins the year by investing exactly equal amounts on all 9 3x ETFs on x00, and then subsequently reinvests returns, etc.
The loss is no big deal but that period should be analyzed closely. x00 is an important buy signal and consistently makes strong returns, but it would be nice if the algorithm could be enhanced to avoid some of the negativity. My top mitigation candidate is range expansion analysis at the moment. Personally I don't consider that a high priority.
Note that the other strategies also have had negative experiences in other years.
Since we are looking at data in a live mode for the first time, the positive returns for xCF over the period stand out in the table. xCF is a theoretically OK signal, the two averages both deteriorated but not enough for price to close below either. The absence of the volatility pop to break below the averages suggests that this most likely is a mild pullback. Sometimes this doesn't work, but at the moment it has been working nicely and consistently over a relatively long period.
Classical 3x Specter/FF5
A table summarizing the performance of Classical 3x Specter/FF5 for the last 4.5 years is shown below:
Created by the author with data from Norgate
SOXL has doubled its money every year for the four years and then doubled again in the extra half year. LABU and TQQQ are currently slightly below doubling their money every year.
Looking at the aggregated matrix, xCF has had an aggregated win of 4.92 for the last 4.5 years. 4.92 is larger than the xFF and xMM returns combined. Playing xCF can significantly enhance profits.
Neo-Classical 3x Specter/FF5
Neo-Classical refers to Specter/FF5 playing six states instead of five. The extra state is xCF.
Created by the author with data from Norgate
Note that Neo-Classical aggregate strategy performance has increased to 26.32 from 21.40 for Classical. 26.32 - 21.40 = 4.92 which is the number in the aggregated matrix for xCF.
Note that in addition to the long state return improvement, the flat state returns have deteriorated.
These aren't chicken dropping improvements that we might expect from traditional strategies. Note that FAS returns doubled. FAS returns almost 3 times what buy and hold on QQQ did. LABU and TQQQ now have returns that are close to Classical SOXL. Neo-Classical SOXL returns an extra $14 for every dollar invested compared to Classical SOXL.
Recent Neo-Classical Signals
Created by the author with data from Norgate
L stands for Long in the last column and F is Flat.
There is a weird one day delay in the signals because the algorithm doesn't want to know about them until the next day's close when the returns are available. The delay confuses me because of the switch to real time even though I designed and wrote the routine.
Signals containing xF sequences are twitchy compared to the other ones. The return distribution between xFF, xCF and xFC seem more unstable than other character combinations. xFF is steadier than xCF and xFC is typically worst.
Pivot Tables - The Timeline
Finite State Accounting documented some of my initial adventures with pivot tables. It didn't happen overnight, but Excel and Pivot Tables are extraordinary human achievements. Commercial market analytic software doesn't offer pivot tables or anything remotely equivalent to my knowledge; Python has a grid as an add-on, but I think that's more money (I'm sure it is a really cool grid).
One of myriad of incredibly good reasons to use natural logarithm returns is that backtesting reinvested returns strategies based on virtual trades are unsound if the timeline is changed. I spent months looking at that after Finite State. Daily natural log returns make that problem go away.
For a correct example, the aggregated matrix is shown below with a Timeline.
Created by the author with data from Norgate
This shows 3x returns from February through June 5, 2020. The green columns are the Neo-Classical numbers.
June has had at least one F every day so far. Even the usually weak xFC is making money. That is why the foxes underperformed in May and June. They stay flat on xFC which had sizable losses in March and April followed by decent returns in May and June. I wouldn't put any big bets on my being able to solve that in the next few weeks.
Final Remarks
My official confidence level in the strategy numbers has gone from "pretty sure" to "real sure" after a careful verification of the signaling logic.
I misread the signals at least three times in the past four-day week (including weekends), in addition to making some other mistakes. For example, I bought LABU on xFF without looking at LABU's returns under that state. LABU does very well with xCF, but it is a coin flip with xFF. Fortunately being wrong to the long side was the right wrong last week.
I've been doing daily signal updates on my web site with a little commentary included. Keep in mind that historically, there is a pretty good chance of me reading the signal wrong; I'm hoping to improve that metric, but you get what you pay for. Ask me what it was a few days ago and I'll be right every time.
The strategy does better if markets are wiggling, but the foxes always stay in the conversation.
This article was written by
Analyst’s Disclosure: I/we have no positions in any stocks mentioned, but may initiate a long position in TQQQ, SOXL, LABU over the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it. I have no business relationship with any company whose stock is mentioned in this article.
Seeking Alpha's Disclosure: Past performance is no guarantee of future results. No recommendation or advice is being given as to whether any investment is suitable for a particular investor. Any views or opinions expressed above may not reflect those of Seeking Alpha as a whole. Seeking Alpha is not a licensed securities dealer, broker or US investment adviser or investment bank. Our analysts are third party authors that include both professional investors and individual investors who may not be licensed or certified by any institute or regulatory body.
Recommended For You
Comments (10)
