Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
KJ Trading Systems Kevin Davey - Ask Me Anything (AMA)
I'd love to have it, but I'm confused as to whether it's GONE GONE GONE as you posted at 3:10 PM or if it's still available, as you posted at 10:30 PM.
Starting with the basics, if we have the two most basic market regimes, i) momentum/trending vs. ii) rangebound/mean-reverting, you would have to predict whether the given market will be one or the other during the planned time in which you will run the system. Then, you apply the appropriate system to the market. If you predict the regime correctly, you will reduce heavy drawdowns that occur when the system runs during anti-ideal conditions. Reducing drawdowns is the goal. If done correctly, you would also avoid limiting big gains—i.e., maximize upside equity-curve potential while simultaneously reducing downside potential (smoothing without limiting). Typically, reducing heavy drawdowns necessarily also means reducing the upside potential to some extent. However, here, we are offloading the need for robustness, taking it out of the system and putting it on the trader’s discretionary shoulders—which, like leverage, is always a double-edged sword.
Using such a simplistic regime structure such as trending vs. mean-reverting will more likely lead to more loss stemming from the trader’s incorrect prediction than would be mitigated from this whole approach. Other market-regime viewpoints (micro-structure philosophies) may be more fruitful but entail exponentially greater effort to get established.
Before I go further, let me digress and say that we need to know how to measure performance. There are two independent performance measures, and any given daily loss must be attributed to one or the other root cause. The two measures are: i) whether the regime was correctly predicted and ii) whether the system performed as expected over the regime that actually occurred. Losses on individual trades are a part of a correct system correctly running in the correct regime, but a particular day should (generally) not end in a net loss if the correct system runs correctly on the correct regime and there is no sort of unexpected news event. The more you tailor a system to one particular regime, the better it will perform in its ideal regime and the worse it will perform outside its ideal regime. Therefore, either a) the trader is fault for incorrectly predicting the regime, b) the system is at fault for failing to perform in its ideal regime, or c) the correct regime was predicted but some outside impetus caused the regime to change or otherwise act improperly. This is where discretion (and cognitive biases) can really bite you. Not only is the development of the system (as an entire whole) subject to cognitive bias, but the execution (or application) of the system is subject to cognitive bias and the resulting performance metrics can be recorded and interpreted subject to cognitive bias.
One other digression on a defense mechanism to remember, the system should programmatically read the current regime and switch itself off if the ideal regime does not present itself. This would be a backup to supplant the trader’s discretion but the whole point of this approach is to avoid the need for programmatically reading regimes and the inherent lag associate with doing that. The backup should not deploy itself on its own unless the trader is not actively watching over the systems.
Turning back to more-performant market regime structures, one key idea is the “day types” defined under market profile theory or auction theory. It is common for equity index tape readers to go into an RTH session with some notions about what kind of day today will be. They may have more than one possibility and adjust based on what the market tells them during the initial balance period and throughout the day. The same is true for overnight trading hours (where most of the movement in equity-index contracts occurs). The point is, under this approach, the trader would make predictions based on the same skills that every other tape reader employs. The trader would keep track of correct and incorrect regime predictions for each day and also keep track of whether the individual systems performed as expected over the given regime that actually occurred.
However, while targeting between trend vs. rangebound is overly simplistic, targeting a market profile day type maybe overly complicated and lead to similar error in predicting the correct regime. I think the most realistically achievable result is to understand one particular market type and find (or define) your own regimes for that particular market type. You might focus on equity index contracts and have a set of regimes for that. Or a set of regimes for currencies, or energy, or precious metals, or softs, etc. And you could even have a regime for big economic calendar days, if you’re into that sort of thing.
As you said, the devil is in the details. Defining (and refining) the set of market regimes is the core of the task. Once you have a set of regimes that reliably appear on a repeated basis, you have to develop a set of systems to attack inefficiencies in each regime as well as some method to reliably predict the regime early enough to apply the correct system or make a change before too much loss occurs. I must assume that if you can identify repeated regimes (in whatever definitional context) then you would also be able to reasonably predict what regime was about to occur or, alternatively, what is second-most likely to occur. When your first prediction proves to be incorrect, hopefully you can recognize it quickly enough and apply the alternative system appropriate for what actually occurred. Of course, all the normal psychological problems of discretionary trading still apply in this type of approach because you cannot get something (e.g., reduced drawdown) for free.
Of note, I was thinking about this idea from listening to a guy named Xiao Qiao on Ep 153 of the Chat With Traders podcast (https://chatwithtraders.com/ep-153-xiao-qiao/). Basically, Xiao runs a stock scanner to determine whether a particular equity is more mean-reverting or more trending. He said he plots his calculation on a cartesian plane, and the stocks that plot at the bottom right of the grid are one type while the top left of the grid are the other type…everything in other places is more random-walk and not reliably trending or mean-reverting. I actually think he’s using some type of Hurst Exponent calculation based on how he described it. But he calls his scanner his “binoculars” and his systems his “rifles” to go hunting with. Each rifle is either trend-following or mean-reverting. I think this is a sound overall strategy, but my goal with the approach I am describing is to reduce drawdowns in order to maximize leverage (or position sizing), which has more of a place in more-highly leveraged instruments.
Edited to Add: the Linda Raschke webinar (set for 12/12) appears to be touching on topics related to this, and she may (or definitely) have it more figured out than I do right now. https://nexusfi.com/webinars/#registernow
I understand what you are saying, and while it might be worth pursuing, realize how daunting PREDICTING the future really is. That is the core of what you are proposing (you mention "predict" a dozen times in your write up), and that is REALLY HARD.
I do not personally know any successful real money trader who thinks he can predict the market (most believe their success is reacting to the market). I do know plenty of wannabe/paper traders who profess such prediction ability, though. Prediction ability and real money success almost seem to be inversely correlated, in my experience.
Of course, the nice thing is that you can test any prediction idea you have, before going live. You can see if it is statistically valid. If it is, then you can combine it with the trend or mean reverting entry/exit.
I hope the well thought out approach you just detailed works out for you. It is certainly interesting. Just really hard too.
I would argue that "prediction" is required to make money in markets, regardless of what you call it. If you look at the field of machine learning, it speaks about "prediction". I have wondered why it is not trivial for everyone to see that. It might be because people tend to assume that if one can predict something that it means one must be able to do it absolutely and without variance. But, even with things that are highly deterministic, such as shooting a basketball, there is obviously variance and uncertainty. In any prediction, you have a confidence level (how sure you are), a measure of precision (what your stop needs to be), and an accuracy (a winning percentage). Some scientist speculate that the sense of volition that people have is merely the result of the ability of our brains to predict future states.
When you talk about "predicting the future", this is a more general form then required. Prediction can also be "empirical". Karl Popper's ideas of regarding "falsification" are important when making predictions: namely you must specify the conditions that will falsify your predictions. Perhaps it is the historic tendency among traditional technical analysts to make non-falsifiable predictions, non predictions, is where these "wrong notions" regarding what prediction means. However, in the conversion of a a prediction or idea into an actual trade, there is information loss. Arbitrary elements are introduced. There is more skill required to do this then what one would suspect.
Profitable speculation is the skilled application of trading a prediction with money. Kevin, I agree also many traders think they react to the market but it's nonsense. Whether or not one reacts to the market or not, one must be able to predict it to profit from it. If one reacts to the market, it merely suggest that one is looking for patterns (predictive patterns) in the data. This is the empirical model. It is still prediction. Now, a speculative model or behavioral model probably incorporates both price and other trader behavior. For example, if one sees a big green candle and buys. They might say they are just reacting to the market. However, they are really saying that a big green candle is predictive of a future bullish condition. If one says they react, it might suggest they are using implicit cognition and not conscious prediction. It is still prediction though. Now, perhaps we are reserving "prediction" to refer to future predictions based on "future unknowns" vs "past knowns". However, all future states are unknown. But, going with this idea it would be the form of prediction such as I think a future event will have a certain effect. This is more speculative trading, of course.
I will agree on one point: if you go too speculative then the consistency should be less. However, what I suspect the highest skilled traders do is a combination of empirically based prediction (pattern recognition) and speculative prediction (if then/hypothesis prediction).
Anyway, I never understood where this idea comes in that one can "react" to the market, profit from it, but not be able to predict it. Now, a lot of what I discussed is opinion: but this is not opinion, you must be predicting something better then chance to be profiting from it. It doesn't have to be direction. It could be expectancy.
I also agree it is very, very difficult in practice.
