I hope you will forgive me if I make a heartfelt plea right at the start: this is an emotive topic for some and this thread is started as a genuine request for discussion and a prompt for thought and the testing of assumptions and hypotheses. May we keep this an acrimony and prejudice free zone.
My recent threads have been designed to penetrate the heart of this topic: I have discussed and back tested random entries at some length, I have queried the use of back testing and I have tried to explain why, over the past 40 years, trend following seems to have become increasingly difficult.
Next, as all who have read the meanderings I have put out over the past decade will know, I am an arts graduate. My lamentably poor “skills” in mathematics come from ill learnt lessons at school at an early age; my almost non existent knowledge of statistics comes from a few rapidly read and poorly understood books glanced at over the past decade. So, I look for advice and guidance here and do not seek to give it.
Professor Tom Gastaldi holds that trend following is a dangerous illusion: A “single threaded” strategy conceived by those “deceiving themselves by believing that the market data is giving out "entry signals". While in the grand scheme of things they are just making random entries.” Tom goes on to say: “The appearance of profitability is most often just curve fitting. In reality, in those cases, PNL is mostly oscillating around 0 (even though with large variance, large enough to create both dreams and defeats.)”
I responded: “You can't really describe random entries as a single threaded strategy I don't think. And in reality not many people design a single threaded strategy do they? Or is the very nature of trend following (running profits and cutting losses) "single threaded"? In your view?”
Tom replied: “ah well, "single-threaded" is a terminology that I use to denote the "class" of strategies which have the following features: * There is one and only one "trade" at a time for each instrument * A "trade" is defined by an "entry" and an "exit". Normally, entry happens on some "signal", and exit on "take profit" or stop (="take loss"). [nothing to do with "thread" in the computer science meaning, clearly.”
My response: “So to take that a stage further, over the long term trend following is a zero sum game and an illusion? And that if one conducted enough tests on random entries using a simple trailing ATR exit, the average CAGR would equate to zero?”
We left it at that (probably very sensibly, knowing how things can get out of hand on Linked In!).
So, does back testing have any validity at all? And if it does, can Tom’s views on trend following be proved or disproved in empirical terms using past data?
In my thread on random entries, I sought to bolster my bias in favour of trend following rather than to destroy it. I sought to test out the theory that random entry, using a wide portfolio of instruments, coupled with a simple ATR based trailing stop would prove profitable using a statistically large sample size of test runs and trades. I thought to show that “cutting your losses and letting your profits run” was a profitable strategy. But I can see how that may be thought of as a “single threaded” approach.
There are people on this forum, Tom included, who DO have a mathematical, scientific or applied science back ground. Whose careers have been pursued in areas where theories are put to the test in a rigorous and valid fashion.
Can those people please add their thoughts and please suggest whether there is any back testing approach which can shed any light on this matter. Can it be proved or disproved, using past data, whether trend following is a profitable reality or a dangerous illusion? I don’t think we want “opinions” here: I would prefer “proof”; if such is the appropriate word to use or indeed a practical possibility.
If “proof” is not possible one way or another, then back testing is indeed “ a tool for fools”.
According to some experts back testing is not a tool for hypothesis validation but for hypothesis rejection.
Ernie Chan (He is considered an expert by many and he is giving seminars about backtesting in England):
"6) Backtest can only be used to reject a strategy, not to predict its success. This echoes the point made by commenter Michael Harris in a previous article. Since historical data will never be long enough to capture all the possible Black Swan events that can occur in the future, we can never know if a strategy will fail miserably. However, if a strategy already failed in a backtest, we can be pretty sure that it will fail again in the future."
Back-testing is an indispensable tool for rejecting bad performers but it won't make anyone rich just by itself.
Useful feedback Basil. I put a link to Harris' website in the web listings section here at TP over the past 10 days but cannot claim to have explored it properly. As I have long suspected, you seem to confirm that there is no "answer". Or rather, if there is an "answer" it has to be that of maximum diversification in every sense of the word: systems, markets, entry points, managers, brokers, custodians, banks the whole lot. Which is why of course for the majority of us the best answer is a form of "indexing" to hedge fund strategies in general, to CTA strategies in particular, to markets and so forth. Everything messes up from time to time and you have to hope that it does'nt all mess up at once. If we do not adopt massive diversification we may well achieve higher returns in the long run but equally we run the risk of ruin if we concentrate too narrowly or indeed physchological difficulties in not being able to handle the consequences.
Of course, it COULD all mess up at once as it very nearly did over the past decade when the entire world financial system was on the brink of collapse. And, taken to logical extremes, when the dinosaurs got knocked on the head by that meteorite 90 million years ago.
Anthony I like your thinking because it is practical. There is more to trading than backtesting and analysis. But pls allow me to say that proper diversification worked during the financial crisis. Long/short commodity funds made up the losses of equities. One problem is that these things are too expensive for me and I have to manage on my own. It is a fact that commodity funds have no correlation to equity funds. The problem of course is finding that fund that will not go belly up and you end up with 100% loss although the CTA index is up 10%. It is a difficult game. Ernie Chan and Michael Harris are two of the best out there in my opinion. They are fully aware of the pitfalls. If you read their works carefully the truth is some place in there but it comes in bits and pieces and hard to recover in whole.
Just to give us all a wry smile take a look at the:
Managed Futures Liquid Index
The Credit Suisse Managed Futures Liquid Index seeks to gain broad exposure to the Managed Futures strategy using a pre-defined quantitative methodology to invest in a range of asset classes including: equities, fixed income, commodities and currencies. The Managed Futures Liquid Index is also a factor within the Credit Suisse Global Strategies Liquid Index.
Note how the back test looked and note subsequent performance. Note the unfortunate start date of "actual trading"!
If “proof” is not possible one way or another, then back testing is indeed “ a tool for fools”.
Hi Anthony, you have done a good job summarizing our discussion. For more details, people interested can see here:
-- About backtesting, let me also report a quote which I think clarifies better my view:
I see mainly the danger in the following "conceptual inversion":
* Data is given, strategy is varied (to search a "fit") [possibly fatal]
* Strategy is given, data is varied (for "strategy robustness analysis") [ok]
Unfortunately several beginners fall in the conceptual trap. Also due to the fact that, as noted, some available tools are conceptually misleading, and naturally introduce them to the first (silly, imho) approach.
[ Note that this has parallels with what all classical statistics is based upon.
For instance, the attractiveness of any estimator is evaluated based on its properties in the space of all possible occurrences. And not on the basis of a few realizations computed in the past ... ]
-- About your sentence "Professor Tom Gastaldi holds that trend following is a dangerous illusion", allow me to to point out that I was not referring to the "trend following" approach in particular. ("Trend" is a concept irrelevant in the context of what I was saying.) Let me clarify:
Take a "single-threaded strategy" as defined above. Let P a "take profit" size and L a "take loss" (stop) size.
If you denote by W the ratio: #winning trades / #losing trades
and by WS the ratio: (avg size of winning trade) / (avg size of losing trade)
I am saying that you will have, in the long term, a "convergence" (take the word in an "intuitive" sense for now), with large variance, to:
W * WS = 1 and therefore PNL = 0
I see this a "necessary condition" for mkt efficiency.
This is independent of the criterions used to "open" the trades.
That is, you can open the trade in the direction of what you believe to be what you call "trend", or you can open it in the opposite direction.
In other words, what I am saying applies to the whole class of strategies above defined, and it is not particularly related to a "trending" or "countertrending" entry style.
@Tommaso: I just wonder what makes you think that your G-Bot strategy is better than random entries. It looks like a Martingale system with option hedging. An excerpt:
A game is essentially resulting from a superposition of 4 sets of "players" (you can imagine "virtual traders") called:
CT Buy, T Sell, CT Sell, T Buy
The "CT players" and the "T players" follow different "rules", dictated by the user (trader/fund manager). While the naming of the 4 sets of players can be seen as arbitrary, it originally came from the the intuitive idea of having Countertrending (CT) positions overlapped Trending (T) position. Clearly, in a more abstract view, there is no need at all to make reference to the ideas of "countertrending" or "trending", but one can just consider them different ways of trading, defined by the user, which can be programmatically overlapped at will by G-BOT. The application will coordinate all the players, pooling their requests into single orders and then "redistributing" the fills according to the requests.
Thus, in summary, a "game" arises by the programmed interaction of the CT and T player rules, and the combinations and interactions are clearly countless.
In intuitive terms, one might have to decide if he wishes to be prevalently contrarian or not, want to use "stops" or not, etc. For instance, if one is mainly "contrarian" (like most of my example games), the CT Buy will be "hedged" by T Sell, and CT Sell will be "hedged" by T Buy. Clearly, one cannot achieve a "total" hedging, as they would lead to a "neutralization" of the positions, but in practice can extend the length of the "adverse" moves we can withstand with a given capital. In this case, the use of options can also be useful (see below) and create an additional effective protective layer.
Now, while a "game" is being played on a "layer" (by the players), multiple layers can also be created to overlay different games.
In addition, the layers can be surrounded with option layers for hedging purposes, as we will explain below.
In essence, there are 2 main approaches to scalping:
- try to constantly "guess" the price direction (trending): in such a case once one determines that the "guess" was wrong, it has to stop.
This approach fails when too many stops and trading costs eat all scalping profits.
- holding even in a unfavorable move (contrarian) increasing (very gradually) the size and waiting that the combination of a large enough size and a large enough reversal yield a profit.
This approach may fail when the sizing is relatively too large, the price is too far away from a reversion point and the capital (or the psychological attitude) is not sufficient to hold the unfavorable move, before the market is ready for a reversal.
The first approach is more natural for many, and in a sense "naive", and it is what most beginners do. The second approach, which is normally more profitable (when carried out appropriately), requires good capital, careful planning and is normally used by more experienced fund managers. Normally, to be viable, it is necessary to make sure that the (first) entry direction be meaningful, to avoid (the possibility of) taking a drawdown as large as the entire price range (which would be normally deadly). For this reason, one would carefully select a folio with suitable instruments which are possibly near the end of the historic range, or at least near the end of a large local move, and possibly force a direction (long, if low, short if high). In alternative, and more generally, one can use options for a more general and sensible form of protection (in this case it would be practically viable birectional scalping, or start scalping at any price level).
Note that the naked "contrarian" play is not for everyone: in fact, if played without options, it is essentially based on potentially large drawdowns, which can be seen as the"investment phase", in fact the process consists in carefully investing in any move (even adverse) in order to reach a good position at the end of it, and to stand ready at any time for the reversal, without any need to "predict" when it will occur. Like in real life we do not expect an investment to necessarily work in the short term, nor that, after deciding to invest in some ventures, we continuosly quickly change our mind: the same attitude may be necessary when trading with this approach.
In any case, any new game devised should be accurately studied with the "robustness analysis" facility in G-BOT, to see if the trading mechanism corresponds to what is really wanted. And, when applied in real trading, a particular attention should be devoted to planning a suitable folio and entry directions (by using the "discretionary bounds" or forcing the entry direction) or to setup a good option structures to "protect" the scalper.
I also see some contradictions in this strategy: "Guessing" the price direction is considered "naive" and wrong. Yet, as "total hedging" is not possible because it would lead to "neutralization", the "experienced" fund manager (purpotedly engaging in guessing out of this experience) is expected to do careful planning, with good capital. For the strategy to be viable "it is necessary to make sure that the (first) entry direction be meaningful, to avoid (the possibility of) taking a drawdown as large as the entire price range (which would be normally deadly). For this reason, one would carefully select a folio with suitable instruments which are possibly near the end of the historic range, or at least near the end of a large local move, and possibly force a direction (long, if low, short if high)."
I also wonder how one can vary the data to test the robustness of the strategy and if such variation has any predictive value better than random entries and again, when applied to real trading, "a particular attention should be devoted to planning a suitable folio and entry directions (by using the "discretionary bounds" or forcing the entry direction) or to setup a good option structures to "protect" the scalper." Again, it seems to me that a lot of "guessing" are required by the "experienced" fund manager.
@Tommaso: I just wonder what makes you think that your G-Bot strategy is better than random entries. It looks like a Martingale system with option hedging. An excerpt: ...
For "martingale", see my post here:
The indication about the possible use of "discretionary bounds" was meant for the user to familiarize initially with real trading, but it is not needed when using options.
It's always good the have the possibility to restrict the trading action in some definite zone (one may choose not to, anyway).
For option setup there is a definite "rule" see here:
There are many other things at work, such self-hedging, cointegration blocking, folio recalibrations, option hedging. All stuff that makes the difference between printing $$$ and blowing up.
Clearly, throwing around the word "martingale" does lead anywhere, as explained in the post above.
[ For a better understanding of what it does, you are most welcome to get your copy: http://www.datatime.eu/public/gbot/#copyrequest&nb... ]
I know most people do not want to hear my opinion about the "single-threaded" class of strategies, and it irritates them beyond measure. In fact, Anthony is witness that I always try to avoid this sort of discussion, because it clearly ends up in a "religious war" between people who evidently have a different understanding of reality.
In this case, I had to make my case clear, because Anthony was very interested in my opinion and... he thinks he can manage a flame war
It's just matter of a "journey". At some time people are ready to hear certain things, at some other times, they are not.
You might have whatever "holy grail" and put it right in the hands of another person, but if he is not "ready" to see it, he will not see it.
This is a generous offer which should not be declined. The opportunity to grow and to improve should not be spurned. I am still grappling with Tom's opinion that a single threaded game like TF is ultimately unprofitable but that will not stop me from inquiring further into possibly better (or additional) approaches.
There are several reasons why I take this attitude. I have come to attach huge importance to admitting I may be wrong and to the realisation that I have no monopoly on "truth"; I guess this is just the lesson that life teaches most of us.
The other reason is that my prime aim for this website is for it to become a centre of excellence where there is room for a wide diversity of views from which we can all learn. I believe none of us are entirely free from bias and preconceptions. And that it is uncomforatble for all of us to have our long held views questioned.
It is "right" for us to feel uncomfortable when our views are challenged - if our views are not challenged we will not grow, we will simply stagnate. Discomfort in this context is good for us.
So, I thank Tom for shaking up my rigid thinking and I look forward to downloading G-Bot later today and learning whatever lessons are there to be learnt.
Here endeth the sermon.
Interesting take on the idea of options as well.
I used to be an old style equity options market maker, and it has similar themes here for the long volatility players (short volatility in this discussion is irrelevant IMHO)
As a long volatility player you are constantly jobbing/scalping/trading gamma in order to pay for the time decay on top of ideally having some edge in regards spreading options to lock in a theoretcial edge. The real value came when you could get an option position working for you whereby either
- the option was effectively paid for and hence upside was unlimited, downside zero
- the move occured so quickly you had no time to hedge, and gaps/crashes/takeovers etc made you a lot of money - the mini black swan events as you knew they would happen at various stages.
- you managed to really be good at the scalping/intraday trading even if the historical = the implied.
This resonates (at least for me, and maybe I am still way off the mark - doh!) the difference between single thread v multi thread classes of strategies, as you can have multiple elements of a postion, be wrong and yet still make money at the same time.
example; long calls, short underlying, net bullish view point, and long delta.....yet, if you are trading in an out of this, the instrument goes down, but you still dont loose money, OR the instrument gaps down and you actually make money.... wrong in position and view, but still able to make money - if only it happened more often and some people could not see this reality.
People even wondered how they could make money being short volatility, and we could make money being long volatility in the same options and instruments......we prefered not to reveal our secrets LOL.
So,........maybe this is missing the point, or confusing the matter, and maybe its just an off topic side note.......but it helps me get a better understanding on the G-bot.
back on topic of trend following.....bascally it gave extra ommph when right and with the trend, protection when wrong against it and the ability to really make the most of those trend like moves......if so it sounds like a pretty handy tool
(I would like to trial/test it - but I am not sure I have the time at present as i have some projects that might take up my time in the next few months, but from what i have seen - it looks at least interesting - thanks Tommaso - I also liked the little quote you had of 'dead past data' it sounded something Italian you could eat and i dyslixically read it as 'bread passata")
A superb article pointed out to me by "Stamo" on the Trading Blox Forum. Very apt to this topic. Yep, I'm afraid this is another of my "must read" recommendations.
"In questioning initially whether I am a great investor, I open the door to question whether other similarly esteemed public icons like Bill Miller are as well. It seems, perhaps, that the longer and longer you keep at it in this business the more and more time you have to expose your Achilles heel – wherever and whatever that might be. "
"What if there is a future that demands that an investor – a seemingly great investor – change course, or at least learn new tricks? Ah, now, that would be a test of greatness: the ability to adapt to a new epoch. "
My intent is not to initiate a religious war, but just to make sure you're not just another brother from a different denomination and with different dogmas. And so I try to breakdown your ideas to simple, understandable concepts, and also checking for inconsistencies and conflicting assymptions. Of course, you're not required to disclose your strategy. So, thanks for the offer. I should take the time to download a copy and do some tests. And I would advise so to any other brothers in faith.
By the way, from fiat lux to Central Banks fiat money and highly professional investment analysis, I do not indeed see much difference between investing/trading and religion. We're all got to have some faith after all.
Mucking around this week with trend analysis techniques; autocorrelation analysis of markets, looking at trend definitions and how they change over time.
At heart I don't think there is any simple answer to the question I raise. Even if we had data stretching back to the Garden of Eden, I don't believe any amount of trend analysis would prove a reliable indicator for the profitability of otherwise of trend following systems going forward.
I think one would be well to listen to Bill Gross and what he says in the article I reference above. There will be periods where one particular strategy makes you look a hero and other periods where it makes a monkey of you. No amount of statistical analysis of past data is going to tip you off in advance about what kind of period the next year, 10 years or 40 will be.
Nonetheless I will publish my analysis in due course.
Looking further into this, it seems that to do a proper test of G-Bot it's necessary to have a paper trading account at Interactive Brokers, which is only available "after your production trading account has been approved and funded".
No, talk to Tom. I have just been hours on-line with him and he has been demonstrating G-Bot to me....think we need to move this to the G-Bot thread if you don't mind. I'll explain. If you don't have an account at IB you can still get data but much of it is, apparently, simulated or synthetic. Which if you believe the markets are a Markov chain doesn't matter a tinkers wotsit.... You can open an account at IB for peanuts ( Iwould imagine) and then get real data to paper trade on.
In any event, chatting to Tom and getting a demo of G-Bot is excellent fun. Even if he wants to persuade you that believeing in auto correlation is a bit like believing in God - IE requires blind faith.