Following the Trend
There’s a good reason why most professionals who apply models similar to trend following to stocks call them momentum models. It’s not just a clever rebranding, it’s really a very different game. To blindly cling to trend following as a religion, disregarding any real world evidence and attacking anyone presenting ideas that differ to the trend following mantra is not only unprofessional, it’s outright dangerous.
I bet you’re wondering about the title of this article. After all, I do employ quantitative models based on trend following logic on single stocks in quite large scale myself in my business. Some models that I’ve been using for many years produce very attractive returns on single stocks. So why am I writing such a provocative title? It’s not only to get you to click on it (though that worked, didn’t it?). It seems as some people stop reading after such a headline, and simply go on an all out counter attack, without bothering to read or understand the rest. Well, I’m guessing they’re no longer reading, so let’s get down to the real deal.
If you apply a standard trend following model on stocks, you will lose.
The operative word here being ‘standard’. Trend following on futures is quite easy in comparison. It’s much more complex to model strategies on equities. Most people simply ignore the difficult parts and hope for the best. That’s not an advisable course of action. Doing really proper strategy modelling on stocks may be outside of the budgets and technical capability of most retail traders.
Even if you get your models right, you can’t treat stocks like futures. There are a few key differences, that require you to adjust your expectations and approach.
Stocks are cash instruments and need to be funded. You have a clear limit to how much exposure you can take on. You do not have a pool of cash anymore to be placed in govvies. You cannot have your client fund his managed account at 20%, putting up 200,000 for a million notional.
Stocks are a very homogeneous group. The internal correlation is massive. They will all go up and down at the same time with some small variation. In a bull market, they all go up. In a bear market they all go down. Diversification becomes much less important. You end up mainly trading beta anyhow. That can be ok, but if you’re under the delusion that you’re a great stock picker for buying high beta stocks in a bull market, you’re in for a nasty surprise when the bull leaves the field.
Stocks are prone to rapid vola expansion in bear markets. Your neatly calculated risk measurements goes right out the window real quick. Suddenly all those stocks that were doing so nicely all fall down hard at the same time.
As you start entering shorts on new lows, the stocks tend to make huge, albeit temporary, jumps up. As you’re forced to shut down positions not to blow your portfolio up, they fall back down. The short side of the single equity game is a veritable nightmare for standard trend following models.
Modelling strategies on equities properly require total return series and dividends details. You need to analyze the total return series, trade the price series and have logic in place for how to handle the dividends when they come in.
The potential for survivorship bias in single stock strategies is massive. If you run a strategy on the S&P 500 stocks for ten years back, base on the current S&P500 constituents, you’ll get an extremely distorted picture. Many of those stocks are in the index because they had a massive price increase. They were not in before. They wouldn’t have been on your radar when they had those returns. Check your data.
Applying standard trend following models on single stocks is dumb. It doesn’t matter whether you use breakout channels, moving averages or other indicators. Toggling parameters up and down won’t help. People who say that you should apply standard trend models on stocks also tend to be the people who lacks experience with professional trading or quant modelling but don’t let that stop them from selling defunct trading system to unsuspecting retail traders at a few thousand bucks a pop.
The most common arguments for applying standard trend following models on stocks is based on anecdotal evidence and classic fallacies. The first kind would be to point out that some hedge fund seems to be doing it with good results, without of course knowing anything about how they have adapted models for stocks, or to point out that someone’s cousin got rich doing it. Hedge funds certainly don’t run standard trend models on single stocks, though the often simplify the marketing pitch by calling it trend or momentum strategies. I can assure you that real hedge funds are a little more sophisticated and are fully aware of the special situation in stocks. As for the cousin, well, going on anecdotal evidence anything is possible. Apparently there are people who made gazillions trading on financial astrology too.
The fallacies are usually about mentioning stocks that went up a few thousand percent, and how trend following models totally would have captured this. Disregarding of course the probabilities of having covered that stock when it was a small cap, the many times you would have been shaken out along the way, the allocation to this stock compared to the many that did less well etc. A massive simplification of the real world. This argument concentrates on the position level, and on the pro side we’re just concerned with portfolio level results.
Does trend following really not work on stocks?
If you’re willing to adapt your models and do something closer to momentum trading, you’ll do just fine. But the return expectations cannot be the same as for futures. Not that it’s necessarily lower, that’s not the point. But you’ll be much more dependent on the overall state of the equity markets. You can’t expect to make a killing in 2008 because you were supposedly short all the stocks. Would be nice if the real world worked like that though.
How do I know that standard trend following does not work on stocks? Besides the common sense arguments of having lost the advantages of diversification and leverage, there’s quite a bit of actual, empiric evidence. I do this for a living. I have no reason to say that something works or does not work, unless that’s based on experience and research. I’ve modelling thousands of iterations of trend following models on every major index in the world. I’ve arrived at models that work and models that don’t. Standard trend models don’t.
So what can be done to make trend following work on stocks?
1. Don’t go short. Kill the short leg of your strategy. Replace it with a short index overlay if you have to.
2. Take the state of the overall markets into account. You can’t keep going long in a bear market and expect to gain.
3. Build a ranking methodology to pick the best stocks. Don’t trade the stocks you read about in the news or heard about from your friends. Automatically analyze a large number of stocks and have the best ones selected.
4. Single stock vola can change dramatically over time. Rebalance your position sizes.
5. Trade fewer stocks with larger positions. Yes, that’s right. Counter intuitive isn’t it? Don’t be fooled into thinking that you’ve got more diversification with 50 stocks than with 20. They’re all beta bets and you just need to get the individual event risk down to reasonable levels. Beyond that, further diversification will worsen your results. Don’t believe me? Run a few hundred iterations of your model and tell me what you find.
6. Expect to have great returns in bull markets and aim to make your strategy lose as little as possible in bear markets.
Religious dogma about a trading methodology is not helping anyone but system salesmen. It’s dangerous to view yourself as a trend follower. A much more pragmatic way would be to look at yourself as a systematic trader. Investigate what works and how it works. Trend following is a great concept but be very aware of its limitations. Adapt your rules to reality and overlay satellite strategies where needed. Hard work, quantitative modelling, research and pragmatism will get you to your goal.
Good post.
As I was reading it, I couldn’t agree with what was written until I read the last part. In my own back testing, I found that I have to do the 6 things you listed for trend following to work on stocks. (except I have not tried #4, will add that to my system)
Thanks,
-Ajesh
Thanks, Ajesh! And thanks for reading all the way to the end. 🙂
The philosophical question of course is whether a strategy that implements all those points is still trend following, or a long momentum strategy. Semantics, I guess…
can you comment on this, I am almost scaring myself. I started picking stocks based on two things, money flow (increasing) with only best performing sector stocks (right now healthcare/electronics) between $2 and $25 and spreading money across 25 stocks. I look for best opportunities with new uptrends and MACD crossovers, but the bottom line is increasing money flow. It must > and climbing new movements past 0.00. I made 25% in the past 3 weeks and that huge drop a week ago didn’t phase my strategy. Where is my Achilles heal?
Meant to add. My target on any stock is 35%. My exist loss on any stock is 30% or a issue that just doesn’t perform for 3-4 weeks. Then I exit that position.
It’s tough to comment on a strategy without properly testing it. You should model and test how your strategy would have worked in the past to better understand the dynamics. Without testing, it’s all guess work.
A couple of brief comments though:
In my view, the share price itself is absolutely irrelevant. Who cares if it’s traded at $2 or $25? That’s a non-factor. If you care about that, a split would impact your trading, which makes no sense. You may want to look at market cap instead.
A universe of 25 stocks seems very limited. I normally hold more stocks like that at any given time in a portfolio.
Targets and stops expressed in percent is usually not optimal. I know, may people like them, but the problem is that this method fails to account for vola. If you only trade stocks with very similar vola, this may be fine. But some stocks can have normal daily swings of 3% while others average a 0.5% move per day. The same percentage target becomes an arbitrary number.
Hi there Andreas,
I really appreciate you writing this article. Everything you said confirmed my findings when I backtested the trend following system (all long positions) on the 500 stocks of S&P since 1998 upto 2013. In terms of portfolio management, the number of positions I have is based on a “fixed %” based on total stocks monitored. In this case, I used 18% of 500 = 90 stocks. Same case that when I monitor 100 stocks from the S&P, I’ll hold 18 positions. My results were as follows. Appreciate if you could share your opinion on these.
(Figures are in Net ROI)
Mean per trade 5.9%
Sdev 13.4%
Max Win 76.4%
Max Loss -12.9%
Sum of Win 715.7%
Sum of Loss -189.1%
Count of Win 75
Count of Loss 42
% Win 64.1%
Ave Win 9.5%
Ave Loss -4.5%
Risk/Reward 2.12
Compounded Annual Growth Rate 12.9%
Hi Andreas,
When I read your post again, it seems that my system uses the standard trend following method. I’m interested in purchasing your book and curious if momentum strategy would really outperform the standard TF. Does it include all the entry/ exit signals for stocks? Thanks. Looking forward to backtest and use your system.
Hi Ed,
It’s hard to evaluate trading models based on a few summary statistics. To really assess the validity of an approach, you need to dig much deeper than the summary screen.
My latest book, just like the previous one, shows all details and all rules. I think it would be dishonest if I showed results from a trading model without properly explaining the rules. All details are in the book, allowing anyone to replicate and verify my research.
Hi Andreas,
Thanks for replying.
Entry: 1) Close above M-day SMA
2) Positive Territory using N-day Donchian Channels
Exit: 1) If Close crosses below Trailing Stop Loss (function of stock’s X-day volatility)
I used this in PH stock market and results were great (market was mostly trending from 2001 to 2015). But when I tried using all S&P 500 individual stocks, results reduced to 13% average annual.
I’m still interested in your book. Please let me know how I can get a copy. Thanks again. I’m trading in the PH by the way.
Hi Andreas,
Thank you so much for this post of yours. My results improved with less drawdowns when I incorporated #2 in my TF system. Again thank you, thank you! =)
Hi Andreas,
I can’t thank you enough for enlightening me with this post. Do let me know me if you’d like to get in touch probably through email so I can share and discuss in details my results/equity curve using S&P stocks and the portfolio/risk management and position sizing that I use. You can verify them and I hope it may be of use to you too.
Very interesting, thanks Andreas.
I’m currently testing TF on US and Canadian stocks, using simple breakout/SMA with ATR for sizing.
Long only, weekly rebalanced, free access on my website, if interested, no pressure.
Even though the 2 portfolios are fairly new (+3 months) I see the correlation effect that you mention so I’m not sold with TF on stocks either. The US portfolio is underperforming the SP500 while the Canadian outperforming (thank you miners.)
I intend to maintain these 2 models for one year, as an experiment.
So far I get better returns with relative strength on index ETFs (stocks, commodities, REIT, …)
AQR is using a 12-momentum strategy with AMOMX fund.
400 stocks though…
Looks interesting. You can get good results out of momentum style approaches to single equities. Just make sur your model takes the vital points into account. Stocks are not futures and you can’t treat them the same, nor can you expect the same return profile.
You’re on the right path here. Just watch the two really tricky points when it comes to single stock strategies. Dividends and survivorship bias. Solving them are expensive and painful. It’s easy to skip over them, but you’ll find that they have a massive impact on results. You’ll need historical dividends data, total return series, historical index joiners/leavers etc, and integrate all of that into your simulations.
Would be interesting to see results of your backtests concerning position size. Obviously correlation is common, but at least since I’ve started tracking various portfolios and strategies,
I read a paper recently that showed even if all you did for a system was buy a stock at an all time high, sell at a trailing stop, you would outperform the market over multiple timeframes going back quite far.
I read that paper too. The drawdowns were soul-crushing as compared to the returns.
Did not finish my first sentence, got distracted. Was going to say I’ve seen striking difference in performances between methods, and I’ve seen divergences between indices that while it might seem like variations in beta, I think it’s an oversimplification.
How can it be that simple rules will kill the index like you demonstrated last month but trend following on stocks is such a disaster? I’m missing something – what’s the difference?
That article was a long only momentum model, in it’s simplest form. And as you see, it differs significantly from standard trend following models, both in types of rules and in results. It’s a relative model highly dependent on the state of the equity markets and it lacks meaningful diversification. That model is an interesting demo, but I wouldn’t trade like that.
Try applying a classic long/short trend model on stocks and watch your portfolio wither and die. It doesn’t matter much if you apply the old turtle model or some more modern variation. It’s just a bad idea.
Thank you, Andreas. I knew something wasn’t quite right when I studied the turtle stuff, but reading your book and your thoughts here on your blog may have saved me from a disaster. I appreciate your patient replies to such basic questions as mine. All best to you with much admiration.
Great post Andreas. Re: #5, number of stocks still needs to be large enough to offset single stock risk. Let’s say you have 10 stocks, one of them suddenly gets busted for fraudulent accounting. The ticker is suspended, you can’t get out of it. One recent example is China Metal Recycling (HK:0773). That position effectively becomes written off to zero. Not sure how the rules work for US listed stocks. In HK at least, stocks are routinely suspended pending important announcements/releases.
#3 is quite difficult to do for retail guys. The investable universe is massive. You’ll need a lot of reliable data and an efficient way to rank and sort the “best” tickers to trade.
Yes, right way of thinking about the problem. For S&P x00 stocks, the probabilities of seeing zero values before you can exit is low enough to take, in my view. You always have to look at two factors, as with all risks in live. What’s the damage if it happens, and multiply that by the probability of it happening. You’d need more stocks in the portfolio if you’re trading more dangerous stocks or markets.
The dinosaurs of course didn’t expect a big rock to hit them all in the head either.
#3 is quite difficult. The largest hurdle for retail I’d say is the lack of total return series / dividends data. Ignoring that will have a massive impact. Survivorship bias is also massive, such as assuming that the stocks in the index today are the same ones that you would have traded five years ago. Historical joiner/leaver data is important, but often out of reach for retail.
In the end, most retail traders simply fly blind and hope for the best. Quantitative stock strategies are very difficult and very expensive to properly simulate. Much more so than futures.
Excellent article, Andreas, some really important points made. And congratulations also on having written probably the best book ever on Trend Following (and I think I’ve read nearly all of them.)
After lots of system testing, plus many years of trial & error, I’ve come to similar conclusions. As a result I trade futures using standard trend-following rules, but for stocks I apply only the trend-following ‘concept’.
Typically I will wait for what I hope is a long-term general index breakout (e.g. July 2009), and then begin buying stocks with promising trends. Unlike most futures trends, the bull market will then often last many years and I aim to hold on to many of the stocks I buy at the beginning of the bull until the very end of the long-term trend, riding out all the large corrections in between.
The drawdowns while sitting through medium-term down-trends are of course very large, but because they come after big upswings I find them tolerable. I do not attempt the same forbearance with futures market corrections – my stops are far tighter – but isn’t that the beauty of the equity markets for trend followers? That they tend to exhibit these vast sweeping bull trends?
I also agree with you that shorting stocks, even in a powerful bear market, is much more difficult to pull off.
so you are saying trend following doesnt work at the stock level but does work at the index level? (ie applying trend following at eg sp500 level just not on individual
stks within the Sp500?)
Professional trend following is very much based on diversification. Without it, the concept fails. Running TF models on any too homogeneous group will cause a major problem. On single stocks you have the additional problem of the extreme behavior in bear markets.
Equity index futures have their place in a diversified trend following portfolio, along with commodities, rates and currencies. Running a TF model on only a group of equity indexes would not be advisable, and on a single index even worse.
In my experience, equity indexes is the worst performing sector for trend following strategies over time. It still makes sense to include them for diversification and they do have a positive expectancy over enough time.
Trend following is about throwing a slightly flawed die. It has a slight advantage, but very minor. To throw the die once is pure gambling. Throw it a thousand times, and you’ll be able to take advantage of that tiny flaw and win the game.
The problem is that many people don’t have the patience for that. They put all the chips on the table and throw once or twice before blowing up.
Great point. It is amazing how often the above is ignored. There is a barrier of entry into trading business and this barrier seems to be moving upwards over time as markets become more efficient (need for diversification over groups, methods, time frames, etc.).
Andreas,
I’d seccond on Glen’s comment on your book, it is one of the most valuable, most complete books on this I’ve ever read. Great works, great effort writing it, I liked reading it (and took some valuable informations to improve my own bottom line).
I’d like to see one quote from you in every book written about trading:
“Religious dogma about a trading methodology is not helping anyone but system salesmen. It’s dangerous to view yourself as a trend follower (replace this with whatever methodology you like). A much more pragmatic way would be to look at yourself as a systematic trader. Investigate what works and how it works. Trend following is a great concept but be very aware of its limitations. Adapt your rules to reality and overlay satellite strategies where needed. Hard work, quantitative modelling, research and pragmatism will get you to your goal.”
Thanks, Martin! Appreciate your comments. Oh, and I’m going to make a killing on the royalties for that quote… 🙂
As they say in English, it’s an “eye opener”. 🙂
>It’s dangerous to view yourself as a trend follower. A much more pragmatic way would be to look at yourself as a systematic trader.
…For the pupil might lose his “presence of heart”…
Andreas,
You have great points. I did a simple test on US stocks with a naive trend system. Hold N stocks at or near all time high; rebalance every month; drop any stock that lose money;hold winner forever. The test starts from 1/2000 to 12/2013. The results are very strong. It turn out I did not set the Factset screen correctly to remove the survivor bias. Still there are some valid points. You can have 90% of your stocks getting stopped out losing money. But a few stocks (i.e Green Mountain Coffee which goes up 30 folders during the test) can have such a huge gain that you still end up doing quite well. Positive expectancy works the same way as in trend trading futures. Except you need to have a very long time frame and shorting does not add any value. It seems to me the game of stcoks is not about trading but filtering out good growth companies.
By the way love your book.
Bo
Hi Andreas,
I’m glad you raised this issue, I have been struggling with the concept momentum trading versus trend trading, many traders/authors use them synonymously, in your mind what is the difference. In my mind momentum trading, is when share is selected on previous months return performance.
Cheers
Robert
Hi Robert,
I view it as fundamentally different approaches. Still, the semantics are of less importance and there are no formal definitions here. My main concern is with misguided blind faith in trend following as a trading methodology. Trend following models are traditionally very simplistic, designed to capture medium to long term trends in a broad set of cross asset futures markets.
The simplicity of this approach has attracted a large number of scam artists, selling trading systems, coaching and other nonsense for inflated prices to the unsuspecting. These scam artists are in almost all cases people who have never traded in a professional environment, lack any sort of quant background and are generally quite clueless. Their common mantra is that Trend Following Works On Everything (if you just buy my trading system)!
That assertion is as dangerous as it is stupid. A professional wouldn’t say something so silly. Still, it sounds cool enough if you put the right spin on it and you can make money pushing that agenda.
The point I was hoping to make in this article was that the real world is a little more complex. To settle on a trading methodology as if it was a religion or a foot ball team, and stick to it no matter what the real world tells you, amounts to financial suicide.
Single stock trading is different. Apply standard trend following models to them at your own peril. I’ll return in futures posts with more specifics on what can be done with single stocks.
Andreas –
Just finished the book – cover to cover in 6 hrs. Great read.
Going live with a newbie systematic fund ( £1m ) next month.
Bit tight for this game I know, but we think we can make it work.
Broadly similar approach, with a smattering of Machine Learning for the counter-trend.
If you’re interested, will keep you posted on progress.
Best
Dave
Congratulations on going live, Dave. Please copy me on your monthly.
Going live with a million quid sounds difficult. Not only due to the difficulty of taking positions sizes in the futures world, but also due to the run-away cost for starting funds these days. Compliance is getting very, very expensive.
Hi Andreas,
How do you think about trend following on major stock indices? Does CTA on indices perform better or worse than on commodities? Thanks,
Leon
Hi Leon,
My experience is that equity indexes, while profitable, is the toughest sector over time. I would include it in a trend following universe, but applying TF only to that sector can be risky. If you would ask me to pick only one sector, I’d go with agricultural commodities.
Mr. Clenow, why is this number 2?
2. Take the state of the overall markets into account. You can’t keep going long in a bear market and expect to gain.
This makes me think you have a different definition of trend following than most trend followers, even though many of your concepts make sense to me. Why would any trend follower keep buying in a bear, this is the opposite of trend following. One should short in a bear.
This was a great article.
Even though the index is in a bear market, you might get buy signals on individual stocks. Most bear markets have strong interim rallies where this can happen. If you don’t use a long term trend filter on the index, you risk getting buy signals in individual stocks during these rallies. Such signals have a very low success rate and my point was just that you’re better off not taking them.
Short trend following on stocks in a bear market is also a bit of a problem. Even in 2008 it was extremely difficult to make money from short trend following. The trend following business had great results that year, but that most mostly due to long bonds, long metals and long energies. Shorting is a very difficult game.
Excellent (dearly needed) article.
I’ve read your book cover to cover a few times in 2013. Like all that you publish its brutally honest (emphasis on honest). I seldom comment on blogs but wished to chime in. One a sidenote, very glad this article comes first in https://www.google.fr/#q=%22trend+following%22+%22stocks%22 .
In no specific order, on the myth of “trend following on stocks for retail joe/jane” :
* Simple is not the same as Easy
Vendors/Authors often either make “trend following” seem easy, or don’t tell the gullible (talking from past experience here) retail trader who’s a large part of their target audience that its do-able, only omitting to tell her/him that the futures market is for the well capitalized and that vanilla R. Dennis trend following sortof assumes you don’t pick instruments except for liquidity concerns.
* Can I Haz simple then ? Welcome to Stock Selection
On stocks on the other hand, retail Jane obviously cannot trade all of them. If trading a basket, what basket and why said basket ? All retail systems I know of which call themselves “Trend following on stocks” (from books, seminars, and not implying their are not good or bad) are derivatives of O’Neil and/or Darvas’ approach, so the assembling of the basket is “rules based discretionary. Then the retail trader slowly comes to realize that the picking (or pure filtering if managed to be done systematically) is the truly difficult/hazardous part.
If the Wilcox/Crittenden 2005 report (page 7), or Meb. Faber’s “Total Lifetime Returns for individual U.S. stocks” (http://www.theivyportfolio.com/wp-content/uploads/2008/12/thecapitalismdistribution.pdf) show something, its that there’s a serious fat tail in the stock market’s returns distribution. Wilcox & Crittenden also state, for obvious reasons, that ” The mechanics behind our portfolio management system are not disclosed (…)”. Bingo. Selection.
* Data & stock market specifics
As there’s no systematic without some degree of backtesting, one has to obtain data. Either go with free data and play backtest roulette, or buy data (retail trader likely had rather buy stocks than buy data). So : Dow stocks ? S&P500 stocks ? Russel3000 stocks ? How about Euronext… If Joe only trades US stocks, good for him ; if not, welcome to a world of pain.
But as you also point out, for retail Joe/Jane, the fun only starts here. Splits, spinoff, cash or stock dividends, delisting, rights offerings and other weird corporate actions…
And not even talking about the fact that Joe/Jane won’t backtest such a complex universe in a drag and drop retail trading software. Learn some C#, python, R at the very least. Easy is long forgotten, and simple begins to take a mythical coloration.
So basically “vanilla trend following on stocks” is hard to even evaluate. Not talking, again, about the larger category of “buy high & sell higher” trend trading methodologies (Darvas, O Neil, …), which I use, and which have a strong discretionary component and are not per say “trend following”.
* Short side
Ah and there’s Bear markets. What’s plan B say for japaneese folks in the past 20 years, because if most of the stocks traders in the Market wizards series shamelessly confess shunning shorting stocks like the plague *even* in a bear market, retail-trend-following-Joe/Jane better think twice 😉
Obviously neither simple nor easy.
Spot on, Suvarow. There’s so much unrealistic nonsense being peddled to retail. Too much money being made by selling romantic dreams of easy cash.
Stock strategies are much more complex than futures. Still, it’s also easier to peddle to the unsuspecting by people who either don’t know better or who don’t care.
Let’s see, if I find the time, I’ll write a book about systematic equity strategies.
Hi Andreas,
I couldn’t agree more with all you have stated above. i have got a few general questions in regards to trend-following .
Over the last 10-15 years trend following funds are having an increasingly more difficult time extracting profits from price velocity/acceleration strategies that trend-following encompasses. AUM employed in some of these funds have grown exponentially, and if they are all doing the same thing wouldn’t that be a concern? Is trend-following popularity the beginning of its own demise? and has it fallen victim to ever changing cycles Neiderhoffer (I know he is very biased against trend-following) speaks off? How does one who uses such strategies protect himself? ( Serial correlation of various indexes and commodities has been shown to be negative! Although its gone through cycles of positive and negative periods.)
On another note, while backtesting trendfollowing strategies its become evident that longer term holding periods have faired much better than short term, wouldn’t you agree? Why do you think that is? (Central bank manipulation ?)
Through some basic backtesting I have found equities and currencies to be much noisier and have yielded unfavourable results in my tests, even on longer holding time periods.
A little while back i remember reading about Ed Thorp and his interest in starting a trend following fund, which didnt happen. None the less he speaks of employing other data filters to his price based entries and exits. Those included term (backwardation and contango) , inventory levels, etc. There is very little about such ideas in the public domain for simpletons like myself to read and digest. Employing volume analysis, COT data, Sentiment, and fundamentals to various commodity markets could create for more interesting trend following models? I have to say that what drew me to exploring trend following is the simple logical concept of limiting your losses and having no limit on your wins.
What a rant! I hope you answer a few questions I have stated above.
Thanks,
Slav D
Hi Slav,
Trend following has been tough for a while. I’d say that the main reason has been the single factor regime we saw for a couple of years. At any given time, there was just one factor that mattered. It could be a potential Greek default, a potential US default, a potential Italian default, or a few other previously unimaginable scenarios that almost happened recently. Everything, all asset classes, depended on a single short term factor and therefore diversification just didn’t exist.
The other factor is of course the ‘manipulation’. That is, central banks deliberately killed volatility, and trend followers need vola to survive. Yes, the evil FED did it and for once, at least in a tiny detail, I’ll side with the conspiracy nutcases who hate the FED.
Many players were knocked out lately, that’s for sure. But there was far too many johnny-come-lately after 2008. That usually happens after something had a great success. Everyone jumps on the bandwagon, and most will fail. That’s business as usual.
Remember the key point: This is not a democracy. It’s not one person, one vote. It’s one dollar, one vote. It takes tens of thousands of small CTA startups to make a dent in Winton’s or even Transtrend’s impact. There’s a group of billion dollar trend followers who make all the rest into a rounding error.
Many also failed to adapt. Markets change. Running a ‘turtle’ strategy from 1985 today is simply dumb. It’s not the strategy’s fault. It was developed in a different era. Things are very different now.
One of my best performing trading models is based on taking advantage of the flaws in standard trend models. Whether this model works because too many people use standard models or not, I can never know. All I know is that it works.
Taking term structure into account is critical. The other factors are very optional.
Hi Andreas,
Thanks for such a quick reply. I applaud you for your effort and ability to adapt to changing regimens. One strategy I am currently investigating is the exploitation of market structures and obvious stop placements. One only needs to look at a few years of data to determine the amount of times previous swing highs and lows have been breached only to force liquidation of weak hands and a resumption of over trend. Limit buy order below obvious swing lows are some of the best places to be a buyer and very counter intuitive to classic technical analysis! Its a tricky game!
Also if I may add, one would be a lot better off investing in stats/programming/ math and econ books to become a better trader than the mumbo jumbo published today!
Slav
PS link to the Ed Thorp interview about trend following.
http://abnormalreturns.com/ed-thorp-on-trend-following-an-excerpt-from-hedge-fund-market-wizards/
Hi Andrew,
I read through your excellent book.
I tried to use a basket of 40 futures, for which I had access to data. I can not seem to get the return you get in your trend following strategies. In fact I am getting about an order of magnitude less. Being a newbie in this area, I must have got a fundamental mistake in my understanding of the position sizing or how futures work.
May be I can pose a queston for you. Taking your position sizing rule of using 0.2% of capital:
units = Capital * 0.002/(atr * point value).
Now let us say for a whole year, I traded 30 contracts and all of them had an atr=0.01 i.e., 1%. The max my strategy can appreciate if all 30 contracts had a 10% price gain is:
number of contracts * units * (point value * delta(price)) = 30 * Capital * 0.002/0.01 * 0.1 = 0.6 * Capital
I.e. a max capital gain of 60% only. Does that sound right?
Cheers.
Hi Sudipta,
Your calculation is technically correct, though more importantly, it implies highly unrealistic assumptions.
You assume:
* Position bought before year start and held the whole year. No trades for a whole year.
* Exactly the same vola for all 30 markets.
* 10% gain on all positions.
These assumptions are very wrong and very important.
With your assumptions, let’s assume for the sake of calculation that we’ve got 30 markets, all with an ATR of 1%, showing a yearly gain of 10% and they all have a point value of 100 and a price of 1,000. Our total capital is 10 million and we’re working with a risk factor of 0.2%.
Target daily variation would then be (0.002 * 10M) 20,000. One contract would have an average daily variation of (0.01 * 100 * 1,000) 1000. We would therefore buy 20 contracts.
The notional exposure of that position is (20 * 100 * 1,000) 2,000,000. If those contracts move up by 10%, we make a gain of (0.1 * 2,000,000) 200,000.
If we now had 30 identical positions, that would mean a gain of 6,000,000. Your total notional exposure, at the start of the year, would be 600%. Yearly return lands at 60%.
Of course, if the vola of the markets were even similar, we wouldn’t need to go through all of these steps to begin with. Try with different asset classes and shorter holding periods and you’ll see how things change. Compare a bond future with a metal future for instance.
andreas ,you live in swiss ,can you talk french? thks sébastien fourne
Je suis désolé, Sébastien, mais mon français ne est pas très bon. Je habitais cinq années à Genève, mais ce était il ya longtemps.
I moved from Geneva almost ten years ago, and my French wasn’t very good even then. Odd city really. Where else can you live for five years without meeting any local people? I speak better German, but my native language was Swedish.
My book’s coming out in French next year though. Hopefully translated by someone with French skills beyond my rusty school book level…
Hi, Andreas…..
Thinking outside the box, and trying to do what others haven’t in the past, I wanted to ask you if you have ran trend following strategies or simulations on non time based charts?
I have been collecting tick data and am currently experimenting on running trend strategies on stocks using bars constructed based on % of float….. One could also use % of shares outstanding or for futures average OI in n-period look back (4months for ES for example). In order to incorporate $ volume one can construct bars based on % of n-period $volume( # of contracts x price at which trade was executed).This seems to produce much clearer looking charts with fewer overlapping bars. I thought this to be a novel idea, and wanted to see what your thought is on this? I cant see the point of looking at trend following through the same lens as everybody else. My only limitations in exploring this on portfolio level backtests is my coding ability. Thinking of using R or Matlab for it.
In addition I also wanted to ask you about the method you use in order to determine favourable trending or non trending environments, in other words how do you rank favourable stocks or quantify trendiness?
Thanks
Slav
I never tried it. Never know until you try. Build and see what you get.
R or MatLab should be fine. I mostly build in RightEdge, but nothing wrong with your choices.
I won’t reveal my ranking criteria at this time. I’m considering writing a new book based on my methodology for that, and I wouldn’t want to give it away too early…
Hi Andreas
Nice site and posts. I discovered your work after listening to a podcast from Kathryn Kaminski on Top Traders Unplugged.
I agree with your ideas about trend following on stocks, and that a ‘momentum’ style seems to work better. I agree with the things you note above about what can be done, too. I noticed another post by you, about the Tetsudo fund, where you listed a few things similar to the above. However, could you please clarify what you mean about the following two points, please:
1. Risk allocation, and re-allocation, is critical.
2. Adapting to market regime makes a world of difference.
Is number 1 relating to stock volatility? Is number 2 relating to taking into account bull and bear markets, eg going to cash in a bear market?
Kind regards
Shawn
Hi Shawn,
Nice of Katy and Niels to mention me. I’ll return the favor and say that her new book is essential reading for serious CTA quants.
Risk allocation: Allocate risk, not capital. Notional amounts is not a good measurement. Allocate based on proper risk measurements, whether simple vola analysis, VaR or similar.
Re-allocation: In retrospect, it would have been more clear if I had used the word ‘rebalance’. Point being that markets are not static. If you allocate a certain risk today, you won’t have the same risk on in a month. To maintain a desired risk level, you need regular rebalancing.
Market regime: You can’t trade the same way in bull markets, bear markets and sideways markets. Equity based models need to adapt to the current environment. Going into cash in bear markets is certainly a valid course of action. There are other ways to adapt as well, but the critical point is to be aware of the market regime and behave accordingly. For instance, buying momentum stocks in a bear market is a horrible idea.
ac
Hello,
When you say trend following, do you also mean technical analysis?
Thanks for the earlier reply, Andreas.
As I mentioned in my first post, I agree with your general premise that general unmodified trend following strats (taken from futures) wont work well on stocks, but I think they do work well when modified. Which is what I believe you infer in your post here, and suggest that it is more semantics – trend following versus momentum. Anyway, I wanted to get your opinion on the following two studies. I think they support what you say – modified strats work, regardless of what we name them.
http://toptradersunplugged.com/wp-content/uploads/2014/05/Trend-Following-on-Stocks.pdf
The pdf located at this url:
http://www.michaelcovel.com/2009/10/22/does-trend-following-work-on-stocks-part-ii/
Kind regards
Shawn
You’re right that the semantics is up for debate, Shawn. I prefer to use different terminology, because I see it as a very different type of strategy. Trend following strategies need to be changed so much for stocks that a different name is warranted.
On the Blackstar/Longboard papers: It’s good work, but highly over simplified. Perhaps they’ve done this on purpose to make it more accessible.
I find it a little odd that professionals like these guys use hobby trader type terminology at times. For instance, they define risk as distance to stops. Try explaining that to an actual risk manager or risk controller and you’ll get a good laugh out of them. “Don’t worry about the VaR, stress testing, vola measurements, covariance matrices etc, just look at my stop distance”… Well, these guys may have deliberately chosen to use hobby level terminology to make the papers more accessible, I don’t know.
On their strategy: The universe definition is a little fussy for my own taste. They start by considering all stocks, then filter for stock price level and liquidity. The assumption that high stock price means big and solid company is extremely US-centric and barely holds up even there. In my view, notional stock price is irrelevant. Also this method will end up in a universe that’s likely far wider than you’d realistically trade. It’s more realistic to consider only members of a selected index on any given historical date. That’s a realistic method to select the stocks going forward, and a solid method of testing it historically.
Entering on all time high doesn’t seem like a very good criterion. Again, they have have chosen this deliberately to simplify things. Such an odd criterion can results in many unwanted scenarios. A stock that took a 90% hit in the 2000 crash and trended up every since may still not be back to where it started. It may still be a very strong stock.
This entry method doesn’t care about the stock’s volatility at all. If a stock just made a 50% jump on a takeover offer, this simplistic method would just go and buy it. That’s not a momentum situation. There’s lots of these situations where ‘buy all time high’ is a bad idea.
I’m very skeptic about the use of ATR based stops for this purpose. In particular such a wide stop as 10ATR. You’d really have to question whether a stop so wide makes any sense to have in the first place. It also has many unwanted implications, one of which can be seen on page 4. If a stock loses momentum and goes sideways, it can go on forever without being kicked out. It will take up space in the portfolio for years without performing. There may be much better stocks out there, but you can’t buy them because you’re stuck until ATR stop hits. No, I’d throw out that ATR stop completely and replace for a required momentum rank compared to peers.
Their methodology lacks any sort of ranking. It just randomly buys whatever stock happens to make an ATH first and holds it until it hits -ATR10. Doesn’t make any sense to me. Your stock selection is random and you keep holding stocks that may not be the best candidates.
They are very fussy about their position sizing. Their terminology implies that they’re using professional methodology, but I didn’t find any details that could be tested. “Multifactor utility function” for instance, that could mean anything. They do imply that they use risk parity sizing with a rebalancing mechanism though, which a what I’d recommend as well.
I’m puzzled by odd statements like ‘limiting portfolio risk to 30%’. Given that industry professionals wrote it, I would have assumed they’re talking about annualized portfolio volatility, but they seem to mean ‘the amount of money we lose if all positions fall down to their stops at once’, which has very little to risk management.
I also wonder how they prioritize trades. If there’s a large move up in the market, and 50 stocks make an ATH on the same day, what do they do? Buy in alphabetical order?
My impression is that they have deliberately oversimplified the ideas to make the research paper more accessible. I couldn’t imagine that they actually trade like this, and I wouldn’t recommend anyone to try it.
I believe that they just want to make a statement about the momentum effect in the equity markets, and as such it’s good work.
Andreas
Thanks for the quick, very comprehensive, and thoroughly awesome reply! Much kudos to you!
I dont really have anything to add, but that I understand and agree with you. I, too, dont believe they would trade in this manner, but as a simple illustration it works to show momentum trading in equities.
Thanks again.
Kind regards
Shawn
Trend following does work on a stock index:
http://sixfigureinvesting.com/2013/09/modified-davis-method/
Note that the above is an out-of-sample test, not an optimized back test.
I’m reading it now, and I’m actually struggling to understand what this approach is about. That’s not a criticism of the approach. I don’t understand enough to have any opinions about it.
Is the strategy trading the SP futures, the SPY ETF or the constituent stocks? How is an entry triggered? How is the position sized? How is the position exited? Any rebalancing logic incorporated? What does it mean to dynamically construct a trend line and how is that done? What is such a trend line based on, and how is it used for the trading rules? What about the short side, how is it handled?
Don’t get me wrong, Frank. I don’t want to criticize your work. What I’m saying is just that from the description, there’s not enough to comment on.
Hello Mr. Clenow,
Thanks for the critique. I guess that I’ve been thinking in this way for so long that it never occurred to me to be more specific than I was in the article. So here are the answers:
Is the strategy trading the SP futures, the SPY ETF or the constituent stocks?
*** The article reports the gains/losses that are obtained trading the Russell 2000 index. Thus the ETF IWM (and its inverse for shorting) can be used, or Russell 2000 futures can be used. However, the method uses no leverage, so I would not recommend trading futures contracts with a greater face value than one has funds.
How is an entry triggered?
*** Only weekly close values are used, which filters out a lot of the “noise” of the daily swings. For a simple example, assume the market is declining and you know that, as always happens, sooner or later it will reverse to the upside. Keep track of the weekly closes, noting the lowest value. When the market turns up, there will be a weekly close that is some percent above the low, and one purchases at that point. The reverse strategy applies when the market is rising; keep track of the highest weekly closing value and sell when there is a weekly close below some percent below the high. Ned Davis used 4% for both thresholds and got fairly good results. My back testing uses different percentages to find the best performance; something other than 4% usually is selected but not too far from it. Whatever the back test selects is used for the next trade. The back test results are not recorded; only the forward test results using the back test parameter values are recorded. For selling, I add the requirement of penetration of the trend line as described below. Davis did not use a trend line.
How is the position sized?
*** Leverage is not used. Long positions are either 50% or 100% of all funds invested (usually the latter). Short positions are always 50%. Thus compounding results.
How is the position exited?
*** Described above, plus the trend line as described below. See also the answer regarding shorting below.
Any rebalancing logic incorporated?
*** Not explicitly, but I guess maybe it does occur implicitly as time goes on and compounding occurs. Maybe you could call the selection of a 50% or 100% sell rebalancing. I don’t think in terms of rebalancing, but maybe you can decide whether it happens based on my description of how the method works.
What does it mean to dynamically construct a trend line and how is that done? What is such a trend line based on, and how is it used for the trading rules?
*** The trend line is used only on long positions. It is drawn forward from the lowest weekly point reached before the buy signal occurs, and has a certain slope based on highs reached after the buy. The slope is one of the parameters used in the back test. If the market is still above the trend line when the required percentage weekly drop occurs, the sell will not be taken. Instead, the method waits until the trend line is penetrated to the downside.
What about the short side, how is it handled?
*** A sell has to occur as described above, plus there has to be a divergence in breadth as given by the cumulative advance-decline line on the NYSE. The trigger is higher price values in the index accompanied by lower values in the cum a-d line. If there is no divergence when the sell signal occurs, only half of the position will be sold. Since the market has declined to a sell, a low point has been established (new lows may be established in subsequent weeks) setting up the possibility of buying again at the buy percent threshold when the market reverses. At this point (weekly close above the buy threshold) a purchase is made using all available cash, thus returning to a 100% invested position. Note that at times the method will be 50% invested with no possibility of further selling. I have not been able to find a way to sell the other 50% that improves results. To me this solidly demonstrates
the value of paying attention to cumulative breadth. In fact, in the current bull market, my method did no shorting between July 2009 and July 2014.
See more at: http://www.followingthetrend.com/2014/04/trend-following-does-not-work-on-stocks/#comment-17418
If you are interested in more detail, send me your email address and I will send you actual trade results. Many test variations are possible. For example, the best parameters were selected by the back test from 1960 through June, 2008. Trading the method forward from July, 2008 to the present using these parameters resulted in an average annual gain of 14.5% with a maximum drawdown on closed positions of less than 10%. You could see exactly what it did around the 1987 crash or the huge 1982-83 bull market, etc. Or, what would have happened if the Russell 2000 had been traded forward from its inception in 1979 using only the parameters selected from the Value Line back test through 1978.
Just another 2 cents’ worth… Davis’s initial algorithm was “Buy on any 4% or greater increase in the weekly close, sell and short on any 4% or greater drop in the weekly close.”
I agree with you that simple rules work just fine, and this is about as simple as you can get. I just found that values other than 4% work a little better, and that the trend line eliminates a lot of false sell signals. Limiting shorts to periods of breadth divergence increases the probability that a short will be profitable. Shorting at the 50% level is easier on the nerves.
I formulated the trend line in 2008. I did not short but sold everything in September. By March 2009 I was back on the long side and since have more than doubled my accounts.
In spite of all of my words, this is a very simple system. You’ve got a great site here and it reinforces my convictions about markets. So thanks for that. If you haven’t seen this, you might get a kick out of it:
https://www.youtube.com/watch?v=LiE1VgWdcQM
Best wishes, Frank
Hi Frank,
A few thoughts on your approach:
Most importantly: Always test on things that can be traded. An index is by definition not tradable, so it doesn’t make sense to run tests on it. Index futures, index ETFs and similar securities based on the index are of course completely valid to use. But testing on an index itself is not advisable as it’s not something you can trade in reality. The results may differ more than one might think.
If I understand correctly, you’re considering using inverse ETFs for trading the short side. I’d very, very strongly advise against that. Short ETFs are rebalanced daily, which means that you’re trading gamma and not delta as the names might imply. The returns are extremely different from inverse index returns. The short ETFs are structured products with primarily gamma exposure. Don’t trade short ETFs unless you have a full understanding of derivatives math.
It still seems to me like the rules are a little difficult to quantify. That’s not to say that it’s a bad approach, just that it’s hard to properly test it.
Andreas
Hi Andreas,
Thanks for your thoughts.
I mentioned inverse ETFs only for illustration. I short using only Russell 2k futures. I often check IWM against RUT. Over time they track very closely. I admit that one will get different results when using vehicles that are merely “based” on an index and not the index itself, but this method does not trade very often and my actual results over many years are close enough to the method’s results for me. There is a lot of “noise” in trading with real money no matter what one does. Precision is impossible.
My command of English is not adequate to explain precisely how my programs work, and I consider some of this proprietary so am leaving out some details. I have spent much time working and reworking this code and testing it so I know that it is doing precisely what I want it to do. And in fact it is far simpler than the writing of all these words implies. I code everything in C; I use no canned software other than the compiler.
Thanks again. If the performance of your new fund is reported here we’ll be able to see how the 2 compare in the future.
Best wishes,
Frank
Dear Andreas,
thank you for your very interesting article and thank you for all the comments. I carefully followed your 6 steps indicated above. But i find difficulties implementing #4 (position size) and i am seeking your expert advice.
I believe i have a solid ranking system (relative strength) for stocks but my results are highly dependant on how much money i allocate on each new position (when some stocks go out and others come in).
what would be your advice on how to proceed to solve my problem?
I am looking for a simple solution as the concept would suggest but, as you said, the most difficult thing is to implement it.
Best regards,
Andrea
Two words: Risk Parity.
A large part of momentum returns come from risk parity sizing.
Thanks
I think i ve got good results, would be happy to share with you
Hi Andreas
I just discovered your site yesterday. I think it is great. I am confused about your term TF does not work for stocks. I am a simple person. I think it is just terminology / definition of TF. There are stocks with great trends, both up and down. To me trend following just means that one discovers a trend, jumps on it and rides it. Obviously there is a vast difference between Stocks and Futures. Jump on signals for futures are easily programmed. Entry signals for stocks are not. I use “eyeballing” after initial screens for high performance stocks, which end up in my watchlists, refreshed every month.
But to me, both are trend following, wether one follows a trend of a stock or a future. Though entry decisions are vastly different. But I guess you have a stricter definition of a trend following system?
Hi Andreas,
I am still very new to the world of algorithmic training and have only recently discovered your site recently. I must say I greatly enjoy your to-the-point, intelligent, humorous style and of course the excellent, inspiring content.
I was wandering if one of the reasons that momentum or trend following with equities is so difficult might be the fact that it is actually not easy to reliably evaluate the trend of a stock, or even a stock index, in the first place.
There are of course many ways of doing so, but the specifics of the estimate should not really change much the information contained in it. If if take the average of a series of log returns as a measure of a trend (since this is in line with the definition of the geometrical random walk process — GRW), we have to be aware that such an assessment of the drift parameter (usually denoted by mu) carries an uncertainty, which is proportional to the standard deviation of the sample (denoted usually by sigma, which is a the same the time the volatility or random component of the stock movement in the GRW process), and inversely proportional to the square root of the size of the sample (N, also “duration” of a series). Suppose we sample with daily frequency and take a years’ worth of data (N = 252). The volatility of a typical stock (or even an index) is unfortunately so great, and the drift usually so small, that the resulting standard error of the mean (the uncertainty of the mu parameter) is about ten times as large as its value! That means that our measure of a stock’s trend has a signal-to-noise ratio of only 0.1. Not very useful, even if we trade a large basket of stocks based on such a criterion.
Do I have a point here? Or is it complete nonsense what I am peddling?
Many thanks in advance for any comments!
Sorry, it’s me again,
Just to summarize the idea of my last post before this one: in whatever way one estimates the trend, one should make sure, I think, that the estimate is statistically significantly larger (or smaller, for shorting) than zero.
I have only ever “worked” with equities (Quantopian). Is the statistical significance (large signal-to-noise ratio) easier to achieve with options? What are the typical S/N values?
Regards,
Tim
Timeo Danaos et dona ferentes.
Hi Andreas,
It was probably not entirely clear, but in my last two posts I was referring to your article entitled “Trend following does not work on stocks”.
Regards,
Tim
Andreas, I have been moving through your material and find it quite insightful. I am new to system type trading and have been working on developing 2 systems. I have been having a problem getting results on individual stocks. I have been having much better luck with ETF’s. I have also had some luck with stocks that are tied directly to a commodity. Do you feel these two categories can work with trend following strategies or do you disagree. Thank you for any feedback. Your information is helping a new system trader learn.
Fred
Hi Fred,
Broad ETFs can be easier, given their generally lower volatility. Commodity stocks can be highly volatile though, so I’m so sure those are typically easier.
I read the entire article, the first paragraph threw me off and second paragraph caught me. However, I dont really agree with any of what I’ve read. As an amateur trend following always works within it’s limitations. Meaning it’s just price action, there’s no guarantee a trend will hold or continue. To me it’s more of a guideline to use, something to find and plot, to follow, to trade within, to hold a position in. It helps with entries and exits and the confidence to act.