I have long found that it is easier to find good (i.e. high Sharpe ratio) mean-reverting strategies than good momentum strategies. Partly, that is because I was mainly a stock trader instead of a futures/currencies trader, and individual stocks mean-revert most of the time. There are exceptions, such as after special corporate events such as earnings announcements, and I have tested momentum strategies based on these events. But the success of even these event-driven strategies has been uneven, especially since more traders become aware of them.
Now that I am focusing more on trading futures and currencies, I have gradually been introduced to the world of momentum investing. There is a good book in this area that deserves to be better known: Joe Duffy's The Ultimate Trading Robot, which is an almost step-by-step guide to constructing futures trending strategies that rely on prices alone. Another example would be the London Breakout strategy mentioned by our reader Bernd in the comments here. After studying these examples, I realized why my previous, rather desultory, search for momentum strategies in the futures and FX markets had been in vain: the overnight gap in these markets seems critical. For futures, the overnight gap is obvious, but in the case of the London Breakout strategy, for example, the trader has the task defining for herself what the optimal closing and opening times are in order to compute the gap. Intraday trend without an overnight breakout does not seem persistent enough to be traded profitably. I also wonder if there is a more elegant (i.e. mathematical) way to quantify such breakout phenomena without using the traditional technical indicators.
If you know of ideas for good momentum strategies, you are most welcome to share and discuss them here!
I generally trust your book reccommendations, but a quick google search on this one looks dubious. Claims of 1000% annualized returns, etc. Are you sure about this one?
I prefer to separate asset class decisions from momentum at the sub-asset class level. For instance, a cyclical industry might rally strongly simply due to its high beta if the market rallies. Take the idiosyncratic returns, calculate the 2-12 month return (first month tends to have some mean reversion), scale that by the idiosyncratic volatility. Once a week/month (they won't change as frequently as your traditional signals), convert these to a Z-score that can be used in some other part of the view construction process or form a portfolio of the top 25%, bottom 25%, and middle 50% and track the performance. You could do this within each asset class or across all the asset classes.
You could also take some views at an asset class level as well with a similar approach. The trick then is methods to combine views together (Black-Litterman/Entropy Pooling). Once you have a method to combine disparate types of views together, you could easily incorporate mean-reversion and momentum strategies into one portfolio.
At SensoBeat (www.sensobeat.com) we assume there is a "momentum" to news items, and we try to track that momentum (stock "buzz"). We do it just for stocks but can be adapted to other fields as well, as long as they can have a "buzz". We thought of using it for algo-trading, which is more relevant to you but making it fully automatic was a big problem. E.g. the sentiment of a news item is positive, but if it misses the expectations, the effect is negative. We decided to go for a decision-helping tool, that the trader does the final decision. Would be interesting to hear what professional algo-traders think of the idea
As I mentioned in my book, I seldom find any published strategy profitable as is. Often, it won't even stand up to backtesting, not to mention live trading. So I won't put too much weight on the 1000% claim. The important take-away from the book is some techniques I didn't know before which I can modify and improve on.
Thanks for your idea. Actually, this reminds me of a whole class of momentum strategies that I read about: basically holding a long-short portfolio based on some simple ranking criteria such as the lagged returns as you suggested. Apparently this works not only in stocks, but in commodities futures too. (Google the paper by Joelle Miffre and Georgios Rallis called "Momentum in Commodity Futures Markets").
The problem for me (but not necessarily for, say, pension funds) is that the holding period is too long, and the return comparatively low. The long holding period necessarily imply that the portfolio suffers interim volatility thus suppressing the Sharpe ratio. Which is not to say that your suggestion necessarily has this problem.
Thanks for sharing your product with us. In this context, I should mention that the company Ravenpack has a similar news sentiment indicator which I believe can be used for algorithmic trading, and Ravenpack's indicators can be integrated into Alphacet Discovery's platform.
Also, if one is interested in news gathered from the internet but not necessarily from financial newswire, the company Recorded Future also offers similar sentiment data through an API suitable for algorithmic trading.
Thanks for pointing me to Ravenpack. They do sentiment analysis which a few other companies do as well (www.thestocksonar.com, www.sentigo.com). They all try to decide if a news item is positive or not. SensoBeat tries to answer a different question: how much has the news item spread (in real time)? As far as we know this information is not available to traders. 2 similar items from 2 different companies can have very different spread and therefore different impact on the stock. When the trader reads a news item from his favorite feed he doesn't know if this news is now starting to spread, is it already "all-over" the internet, and so on.
That is certainly an interesting feature. Good to know this product exists!
The thing with momentum is that it can keep on going and going or it can be a dud. The best rule that I find to trading momentum strategies is just manage your exits and never set a target. The saying "limit your losses and let your profit run" may be simplistic but it's so true.
Another idea, which I picked up from TraderFeed long ago, is to:
1) identify a trend; and
2) enter at a counter-trend.
Effectively, buying at local minimums in a bull market, for example.
This is such a timely article! I was re-reading your book again and you suggested that if one has a low level of capital, strategies with leverage (such as futures and forex) are probably the best to start out with. However, having had no experience in trading futures or forex, what would you recommend to be the best book / website to start out with?
Yes, with momentum strategies I like to have stop loss but no profit target. Conversely, with reversal strategies I like to have profit target but no stop loss.
However, just as a frequently recomputed profit target can actually act as a stop loss, a frequently recomputed stop loss can become a profit target as well.
In terms of trading futures, you can start with Joe Duffy's book as I recommended.
For FX, I learned everything I know "on the job" and from my ex-partner in my hedge fund. Maybe some readers here can suggest a good book?
Thanks Ernie. Just ordered the book today so hopefully will get it in a week's time or so.
Found a website that's good as it really starts from the basics.
Ernie, do you feel that mean reversion isn't a good strategy in forex? I tried pretty hard backtesting EUR/USD data looking for mean reversion at various timescales, using mixtures of oscillators, and did not find anything useful. Perhaps it is because currency markets are so big they are only really moved by actual news, not stochastic trading patterns.
There are mean-reverting strategies that work in FX, but EURUSD is not a good candidate. One has to look for countries whose economic indicators are more strongly cointegrated.
For gaining more momentum on your Forex trading,visit http://ultimatesignals.com/members/go.php?r=38&i=l0
have you ever looked into Gaussian functions?
Many of us quants have used Gaussians in many forms, but maybe you can be more specific as to its usage in the momentum context? Maybe point to a online reference?
Hi Ernie, I am a big fan of your book and this blog, but I couldn't find any way to track your fund's performance. Where can I go to see that please ?
Please email me privately.
Hey Ernie you just made me famous :D!
I have been reading mostly about Forex for the last couple of years, as well as practicing different momentum-techniques with a demo account with no consistent success.
The only book I found "more or less" interesting to read, to get a good grasp of all the stuff was "Day trading and swing trading the Currency Market", by Kathy Lien, but again, it is just a "learn the basics" book.
About momentum-techniques to trade the fx market... I guess purely mechanical methods per se wont work (aka trading systems you usually find published in many fxforums). You need to 1st identify moments or situations when a trend is likely to happen due to external events that need to be taken into account ( after the London open/ after NY open /after news release like NFP...), or due to "technical" events like "Master candles" ( price ranging without trespassing previous bars maximums or minimums, then breaking them out violently). Maybe there are other ways to identify trends, but am not aware of them, so if anyone knows, I´d be really happy to know.
Am no expert, but after reading Ernie´s complete blog, I find the pair trading approach much more solid for my taste,and if applied to Forex, I guess it could be a nice way for the small speculator to try without a big requirement of capital ( because some brokers allow you to trade with mini and micro lots (mini = 10.000 € / micro= 1000 € i.e.)
I will mail you too Ernie ;)
What is your take on catalysts such as earnings releases when it comes to mean reversion stock swing trading systems? Have you done any statistical testing and decided to:
1) Do not enter stocks that are to report earnings before you'd expect to exit.
2) Enter smaller positions still hoping for mean reversion.
3) Enter no matter what based on price action ignoring any news.
I would avoid entering into positions of stocks that have announced or are expected to announce earnings for mean-reverting strategies.
>> "I would avoid entering into positions of stocks that have announced or are expected to announce earnings for mean-reverting strategies."
I have been avoiding earnings.. but my guess would be that there's still positive expectancy there.. just a lot more volatility. I had a hard time getting earnings dates for a large enough data set in order to actually test that - were you able to backtest this?
Free! Trade the odds. Complete statistical center for seasonal and statistical patterns for Dow, SP, Nasdaq, Dax... Search your best trading patterns choosing month, day of month, expiration weeks, moon phase, presidential cycle, politics etc. Extra tools: 1) What if... (Return n days after if change is...) 2) Amazing and profitable intraday statistics. 3) Day forecast for Dax and Nasdaq. Try it and profit.
Comments and suggestions are welcome.
Have you heard of PEAD: Post Earnings Announcement Drift? Research indicates that price will not mean-revert after earnings announcement.
I backtested such situations by web-scrapping data from earnings.com.
Thank you for your responses, Ernie.
When it comes to PEAD and mean reversion testing with earnings.com scraped data, what was
a) the average hold time for your strategy
b) and how many days before or after earnings would entry be excluded?
Most of the PEAD research I read about talks about a drift lasting 3-12 months, while my mean reversion swing trades don't last longer than 4 days.
A similar question to mine was raised at your blog at http://epchan.blogspot.com/2007/07/more-on-news-driven-trading.html by "vivkrish"
I can't disclose to you the exact holding period of my strategy, but I can tell you that the time scale is quite similar to your mean-reverting strategies.
PEAD momentum cannot possibly last more than 3 months, since there is an earnings announcement every 3 months which will trigger a new trend.
Ernie, I find that profitable momentum trading strategies for *portfolios* of futures, are not impossibly difficult to find. Typically they have an average winning trade hold time of 25-100 days and an average losing trade hold time of 5-25 days. (Because they cut losers and let winners run.)
Even the textbook triple moving average system is solidly profitable, even with punishingly large commissions and slippage, when tested on a diversified portfolio of 50 futures markets. (Be sure to use a *globally* diversified portfolio, to get more of that free-lunch noncorrelation). Adjust parameters to get >75 day hold times for winning trades, voila: profits.
Another simple and profitable momentum system for futures appears on Ed Seykota's website. He calls it "Support and Resistance" but it's actually a classical Breakout system: go long when price breaks through (above) resistance, etc. http://bit.ly/e5tTRo
With what sort of capital did you find it possible start prop trading (day trading) for a living?
With some upfront capital needed just to be able to day-trade in most exchanges and many macho hedge funds being happy with 4% above 3-month LIBOR these days (mentioning it as an indicator of an ambitious yet possibly realistic of performance expectation - note: LIBOR is pretty low these days as well), realistically do you think it's a bad period and fundamentaly different to the time you set up your own business?
Would we be talking about a bare minimum of 100-150k available purely for starting up?
Thanks much for your references. They sound very interesting.
It is possible to make a living trading $100-150K capital, but obviously not if the levered return is 5%. It would be a simple calculation to find the returns needed for survival once you define the profit you need.
Thanks for discussing Duffy's book. I have found some interesting wrinkles there.
Many highly directional/low vol trends occur in the overnight trading session of index futures, for example the 9:15pm to to 7am BST for S&P 500 Index futures, typically as reversal after very large moves, so go long if the trend/move is largely down, and go short if the trend/move is largely up.
Sorry i meant GMT! the strategy is correct though.
Thanks for the futures tip: will backtest this sometime!
from my experience, momentum are much easier to find than mean reversion strategies for commodities (futures)
Yes, I agree with you. Mean-reversion mainly works for individual stocks, while momentum mainly works for futures and currencies.
I can appreciate why you might no be inclined towards using adaptive strategies, especially if one is employing off the shelf tools such as the neural tool box in Matlab. Their neural nets mostly train in back propagation mode, burn through years of data, and your out-of-sample dataset is meaningless as the model is static and cannot adapt itself.
While trying to solve the overnight gap in futures and currencies, I decided to give neural nets another shot, and proceeded to design my own. Using compression technology, I managed to build one that burns through the minimum amount of data, e.g. about 2 months using hourly bars with 5 years of data, and then every new prediction is out-of-sample and the net integrates new bars toward the next prediction. At the end of the day, the net retrains itself thoroughly, and try to predict the opening bar for the next morning, and then goes back to predicting intraday closing bars.
I view mechanical systems as saying to a sprinter to train intensively for 4 years, then drop out of sight and go on a binge for a year, reappear, go back immediately on the track and you're expected to win the gold medal. Unlikely to happen. We need to constantly upgrade our knowledge to reflect upon the future.
Anyhow, ever since, it's been smooth sailing for overnight gaps in FX and futures trading.
Thanks much for sharing your improved neural net training method with us!
It is gratifying to hear that someone has actually made adaptive methods work to profit in trading. However, there are quite a few people who told me they have used quite simple technical indicators to trade the overnight gap in futures and FX, and in fact I have backtested one such method which produces very nice results too. So maybe sophisticated machine learning methods are not strictly necessary to produce consistent results for this particular market opportunity.
In today's Financial Times....
Please respect FT.com's ts&cs and copyright policy which allow you to: share links; copy content for personal use; & redistribute limited extracts. Email email@example.com to buy additional rights or use this link to reference the article - http://www.ft.com/cms/s/2/a29d2b4a-60b7-11e0-a182-00144feab49a.html#ixzz1J8QME6Pt
As the new money came pouring in, Madoff insists he planned to continue using his legitimate investment strategy, which was based around a so-called “black box” – a complex technique that relies on computing algorithms to select trades. “Before, I had helped develop products for the Chicago Board Options Exchange for index trading – I had built a model for that business,” he says. “So I thought I would put together a portfolio of S&P 500 stocks, with 85 per cent exposure, then used OEX [the S&P 100 index] positions as a hedge.”
This type of jargon sounds unintelligible to non-bankers, but is entirely typical – and credible – on Wall Street, and Madoff delivers it without missing a beat. Is he lying? It is impossible to tell; but as he speaks, he becomes so animated that colour flushes into his cheeks.
“But the problem [with my black box] was that to make it work you need to have volatility, volume and momentum. And, of course, we didn’t get that.” Soon after Madoff took in this new influx of capital, the markets became becalmed – which prevented his “black box” strategy from producing profits. Yet his new clients expected generous returns, and were soon demanding redemptions.
Hi Ernie, can you please share some of the techniques to trade the FX overnight gaps?
PS: i purchased your book and awaiting delivery.
spritrig: Ah yes, don't we all hate becalmed markets?
An example of the overnight gap momentum is the London Breakout strategy discussed in the comment by Bernd referenced in my blog post.
Ok so let me put myself in the shoes of a new trader with not so much capital and not a lot of experience, let´s say 10 or 20k, just trying to get a nice return on his savings, not making a living off trading
The trader finds a model that is profitable, he/she does not have the resources to automate his/her system using matlab ( needs to pay for it + making it able to interact with the broker platform )
The trader will develop its activity in Forex, for example, because of the better conditions to leverage his/her capital ( a 20% unlevered return in forex --> 40% if the leverage is 1:2, which is a quite conservative leverage...)
What would be the best choice for this trader to backtest the strategies? If this person trades part time and does it in the 4hr timeframe for example, will it be likely to achieve high sharpe ratios? or is that just inversely correlated to the timeframe?
I ask about this because, when you have 500k or 1Million or more, it can be profitable to invest 10 or 15k in automating your operations, even more, but if you are a 20k trader, that would just drain your capital...
Thanks in advance Ernest!
hello M chan,
I have been developing trading strategies on close to close data for about a year and i'm looking to start trading intraday (1 hour bars).
Do you know of any book were I could find the basics of the technics involved. For exemple what are the slippage assumptions? What kind of order execution should I use for backtest (trade on next bar opening price, VWAP)? etc...
Thanks in advance.
I assume that when you said "does it in 4 hr timeframe", you mean this trader research and send in an order with this 4 hours? Not that the trader execute many trades within this 4 hours? If so, then the trader can use Excel, or a standard FX automation program like Metatrader to automate the strategy. In fact, if the trader is good at programming but short of cash, she can use R instead.
Actually, you can just backtest what order types will produce the best backtest results.
As for slippage, it is equal half-of the bid-ask spread, assuming that your order size is no bigger than the typical bid/ask size.
Any recomended reading? (I'm not looking for strategies but for methods)
I learn most of these execution-related issues from actual trading. Few books will go down to such details. However, you can check out the Trading and Exchanges book on my Recommended List on my blog's right sidebar -- it does a good job of explaining the market microstructure.
Do you think that it is possible to find good mean reverting strategies in futures, that's a good question
Yes, there are good mean-reverting strategies in equity index futures.
i just started blogging on this platform a couple of days ago and was looking for like minded people to read and follow. this looks like a great blog.
you're more than welcome to comment on my page and i'll be looking forward to reading more stuff from you.
You can see how the momentum works in the fx from this program:
Do you have any experience looking for co-integrated currency pairs? Do you think we can apply the concept of pair trading to co-integrated currency pairs, just like stocks/etfs?
Sure, you can find cointegrating FX pairs as well.
There seems to be many studies on the profitability of Pair trading for stocks/etfs but not for FX.
Do you have any references to papers that have conducted such studies for FX pair trading?
It seems Pair Trading using stocks/etf seems more straight-forward than FX, in terms of position sizing.
Say we find a cointegrated FX pair using different base currencies, AUD.CAD and NZD.JPY. If we want to risk say just USD10000 on each long/short leg, how many lots should we get for each leg?
Hope to get your advice on this. Tks!
If NZD.USD=0.75, then
US$10,000 is equivalent to 13,333 units of NZD.JPY.
You have to convert both sides of the pair to USD first before running them through the usual pair trading strategies.
Instead of reading papers on FX pairs trading, I recommend reading up on basic FX trading. For e.g. study materials for FINRA Series 34 exam at http://thectr.com.
first, thanks for producing a very informative blog. i'm struggling a bit with how to find cointegrated pairs and triplets in futures but you're last comment re: needing first to convert to $ value in forex may have helped. before testing for cointegration (or even Paerson's r), should i first multiply the various contracts by their dollar value in order to get them into dollar terms? for example, multiply the ES contract by $50, and the ENQ by $20. i would then apply a hedge ratio to these values before testing. i've been getting hung up when trying to compare an equity index to a currency or commodity.
When the multiplier is a constant (as is the case for a future or ETF traded on a US exchange), the hedge ratio will take care of it automatically.
If the multiplier varies (such as a foreign currency where the "quote currency" is not USD), then you have to convert the time series using the FX rate back to USD first, because the P&L of this pair is denominated in the quote currency.
Could you elaborate on why "For futures, the overnight gap is obvious?" Many futures contracts trade almost 24 hours on Globex. Is there a "consensus" definition of the open and close in these markets in order to define gaps?
The gap in futures refers to the open and close of the pit trading.
Do you think is a good idea to apply momentum strategy during events like, for instance nonfarm payrolls announcement? I know that many traders use this technique to trade manually. From the other side there are many high frequency traders that do the same and using low latency technology. What is the point to compete with them if this guys are always trade faster?
Indeed, I haven't found much alpha at this short time scale. But that's because we never pretend to be HFT!
The Logical Trader by Mark Fisher has some SIGNALS to play around for Momentum Trading Strategies.
Paul Tudor Jones recommends this as one of his favorite trading books and has an excerpt at the beginning of the book.
You can also follow a blog on ELITE Trader: the ACD Method - this is the longest thread on any strategy on Elite Trader. The guys on this blog are manual traders, but automated traders like can take many of these ideas and systematize it.
Thanks for the tip, Harry!
I will check that out.
The only model which closely approximates financial markets is Geometric Brownian movement(GBM).Distance travelled under GBM is proportional to square root of time interval. Positive & negative shocks cancel each other over time in A diversified portfolio of stocks. On A net basis one can rarely beat the markets. According to option formula for A given stock S, if one month option costs 1 dollar then 4 month option on the same stock costs only 2 dollars because square root of 4 is two. Indirect way of stating this is that for A given time period chances that this stock would travel distance of 1d is 4 times compared to travelling distance of 2d.Option formulas may not be perfect 100%, but are damn good because trillions of dollars of derivatives are traded every day based on option formulas & market makers do not go bankrupt—whether they make market in puts or calls & stay out of speculation.
My question is whether following strategy is possibly sound in trading using computerized trading by A fund manager--
Computer puts in following order on stock “ S”.On the same ticket take profit & stop loss orders are always on the same side of current market price that day & not on opposite sides of current stock price.
1) Below the current price “P” put an order to buy that stock at “ P minus 1d” with take profit at “P minus1/2 d” & a stop loss at “P minus 2d”.This order is entered every day based on current price that day until executed whether at profit or with a loss--& same process is repeated on diversified portfolio of stocks all by computer with no human intervention. Similar orders are placed on the upside to sell short every day based on current prices that day using the same principals by the computer.No directional bet is ever made.
2)Stock prices go through noise every day on intraday basis. Chances that buy order would get filled at distance of “P minus 1D” is 4 times compared to hitting stop loss at “ P minus 2D” within same period of time on the same ticket order. With intraday noise, reversion to the mean, take profit order would get hit more times than stop loss on the same ticket order.
3) Under GBM, out of 4 episodes, 3 times there would be profit earned of “1/2d” each & one time there would be loss of “ 1d”with net profit of “½ d” on these 4 executions over & over again both on the downside as well as on the upside. Unfilled orders are cancelled every day when stock exchange closes. New orders are entered every morning based on CURRENT PRICE of the stock that day. Distance d is adjusted depending upon historical volatility of the stock so that decent number of orders are getting executed—if too many orders are getting executed then value of “d” is increased to slow down executions.With decent number of executions laws of averages would apply. Risk is controlled by controlling how many stock orders are placed both on the upside & downside. No directional bet any time—all orders are non-directional ,automatic & computer generated based on current volatility.Risk is also controlled by trading smaller amount of fund assets relative to total assets.
With low transactional costs ,fund manager would make money.
I would greatly appreciate your input into this strategy
I believe your strategy is a fairly classical mean reversion strategy. As such, it will only work if the market in question is NOT executing a GBM, but an Ornstein Unlenbeck (OU) process.
Post a Comment