Quant analytics: Does Trend-Following Really Work?
Bill Eckhardt one of the pioneers of the Turtle trading system is a firm beleiver. In a recent interview in Futures magazine he notes they are “flying in the face of a lot of data.” What do you think?
This issue is linked to the volatility arbitrage. If volatility is very low then the tren following is a good strategy, on the contrary with a high volatility, better to be contrarian to take profit on ups and downs.
Now when the observation frequency decreases,so does the volatility. Hence Trend-Following should be a long term strategy.
I guess the biggest drawback I saw to the turtle system is that you require a fair amount of buffer in terms of capital to withstand the short term dips in longer term trends. You could look for shorter term opportunities but the risk becomes higher and the data is less sound. Where does this leave an individual investor?
I have written several trend algos I can say that it works!
but the trend you are looking at should be very short.
20 minutes or less.
Jonathan, the trend should be short, the trend gives you the side and the momentum gives you the signals to go in and out!
Yes it works and quite well. Just one of many forms of providing returns. Has it’s own unique risk to reward and robustness. Many people don’t like the drawdowns that can result but long term compounding is the value the strategy brings. Lots of different types of trend following hard to lump them into one generic meaning.
. This is primarily dependent on the time horizon. 15m bars seems to be the norm in FX. I would think the ideal time horizon (chosen) be dependent on the volatility itself. Hence, FX traders seek large moves, while Futures trader seek large gaps.
I found better results with 5 minutes bars with most levels of volatility
5m is often overlooked. I typically zoom in from 15m to 5m bars confirm momentum to any ceil & floor observed in the 15m chart. I then zoom out to maybe 45m – 1h to establish a limit/stop. My EUR/USD trades could be held anywhere from 1hour to 4-6 hours depending on announcements or the market.
Interesting that you look at the business side, how about another issue due to speed demanded to act on possible arbitrage scenarios most systems are inherently a state engine. Obvisly switching between diferent states, it can be problematic about what data and how much data you are processing. In addition due to the fact that you mechanically swithing between these states you should be able to observe a trend in the market. The addition of different data types and origin can complicate the complexity of the state engine but it obviosly must be observable. I think the short lifecycle are possibly evidence of limited state engines acting on market events and inherently driving a limited lifecyle of a given trend
http://www.personal.psu.edu/nuk126/IEEE_affordance-FSA_nhkim.pdf this looks fairly close to what I’m refering to related to finate state machines and trading trends being observable and therefore it should be possible to trade against them. The problem obviously becomes more interesting if multiple algos are running in parrallel and switching to the most profitable rules, so confusing any trend monitoring. Should of put this in the bit above….
Yes, trend following is really good thing. Of course it should be modelled in a proper way. So here are the points one should take into account while taking decision:
1) system of trend estimation (shortly – the way of calculating entry and exit angles, there is a big space for this choice)
2) volatility, indeed there are ways to deal with different volatility and point 1), where formulas for angles is designed can take this point into account
3) Because I used history based optimization if one also does it’s better to use cross-validation also in order to avoid risk. So the decision about best pair (entry and exit angles) should be chosen not only as a pure optimum but kind of convolution of optimum and stability
If one is interested in some technical details of trend following, welcome.
• Yes. It works when it used with good risk management practices.
Some Trend following strategies will work, others won’t. From our experience, success over time depends on how entries are triggered (SMA Crossing, Break out on prices, Triangle exit, …) how they are filtered, what stop loss and what position management is applied … and last but not least, as Daniel Byrnes said, on which time frames the strategies are running.
I think that on a persistent time series trend following is the best way to capture a large part of the move. About time frames in fx: in general I think it’s fair to say that the higher the time frame, the more persistent the price action. This can be measured with a variety of statistical tools.
I now post my TRADING ALERTS
into my personal FACEBOOK ACCOUNT
. Don't worry as I don't post stupid cat videos or what I eat!