What Is a Trend?

This all sounds very wonderful and intelligent but how are you applying this to make money from the markets?

The problem is by defining a trend you are defining something that has already occurred. Unless you are entering at the beginning of the trend the greater the likelihood that you are entering into a weakening trend. The later you get in the higher the risk of failure and the less probability of the trade working in your favour. Surely, it is no use understanding what a trend is and how to measure it. It is more useful to understand how a trend begins and how it ends.

You briefly mentioned supply and demand but no mention of multiple time frames. All trends begin and end at levels of supply and demand. By understanding where smaller time frame price action is within bigger picture supply and demand (horizontal), the trader is able to trade within the context of actual market forces and know where he is trading in the larger supply/demand curve the beginning and possible end to the trend.

I congratulate you effort to place an objective definition into something that is extremely subjective. However I think you may find that the market rarely cares about the math as it is made up of people and emotion not logic.

Even though correlated trend may point in one direction the market can and will move in the opposite direction. By the time the mathematical model reflects this change you have lost the opporunity for a low risk entry and must now place a much WIDER stop or wait for another change.

I too struggle to make trading more objective but in a world wide market filled with illogical people, the math models are either too slow to keep up or end up putting you in the wrong direction at the wrong time.

Thanks for your post. My objective in providing this analysis was to elevate the discussion. The thing that has bugged me ever since I read twoblink's book is the woeful understanding of statistics and a general ignorance on the subject of trends. Why do two points make a trend under traditional TA? If trading is a probabilities game then why do so many traders go about trading using heuristics instead of statistical measures? It makes no sense to me.

I agree that the correlation measure isn't a panacea but I think by using a shorter term measure of correlation (because it is prone to lag due to the fact that you're comparing price vs. an average which is inherently lagging) or simply use the correlation chart in a different way. Like thinking in terms of a change in the correlation trend coinciding with a change in the price trend. You get direction from price and the confirmation from the correlation measure. And there are better ways to measure this stuff but surely it trumps the vaunted ADX indicator for trend (which is even worse lag wise).

Trend traders must identify the trend. Most don't have objective means to do so. The method outlined will help to overcome that. The whole point was to make it objective. If objectivity doesn't work for you then by all means don't use it. The purpose of using the increasing correlation factor is to help identify trends that are increasing in strength. Are there better ways to do that than the one that I put forward? I'm all ears.

Yes clearly it's of no use to understand what a trend is. And trend measurement is of zero value. But if that's so then how on earth do you plan to find the beginning and endings of those trends if you don't a) understand what they are or b) have no way to measure them? Apologies for sounding intelligent, I'll try and dumb it down from here on out.


I'll give a straight answer because this part of your question was straightforward. The purpose was to propose some tools that can be used to do exactly what you describe - enter trends early, and measure those trends so it's possible to see the end coming before it happens. In the spreadsheet I posted a series of charts with some analysis that attempts to do just that. I notice that the download to view ratio is rather low. I'm sure as discussion kicks in here more will have a chance to read through the entire material and study the Excel sheet in the first post and play around with the trend metrics in Excel. Then hopefully the discussion will advance beyond the theoretical into the practical and application of how to do it or how to do it better.

Good work .... eager to hear your thoughts and how you use it with trading.


V

Here is one possible application.

http://www.meyersanalytics.com/papers.php

To the 3 individuals (all family relations no doubt) who have followed along and read the material, downloaded the Excel sheet and tried to follow along with my methodology you might feel a bit betrayed if you tried to reproduce the research. I know I do - just a bit. I seriously doubt anyone has spent the time to study out the material in detail.

But if you had spent the time you would have found as I have found that it looks great on paper but doesn't seem to work as well as it should work on the hard right edge. So I've been pondering on why this might be. I think it comes down to assumptions once again. The statistics-based trend lines are supposed to be drawn on data that exhibits an underlying tendency. Or in other words a trend. Catch-22. As I've dug deeper into the anatomy of the market - its structure at the atomic level if you will, I've found some interesting things.

Lots of people have different ways of talking about this underlying structure. Some say the market is random. Others say it isn't. I think they're both right and both wrong. Maybe after I read Mandelbrot's books I'll have an even more confused or chaotic view or clarified, who knows?

But let's use some different terms so we can create a more useful paradigm. Let's call the predictable component of market movement trend. That's the component that is generally directional and persistent that our eyes see on a chart going up or down. Let's call the component that doesn't move in a persistent directional line cyclical. Some might call this component random but for our purposes let's think of it as an oscillating cycle.

I recently started reading the book "Cybernetic Analysis for Stocks and Futures" by John Ehlers. It is no coincidence that Ehlers provides tools that attempt to divide up the trend component in price from the cyclical component in price by mathematical means. I read some of Ehlers' papers way back before most of you were born as traders. Most of it went over my head. But I've filled in lots of empty spaces in my education and the things he writes now make a lot more sense to me. I've scanned the book and am now going in for a more in-depth look including coding up some of his sample code. Ehlers makes some assumptions too and I can see already that at least one of his main assumptions suffers from the same problem as mine (detailed above). But I look forward to learning about his tools and then will attempt to make the right assumptions so that I can get the best use out of them. I have no allusion that his work contains any type of magical grail, nor am I looking for it. I do think understanding his tools and methodology will help me know the next avenue of my research and development - what tools to use and when to apply them and how to interpret those tools.

An update on the John Ehlers book mentioned in the previous post, "Cybernetic Analysis for Stocks and Futures" - I'd pass on it. To me it exemplifies what is wrong with the books "for traders" and written "by traders". While this book has lots of fancy new ways of spotting trends and cycles with various iterations of digital filters, transforms and smoothers, it seems like he misses the mark over and over - applying them to the wrong data, layering filters over filters just for the sake of doing something fancy or to create a strategy that has no logical basis. There seems to be no understanding of the underlying stochastic data and how to deal with it in its native form. Ehlers talks a good game, but when I got to the section on Cycles that attempted to justify how adding 8 bars of lag to a target 20 bar cycle was not that bad compared to other methods that use cycle warm up periods, I just about tossed the book at the wall. 8 bars of lag! Seriously? (chapter 9, p118-119) And this was after all his crowing in previous chapters about how little lag his work introduces. Good thing it was an ebook or I might have some drywall work. The kicker for me was reading a recent article by said author that seemed to grab a important sounding buzzword of the day and then proceeded to present something that bears no relation to that buzzword. None whatsoever! If I hadn't already studied the subject, I might have been fooled and looked no further. It makes me seriously doubt the validity of the rest of his work. He should know better by now. And so should I.

Personal note: from this point onward I'm back to doing all my own original research or relying on firsthand research from creditable sources. Enough of this drinking through a filtered straw or relying on the conclusions of others. I stopped reading these types of books a long time ago and now I remember why I stopped. They seem to leave a bitter aftertaste in the mouth and an empty feeling in the stomach. This means more fun-filled hours studying arcane looking scholarly papers with their esoteric formulas and stilted jargon. Oh joy, but at least when I digest a scholarly paper I feel like I've actually learned something.

FXEZ...I can hear the frustration and disappointment in your last post. I relate to it well as I too have fought this battle. Trading is a highly subjective game and repleat with inaccuracies. Thus logically speaking, if we remove as much of the subjective nature as possible, we can improve accuracy. I'm not sure if you're ready to admit it yet but this premis is false.

Greater accuracy does NOT follow greater objectivity. In fact its quite the opposite. This is and always will be counter intuitive, but true. To demonstrate this to yourself, review your records (I imagine you've kept good ones) and check to see when your trading was at its most accurate. Now if you conclude from that review that it was only during trending markets that this occurred, then you must ask (as you've been doing) what is the most objectively accurate way to determine (a) if the market is trending and (b) which direction its trending.

Now ask youself this question. During those trending markets were you worried about defining trends or detecting them? Or where you simply trading what you saw? This is because at the time the market was moving in an obvious pattern that you could readily see. The pattern I dare say was visually recognizable and most likely envoked an emotional (gut) response, allowing you to take advantage of what you recognized as an opportunity.

This process was inside your head. I doubt it can be quantified, qualified, or objectively examined in any way. BUT it can be repeated with greater and greater accuracy. The process was subjective as only YOU were able to take the various inputs and arrive at a successful conclusion. This is extremely subjective. The most difficult part to reconcile is that every trader who participates in the market does something very similar. It is extremely ILlogical but it works.

Struggling to remove the subjective nature of this business will have to trying to empty an ocean with a teaspoon.

Trends are good as they generally represent the "flow" of where the market is likely to go, but as you know this is not always true. Most often by the time a trend becomes clearly visible and well "defined" it is almost exhausted. Ask youself WHY you need to define and quatify a trend. If the answer is to know which direction to trade, then ask youself WHY you need to know which direction to trade...continue the process until you reach the core quetion which, I am willing to guess will be something like..."how can I prevent losing".

This core question cannot be answered by math or objective science. This is my first point.

My second point is I have personally found a great deal of "sense" in learning to "read" the psychology of the market. While this is not nearly as objective as you are wanting to achieve here, it is very accurate and keeps my on the right side of the market 65%-80% of the time. I generally enter long before any "trend" is visible and by the time it does develop I am looking to get out.

It is not math that drives this...it is learning to interpret the subjective nature of the market as seen in the candlesticks you trade. I'm sure you already have the book but if you don't may I suggest you read, "Japanese Candlestick Charting Techniques" by Steve Nison. But instead of looking for a method or "system" read it from the stand point of how to gage the psychology present in the market.

Then I believe your clearly analytical mind will develop an extremely powerful AND objective way to find entries and exits into the market. In fact you may even find it boring...if you do, don't worry that's the finish sign...you've made it.

I don't know if this was what you're looking for but I hope you will consider it. It is difficult to see such an obviously intelligent mind struggle with such a simple concept without trying to point you in the right direction. I hope you will at least consider my two points.

Thanks for your thought-provoking post gspajon. I'm really not as much of a basket case as I apparently come off as on the forum. You are correct in that I have a very analytical mind and approach things in a very logical manner, exploring all avenues and discarding those things that don't fit while socking away those few gems that have a +ev. My manual discretionary trading is quite horrible. But my best success has actually been while following an automated strategy - when the strategy fit the market being traded. I've tried both trend following and fading and both do well when they fit the market. But it's the switching between the trend mode and the nontrend mode that do in such lone strategies. So identifying the trend is a means to an end for me - the end being a switching algorithm.

I have looked at candles before and yes have done extensive testing on them. Your assessment is correct in that none of the patterns proved +ev. Initially when I read your description of reading the psychology of the market, with regards to candles I thought of high volatility displayed on the charts. I have looked extensively into volatility and trend reversals - there is a link but it isn't quite in the percentile you mention. Then I saw this interesting thread and thought this might be more the type of analysis you're thinking about with regards to market psychology/sentiment, but I'm not so sure.

http://www.forexfactory.com/showthread.php?t=227948

Appreciate your comments!

I've just been back through reading this thread from a couple of years ago. It is interesting to see how my thoughts have progressed from that period to the present. When I started this thread, I was hoping to explore more deeply my understanding of trends to come to a final conclusion on the subject that would help in system development. I didn't expect where this line of inquiry would lead. The realization starts to kick in on post #27:
I had just started reading some information regarding the fact that the markets might in fact be random, or effectively so. I hadn't made up my mind at the time since it went against what I had been taught at the outset of learning about the markets. But since then I've had to alter my beliefs as I have come to the conclusion that effectively market data is random. If market data is effectively random or non-deterministic then the search for a trend is pointless due to the fact that the trends will come and go without warning as there is no predictive component in the data, which consists of stochastic trend or random trend. (Most people will stop reading at this point because what I've just posted goes against the beliefs of most FF members. But the best insights follow for those with an open mind.)
There are trends in the data (visible to the eye), but those trends tend to be random in nature or in other words non-predictive a priori. Random isn't really the right word, but non-predictive or non-deterministic. This means that in advance you can't know exactly where price will end up. In a deterministic system it is possible to know in advance where price will end up. An example of a deterministic system is a sine wave. Price data or financial time series are anything but deterministic. This excerpt exemplifies a cathartic moment and one of the best decisions I've ever made: Since this time I've spent the past (nearly) two years monk-like studying the nature of the markets, and going ever deeper down the rabbit hole. I took the red pill (or was it really the blue pill?) and there is no turning back now.

One of the characteristics of stochastic trend data is that price can have unlimited variance. In a deterministic system, price has finite and bounded variance. Meaning the standard deviation is fixed and statistically valid. The other component of stochastic trend data is that it has a varying mean. If you've ever put a moving average on a chart you will note that the average does NOT remain at the same price. This meets the definition of a varying mean. In a deterministic system, price has a fixed mean. In this way with a fixed mean and fixed variance, it is relatively simple to set up a mechanical strategy to exploit that predictable system. A simple standard deviation based system would do the trick.

If you've managed to read this far and are scratching your head wondering what next, let me start you on your journey. Best of luck!

I just stumble on this thread yesterday. If you dig deep enough you find gems in FF.

I'd like first to quickly clarify something about the term randomness with an example. Say you see a car in the street. It stops at the road crossing. Is it going to continue ahead, turn right or turn left? To your best knowledge the direction the car will go is a random event. What about the driver? He perfectly knows where he is going and his direction is not at all random. It is perfectly deterministic. As long as you don't know the system generating the event it is random for YOU. Random market doesn't mean there is no deterministic components. It could even be perfectly deterministic it would still be as random for you as the car above was.

A trader makes or loses money from the difference of the price at the open time and at the close time. A trader doesn't makes/loses money from the reasons of this move. Analysts do. Even if a trend is 100% stochastic doesn't make the trend following a idiotic thing. It only make the forecasting of the trend or of its continuation a fool's attempt. You can mesure the past movement of the price to see if the trend is gone or not yet. Trend following, not trend forecasting.

Sept 6th 2011, 8:00am UTC. SNB forces EUR/CHF to 1.20. To do so they need liquidity. They certainly sold gold, USD or GBP they had. So E/U and swissy have moved of course. Gold has moved. Just because of this move in gold, USD varied. Because USD varying all the currencies moved as they are quoted against USD. How much Mexican Peso or Polish Zloti do you think SBN sold to have liquidity? So why did PLN/MXN moved 520 pips?

I read somewhere (couldn't find it back) that if a time series has a non Normal innovation process it is correlated with something. If EUR/CHF has an impact on PLN/MXN it means they are somehow linked. Correlated. We can see that any pair is correlated to any other pair. Therefore EUR/CHF impacts PLN/MXN but PLN/MXN also impacts EUR/CHF. Conclusion EUR/CHF impacts EUR/CHF. It is also correlated with itself! If a random process is auto-correlated trends will certinly appear, won't they?

I don't get why this is true for price (except for limited resources). In general it isn't true. f(t)=t.sin(t) has infinite variance while it is deterministic.

Here also I don't get why. For instance, hyperinflations usually result from a deterministic politic and never mean-revert. Also the mean reversion doesn't impose any limit on the time it takes. I could argue that E/U is mean reverting in a 9000 pips ranges. Just wait 50 more years to see it better.

Back to the trend problem. The first problem I had with the trend, and certainly anybody had the same, is the THE in THE trend. Why do thoses guys say the trend is up I see here the price is going down. Is your chart really showing three trends UP, DOWN, UP or just one single volatile trend? Why is the linear regression a so bad estimator then? The most non-objective part of the definition is this THE. Intuitively, the trend is the persistent simple and smooth part of the move. Something you see at a glace after the fact. If I search the (some) trend I shall then remove the high frequency noise and the cycles. They are not smooth and simple. To remove the cycles I need find them. One method to do this is the Empirical Mode Decomposition. This technique mainly searches for highs and lows to detect cycles. So the HHHL/LHLL TA definition becomes sound. It lags by the time a new swing appears and is nothing continuous but it's also only a quick manual method. Unfortunatety all the mathematic decompositions methods like EMD or wavelet fail at the right end of the chart. The only part we are interrested in.

The only hope is then in the persistent behavior of the trend. It is even its definition. By the time we detect the trend it may already be over. To be useful it must last some more time. The first and obvious idea we have is to find a way to detect the trend with extremely low lag. Spotting the trend right at its beginning. But the definition of the trend is a continuation of the move. To continue it has to have already begun. A trend only exists after it has started to exist! I call this the "beginning of the trend paradox". There is no gold pot at the end of a rainbow nor there is at the beginning of a trend, simply because thoses places don't exist!

The second choice is to find a detector that detects longer term trends. This way even of we are late we still have time to exploit the move. Because the prediction part is a dead-end, the detector will be a probability based estimator relying on the autocorrelation. The time scale of this estimator is now part of the definition of what it is supposed to detect!

The longer the time to detect the longer the wavelength of the trend must be. If the wavelength is long we have a large bandwidth. Then there is plenty of room for the higher frequencies to be present. The trend will be noisy! This is good news. Good news? Yes because we are both late and unsure (probability based estimator!). We need entering on a pullback to have a small stop. We need a pullback. We need noise. A part of this noise is now in the medium frequencies. This thread isn't about trading strategies but here we have one: when the detector detects a trend, we remove the trend from the price. We go lower TF. We run the detector again. When it find an opposite trend, part of the medium frequencies cycles, we know this trend is too short to be traded. We are late. But we also know it is a counter-trend for the above estimator. We take the opportunity of beeing late to fade it! Quantitative swing trading...

No holy grail yet. There is another issue. The definition of the trend also lacks the definition of the slope of a trend. If you develop an instrument to mesure the trend, it will ALWAYS mesure something. Even when the trend is horizontal. A range bound market doesn't suite well to a trend follower... How many pips per bar is the frontier between a trend and a range? I mean OBJECTIVELY.

Yes this very nicely sums up my thoughts on the difference between deterministic and stochastic trend.

Yes I think that is correct. The foreign exchange market is a closed/zero-sum system in that all currencies affect all other currencies. One currency cannot move without it affecting every other currency pair in some way. There are many ripples after a stone is thrown in the fx pond. It is sort of like the "whack a mole" game at some carnivals. You hit a mole with a club and another mole immediately pops up somewhere else.

I think this is a simple misunderstanding. I should have been more clear, such as:

I'm not sure about your second point. All my sine waves have finite and fairly predictable variance. Can you give me an example of a sine wave with infinite variance?

Agree with your last point regarding EMD, FFT, etc, but I think the discussion of THE sort of went over my head.

Yes that's right, all the methods I know regarding trend following involve some sort of confirmation by price. The start of the identifiable trend is also where whipsaws live. Domino has some interesting ideas on how to use price compression to lead into a breakout.

I would say, it depends on a lot of things. Any trend method should be able to define this for a given market / condition by where it enters.

I said T*sin(T)

Right but the sine wave itself has stable variance, no? At any rate, stable variance is not even necessary if you have a fixed mean as in the above chart. A pair of bollinger bands would harvest those swings quite efficiently if this were a real market. Infinite variance would be infinite profit! The key is that the system is deterministic. No need to predict, it is a simple calculation to determine where the curve will go next... assuming it doesn't completely change in the future.

And a stable mean with drift is no problem either for making profits as averages can move. It is the changing of direction or unstable means that cause most of the headaches for traders.

I just got an idea about the trend/range/other distinction. After all a range bound market is no more than a volatile trend which is running horizontally. Flat is a slope like any other, just a special case. Why do we want to make the distinction? Simply because of what we want to do with it.

A counter trend trade will have more or less the same SL as an in-trend but the potential profit is smaller (the retracement is smaller than the previous swing). A trend is a persistent behavior: it has a higher chance to resume than to reverse. The probability of winning is then also smaller. A counter-trend is less profitable than the in-trend, expectancy wise.

But in a range we can go both directions with the same expectation of profit. The probability of a breakout by the up side or the down side is the same. The SL at the support is the same as the SL at the resistance and the potential profit, the opposite side, is the same.

So to distinguish a trend from a range we have to define some limit after which going in the opposite direction of the slope of the determined trend is no longer acceptable. This will be a function of the trend slope, the width of the trend channel, the volatility (expected height of one bar), perhaps the spread too. A point of negative return of some sort...

a break out of the range must have happened if you don't want to trade the opposite way.

What "the" range are talking about? The goal is to define IF there is a range or a trend...

Trend traders are essentially attempting to buy high and sell higher. Counter trend or mean reversion (MR) traders are looking to buy low and sell high. The markets are known to have fat tails which means that there is an advantage to trend trading over MR trading in terms of market structure (in the long run). Think of it like how in the stock market there is a general tendency for markets to go up over time giving a general advantage toward buying. But bear markets exist or in other words extended down cycles that make buy and hold anything but a sure thing for stock traders. In like manner, some say 70% of the time FX markets aren't trending. The low probability of success for trend traders leads to whipsaws and drawdowns if they follow a naive trend following approach.

But essentially, trend traders are playing for continuation, while MR traders are playing for reversals. A trend trader never knows if prices are too high or too low, they just keep buying or selling (or holding).

A MR trader can identify when a market is too high or too low but these assumptions are typically based on the assumption that price is a deterministic system (stable), that it must or will revert. MR traders accept limited profits and losses with a higher hit rate. However, an open loss for a mean reversion trader seems like the value has improved and should indicate they should buy / sell more. This of course, can lead to ruin as it is essentially a Martingale. So MR traders are incented to find or build deterministic or quasi deterministic systems where the mean is at least fairly stable and the variance is fairly well behaved. But of the two as was previously pointed out, a stable mean is far more important. One method to do this is Cointegration. But FX markets don't cointegrate very well so either more complex methods must be used, better moving window and underlying trading strategies, or better markets for cointegration must be sought. The irony is that the FX market is a closed system where movement in one currency affects all the others. This should indicate some hope for creating a deterministic system from the currency components. (should).

Oops. I shall stop saying I'm trend follower in this case... I try to buy the retracements and exit when the swing is (expected to be) exhausted and repeat. Always in the direction of the (most probable) trend. More a swing-in-the-direction-that-follows-the-trend trader. That's longer to say.

My Bayesian point of view on this: http://www.forexfactory.com/showthre...06#post6772106