What Is a Trend?
Page 1 of 502 1 2 LastLast
Results 1 to 10 of 12

Thread: What Is a Trend?

  1. #1

    What Is a Trend?

    What is a fad? The question seems simple enough but you'll realize the definitions are varying, sometimes obscure and contradictory or illogical if you do any amount of research. The most common response given is a downward trend consists of lows and lower highs and an upward trend consists of highs and lows.

    Wikipedia, under its entry for Trend (technical analysis) says the following:

    Quote Originally Posted by ;
    To a trader that uses technical analysis, the tendency in price of a financial security is, intuitively, the general direction of its movement. Loosely, if a person who looked in the price chart would generally state the price is moving up, the tendency is upward, and when a person would generally state the price is moving down, then the trend is down. Oftentimes an observer would find it hard to determine whether the price was generally moving up or going downin this case the tendency might be stated to be unclear. When price is oscillating back and forth across a range, the tendency is often said to be.
    The general direction is surely broad enough and general enough to cover nearly all facets of a fashion. However, I believe we could and should do. The CFTCs definition is no less ambiguous:

    Quote Originally Posted by ;
    Trend: The general direction, either upward or downward, where prices are moving.
    There are a range of books that discuss the trend however few of these give a concrete and goal definition beyond these nonspecific descriptions merely mentioned. One publiion that claims to provide a fashion definition that is concrete, and offers mathematical evidence concerning its exactitude is Trends and Trendlines from the writer, who happens to post on this forum from time to time. A survey of the definitions or review let me state the fact of a set of assumptions rests on the truth of the assumptions without turning this. It is impossible to reach the conclusion, if you start with erroneous assumptions. But if you start with first assumptions that are sound, you have a chance.

    I will endeavor to explain in an objective and statistically measurable way how I specify the fad or a single way you can do it too. It's important to specify the conditions correctly so that a statistically legitimate trading egy could be built up in the base. But why everyone appears to have the wrong idea, Im going to explain about trends and put a base.

    ---------------------------
    Notice: this isn't an open discussion about what you believe defines tendency. Please keep the discussion on this issue of trends exhibited in the framework . I placed a three voucher minimum after reading the content, so the more seasoned will chime in about the topic. Feel free to PM me your questions I believe this material fairly well explains itself and though I can't guarantee a reply due to time constraints.

    It's unfortunate but some will question my motives so let me put it in the first post. What is my motivation for submitting this info? I'm tired and sick of the misinformation that accepted by the masses and is passed around under the guise of both TA fact. Books have regurgitated this misinformation? How can anybody possibly get the ideal response if they're basing their question? Nowhere have I seen advice regarding how to statistically establish a fashion. I get nothing other than the good feeling from knowing I have made my one small attempt to shine some light to the topic of trends. This thread will become the most and most famous forgotten or the hottest ribbon on FF. I doubt it's going to be the latter given but I hope that at least one individual accomplishes something. That person will be me.

    I've attached an Excel sheet. It'll be used during the discussion and you won't be able to follow along unless you examine it when reading a few of the sections and download it.
    https://www.cliqforex.com/attachment...8868751625.xls

  2. #2
    I can give you a few Details here:Currency pairs tend to tendency, tails are shown by their distribution of returns and their price show reveal autocorrelation. I advice statistical tests to run until you think a tendency exists. There are several ways but you may look and examine whether a price series statistically looks like a random walk with a drift component. This will let you know the likelihood where there exists really a component that is trending. Hurst exponents and auto-correlations, both of these are important when measuring tendency power. You need about 10 egies to measure fad statistically to make accurate decisions. I worked for a large bank, we tracked 90 tools from that we'd 20 FX pairs. From these the best were pairs from Latin America, etc.. Exotics sometimes trend a lot greater than major currency pairs or even crosses.
    Finally. I advice you to learn statistics, if you do not, take a course on coursera for God sake. It will help out you.

  3. #3
    How can you estimate the Hurst exponent? I know two methods: Re-scale range and wavelet. I am aware there are a couple more but don't know how they work. Except for RS them split with tail distributions. Wavelet needs gt;50K samples to give a reading that is trustable. RS appears to be the quickest it merely needs ~1000 samples to converge. Issue using the autoregressive models. An ARIMA does not converge before hundreds of samples. GARCH needs tens of thousands... That is a tiny bit too much of lag for my taste. Please post if you know methods which converge very fast.

    Quote Originally Posted by ;
    There are many ways to specify a trend but you can look and examine if a price series statistically resembles a random walk with a float element. This will tell you the likelihood with which a trending element really exists.
    I use a Kalman estimator using another order polynomial model (ax^2 bx c sound ). The choice of this model was driven by the simple fact that it encodes the price, the velocity (trend) and the acceleration to ch up quickly after a shock. Because the market is not Gaussian while Kalman relies on Normal distribution, I correct this estimator with a different KF (A continuous noise model) that feedback the error (by increasing the elements of the variance matrix). That is because using a Cauchy or Student distribution creates the prior/posterior untractable (to me ). Do you believe this is a bad method of doing things?

  4. #4
    Quote Originally Posted by ;
    Do you think this is a good or a bad method of doing things?
    Very good method of doing this. Part of the analysis included Kalman estimators. However Hurst exponent and auto-correlations are crucial. Take ideas from here: http://ccsenet.org/journal/index.php...cle/view/18513.

  5. #5
    Hmmm the paper clearly indies that H-values are found to vary widely from period to period (their words) with just 60 observations. Is it the Hurst exponent does really change that quickly or because the trials size is too small? Like RS and wavelet, a stationnary period collection is expected by the DFA method. Financial time series are fractionally integrated. The yields (1st diff) is still not stationnary generally. These show are corrupted with non gaussian noise. It is very strange for me... How can a process be revealing long-term memory now if it did not yesterday and will no longer tomorrow?

    Also the 12 indexes are highly correlated and tested over precisely the exact same period of time. This is almost equivalent as a single observation. No random walk generated series was tested to create a witness hypothesis. I highly suspect their conclusion.

  6. #6
    Quote Originally Posted by ;
    I can provide you a few facts here: Currency pairs have a tendency to tendency, their supply of returns reveals fat tails and their price series reveal autocorrelation. I advice you to run statistical tests until you believe there exists a tendency. There are many ways however, you can look and examine whether a price series statistically resembles a random walk with a float element. This will let you know the likelihood with which there exists really a component that is trending. Hurst exponents and auto-correlations, these two are very important when measuring tendency strength....
    Quote:
    You likely need about 10 ways to quantify trend statistically to make accurate conclusions.

    Would you name them and say a little more info about them?
    Thanks.

  7. #7
    Erik.
    Quote Originally Posted by ;
    I will provide you a couple of facts here: Currency pairs tend to trend, their supply of returns shows fat tails and their price series show autocorrelation. I advice statistical tests to run until you think a trend exists.
    It's well-known that market distributions generally have fat tails. I presume you're speaking about autocorrelation of price returns because autocorrelation of a price series would appear to be not quite meaningful.
    Quote Originally Posted by ;
    There are several ways to specify a fashion but you can look and examine whether a price series statistically looks like a random walk with a float element. This will let you know the likelihood with which there exists actually a element that is trending.
    How does comparing the price series statistically to a Random Walk with Drift (Yt = #945; Yt-1 #949;t ) let you know the likelihood that a trending element is present? Which test(s) should be carried out? Can this analysis also suitable for shorter timeframes, i.e., sub-one hour intervals?
    Quote Originally Posted by ;
    Hurst exponents and auto-correlations, these two are extremely important when measuring trend power. You need about 10 ways to quantify fad statistically to make accurate conclusions.
    What sort of precision are you speaking about? I ask because autocorrelation is a fairly transitory occurrence in price returns when trying to use prior period autocorrelations on period data. I've utilized an indiator called iVAR which may be associated with Hurst and is a measure for trend/non-trend. The principal difficulty with measures of the type is lag.
    Quote Originally Posted by ;
    I worked for a large bank, we monitored 90 tools from that we'd 20 FX pairs. From these the best were pairs from Latin America, etc.. Exotics sometimes trend than major currency pairs or even crosses.
    It's a pity retail traders don't have general access to the majority of Latin American markets.
    Quote Originally Posted by ;
    Finally. I advice you to learn statistics, if you don't, have a course on coursera for God sake. It is going to help out you.
    Good advice for anyone! I also highly recommend taking a system learning course (also offered on coursera). For your own benefit, although not for God's sake!

  8. #8
    I tried to improve my Kalman filter by comparing it and incorporating another estimator of this trend yesterday. Then I mastered the variance matrix according to how both estimations were different (like I already do for the price). To make things was deceiving. As usual the downturn can be reduced at the cost of the smoothness. That made me understand something. The issue doesn't come in the filter itself however the model utilized (here a 2nd order polynomial). Whatever model is used such as a MA, linear/polynomial regression, a Kalman filter, a random walk with drift, an ARIMA model, spectral decomposition, etc.. . All of these models make the assumption that the trend is a stable drift of the price. This is very accurate in the middle of the trend however, the trend can change quite suddenly. We've got no way to detect this shift.

    So I begin wondering whether it wouldn't be a lot easier to have a phenomenological approach rather. We realize that there are tendencies but without trying to know whether a given trend is a movement or if it is statistically valid or where it is coming from. Just that it is a trend. Like Erik indies we use versions and/or heuristics and a likelihood to assess the trend with some confidence period and combine them together. We answer what's a trend? But what's the most likely current estimation of this trend? (for a given time scale).

  9. #9
    Yes that's my takeaway in the statement:

    Quote Originally Posted by ;
    You probably need about 10 egies to measure fad statistically to make accurate decisions.
    It sounds like using a way where many methods are utilized to identify the likely trend, and then bringing these quotes together to produce a consensus view.

  10. #10
    I tried this descriptive approach.

    I used a Kalman filter (2nd order polynomial model). The result is shown on the picture. This estimator is meant to denoise the signal. This denoises the price quite but this doesn't give a good estimation of the trend.

    I gave another opportunity into the wavelet decomposition. On the second picture I show the smooth part with 32 (red), 16 (blue) and 8 (green) bars delay. 16 and the 32 are drawn back previously. The green line (8 pubs delay) is drawn in the time it's figured. Adding more delay doesn't enhance beyond what the red line shows. The line provides a good estimate of what we'd anticipate when we state the trend. The green line is near (in value) of this red one but is so wavy that it makes it useless for a good estimation of this trend.

    So I tried to feed the green line into a Kalman filter (constant model):

    I also tried to feed the green line to the initial Kalman filter (2nd order polynomial model), easy although it overshoots:

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
This website uses cookies
We use cookies to store session information to facilitate remembering your login information, to allow you to save website preferences, to personalise content and ads, to provide social media features and to analyse our traffic. We also share information about your use of our site with our social media, advertising and analytics partners. Read more.