quantitative trend determination

Discussion in 'Strategy Development' started by afx111, Sep 22, 2009.

  1. afx111


    Hi all,

    Assuming i have no access to charts.

    How can i determine the trend (up, flattening, down) and steepness of that trend quantitatively based on a stock's closing prices?

    This is an excel question. hope u guys can help. tks!
  2. Hugin


    We have used a trend indicator based on regression. The basic input is the steepness of the regression line using the price data. One problem with trend indicators is that a change in volatility may be interpreted as a trend which results in false indications. To handle this the slope is adjusted by dividing with the standard error of the regression. This results in a trend indicator less sensitive to volatility shifts. There is a paper from Amplitude Capital called "Examining Trend Characteristics" that describes the details (google should be able to find it).
  3. Hi Hugin,

    Have you compared the paper's approach (could only find the abstract on the web), with robust statistical methods?
    Like LMS or LTS?

    Also, how are you determining the window length of your trend? Fixed or adaptive?
  4. ok, I found the paper.
    I get what you meant about the trend indicator now. It's not a type of regression, but a measurement related to regression parameters.

    Did you find any use in their conclusions?
  5. Hugin


    I use an adaptive, rather short, window of up to 40 days. The main benefit over most other standard trend indicators is that it addresses fake trends due to volatility changes. I've done some tests on various time series in Excel with various volatility and most time my eyes agree with the conclusion of the indicator.

    Regarding the conclusions they make I suspect the paper was released in order to create some interest around their funds. I still find the result that correlation between their trend indicator and volatility breaks at high vola quite interesting (even though it may be specific to Forex of which I know next to nothing). Their approach may be interesting to equities as well but I haven't done anything yet.
  6. rosy2


    calculate the hamming distance between 2 vectors

    A= binary vector of market ups/downs
  7. Interesting approach, but doesn't take into consideration magnitude of vector elements.

    For example, suppose you had sequence
    10% 10% 10% 10%
    -1% -1% -1% -1%
    repeated several times

    now suppose you had
    10% 10% 10% 10%
    -10% -10% -10% -10%
    repeated same number of times

    Hamming distance would be same, but trend characteristics are completely
    different. One is perfectly trending, other
    is perfectly sideways. Not a good measurement for trend.

    Might work for coin tosses; not market.
  8. Karsten and his team have some problems to identify "the trend" this year...maybe adapt the model ?


  9. "One problem with trend indicators is that a change in volatility may be interpreted as a trend which results in false indications."

    I disagree with this. It's not a problem with the trend indicator. The trend indicator is doing exactly what trend indicators are supposed to. It is using an algorithm to try to determine whether a sequence of price movements is weighted in one particular direction and to quantitatively score that weighting. The fact that the timescale over which the price movements occur has no bearing on whether they constitute a trend or not. I think what you are actually saying is that you wish to filter out trends which occur on short timescales (which I think you consider to be just volatility induced noise).

    Nothing wrong with that, all I'm really saying is that a trend is a trend is a trend...timescales have nothing to do with their definition. They are defined solely by price movements.
    #10     Sep 25, 2009