Technical Analysis alone will not give you an edge
Good evening my fellow traders, I have been a student of the market since 2011 when I combined FF. The whole concept of investing, and obtaining profit through fluctuations in price movement has fascinated me. I have not been fortunate enough to blow accounts as many of you have, I enjoyed relative success in the markets earlier on which fueled my hubris. I surmised that observing charts, studying patterns, utilizing some type of money management subject and system has been responsible for my success. I have studied Technical Analysis tools broadly, Parabolic SAR, from MACD, Stochastics, RSI, S/D, S/R, candle stick patterns, you name it.
I have began taking trading seriously recently, not just as an interest and a passion except to dedie my time like this a second carrier. I see many here purport to have the ability to forecast price, to stick to the market, many here will inform me that you've learned the skill to do this successfully. I'll respectfully disagree with you, I go as far as to call it Bull shit. In almost any area of human history where a phenomenon is not well known, not understood, or plainly random... there'll be superstition, there'll be an array of different methods, some contradictory that is only going to work because of pure chance. I claim yours is not any different.
When analyzing such a happening, the very best tools at our disposal is the scientific method... 1- ask a question, two - do study, 3- formulate a theory, 4- test your theory, 5- evaluate your outcomes, 6- communie your findings.
From readying probability theory and statistics I claim you cannot profit from arbitrary data, do you concur with my claim? If you do not let us talk about your ideas here as I am prepared to change my thoughts given the signs.
In case you do not contest my claim then the very first question one needs to ask is.... Is the market arbitrary? Then hypothesis should be that it is; unless the evidence points to it not being arbitrary.
So how do we test for randomness? Let us begin with asking this questions in the statistical sense.... For the data to be arbitrary a distribution that is random must be followed by it? Can you concur? Below are just two distributions
1- fitted into the euro/usd named DEXUSEU in the Saint Lous FRED, this data is a complete of 4227 euro logarithmic returns
https://www.nigeriaforextrading.com/...1707168773.png
here is the density of a randomly generated data of a sample size of 4227
https://www.nigeriaforextrading.com/...7735939595.png
when we do a QQ plot of euro.log, normally distributed data we can see the fit
https://www.nigeriaforextrading.com/...0219429343.png
My ideas: the euro log density seems to possess more peakedness and thicker tails, this however just shows that the volatility of the data differs not that it is or isn't predictable. Therefore, how do we do a test for randomness?
Statistically we would see if there is any serial correlation (auto correlation) significance: Autocorrelation, also called serial correlation or cross-autocorrelation, https://en.wikipedia.org/wiki/Autoco...on#cite_note-1 is the https://en.wikipedia.org/wiki/Cross-correlation of a https://en.wikipedia.org/wiki/Signal...mation_theory) with itself at different points in time (that is what the cross stands for). Informally, it is the similarity between observations as a function of the time lag between these. It's a mathematical tool for finding repeating patterns, like the existence of a periodic sign obscured by sound, or identifying the https://en.wikipedia.org/wiki/Missing_fundamental frequency in a sign implied by its https://en.wikipedia.org/wiki/Harmonic frequencies. It's frequently used in https://en.wikipedia.org/wiki/Signal_processing for assessing functions or series of values, for example https://en.wikipedia.org/wiki/Time_domain signals. -- wikipedia
https://www.nigeriaforextrading.com/...1159746612.png
here is your auto correlation of a random information
https://www.nigeriaforextrading.com/...1467713448.png
here is also the Partial Auto correlation function:
In https://en.wikipedia.org/wiki/Time_series_analysis, the partial autocorrelation function (PACF) gives the https://en.wikipedia.org/wiki/Partial_correlation of a time string with its own lagged values, controlling for the worth of the time series at all shorter lags. It contrasts with an https://en.wikipedia.org/wiki/Autocorrelation_function, which does not control for other lags.
This function has an important role in data analyses aimed at identifying the extent of the lag within an https://en.wikipedia.org/wiki/Autoregressive_model. The use of this function has been introduced as a member of the https://en.wikipedia.org/wiki/Box–Jenkins method of time series modelling, where by plotting the partial autocorrelative works one could determine the appropriate lags de within an AR (p) https://en.wikipedia.org/wiki/Autoregressive_model or within an extended https://en.wikipedia.org/wiki/Autore...moving_average (p,d,q) model.
PACF of the Euro
https://www.nigeriaforextrading.com/...7351087946.png
PACF of arbitrary information
https://www.nigeriaforextrading.com/...0261100552.png
Decision: as you can see that the Euro does not breach the confidence intervals, the data does not seem to be connected which affirms the hypothesis of arbitrary data. Below are the results of the correlation
Autocorrelations of series'x', by lag
0 1 2 3 4 5 6 7 8 9 10 11
1.000 0.012 -0.013 0.000 0.035 -0.035 0.016 0.028 -0.002 -0.003 -0.019 0.006
12 13 14 15 16 17 18 19 20 21 22 23
0.013 0.018 -0.004 -0.012 -0.006 0.013 0.005 0.016 0.008 -0.012 0.011 -0.012
24 25 26 27 28 29 30 31 32 33 34 35
0.008 -0.004 0.007 0.002 0.005 0.017 -0.004 -0.002 -0.003 -0.008 -0.007 0.007
36
-0.011
gt; acf(rnorm(4227), plot=FALSE)
Autocorrelations of string'rnorm(4227)', by lag
0 1 2 3 4 5 6 7 8 9 10 11
1.000 -0.029 0.034 -0.034 0.010 0.007 0.004 -0.006 -0.002 -0.024 0.009 0.000
12 13 14 15 16 17 18 19 20 21 22 23
-0.015 -0.020 0.006 -0.006 0.013 0.005 -0.003 0.001 0.010 0.002 0.014 -0.002
24 25 26 27 28 29 30 31 32 33 34 35
-0.006 0.011 -0.001 -0.008 -0.013 0.004 0.017 0.004 0.020 -0.012 0.014 0.002
36
-0.007
let us look for trend and volatility in the data that the first one is euro, the next is a random normally distributed data
https://www.nigeriaforextrading.com/...3136477332.png
https://www.nigeriaforextrading.com/...1782931994.png
I will see a gap in the distribution, the top red and lower red lines are at sigma a single (standard deviations 1 and -1) the red line at the middle is that the SMA of 22 = monthly, the yellow lines is that the SMA 22*12= 264 which will signify a yearly smoothing, the blue line is to illue H0
Brain sees a gap in the distribution but the monthly smoothing looks similar I can see either actual or imagined signs of tendency in the data.
Lets look at the Hurst Exponent
gt; hurstexp(x)
Simple R/S Hurst estimation: 0.5614027
Corrected R over S Hurst exponent: 0.5815521
Empirical Hurst exponent: 0.5223035
Corrected empirical Hurst exponent: 0.5014898
Theoretical Hurst exponent: 0.5211121
gt; hurstexp(rnorm(4227))
Simple R/S Hurst estimation: 0.5169296
Corrected R over S Hurst exponent: 0.5302739
Empirical Hurst exponent: 0.5114854
Corrected empirical Hurst exponent: 0.4897964
Theoretical Hurst exponent: 0.5211121
the Hurst exponent is just another measure of randomness especially Persistence, positive where Hgt;50 Hlt;1 would indie a trending element Whilst Hgt;0 Hlt;50 would reveal mean reversion.
The Hurst exponent is used as a measure of https://en.wikipedia.org/wiki/Long-range_dependency of https://en.wikipedia.org/wiki/Time_series. It is related to the https://en.wikipedia.org/wiki/Autocorrelation of the time series, and the rate at which these decrease since the lag between pairs of values increases. Studies between the Hurst exponent were originally developed in https://en.wikipedia.org/wiki/Hydrology for its practical matter of discovering optimal dam sizing for its https://en.wikipedia.org/wiki/Nile_river's volatile rain and drought conditions that was detected during a lengthy period of time. Https://en.wikipedia.org/wiki/Hurst_...nt#cite_note-1https://en.wikipedia.org/wiki/Hurst_...nt#cite_note-2 The title Hurst exponent, or Hurst coefficient, derives from https://en.wikipedia.org/wiki/Harold_Edwin_Hurst (1880--1978), who had been the lead writer in these types of studies; the use of the typical notation H to its coefficient relates to his title too.
In https://en.wikipedia.org/wiki/Fractal_geometry, The generalized Hurst exponent was denoted by https://en.wikipedia.org/wiki/H_(disambiguation) or Hq in honor of both Harold Edwin Hurst and https://en.wikipedia.org/wiki/Otto_Ludwig_Holder (1859--1937) by https://en.wikipedia.org/wiki/Benoît_Mandelbrot (1924--2010). Https://en.wikipedia.org/wiki/Hurst_...nt#cite_note-3 H is directly related to https://en.wikipedia.org/wiki/Fractal_dimension, D, and is a measure of a data collection' wild or mild randomness. Https://en.wikipedia.org/wiki/Hurst_...nt#cite_note-4
The Hurst exponent is popularly called the index of addiction or index of long-range dependence. It quantifies the relative trend of a time series to cluster in a way or to regress closely to the expression. Https://en.wikipedia.org/wiki/Hurst_...nt#cite_note-5 A value H in the range 0.5--1 indies a time string with long-term positive autocorrelation, meaning that a high value in the show will probably be followed by another high value and the values a very long time into the future will also tend to be high. A value in the range 0 -- 0.5 indies a time show with long-term shifting between low and high values in adjacent pairs, meaning that a single high value will probably be followed by a low value and the value then will tend to be higher, with this inclination to switch between low and high values lasting quite a while later on. A value of H=0.5 can indie a completely uncorrelated show, but in fact it is the value applicable to series where the autocorrelations at little time lags could be negative or positive but where the absolute values of the autocorrelations decay exponentially rapidly to zero. This compared to the typically https://en.wikipedia.org/wiki/Power_law rust for the 0.5 lt; H lt; 1 and 0 lt; H lt; 0.5 instances. --wikipedia
now we can assess the Runs Test for randomness
gt; runs.test(x,plot=TRUE)
Runs Test
data: x
statistic = 2.1203, runs = 2163, n1 = 2080, n2 = 2107, n = 4187, p-value =
0.03398
alternative theory: nonrandomness
gt; runs.test(rnorm(4227))
Runs Test
data: rnorm(4227)
statistic = 1.3846, runs = 2159, n1 = 2113, n2 = 2113, n = 4226, p-value =
0.1662
alternative theory: nonrandomness
since you can observe the P-value about the data x =eur/usd seems be contrary to the hypothesis of randomness
to complete my opinion, I assert that with the statistical instruments at our disposal it is quite tough to inform the Eurusd data from a Random normally distributed data, so T/A alone would be similarly hard to differentiate from a brownian motion monte carlo simulation and the real market. I assert you cannot profit from T/A alone the allegations of profiting are likely to be false.
