Scribd Logo
Upload

Welcome to Scribd!

  • 84 views

    Pearson Correlation Coefficient

    Uploaded by

    Ronitsinghthakur Singh

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd

    Pearson Correlation Coefficient

    Uploaded by

    Ronitsinghthakur Singh
    84 views7 pages

    Original Title

    Pearson correlation coefficient.docx

    Copyright

    © © All Rights Reserved

    Available Formats

    DOCX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Download as docx, pdf, or txt
    84 views7 pages

    Pearson Correlation Coefficient

    Uploaded by

    Ronitsinghthakur Singh

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Download as docx, pdf, or txt
    of 7

    Pearson correlation coefficient measures the linear correlation between two variables.

    It has a value between +1 and −1, where 1 is total positive linear correlation, 0 is no
    linear correlation and −1 is total negative linear correlation. It’s often denoted by r for
    sample correlation and ρ for population correlation.

    Note: Pearson Correlation only measures the linear relationship between two variables,


    such as y = a*x + b. There are other measurements for nonlinear correlations.

    When option “R” for "Correlation variants" is chosen, the value would be the same as
    TradingView's built in correlation() function. For "Adjusted R ", the calculation is based on
    the traditional Pearson. The sample r is a biased estimate of population ρ. The adjusted r
    gets rid of some of the bias but not all. As the sample size or lookback period increases,
    adjusted r will be closer to r.

    The confidence interval is computed for population ρ estimation based on sample


    r. Correlation coefficient itself doesn’t follow a normal distribution. Fisher
    transformation is applied to transform the data into an approximately normal
    distribution. We compute the standard error based on the transformed data, then use an
    inverse fisher transform to transform back the standard error in terms of r.

    Note: the confidence interval band is an approximation of population, it proposes a


    range of plausible r values (instead of a point). The confidence level represents the
    frequency (i.e. the proportion) of possible confidence intervals that contain the true
    value of the unknown population parameter. The proportion of those intervals that
    contain the true value of the parameter will be equal to the confidence level. For
    example, if the confidence level is 95% then in hypothetical indefinite data collection, in
    95% of the samples the interval estimate will contain the population parameter. The
    default setting is 1.96* standard error which is 95% confidence interval.

    The most important and distinguishable feature of this indicator is the p-


    value provided along with the correlation.

    The value of Correlation Coefficient alone doesn’t provide any information regarding


    its statistical significance. For example, two sets of independent samples have 0
    correlation in theory. However, your correlation coefficient on these samples will never
    actually show 0 correlation (small correlation value but not 0). Therefore without a
    significance test, one would be fooled by the value of r when there’s no linear
    relationship at all.

    In statistical hypothesis testing, the p-value or probability value is the probability of


    obtaining test results at least as extreme as the results actually observed during the test,
    assuming that the null hypothesis is correct. The smaller the p-value, the stronger the
    evidence that the null hypothesis should be rejected and that the alternate hypothesis
    might be more credible. Since one could be deceived by r showing values while
    correlation is actually 0. The null hypothesis here is the “r is 0”. The alternative
    hypothesis is “ r is not 0”. The default setting for p critical value is 0.05. It means that
    when p is lower than 0.05, there’s less than 5% chance that correlation is 0, and we
    consider that to be "significant correlation". To get the p-value, We use a t distribution
    with n – 2 degrees of freedom to find the probability. P-value will adjust automatically
    when the sample size or lookback changes.

    Displays :
    When p is lower than 0.05 and r > 0, correlation coefficient shows red, p-value shows
    yellow, panel shows “Significant Positive Correlation”.
    When p is lower than 0.05 and r < 0, correlation coefficient shows green, p-value shows
    yellow, panel shows “Significant Negative Correlation”.
    When p is higher than 0.05, correlation, correlation coefficient shows white, p-value
    shows grey, panel shows “Insignificant Correlation”.

    r² (r squared) also known as the coefficient of determination, is the square of correlation


    r. r² measures how well the data fit the linear regression model used in correlation. When
    two assets show significant correlation, r squared can be used to compare which one fits
    the data better. r² is displayed on the panel and has a different lookback by default than
    the correlation coefficient .

    Contributors : Pig (ideas, code, math and design), Balipour (ideas),


    midtownsk8rguy(applying/employing Pine etiquette).
    May 9
    Release Notes: minor update to add public license.
    May 9
    Release Notes: minor fix to adjust panel's positioning.
    May 10
    Release Notes: added feature to assist the visually impaired.
    May 11
    Release Notes: Modified the security() function to avoid lookahead bias.
    May 11
    Release Notes: removed the mtf function.
    May 15
    Release Notes: update to modify default position of the panel.

    =========================

    // This source code is subject to the terms of the Mozilla Public License 2.0 at
    https://mozilla.org/MPL/2.0/

    // © balipour
    //@version=4

    study("Correlation with P-Value & Confidence Interval [pig]", "BA🐷 CC", false, format.price, 3)

    var invisible = color(na)

    bgcolor(#000000c0)

    var cPI = 2.0 * asin(1.0) // 3.1415926536 Constant

    //===== Functions =====//

    cc(x, y, len) => // Correlation Coefficent function

    lenMinusOne = len - 1

    meanx = 0.0, meany = 0.0

    for i=0.0 to lenMinusOne

    meanx := meanx + nz(x[i])

    meany := meany + nz(y[i])

    meanx := meanx / len

    meany := meany / len

    sumxy=0.0, sumx=0.0, sumy=0.0

    for i=0 to lenMinusOne

    sumxy := sumxy + (nz(x[i]) - meanx) * (nz(y[i]) - meany)

    sumx := sumx + pow(nz(x[i]) - meanx, 2)

    sumy := sumy + pow(nz(y[i]) - meany, 2)

    sumxy / sqrt(sumy * sumx)

    adj(r, n) => // Unbiased Adjusted R Estimation Approximate Function

    (1 + (1 - pow(r, 2)) / (2 * n)) * r

    Round(src, digits) => // Round Function


    p = pow(10, digits)

    round(abs(src) * p) / p * sign(src)

    xp(offset) => // Offset

    time + round(change(time) * offset)

    _label(offset,P, T, s, color_PnL) => // Info Panel

    label PnL_Label = na

    label.delete(PnL_Label[1])

    PnL_Label := label.new(xp(offset), P, text=T, color=color_PnL, textcolor=color.white, style=s,


    yloc=yloc.price, xloc=xloc.bar_time, size=size.normal)

    //===== Inputs =====//

    src = input( close, "========= Source =========", input.source )

    sec1in = input( "SPX", "Comparison Security", input.symbol , confirm=true )

    mode = input("Adjusted R", "Correlation Variants ", input.string , options=["R", "Adjusted R"])

    len = input( 20, "Correlation Lookback Length", input.integer, minval=2)

    //Stats Settings

    sc = input(true, "Show Confidence Interval for Population", input.bool )

    csd = input(1.96, "Confidence Interval SD Multiplier", input.float , minval=0.1, step=0.1) //Default


    95%

    sp = input(true, "Show P-Values", input.bool )

    cp = input(0.05, "P-Value Significant Confidence Level", input.float , minval=0.0, step=0.01) //Default


    = 1- 0.05 = 95%

    pan = input(true, " Show Information Panel", input.bool )

    rlen = input( 50, "  R Squared Length", input.integer, minval=2)

    os = input( 40, "  Panel Position Offset", input.integer, minval=0)

    lT = input( 1, "--- Line Thickness ---", input.integer, options=[1,2,3])


    sec1 = security(sec1in, timeframe.period, close)

    //===== Calculations =====//

    R = cc(src, sec1, len) // Traditional Pearson

    adjr = adj( R, len) // Adjusted R

    float r = na

    if(mode == "R")

    r := R

    if(mode == "Adjusted R")

    r := adjr

    R2 = pow( cc(src, sec1, rlen) , 2) // R Squared

    adjR2 = pow(adj(cc(src, sec1, rlen), rlen), 2) // R Sqaured Based on Adjusted R

    float r2 = na

    if(mode == "R")

    r2 := R2

    if(mode == "Adjusted R")

    r2 := adjR2

    // Fisher Transform

    z = 0.5 * log((r + 1.0) / (1.0 - r)) // Fisher

    se = 1.0 / sqrt(len - 3) // Standard Error

    zl = z - csd * se // Lower Limit for fisher z

    zu = z + csd * se // Upper Limit for fisher z 95% confidence

    // Inverse Fisher Transform to Transform Back to r


    rl = (exp(2.0 * zl) - 1.0) / (exp(2.0 * zl) + 1.0) // Lower limit for r

    ru = (exp(2.0 * zu) - 1.0) / (exp(2.0 * zu) + 1.0) // Upper limit for r

    // P Test

    t = sqrt((1.0 - pow(r , 2)) / (len - 2)) // T value based on t distribution degrees of freedom 2

    Pvalue = exp(-0.5 * pow(r / t, 2)) / (t * sqrt(2.0 * cPI)) // P Value

    pro = Pvalue > 1.0 ? 1.0 : Pvalue // Limit P value overshoot 1

    //===== Plotting =====//

    colorCC = (pro < cp and r < 0.0) ? #FF0000ff :

    (pro < cp and r > 0.0) ? #00FF00ff :

    #FFFFFFff

    colorP = (pro < cp) ? #FFFF00ff : #C0C0C040

    plot( sp ? pro : na, color=colorP, title="P Value", style=plot.style_columns)

    plotUpper = plot( rl, color=sc ? color.new(#00C0FFff,100) : invisible, style=plot.style_linebr,


    title="Confidence Interval Lower" )

    plotLower = plot( ru, color=sc ? color.new(#00C0FFff,100) : invisible, style=plot.style_linebr,


    title="Confidence Interval Higher")

    fill(plotUpper, plotLower, color=sc ? color.new(#00C0FFff, 85) : invisible)

    plot( r, linewidth=lT , color=colorCC , style=plot.style_linebr, title="🐷 Correlation")

    plot(sp ? cp : na, color=color.new(#C0C0C0ff,30), trackprice=true, show_last=1, title="P value


    Threshold", style=plot.style_linebr)

    plot(sp ? na : 0, color=color.new(#C0C0C0ff,30), trackprice=true, show_last=1, title="Zero Line")

    hline( 1.0, color=color.new(#00FFFFff,30))

    hline( -1.0, color=color.new(#FF00FFff,30))

    // Information Panel
    sig() =>

    return = (pro < cp) and r > 0 ? "Significant Positive Correlation" :

    (pro < cp) and r < 0 ? "Significant Negative Correlation" :

    "Insignificant Correlation"

    if(pan)

    txt = "R : " + tostring(Round(r ,3)) +

    "\n\n R Squared : " + tostring(Round(r2 ,4)) +

    "\n\n P Value : " + tostring(Round(pro,4)) +

    "\n\n" + sig()

    You might also like