Scribd Logo
Upload

Welcome to Scribd!

  • 33 views

    On Using These Lecture Notes

    Uploaded by

    kfoster5126
    The document provides an overview of using covariance and correlation to measure the relationship between two variables. It explains that covariance measures how two variables move together or apart from their means, while correlation scales covariance by the variables' standard deviations to produce a value between -1 and 1. Taking notes during lectures is important because it forces students to process and internalize the information in their own words, rather than just passively listening. Graphs are used to illustrate strong and weak relationships between variable pairs.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd

    On Using These Lecture Notes

    Uploaded by

    kfoster5126
    33 views6 pages
    The document provides an overview of using covariance and correlation to measure the relationship between two variables. It explains that covariance measures how two variables move together or apart from their means, while correlation scales covariance by the variables' standard deviations to produce a value between -1 and 1. Taking notes during lectures is important because it forces students to process and internalize the information in their own words, rather than just passively listening. Graphs are used to illustrate strong and weak relationships between variable pairs.

    Original Title

    Lecture Notes 2

    Copyright

    © Attribution Non-Commercial (BY-NC)

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document provides an overview of using covariance and correlation to measure the relationship between two variables. It explains that covariance measures how two variables move together or apart from their means, while correlation scales covariance by the variables' standard deviations to produce a value between -1 and 1. Taking notes during lectures is important because it forces students to process and internalize the information in their own words, rather than just passively listening. Graphs are used to illustrate strong and weak relationships between variable pairs.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    33 views6 pages

    On Using These Lecture Notes

    Uploaded by

    kfoster5126
    The document provides an overview of using covariance and correlation to measure the relationship between two variables. It explains that covariance measures how two variables move together or apart from their means, while correlation scales covariance by the variables' standard deviations to produce a value between -1 and 1. Taking notes during lectures is important because it forces students to process and internalize the information in their own words, rather than just passively listening. Graphs are used to illustrate strong and weak relationships between variable pairs.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 6

    Lecture Notes 2

    Econ 29000, Principles of Statistics


    Kevin R Foster, CCNY
    Spring 2012


    On using these Lecture Notes:
    We sometimes don't realize the real reason why our good habits work. In the
    case of taking notes during lecture, this is probably the case. You're not writing
    it, in order to have some information later. If you took your day's notes, ripped
    them into shreds, and threw them away, you would still learn the material much
    better than if you hadn't taken notes. The process of listening, asking "what are
    the important things said?," answering this, then writing out the answer in your
    own words that's what's important! So even though I give out lecture notes,
    don't stop taking notes during class. Notes are not just are not primarily! a
    way to capture the fleeting knowledge that the instructor just said, before the
    information vanishes. Instead they are a way to process the information in a
    more thorough and more profound way. So keep on taking notes, even if it
    seems ridiculous. The reason for note-taking is to take in the material, put it
    into your own words, and output it. That's learning.

    Two Variables
    In a case where we have two variables, X and Y, we want to know how or if they are related, so
    we use covariance and correlation.

    Suppose we have a simple case where X has a two-part distribution that depends on another
    variable, Y. Suppose that Y is what we call a "dummy" variable: it is either a one or a zero but
    cannot have any other value. (Dummy variables are often used to encode answers to simple
    "Yes/No" questions where a "Yes" is indicated with a value of one and a "No" corresponds with
    a zero. Dummy variables are sometimes called "binary" or "logical" variables.) X might have a
    different mean depending on the value of Y.

    There are millions of examples of different means between two groups. GPA might be
    considered, with the mean depending on whether the student is a grad or undergrad. Or
    income might be the variable studied, which changes depending on whether a person has a
    college degree or not. You might object: but there are lots of other reasons why GPA or
    income could change, not just those two little reasons of course! We're not ruling out any
    further complications; we're just working through one at a time.

    In the PUMS data, X might be "wage and salary income in past 12 months" and Y would be
    male or female. Would you expect that the mean of X for men is greater or less than the mean
    of X for women?

    Run this on SPSS ...

    In a case where X has two distinct distributions depending on whether the dummy variable, Y,
    is zero or one, we might find the sample average for each part, so calculate the average when Y
    is equal to one and the average when Y is zero, which we denote
    ( ) ( )
    0 1
    0 , 1 ,
    Y Y
    X Y X Y or X X
    = =
    = = . These are called conditional means since they give the
    mean, conditional on some value.

    In this case the value of 1 X Y = is the same as the average of X Y .
    { } { } { }
    1
    1 1 1 1
    1 1 1 1
    1 0 1
    N N N N
    i i i i i Y
    i i i i
    XY X Y X Y X Y X Y X
    N N N N
    =
    = = = =
    = = = + = = = =

    .
    This is because the value of anything times zero is itself zero, so the term { }
    1
    0
    n
    i
    i
    X Y
    =
    =

    drops
    out. While it is easy to see how this additional information is valuable when Y is a dummy
    variable, it is a bit more difficult to see that it is valuable when Y is a continuous variable why
    might we want to look at the multiplied value, X Y ?

    Use Your Eyes
    We are accustomed to looking at graphs that show values of two variables and trying to
    discern patterns. Consider these two graphs of financial variables.

    This plots the returns of Hong Kong's Hang Seng index against the returns of Singapore's
    Straits Times index (over the period from Dec 29, 1989 to Sept 1, 2010)



    This next graph shows the S&P 500 returns and interest rates (1-month Eurodollar) during Jan
    2, 1990 Sept 1, 2010.



    You don't have to be a highly-skilled econometrician to see the difference in the relationships.
    It would seem reasonable to state that the Hong Kong and Singapore stock indexes are closely
    linked; while US stock returns are not closely related to US interest rates.

    We want to ask, how could we measure these relationships? Since these two graphs are rather
    extreme cases, how can we distinguish cases in the middle? And then there is one farther,
    even more important question: how can we try to guard against seeing relationships where, in
    fact, none actually exist? The second question is the big one, which most of this course (and
    much of the discipline of statistics) tries to answer. But start with the first question.

    How can we measure the relationship?
    Correlation measures how/if two variables move together.

    Recall from above that we looked at the average of X Y when Y was a dummy variable taking
    only the values of zero or one. Return to the case where Y is not a dummy but is a continuous
    variable just like X. It is still useful to find the average of X Y even in the case where Y is from
    a continuous distribution and can take any value,
    1
    1
    n
    i i
    i
    XY X Y
    n
    =
    =

    . It is a bit more useful if we
    re-write X and Y as differences from their means, so finding:
    ( )( )
    1
    1
    N
    i i
    i
    X X Y Y
    N
    =

    .
    This is the covariance, which is denoted cov(X,Y) or
    XY
    .

    A positive covariance shows that when X
    is above its mean, Y tends to also be
    above its mean (and vice versa) so either
    a positive number times a positive
    number gives a positive or a negative
    times a negative gives a positive.

    A negative covariance shows that when X
    is above its mean, Y tends to be below its
    mean (and vice versa). So when one is
    positive the other is negative, which gives
    a negative value when multiplied.

    A bit of math (extra):
    ( )( )
    1
    1
    N
    i i
    i
    X X Y Y
    N
    =

    =
    ( )
    1
    1
    N
    i i i i
    i
    X Y XY X Y XY
    N
    =
    +

    =
    1 1 1 1
    1 1 1 1
    N N N N
    i i i i
    i i i i
    X Y XY X Y XY
    N N N N
    = = = =
    +

    =
    1 1 1 1
    1 1 1 1
    1
    N N N N
    i i i i
    i i i i
    X Y X Y Y X XY
    N N N N
    = = = =
    +

    =
    1
    1
    N
    i i
    i
    X Y XY YX XY
    N
    =
    +

    =
    1
    1
    N
    i i
    i
    X Y XY
    N
    =


    (a strange case because it makes FOIL look like just FL!)


    Covariance is sometimes scaled by the standard deviations of X and Y in order to eliminate
    problems of measurement units, so define the correlation as:

    ( )( )
    1
    1
    N
    i i
    i XY
    XY
    X Y X y
    X X Y Y
    N o

    o o o o
    =

    = =

    or Corr(X,Y),
    where the Greek letter "rho" denotes the correlation coefficient. With some algebra you can
    show that is always between negative one and positive one; 1 1
    XY
    s s .

    Two variables will have a perfect correlation if they are identical; they would be perfectly
    inversely correlated if one is just the negative of the other (assets and liabilities, for example).
    Variables with a correlation close to one (in absolute value) are very similar; variables with a
    low or zero correlation are nearly or completely unrelated.

    Sample covariances and sample correlations
    Just as with the average and standard deviation, we can estimate the covariance and
    correlation between any two variables. And just as with the sample average, the sample
    covariance and sample correlation will have distributions around their true value.

    Go back to the case of the Hang Seng/Straits Times stock indexes. We can't just say that when
    one is big, the other is too. We want to be a bit more precise and say that when one is above
    its mean, the other tends to be above its mean, too. We might additionally state that, when
    the standardized value of one is high, the other standardized value is also high.

    Tthe standardized value of a variable, X, is
    ,
    i
    X i
    X
    X X
    Z
    s

    = , and the standardized value of Y is


    ,
    i
    Y i
    Y
    Y Y
    Z
    s

    = .

    Multiplying the two values together,
    , , X i Y i
    Z Z , gives a useful indicator since if both values are
    positive then the multiplication will be positive; if both are negative then the multiplication will
    again be positive. So if the values of Z
    X
    and Z
    Y
    are perfectly linked together then multiplying
    them together will get a positive number. On the other hand, if Z
    X
    and Z
    Y
    are oppositely
    related, so whenever one is positive the other is negative, then multiplying them together will
    get a negative number. And if Z
    X
    and Z
    Y
    are just random and not related to each other, then
    multiplying them will sometimes give a positive and sometimes a negative number.

    Sum up these multiplied values and get the (population) correlation,
    , ,
    1
    1
    N
    X i Y i
    i
    Z Z
    N
    =

    .

    This can be written as
    ( )( )
    , ,
    1 1 1
    1 1 1 1
    N N N
    i i
    X i Y i i i
    i i i
    x Y x Y
    X X Y Y
    Z Z X X Y Y
    N N s s N s s
    = = =
    | || |
    = =
    | |
    \ . \ .

    .
    The population correlation between X and Y is denoted
    XY
    ; the sample correlation is
    XY
    r .
    Again the difference is whether you divide by N or (N 1). Both correlations are always
    between -1 and +1; 1 1; 1 1 r s s s s .

    We often think of drawing lines to summarize relationships; the correlation tells us something
    about how well a line would fit the data. A correlation with an absolute value near 1 or -1 tells
    us that a line (with either positive or negative slope) would fit well; a correlation near zero tells
    us that there is "zero relationship."

    The fact that a negative value can infer a relationship might seem surprising but consider for
    example poker. Suppose you have figured out that an opponent makes a particular gesture
    when her cards are no good you can exploit that knowledge, even if it is a negative
    relationship. In finance, if a fund manager finds two assets that have a strong negative
    correlation, that one has high returns when the other has low returns, then again this
    information can exploited by taking offsetting positions.

    You might also like

    pFad - Phonifier reborn

    Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

    Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


    Alternative Proxies:

    Alternative Proxy

    pFad Proxy

    pFad v3 Proxy

    pFad v4 Proxy