Asimetría del logaritmo de una variable aleatoria gamma

16

Considere la variable aleatoria gamma XΓ(α,θ) . Hay fórmulas claras para la media, la varianza y la asimetría:

E[X]=αθVar[X]=αθ2=1/αE[X]2Skewness[X]=2/α

Considere ahora una variable aleatoria transformada logarítmica Y=log(X) . Wikipedia da fórmulas para la media y la varianza:

E[Y]=ψ(α)+log(θ)Var[Y]=ψ1(α)

mediante funciones digamma y trigamma que se definen como la primera y la segunda derivada del logaritmo de la función gamma.

¿Cuál es la fórmula para la asimetría?

¿Aparecerá la función tetragamma?

(Lo que me hizo preguntarme sobre esto es una elección entre las distribuciones lognormal y gamma, ver Distribuciones gamma vs. lognormal . Entre otras cosas, difieren en sus propiedades de asimetría. En particular, la asimetría del logaritmo de lognormal es trivialmente igual a cero. Mientras que la asimetría del logaritmo de gamma es negativa. Pero, ¿qué tan negativa? ..)

ameba dice Reinstate Monica
fuente
1
¿ Esto ayuda? O esto ?
S. Kolassa - Restablece a Mónica el
No estoy muy seguro de qué es la distribución log-gamma. Si está relacionado con gamma ya que lognormal está relacionado con lo normal, entonces estoy preguntando sobre otra cosa (porque "lognormal", confusamente, es la distribución de exp (normal) no de log (normal)).
ameba dice Reinstate Monica
1
@Glen_b: Para ser honesto, diría que llamar a lo exponencial de lo normal un "lognormal" es mucho más inconsistente y confuso. Aunque, lamentablemente, más establecido.
S. Kolassa - Restablece a Mónica el
2
@Stephan véase también el log-logística, log-Cauchy, log-Laplace, etc, etc Es una convención establecida con más claridad que lo contrario
Glen_b -Reinstate Mónica
1
Sí; He tenido cuidado de no decir "log-gamma" en ninguna parte en relación con esta distribución por este motivo. (Lo he usado en el pasado de una manera consistente al log-normal)
Glen_b -Reinstate Monica

Respuestas:

12

La función generadora de momentos de Y = ln X es útil en este caso, ya que tiene una forma algebraica simple. Por la definición de mgf, tenemos M ( t )M(t)Y=lnX

M(t)=E[etlnX]=E[Xt]=1Γ(α)θα0xα+t1ex/θdx=θtΓ(α)0yα+t1eydy=θtΓ(α+t)Γ(α).

Verifiquemos la expectativa y la varianza que diste. Tomando derivados, tenemos

M(t)=Γ(α+t)Γ(α)θt+Γ(α+t)Γ(α)θtln(θ)
and
M(t)=Γ(α+t)Γ(α)θt+2Γ(α+t)Γ(α)θtln(θ)+Γ(α+t)Γ(α)θtln2(θ).
Hence,
E[Y]=ψ(0)(α)+ln(θ),E[Y2]=Γ(α)Γ(α)+2ψ(0)(α)ln(θ)+ln2(θ).
It follows then
Var(Y)=E[Y2]E[Y]2=Γ(α)Γ(α)(Γ(α)Γ(α))2=ψ(1)(α).

To find the skewness, note the cumulant generating function (thanks @probabilityislogic for the tip) is

K(t)=lnM(t)=tlnθ+lnΓ(α+t)lnΓ(α).
The first cumulant is thus simply K(0)=ψ(0)(α)+ln(θ). Recall that ψ(n)(x)=dn+1lnΓ(x)/dxn+1, so the subsequent cumulants are K(n)(0)=ψ(n1)(α), n2. The skewness is therefore
E[(YE[Y])3]Var(Y)3/2=ψ(2)(α)[ψ(1)(α)]3/2.

As a side note, this particular distribution appeared to have been thoroughly studied by A. C. Olshen in his Transformations of the Pearson Type III Distribution, Johnson et al.'s Continuous Univariate Distributions also has a small piece about it. Check those out.

Francis
fuente
3
You should differentiate K(t)=log[M(t)]=tlog[θ]+log[Γ(α+t)]log[Γ(α)] instead of M(t) as this is the cumulant generating function - more directly related to central moments - skew=K(3)(0)=ψ(2)(α) where ψ(n)(z) is the polygamma function
probabilityislogic
1
@probabilityislogic: very good call, changed my answer
Francis
@probabilityislogic This is a great addition, thanks a lot. I just want to note, lest some readers be confused, that skewness is not directly given by the third cumulant: it's the third standardized moment, not the third central moment. Francis has it correct in his answer, but the last formula in your comment is not quite right.
amoeba says Reinstate Monica
13

I. Direct computation

Gradshteyn & Ryzhik [1] (sect 4.358, 7th ed) list explicit closed forms for

0xν1eμx(lnx)pdx
for p=2,3,4 while the p=1 case is done in 4.352 (assuming you regard expressions in Γ,ψ and ζ functions as closed form) -- from which it is definitely doable up to kurtosis; they give the integral for all p as a derivative of a gamma function so presumably it's feasible to go higher. So skewness is certainly doable but not especially "neat".

Details of the derivation of the formulas in 4.358 are in [2]. I'll quote the formulas given there since they're slightly more succinctly stated and put 4.352.1 in the same form.

Let δ=ψ(a)lnμ. Then:

0xa1eμxlnxdx=Γ(a)μa{δ}0xa1eμxln2xdx=Γ(a)μa{δ2+ζ(2,a)}0xa1eμxln3xdx=Γ(a)μa{δ3+3ζ(2,a)δ2ζ(3,a)}0xa1eμxln4xdx=Γ(a)μa{δ4+6ζ(2,a)δ28ζ(3,a)δ+3ζ2(2,a)+6ζ(4,a))}

where ζ(z,q)=n=01(n+q)z is the Hurwitz zeta function (the Riemann zeta function is the special case q=1).

Now on to the moments of the log of a gamma random variable.

Noting firstly that on the log scale the scale or rate parameter of the gamma density is merely a shift-parameter, so it has no impact on the central moments; we may take whichever one we're using to be 1.

If XGamma(α,1) then

E(logpX)=1Γ(α)0logpxxα1exdx.

We can set μ=1 in the above integral formulas, which gives us raw moments; we have E(Y), E(Y2), E(Y3), E(Y4).

Since we have eliminated μ from the above, without fear of confusion we're now free to re-use μk to represent the k-th central moment in the usual fashion. We may then obtain the central moments from the raw moments via the usual formulas.

Then we can obtain the skewness and kurtosis as μ3μ23/2 and μ4μ22.


A note on terminology

It looks like Wolfram's reference pages write the moments of this distribution (they call it ExpGamma distribution) in terms of the polygamma function.

By contrast, Chan (see below) calls this the log-gamma distribution.


II. Chan's formulas via MGF

Chan (1993) [3] gives the mgf as the very neat Γ(α+t)/Γ(α).

(A very nice derivation for this is given in Francis' answer, using the simple fact that the mgf of log(X) is just E(Xt).)

Consequently the moments have fairly simple forms. Chan gives:

E(Y)=ψ(α)

and the central moments as

E(YμY)2=ψ(α)E(YμY)3=ψ(α)E(YμY)4=ψ(α)

and so the skewness is ψ(α)/(ψ(α)3/2) and kurtosis is ψ(α)/(ψ(α)2). Presumably the earlier formulas I have above should simplify to these.

Conveniently, R offers digamma (ψ) and trigamma (ψ) functions as well as the more general polygamma function where you select the order of the derivative. (A number of other programs offer similarly convenient functions.)

Consequently we can compute the skewness and kurtosis quite directly in R:

skew.eg <- function(a) psigamma(a,2)/psigamma(a,1)^(3/2)
kurt.eg <- function(a) psigamma(a,3)/psigamma(a,1)^2

Trying a few values of a (α in the above), we reproduce the first few rows of the table at the end of Sec 2.2 in Chan [3], except that the kurtosis values in that table are supposed to be excess kurtosis, but I just calculated kurtosis by the formulas given above by Chan; these should differ by 3.

(E.g. for the log of an exponential, the table says the excess kurtosis is 2.4, but the formula for β2 is ψ(1)/ψ(1)2 ... and that is 2.4.)

Simulation confirms that as we increase sample size, the kurtosis of a log of an exponential is converging to around 5.4 not 2.4. It appears that the thesis possibly has an error.

Consequently, Chan's formulas for central moments appear to actually be the formulas for the cumulants (see the derivation in Francis' answer). This would then mean that the skewness formula was correct as is; because the second and third cumulants are equal to the second and third central moments.

Nevertheless these are particularly convenient formulas as long as we keep in mind that kurt.eg is giving excess kurtosis.

References

[1] Gradshteyn, I.S. & Ryzhik I.M. (2007), Table of Integrals, Series, and Products, 7th ed.
Academic Press, Inc.

[2] Victor H. Moll (2007)
The integrals in Gradshteyn and Ryzhik, Part 4: The gamma function
SCIENTIA Series A: Mathematical Sciences, Vol. 15, 37–46
Universidad Técnica Federico Santa María, Valparaíso, Chile
http://129.81.170.14/~vhm/FORM-PROOFS_html/final4.pdf

[3] Chan, P.S. (1993),
A statistical study of log-gamma distribution,
McMaster University (Ph.D. thesis)
https://macsphere.mcmaster.ca/bitstream/11375/6816/1/fulltext.pdf

Glen_b -Reinstate Monica
fuente
1
Cool. Thanks a lot! According to the encyclopedia entry that Stephan linked to above, the final answer for skewness is ψ(α)/ψ(α)3/2 (which almost qualifies as "neat"!). So it seems that all the scary zetas will have to cancel out.
amoeba says Reinstate Monica
1
Sorry only just now saw your comment (I've been editing for about an hour or so); that's correct, though if the Encyclopedia gives kurtosis the way Chan gives it in his thesis, it seems that it's wrong (as given above), but readily corrected. The neat formulas appear to be for cumulants rather than standardized central moments.
Glen_b -Reinstate Monica
Yes, the Encyclopedia does give the same formula for kurtosis.
amoeba says Reinstate Monica
Hmm, I mean to refer to the things normally denoted γ1 and γ2. I will fix.
Glen_b -Reinstate Monica
2
I should probably add the note that the Hurwitz zeta function can be expressed in terms of the polygamma function, and vice versa:
ψ(n)(z)=(1)n+1Γ(n+1)ζ(n+1,z)
So, the answer to the @amoeba's question of "will the tetragamma function appear?" is YES.
J. M. is not a statistician