In stochastics, the formula of Wald or Wald's identity is an equation with the help of which the expected value of sums of random variables with a random number of summands can be calculated. It was proven in 1944 in a work by the mathematician Abraham Wald .
formulation
Let it be a sequence of independent, identically distributed , integrable random variables and a -value random variable with , which is independent of the sequence . Then applies
(
X
n
)
n
≥
1
{\ displaystyle (X_ {n}) _ {n \ geq 1}}
T
{\ displaystyle T}
N
{\ displaystyle \ mathbb {N}}
E.
(
T
)
<
∞
{\ displaystyle \ operatorname {E} (T) <\ infty}
(
X
n
)
{\ displaystyle (X_ {n})}
E.
(
∑
k
=
1
T
X
k
)
=
E.
(
T
)
E.
(
X
1
)
{\ displaystyle \ operatorname {E} \ left (\ sum _ {k = 1} ^ {T} X_ {k} \ right) = \ operatorname {E} (T) \ operatorname {E} (X_ {1}) }
.
proof
Because is independent of the sequence , it follows from conditions on the value of :
T
{\ displaystyle T}
(
X
n
)
{\ displaystyle (X_ {n})}
T
{\ displaystyle T}
E.
(
∑
k
=
1
T
X
k
|
T
=
n
)
=
E.
(
∑
k
=
1
n
X
k
)
=
∑
k
=
1
n
E.
(
X
k
)
=
n
E.
(
X
1
)
{\ displaystyle \ operatorname {E} \ left (\ sum _ {k = 1} ^ {T} X_ {k} \; {\ Big |} \; T = n \ right) = \ operatorname {E} \ left (\ sum _ {k = 1} ^ {n} X_ {k} \ right) = \ sum _ {k = 1} ^ {n} \ operatorname {E} (X_ {k}) = n \ operatorname {E } (X_ {1})}
,
so
E.
(
∑
k
=
1
T
X
k
|
T
)
=
T
E.
(
X
1
)
{\ displaystyle \ operatorname {E} \ left (\ sum _ {k = 1} ^ {T} X_ {k} \; {\ Big |} \; T \ right) = T \ operatorname {E} (X_ { 1})}
.
By applying the expectation to this equation, one finally obtains
E.
(
∑
k
=
1
T
X
k
)
=
E.
(
E.
(
∑
k
=
1
T
X
k
|
T
)
)
=
E.
(
T
E.
(
X
1
)
)
=
E.
(
T
)
E.
(
X
1
)
{\ displaystyle \ operatorname {E} \ left (\ sum _ {k = 1} ^ {T} X_ {k} \ right) = \ operatorname {E} \ left (\ operatorname {E} \ left (\ sum _ {k = 1} ^ {T} X_ {k} \; {\ Big |} \; T \ right) \ right) = \ operatorname {E} (T \ operatorname {E} (X_ {1})) = \ operatorname {E} (T) \ operatorname {E} (X_ {1})}
.
If they are all valued, the elementary proof can also take place via probability-generating functions using the chain rule .
X
i
{\ displaystyle X_ {i}}
N
0
{\ displaystyle \ mathbb {N} _ {0}}
Generalization to stop times
Let it now be a sequence of identically distributed, integrable random variables that is adapted to a filtering process , i.e. it is measurable for all . If by independently for all and an integrable stop time with respect , so also the formula is of forest:
(
X
n
)
n
≥
1
{\ displaystyle (X_ {n}) _ {n \ geq 1}}
(
F.
n
)
n
{\ displaystyle ({\ mathcal {F}} _ {n}) _ {n}}
n
{\ displaystyle n}
X
n
{\ displaystyle X_ {n}}
F.
n
{\ displaystyle {\ mathcal {F}} _ {n}}
X
n
+
1
{\ displaystyle X_ {n + 1}}
F.
n
{\ displaystyle {\ mathcal {F}} _ {n}}
n
∈
N
{\ displaystyle n \ in \ mathbb {N}}
T
{\ displaystyle T}
(
F.
n
)
n
{\ displaystyle ({\ mathcal {F}} _ {n}) _ {n}}
E.
(
∑
k
=
1
T
X
k
)
=
E.
(
T
)
E.
(
X
1
)
{\ displaystyle \ operatorname {E} \ left (\ sum _ {k = 1} ^ {T} X_ {k} \ right) = \ operatorname {E} (T) \ operatorname {E} (X_ {1}) }
.
Related concepts
Similar statements about the variance of composite distributions can be made with the Blackwell-Girshick equation .
Individual evidence
↑ Abraham Wald: On Cumulative Sums of Random Variables. In: The Annals of Mathematical Statistics No. 15, Vol. 3, pp. 283-296, doi : 10.1214 / aoms / 1177731235 .
↑ David Meintrup, Stefan Schäffler: Stochastics. Theory and applications. Springer, Berlin / Heidelberg 2005, ISBN 3-540-21676-6 , p. 287.
↑ Heinz Bauer : Probability Theory. 5th edition. De Gruyter textbook, Berlin 2002, ISBN 3-11-017236-4 , chapter 17.
<img src="https://de.wikipedia.org/wiki/Special:CentralAutoLogin/start?type=1x1" alt="" title="" width="1" height="1" style="border: none; position: absolute;">