18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline...

51
18.175: Lecture 28 Even more on martingales Scott Sheffield MIT 18.175 Lecture 28

Transcript of 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline...

Page 1: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

18.175: Lecture 28

Even more on martingales

Scott Sheffield

MIT

18.175 Lecture 28

Page 2: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Outline

Recollections

More martingale theorems

18.175 Lecture 28

Page 3: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Outline

Recollections

More martingale theorems

18.175 Lecture 28

Page 4: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Recall: conditional expectation

I Say we’re given a probability space (Ω,F0,P) and a σ-fieldF ⊂ F0 and a random variable X measurable w.r.t. F0, withE |X | <∞. The conditional expectation of X given F is anew random variable, which we can denote by Y = E (X |F).

I We require that Y is F measurable and that for all A in F ,we have

∫A XdP =

∫A YdP.

I Any Y satisfying these properties is called a version ofE (X |F).

I Theorem: Up to redefinition on a measure zero set, therandom variable E (X |F) exists and is unique.

I This follows from Radon-Nikodym theorem.

I Theorem: E (X |Fi ) is a martingale if Fi is an increasingsequence of σ-algebras and E (|X |) <∞.

18.175 Lecture 28

Page 5: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Recall: conditional expectation

I Say we’re given a probability space (Ω,F0,P) and a σ-fieldF ⊂ F0 and a random variable X measurable w.r.t. F0, withE |X | <∞. The conditional expectation of X given F is anew random variable, which we can denote by Y = E (X |F).

I We require that Y is F measurable and that for all A in F ,we have

∫A XdP =

∫A YdP.

I Any Y satisfying these properties is called a version ofE (X |F).

I Theorem: Up to redefinition on a measure zero set, therandom variable E (X |F) exists and is unique.

I This follows from Radon-Nikodym theorem.

I Theorem: E (X |Fi ) is a martingale if Fi is an increasingsequence of σ-algebras and E (|X |) <∞.

18.175 Lecture 28

Page 6: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Recall: conditional expectation

I Say we’re given a probability space (Ω,F0,P) and a σ-fieldF ⊂ F0 and a random variable X measurable w.r.t. F0, withE |X | <∞. The conditional expectation of X given F is anew random variable, which we can denote by Y = E (X |F).

I We require that Y is F measurable and that for all A in F ,we have

∫A XdP =

∫A YdP.

I Any Y satisfying these properties is called a version ofE (X |F).

I Theorem: Up to redefinition on a measure zero set, therandom variable E (X |F) exists and is unique.

I This follows from Radon-Nikodym theorem.

I Theorem: E (X |Fi ) is a martingale if Fi is an increasingsequence of σ-algebras and E (|X |) <∞.

18.175 Lecture 28

Page 7: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Recall: conditional expectation

I Say we’re given a probability space (Ω,F0,P) and a σ-fieldF ⊂ F0 and a random variable X measurable w.r.t. F0, withE |X | <∞. The conditional expectation of X given F is anew random variable, which we can denote by Y = E (X |F).

I We require that Y is F measurable and that for all A in F ,we have

∫A XdP =

∫A YdP.

I Any Y satisfying these properties is called a version ofE (X |F).

I Theorem: Up to redefinition on a measure zero set, therandom variable E (X |F) exists and is unique.

I This follows from Radon-Nikodym theorem.

I Theorem: E (X |Fi ) is a martingale if Fi is an increasingsequence of σ-algebras and E (|X |) <∞.

18.175 Lecture 28

Page 8: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Recall: conditional expectation

I Say we’re given a probability space (Ω,F0,P) and a σ-fieldF ⊂ F0 and a random variable X measurable w.r.t. F0, withE |X | <∞. The conditional expectation of X given F is anew random variable, which we can denote by Y = E (X |F).

I We require that Y is F measurable and that for all A in F ,we have

∫A XdP =

∫A YdP.

I Any Y satisfying these properties is called a version ofE (X |F).

I Theorem: Up to redefinition on a measure zero set, therandom variable E (X |F) exists and is unique.

I This follows from Radon-Nikodym theorem.

I Theorem: E (X |Fi ) is a martingale if Fi is an increasingsequence of σ-algebras and E (|X |) <∞.

18.175 Lecture 28

Page 9: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Recall: conditional expectation

I Say we’re given a probability space (Ω,F0,P) and a σ-fieldF ⊂ F0 and a random variable X measurable w.r.t. F0, withE |X | <∞. The conditional expectation of X given F is anew random variable, which we can denote by Y = E (X |F).

I We require that Y is F measurable and that for all A in F ,we have

∫A XdP =

∫A YdP.

I Any Y satisfying these properties is called a version ofE (X |F).

I Theorem: Up to redefinition on a measure zero set, therandom variable E (X |F) exists and is unique.

I This follows from Radon-Nikodym theorem.

I Theorem: E (X |Fi ) is a martingale if Fi is an increasingsequence of σ-algebras and E (|X |) <∞.

18.175 Lecture 28

Page 10: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Martingales

I Let Fn be increasing sequence of σ-fields (called a filtration).

I A sequence Xn is adapted to Fn if Xn ∈ Fn for all n. If Xn isan adapted sequence (with E |Xn| <∞) then it is called amartingale if

E (Xn+1|Fn) = Xn

for all n. It’s a supermartingale (resp., submartingale) ifsame thing holds with = replaced by ≤ (resp., ≥).

18.175 Lecture 28

Page 11: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Martingales

I Let Fn be increasing sequence of σ-fields (called a filtration).

I A sequence Xn is adapted to Fn if Xn ∈ Fn for all n. If Xn isan adapted sequence (with E |Xn| <∞) then it is called amartingale if

E (Xn+1|Fn) = Xn

for all n. It’s a supermartingale (resp., submartingale) ifsame thing holds with = replaced by ≤ (resp., ≥).

18.175 Lecture 28

Page 12: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Two big results

I Optional stopping theorem: Can’t make money inexpectation by timing sale of asset whose price is non-negativemartingale.

I Proof: Just a special case of statement about (H · X ) ifstopping time is bounded.

I Martingale convergence: A non-negative martingale almostsurely has a limit.

I Idea of proof: Count upcrossings (times martingale crosses afixed interval) and devise gambling strategy that makes lots ofmoney if the number of these is not a.s. finite. Basically, youbuy every time price gets below the interval, sell each time itgets above.

18.175 Lecture 28

Page 13: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Two big results

I Optional stopping theorem: Can’t make money inexpectation by timing sale of asset whose price is non-negativemartingale.

I Proof: Just a special case of statement about (H · X ) ifstopping time is bounded.

I Martingale convergence: A non-negative martingale almostsurely has a limit.

I Idea of proof: Count upcrossings (times martingale crosses afixed interval) and devise gambling strategy that makes lots ofmoney if the number of these is not a.s. finite. Basically, youbuy every time price gets below the interval, sell each time itgets above.

18.175 Lecture 28

Page 14: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Two big results

I Optional stopping theorem: Can’t make money inexpectation by timing sale of asset whose price is non-negativemartingale.

I Proof: Just a special case of statement about (H · X ) ifstopping time is bounded.

I Martingale convergence: A non-negative martingale almostsurely has a limit.

I Idea of proof: Count upcrossings (times martingale crosses afixed interval) and devise gambling strategy that makes lots ofmoney if the number of these is not a.s. finite. Basically, youbuy every time price gets below the interval, sell each time itgets above.

18.175 Lecture 28

Page 15: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Two big results

I Optional stopping theorem: Can’t make money inexpectation by timing sale of asset whose price is non-negativemartingale.

I Proof: Just a special case of statement about (H · X ) ifstopping time is bounded.

I Martingale convergence: A non-negative martingale almostsurely has a limit.

I Idea of proof: Count upcrossings (times martingale crosses afixed interval) and devise gambling strategy that makes lots ofmoney if the number of these is not a.s. finite. Basically, youbuy every time price gets below the interval, sell each time itgets above.

18.175 Lecture 28

Page 16: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Problems

I Assume Intrade prices are continuous martingales. (Forgetabout bid-ask spreads, possible longshot bias, this year’sbizarre arbitrage opportunities, discontinuities brought aboutby sudden spurts of information, etc.)

I How many primary candidates does one expect to ever exceed20 percent on Intrade primary nomination market? (Asked byAldous.)

I Compute probability of having a martingale price reach abefore b if martingale prices vary continuously.

I Polya’s urn: r red and g green balls. Repeatedly samplerandomly and add extra ball of sampled color. Ratio of red togreen is martingale, hence a.s. converges to limit.

18.175 Lecture 28

Page 17: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Problems

I Assume Intrade prices are continuous martingales. (Forgetabout bid-ask spreads, possible longshot bias, this year’sbizarre arbitrage opportunities, discontinuities brought aboutby sudden spurts of information, etc.)

I How many primary candidates does one expect to ever exceed20 percent on Intrade primary nomination market? (Asked byAldous.)

I Compute probability of having a martingale price reach abefore b if martingale prices vary continuously.

I Polya’s urn: r red and g green balls. Repeatedly samplerandomly and add extra ball of sampled color. Ratio of red togreen is martingale, hence a.s. converges to limit.

18.175 Lecture 28

Page 18: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Problems

I Assume Intrade prices are continuous martingales. (Forgetabout bid-ask spreads, possible longshot bias, this year’sbizarre arbitrage opportunities, discontinuities brought aboutby sudden spurts of information, etc.)

I How many primary candidates does one expect to ever exceed20 percent on Intrade primary nomination market? (Asked byAldous.)

I Compute probability of having a martingale price reach abefore b if martingale prices vary continuously.

I Polya’s urn: r red and g green balls. Repeatedly samplerandomly and add extra ball of sampled color. Ratio of red togreen is martingale, hence a.s. converges to limit.

18.175 Lecture 28

Page 19: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Problems

I Assume Intrade prices are continuous martingales. (Forgetabout bid-ask spreads, possible longshot bias, this year’sbizarre arbitrage opportunities, discontinuities brought aboutby sudden spurts of information, etc.)

I How many primary candidates does one expect to ever exceed20 percent on Intrade primary nomination market? (Asked byAldous.)

I Compute probability of having a martingale price reach abefore b if martingale prices vary continuously.

I Polya’s urn: r red and g green balls. Repeatedly samplerandomly and add extra ball of sampled color. Ratio of red togreen is martingale, hence a.s. converges to limit.

18.175 Lecture 28

Page 20: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Outline

Recollections

More martingale theorems

18.175 Lecture 28

Page 21: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Outline

Recollections

More martingale theorems

18.175 Lecture 28

Page 22: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Lp convergence theorem

I Theorem: If Xn is a martingale with supE |Xn|p <∞ wherep > 1 then Xn → X a.s. and in Lp.

I Proof idea: Have (EX+n )p ≤ (E |Xn|)p ≤ E |Xn|p for

martingale convergence theorem Xn → X a.s. Use Lp maximalinequality to get Lp convergence.

18.175 Lecture 28

Page 23: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Lp convergence theorem

I Theorem: If Xn is a martingale with supE |Xn|p <∞ wherep > 1 then Xn → X a.s. and in Lp.

I Proof idea: Have (EX+n )p ≤ (E |Xn|)p ≤ E |Xn|p for

martingale convergence theorem Xn → X a.s. Use Lp maximalinequality to get Lp convergence.

18.175 Lecture 28

Page 24: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Orthogonality of martingale increments

I Theorem: Let Xn be a martingale with EX 2n <∞ for all n. If

m ≤ n and Y ∈ Fm with EY 2 <∞, thenE((Xn − Xm)Y

)= 0.

I Proof idea: E((Xn − Xm)Y

)= E

[E((Xn − Xm)Y |Fm

)]=

E[YE((Xn − Xm)|Fm

)]= 0

I Conditional variance theorem: If Xn is a martingale withEX 2

n <∞ for all n thenE((Xn − Xm)2|Fm

)= E (X 2

n |Fm)− X 2m.

18.175 Lecture 28

Page 25: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Orthogonality of martingale increments

I Theorem: Let Xn be a martingale with EX 2n <∞ for all n. If

m ≤ n and Y ∈ Fm with EY 2 <∞, thenE((Xn − Xm)Y

)= 0.

I Proof idea: E((Xn − Xm)Y

)= E

[E((Xn − Xm)Y |Fm

)]=

E[YE((Xn − Xm)|Fm

)]= 0

I Conditional variance theorem: If Xn is a martingale withEX 2

n <∞ for all n thenE((Xn − Xm)2|Fm

)= E (X 2

n |Fm)− X 2m.

18.175 Lecture 28

Page 26: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Orthogonality of martingale increments

I Theorem: Let Xn be a martingale with EX 2n <∞ for all n. If

m ≤ n and Y ∈ Fm with EY 2 <∞, thenE((Xn − Xm)Y

)= 0.

I Proof idea: E((Xn − Xm)Y

)= E

[E((Xn − Xm)Y |Fm

)]=

E[YE((Xn − Xm)|Fm

)]= 0

I Conditional variance theorem: If Xn is a martingale withEX 2

n <∞ for all n thenE((Xn − Xm)2|Fm

)= E (X 2

n |Fm)− X 2m.

18.175 Lecture 28

Page 27: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I We know X 2n is a submartingale. By Doob’s decomposition,

an write X 2n = Mn + An where Mn is a martingale, and

An =n∑

m=1

E (X 2m|Fm−1)−X 2

m−1 =n∑

m=1

E((Xm−Xm−1)2|Fm−1

).

I An in some sense measures total accumulated variance bytime n.

I Theorem: E (supm |Xm|2) ≤ 4EA∞I Proof idea: L2 maximal equality gives

E (sup0≤m≤n |Xm|2) ≤ 4EX 2n = 4EAn. Use monotone

convergence.

18.175 Lecture 28

Page 28: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I We know X 2n is a submartingale. By Doob’s decomposition,

an write X 2n = Mn + An where Mn is a martingale, and

An =n∑

m=1

E (X 2m|Fm−1)−X 2

m−1 =n∑

m=1

E((Xm−Xm−1)2|Fm−1

).

I An in some sense measures total accumulated variance bytime n.

I Theorem: E (supm |Xm|2) ≤ 4EA∞I Proof idea: L2 maximal equality gives

E (sup0≤m≤n |Xm|2) ≤ 4EX 2n = 4EAn. Use monotone

convergence.

18.175 Lecture 28

Page 29: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I We know X 2n is a submartingale. By Doob’s decomposition,

an write X 2n = Mn + An where Mn is a martingale, and

An =n∑

m=1

E (X 2m|Fm−1)−X 2

m−1 =n∑

m=1

E((Xm−Xm−1)2|Fm−1

).

I An in some sense measures total accumulated variance bytime n.

I Theorem: E (supm |Xm|2) ≤ 4EA∞I Proof idea: L2 maximal equality gives

E (sup0≤m≤n |Xm|2) ≤ 4EX 2n = 4EAn. Use monotone

convergence.

18.175 Lecture 28

Page 30: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I We know X 2n is a submartingale. By Doob’s decomposition,

an write X 2n = Mn + An where Mn is a martingale, and

An =n∑

m=1

E (X 2m|Fm−1)−X 2

m−1 =n∑

m=1

E((Xm−Xm−1)2|Fm−1

).

I An in some sense measures total accumulated variance bytime n.

I Theorem: E (supm |Xm|2) ≤ 4EA∞

I Proof idea: L2 maximal equality givesE (sup0≤m≤n |Xm|2) ≤ 4EX 2

n = 4EAn. Use monotoneconvergence.

18.175 Lecture 28

Page 31: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I We know X 2n is a submartingale. By Doob’s decomposition,

an write X 2n = Mn + An where Mn is a martingale, and

An =n∑

m=1

E (X 2m|Fm−1)−X 2

m−1 =n∑

m=1

E((Xm−Xm−1)2|Fm−1

).

I An in some sense measures total accumulated variance bytime n.

I Theorem: E (supm |Xm|2) ≤ 4EA∞I Proof idea: L2 maximal equality gives

E (sup0≤m≤n |Xm|2) ≤ 4EX 2n = 4EAn. Use monotone

convergence.

18.175 Lecture 28

Page 32: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I Theorem: limn→∞ Xn exists and is finite a.s. on A∞ <∞.I Proof idea: Try fixing a and truncating at time

N = infn : An+1 > a2, use L2 convergence theorem.

18.175 Lecture 28

Page 33: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I Theorem: limn→∞ Xn exists and is finite a.s. on A∞ <∞.

I Proof idea: Try fixing a and truncating at timeN = infn : An+1 > a2, use L2 convergence theorem.

18.175 Lecture 28

Page 34: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Square integrable martingales

I Suppose we have a martingale Xn with EX 2n <∞ for all n.

I Theorem: limn→∞ Xn exists and is finite a.s. on A∞ <∞.I Proof idea: Try fixing a and truncating at time

N = infn : An+1 > a2, use L2 convergence theorem.

18.175 Lecture 28

Page 35: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Uniform integrability

I Say Xi , i ∈ I , are uniform integrable if

limM→∞

(supi∈I

E (|Xi |; |Xi | > M))

= 0.

I Example: Given (Ω,F0,P) and X ∈ L1, then a uniformlyintegral family is given by E (X |F) (where F ranges over allσ-algebras contained in F0).

I Theorem: If Xn → X in probability then the following areequivalent:

I Xn are uniformly integrableI Xn → X in L1

I E |Xn| → E |X | <∞

18.175 Lecture 28

Page 36: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Uniform integrability

I Say Xi , i ∈ I , are uniform integrable if

limM→∞

(supi∈I

E (|Xi |; |Xi | > M))

= 0.

I Example: Given (Ω,F0,P) and X ∈ L1, then a uniformlyintegral family is given by E (X |F) (where F ranges over allσ-algebras contained in F0).

I Theorem: If Xn → X in probability then the following areequivalent:

I Xn are uniformly integrableI Xn → X in L1

I E |Xn| → E |X | <∞

18.175 Lecture 28

Page 37: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Uniform integrability

I Say Xi , i ∈ I , are uniform integrable if

limM→∞

(supi∈I

E (|Xi |; |Xi | > M))

= 0.

I Example: Given (Ω,F0,P) and X ∈ L1, then a uniformlyintegral family is given by E (X |F) (where F ranges over allσ-algebras contained in F0).

I Theorem: If Xn → X in probability then the following areequivalent:

I Xn are uniformly integrableI Xn → X in L1

I E |Xn| → E |X | <∞

18.175 Lecture 28

Page 38: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Uniform integrability

I Say Xi , i ∈ I , are uniform integrable if

limM→∞

(supi∈I

E (|Xi |; |Xi | > M))

= 0.

I Example: Given (Ω,F0,P) and X ∈ L1, then a uniformlyintegral family is given by E (X |F) (where F ranges over allσ-algebras contained in F0).

I Theorem: If Xn → X in probability then the following areequivalent:

I Xn are uniformly integrable

I Xn → X in L1

I E |Xn| → E |X | <∞

18.175 Lecture 28

Page 39: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Uniform integrability

I Say Xi , i ∈ I , are uniform integrable if

limM→∞

(supi∈I

E (|Xi |; |Xi | > M))

= 0.

I Example: Given (Ω,F0,P) and X ∈ L1, then a uniformlyintegral family is given by E (X |F) (where F ranges over allσ-algebras contained in F0).

I Theorem: If Xn → X in probability then the following areequivalent:

I Xn are uniformly integrableI Xn → X in L1

I E |Xn| → E |X | <∞

18.175 Lecture 28

Page 40: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Uniform integrability

I Say Xi , i ∈ I , are uniform integrable if

limM→∞

(supi∈I

E (|Xi |; |Xi | > M))

= 0.

I Example: Given (Ω,F0,P) and X ∈ L1, then a uniformlyintegral family is given by E (X |F) (where F ranges over allσ-algebras contained in F0).

I Theorem: If Xn → X in probability then the following areequivalent:

I Xn are uniformly integrableI Xn → X in L1

I E |Xn| → E |X | <∞

18.175 Lecture 28

Page 41: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Submartingale convergence

I Following are equivalent for a submartingale:

I It’s uniformly integrable.I It converges a.s. and in L1.I It converges in L1.

18.175 Lecture 28

Page 42: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Submartingale convergence

I Following are equivalent for a submartingale:I It’s uniformly integrable.

I It converges a.s. and in L1.I It converges in L1.

18.175 Lecture 28

Page 43: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Submartingale convergence

I Following are equivalent for a submartingale:I It’s uniformly integrable.I It converges a.s. and in L1.

I It converges in L1.

18.175 Lecture 28

Page 44: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Submartingale convergence

I Following are equivalent for a submartingale:I It’s uniformly integrable.I It converges a.s. and in L1.I It converges in L1.

18.175 Lecture 28

Page 45: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Backwards martingales

I Suppose E (Xn+1|Fn) = X with n ≤ 0 (and Fn increasing as nincreases).

I Theorem: X−∞ = limn→−∞ Xn exists a.s. and in L1.

I Proof idea: Use upcrosing inequality to show expectednumber of upcrossings of any interval is finite. SinceXn = E (X0|Fn) the Xn are uniformly integrable, and we candeduce convergence in L1.

18.175 Lecture 28

Page 46: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Backwards martingales

I Suppose E (Xn+1|Fn) = X with n ≤ 0 (and Fn increasing as nincreases).

I Theorem: X−∞ = limn→−∞ Xn exists a.s. and in L1.

I Proof idea: Use upcrosing inequality to show expectednumber of upcrossings of any interval is finite. SinceXn = E (X0|Fn) the Xn are uniformly integrable, and we candeduce convergence in L1.

18.175 Lecture 28

Page 47: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

Backwards martingales

I Suppose E (Xn+1|Fn) = X with n ≤ 0 (and Fn increasing as nincreases).

I Theorem: X−∞ = limn→−∞ Xn exists a.s. and in L1.

I Proof idea: Use upcrosing inequality to show expectednumber of upcrossings of any interval is finite. SinceXn = E (X0|Fn) the Xn are uniformly integrable, and we candeduce convergence in L1.

18.175 Lecture 28

Page 48: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

General optional stopping theorem

I Let Xn be a uniformly integrable submartingale.

I Theorem: For any stopping time N, XN∧n is uniformlyintegrable.

I Theorem: If E |Xn| <∞ and Xn1(N>n) is uniformlyintegrable, then XN∧n is uniformly integrable.

I Theorem: For any stopping time N ≤ ∞, we haveEX0 ≤ EXN ≤ EX∞ where X∞ = limXn.

I Fairly general form of optional stopping theorem: IfL ≤ M are stopping times and YM∧n is a uniformly integrablesubmartingale, then EYL ≤ EYM and YL ≤ E (YM |FL).

18.175 Lecture 28

Page 49: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

General optional stopping theorem

I Let Xn be a uniformly integrable submartingale.

I Theorem: For any stopping time N, XN∧n is uniformlyintegrable.

I Theorem: If E |Xn| <∞ and Xn1(N>n) is uniformlyintegrable, then XN∧n is uniformly integrable.

I Theorem: For any stopping time N ≤ ∞, we haveEX0 ≤ EXN ≤ EX∞ where X∞ = limXn.

I Fairly general form of optional stopping theorem: IfL ≤ M are stopping times and YM∧n is a uniformly integrablesubmartingale, then EYL ≤ EYM and YL ≤ E (YM |FL).

18.175 Lecture 28

Page 50: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

General optional stopping theorem

I Let Xn be a uniformly integrable submartingale.

I Theorem: For any stopping time N, XN∧n is uniformlyintegrable.

I Theorem: If E |Xn| <∞ and Xn1(N>n) is uniformlyintegrable, then XN∧n is uniformly integrable.

I Theorem: For any stopping time N ≤ ∞, we haveEX0 ≤ EXN ≤ EX∞ where X∞ = limXn.

I Fairly general form of optional stopping theorem: IfL ≤ M are stopping times and YM∧n is a uniformly integrablesubmartingale, then EYL ≤ EYM and YL ≤ E (YM |FL).

18.175 Lecture 28

Page 51: 18.175: Lecture 28 .1in Even more on martingalesmath.mit.edu/~sheffield/175/Lecture28.pdfOutline Recollections More martingale theorems 18.175 Lecture 28 Outline Recollections More

General optional stopping theorem

I Let Xn be a uniformly integrable submartingale.

I Theorem: For any stopping time N, XN∧n is uniformlyintegrable.

I Theorem: If E |Xn| <∞ and Xn1(N>n) is uniformlyintegrable, then XN∧n is uniformly integrable.

I Theorem: For any stopping time N ≤ ∞, we haveEX0 ≤ EXN ≤ EX∞ where X∞ = limXn.

I Fairly general form of optional stopping theorem: IfL ≤ M are stopping times and YM∧n is a uniformly integrablesubmartingale, then EYL ≤ EYM and YL ≤ E (YM |FL).

18.175 Lecture 28