An Introduction to Extreme Value Analysis

Transcription

1 An Introduction to Extreme Value Analysis Graduate student seminar series Whitney Huang Department of Statistics Purdue University March 6, 2014 Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

2 Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

3 Outline 1 Motivation 2 Extreme Value Theorem 3 Example: Fort Collins Precipitation Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

4 Usual vs Extremes Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

5 Why study extremes? Although infrequent, extremes usually have large impact. Goal: to quantify the tail behavior often requires extrapolation. Applications: hydrology: flooding climate: temperature, precipitation, wind, finance insurance/reinsurance engineering: structural design, reliability Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

6 Outline 1 Motivation 2 Extreme Value Theorem 3 Example: Fort Collins Precipitation Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

7 Probability Framework Let X 1,, X n iid F and define Mn = max{x 1,, X n } Then the distribution function of M n is P(M n x) = P(X 1 x,, X n x) = P(X 1 x) P(X n x) = F n (x) Remark { F n (x) === n 0 if F (x) < 1 1 if F (x) = 1 the limiting distribution is degenerate. Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

8 Asymptotic: Classical Limit Laws Recall the Central Limit Theorem: S n nµ nσ d N(0, 1) rescaling is the key to obtain a non-degenerate distribution Question: Can we get the limiting distribution of M n b n a n for suitable sequence {a n } > 0 and {b n }? Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

9 Asymptotic: Classical Limit Laws Recall the Central Limit Theorem: S n nµ nσ d N(0, 1) rescaling is the key to obtain a non-degenerate distribution Question: Can we get the limiting distribution of M n b n a n for suitable sequence {a n } > 0 and {b n }? Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

10 Theorem (Fisher Tippett Gnedenko theorem) If there exist sequences of constants a n > 0 and b n such that, as n ( ) Mn b n d P x G(x) a n for some non-degenerate distribution G, then G belongs to either the Gumbel, the Fréchet or the Weibull family Gumbel: G(x) = { exp(exp( x)) < x < ; 0 x 0, Fréchet: G(x) = exp( x α ) x > 0, α > 0; { exp( ( x) Weibull: G(x) = α ) x < 0, α > 0, 1 x 0; Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

11 Generalized Extreme Value Distribution (GEV) This family encompasses all three extreme value limit families: where x + = max(x, 0) { [ G(x) = exp 1 + ξ( x µ σ ] 1 } ) ξ + µ and σ are location and scale parameters ξ is a shape parameter determining the rate of tail decay, with ξ > 0 giving the heavy-tailed (Fréchet) case ξ = 0 giving the light-tailed (Gumbel) case ξ < 0 giving the bounded-tailed (Weibull) case Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

12 Max-Stability and GEV Definition A distribution G is said to be max-stable if G k (a k x + b k ) = G(x), k N for some constants a k > 0 and b k Taking powers of a distribution function results only in a change of location and scale A distribution is max-stable it is a GEV distribution Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

13 Quantiles and Return Levels Quantiles of Extremes { [ G(x p ) = exp x p = µ σ ξ 1 + ξ( x p µ ) σ ] 1 ξ + } = 1 p [ 1 { log(1 p) ξ }] 0 < p < 1 In the extreme value terminology, x p is the return level associated with the return period 1 p Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

14 Block Maximum Approach 1 Determine the block size and compute maximal for blocks 2 Fit the GEV to the maximal and assess fit 3 Perform inference for return levels, probabilities, etc Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

15 Block Maximum Approach 1 Determine the block size and compute maximal for blocks 2 Fit the GEV to the maximal and assess fit 3 Perform inference for return levels, probabilities, etc Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

16 Block Maximum Approach 1 Determine the block size and compute maximal for blocks 2 Fit the GEV to the maximal and assess fit 3 Perform inference for return levels, probabilities, etc Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

17 Threshold Exceedance Approach Motivation: The block maximum method ignores much of the data which may also relevant to extreme we would like to use the data more efficient. Alternatives: Threshold Exceedance Approach 1 Select an appropriate threshold u, extract the exceedances from the dataset 2 Fit an appropriate model to exceedances 3 Perform inference for return levels, probabilities, etc Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

18 Threshold Exceedance Approach Motivation: The block maximum method ignores much of the data which may also relevant to extreme we would like to use the data more efficient. Alternatives: Threshold Exceedance Approach 1 Select an appropriate threshold u, extract the exceedances from the dataset 2 Fit an appropriate model to exceedances 3 Perform inference for return levels, probabilities, etc Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

19 Threshold Exceedance Approach Motivation: The block maximum method ignores much of the data which may also relevant to extreme we would like to use the data more efficient. Alternatives: Threshold Exceedance Approach 1 Select an appropriate threshold u, extract the exceedances from the dataset 2 Fit an appropriate model to exceedances 3 Perform inference for return levels, probabilities, etc Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

20 Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

21 GPD for Exceedances P(X i > x + u X i > u) = np(x i > x + u) np(x i > u) ( 1 + ξ x+u b n a n = ( ξ u bn a n ) 1 ξ ξx a n + ξ(u b n ) Survival function of generalized Pareto distribution ) 1 ξ Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

22 Theorem (Pickands Balkema de Haan theorem) Let X 1, iid F, and let F u be their conditional excess distribution function. Pickands (1975), Balkema and de Haan (1974) posed that for a large class of underlying distribution functions F, and large u, F u is well approximated by the generalized Pareto distribution GPD. That is: F u (y) GPD ξ,σ (y) u where GPD ξ,σ (y) = { 1 (1 + ξy σ ) 1 ξ ξ 0, 1 exp( y σ ) ξ = 0; Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

23 Threshold Selection Bias variance trade off: threshold too low bias because of the model asymptotics being invalid; threshold too high variance is large due to few data points Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

24 Outline 1 Motivation 2 Extreme Value Theorem 3 Example: Fort Collins Precipitation Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

25 Example: Fort Collins Precipitation Spike corresponds to 1997 event, recording station not at center of storm. Question: How unusual was event? Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

26 How unusual was the Fort Collins event? Measured value for 1997 event is 6.18 inches. Take a look at data preceding the event ( ) and try to estimate the return period associated with this event It is equivalent to ask What is the probability the annual maximum event is larger than 6.18 inches? Requires extrapolation into the tail. Largest observation ( ) is 4.09 inches inches. We will approach this problem in two ways: 1 Model all (non-zero) data 2 Model only extreme data Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

27 Modeling all precipitation data Let Y t be the daily summer (April October) precipitation for Fort Collins, CO. { Yt > 0 with probability p Assume ˆp = Y t = 0 with probability 1 p Model: Y t Y t > 0 Gamma(α, β) Maximum likelihood estimates: ˆα = 0.784, ˆβ = 3.52 Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

28 Modeling all precipitation data cont d P(Y t > 6.18) = P(Y t > 6.18 Y t > 0)P(Y t > 0) = (1 F gamma(ˆα, ˆβ) (6.18))(0.218) = ( )(0.218) = Let M be the annual maximum precipitation P(M > 6.18) = 1 P(M < 6.18) = 1 P(Y t < 6.18) 214 = 1 (1 P(Y t > 6.18)) 214 = 1 ( ) 214 = Return period estimate = ( ) 1 = 145, 815, 245 years Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

29 Modeling all precipitation data: Diagnostics Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

30 Modeling Annual Maximal Let M n = max X t. Assume M n GEV (µ, σ, ξ) t=1,,n P(M n x) = exp { [ 1 + ξ( x µ σ ) ] 1 ξ Maximum Likelihood estimates: ˆµ = 1.11, ˆσ = 0.46, ˆξ = P(ann max > 6.18) = 1 P(M n 6.18) = Return period estimate = 1 = 121 years } Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

31 Modeling Annual Maximal: Diagnostics Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

32 Modeling Annual Maximal: Diagnostics Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

33 Temporal Dependence Question: Is the GEV still the limiting distribution for block maxima of a stationary (but not independent) sequence {X i }? Answer: Yes, so long as mixing conditions hold. (Leadbetter et al., 1983) What does this mean for inference? Block maximum approach: GEV still correct for marginal. Since block maximum data likely have negligible dependence, proceed as usual Threshold exceedance approach: GPD is correct for the marginal. If extremes occur in clusters, estimation affected as likelihood assumes independence of threshold exceedances Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

34 Temporal Dependence Question: Is the GEV still the limiting distribution for block maxima of a stationary (but not independent) sequence {X i }? Answer: Yes, so long as mixing conditions hold. (Leadbetter et al., 1983) What does this mean for inference? Block maximum approach: GEV still correct for marginal. Since block maximum data likely have negligible dependence, proceed as usual Threshold exceedance approach: GPD is correct for the marginal. If extremes occur in clusters, estimation affected as likelihood assumes independence of threshold exceedances Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

35 Temporal Dependence Question: Is the GEV still the limiting distribution for block maxima of a stationary (but not independent) sequence {X i }? Answer: Yes, so long as mixing conditions hold. (Leadbetter et al., 1983) What does this mean for inference? Block maximum approach: GEV still correct for marginal. Since block maximum data likely have negligible dependence, proceed as usual Threshold exceedance approach: GPD is correct for the marginal. If extremes occur in clusters, estimation affected as likelihood assumes independence of threshold exceedances Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

36 Temporal Dependence Question: Is the GEV still the limiting distribution for block maxima of a stationary (but not independent) sequence {X i }? Answer: Yes, so long as mixing conditions hold. (Leadbetter et al., 1983) What does this mean for inference? Block maximum approach: GEV still correct for marginal. Since block maximum data likely have negligible dependence, proceed as usual Threshold exceedance approach: GPD is correct for the marginal. If extremes occur in clusters, estimation affected as likelihood assumes independence of threshold exceedances Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

37 Temporal Dependence Question: Is the GEV still the limiting distribution for block maxima of a stationary (but not independent) sequence {X i }? Answer: Yes, so long as mixing conditions hold. (Leadbetter et al., 1983) What does this mean for inference? Block maximum approach: GEV still correct for marginal. Since block maximum data likely have negligible dependence, proceed as usual Threshold exceedance approach: GPD is correct for the marginal. If extremes occur in clusters, estimation affected as likelihood assumes independence of threshold exceedances Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

38 Remarks on Univariate Extremes To estimate the tail, EVT uses only extreme observations Tail parameter ξ is extremely important but hard to estimate Threshold exceedance approaches allow the user to retain more data than block-maximum approaches, thereby reducing the uncertainty with parameter estimates Temporal dependence in the data is more of an issue in threshold exceedance models. One can either decluster, or alternatively, adjust inference Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

39 Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31

40 For Further Reading S. Coles An Introduction to Statistical Modeling of Extreme Values. Springer, J. Beirlant, Y Goegebeur, J. Segers, and J Teugels Statistics of Extremes: Theory and Applications. Wiley, L. de Haan, and A. Ferreira Extreme Value Theory: An Introduction. Springer, S. I. Resnick Heavy-Tail Phenomena: Probabilistic and Statistical Modeling. Springer, Whitney Huang (Purdue University) An Introduction to Extreme Value Analysis March 6, / 31