Optimal forecast reconciliation for big time series data

Size: px
Start display at page:

Download "Optimal forecast reconciliation for big time series data"

Transcription

1 Rob J Hyndman Optimal forecast reconciliation for big time series data Total A AA AB AC B BA BB BC C CA CB CC

2 Follow along using R Requirements Install the hts package and its dependencies. Optimal forecast reconciliation for big time series data 2

3 Outline 1 Hierarchical and grouped time series 2 BLUF: Best Linear Unbiased Forecasts 3 Application: Australian tourism 4 Temporal hierarchies 5 References Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 3

4 Labour market participation Australia and New Zealand Standard Classification of Occupations 8 major groups 43 sub-major groups 97 minor groups 359 unit groups * 1023 occupations Example: statistician 2 Professionals 22 Business, Human Resource and Marketing Professionals 224 Information and Organisation Professionals 2241 Actuaries, Mathematicians and Statisticians Statistician Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 4

5 Labour market participation Australia and New Zealand Standard Classification of Occupations 8 major groups 43 sub-major groups 97 minor groups 359 unit groups * 1023 occupations Example: statistician 2 Professionals 22 Business, Human Resource and Marketing Professionals 224 Information and Organisation Professionals 2241 Actuaries, Mathematicians and Statisticians Statistician Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 4

6 Australian tourism demand Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 5

7 Australian tourism demand Quarterly data on visitor night from 1998:Q1 2013:Q4 From: National Visitor Survey, based on annual interviews of 120,000 Australians aged 15+, collected by Tourism Research Australia. Split by 7 states, 27 zones and 76 regions (a geographical hierarchy) Also split by purpose of travel Holiday Visiting friends and relatives (VFR) Business Other 304 bottom-level series Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 5

8 Hierarchical time series A hierarchical time series is a collection of several time series that are linked together in a hierarchical structure. Total A AA AB AC B BA BB BC C CA CB CC Examples Labour turnover by occupation Tourism by state and region Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 6

9 Hierarchical time series A hierarchical time series is a collection of several time series that are linked together in a hierarchical structure. Total A AA AB AC B BA BB BC C CA CB CC Examples Labour turnover by occupation Tourism by state and region Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 6

10 Hierarchical time series A hierarchical time series is a collection of several time series that are linked together in a hierarchical structure. Total A AA AB AC B BA BB BC C CA CB CC Examples Labour turnover by occupation Tourism by state and region Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 6

11 Grouped time series A grouped time series is a collection of time series that can be grouped together in a number of non-hierarchical ways. Total Total A B X Y AX AY BX BY AX BX AY BY Examples Labour turnover by occupation and state Tourism by state and purpose of travel Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 7

12 Grouped time series A grouped time series is a collection of time series that can be grouped together in a number of non-hierarchical ways. Total Total A B X Y AX AY BX BY AX BX AY BY Examples Labour turnover by occupation and state Tourism by state and purpose of travel Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 7

13 Grouped time series A grouped time series is a collection of time series that can be grouped together in a number of non-hierarchical ways. Total Total A B X Y AX AY BX BY AX BX AY BY Examples Labour turnover by occupation and state Tourism by state and purpose of travel Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 7

14 Hierarchical time series Total A B C y t : observed aggregate of all series at time t. y X,t : observation on series X at time t. b t : vector of all series at bottom level in time t. Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 8

15 Hierarchical time series Total A B C y t : observed aggregate of all series at time t. y X,t : observation on series X at time t. b t : vector of all series at bottom level in time t. Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 8

16 Hierarchical time series Total A B C y t = [y t, y A,t, y B,t, y C,t ] = y t : observed aggregate of all series at time t. y X,t : observation on series X at time t. b t : vector of all series at bottom level in time t y A,t y B,t y C,t Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 8

17 Hierarchical time series Total A B C y t = [y t, y A,t, y B,t, y C,t ] = y t : observed aggregate of all series at time t. y X,t : observation on series X at time t. b t : vector of all series at bottom level in time t y A,t y B,t y C,t }{{} S Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 8

18 Hierarchical time series Total A B C y t = [y t, y A,t, y B,t, y C,t ] = y t : observed aggregate of all series at time t. y X,t : observation on series X at time t. b t : vector of all series at bottom level in time t y A,t y B,t y C,t }{{}}{{} b t S Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 8

19 Hierarchical time series Total A B C y t = [y t, y A,t, y B,t, y C,t ] = yt = Sb t y t : observed aggregate of all series at time t. y X,t : observation on series X at time t. b t : vector of all series at bottom level in time t y A,t y B,t y C,t }{{}}{{} b t S Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 8

20 Hierarchical time series Total A B C y t = AX AY AZ y t y A,t y B,t y C,t y AX,t y AY,t y AZ,t y BX,t y BY,t y BZ,t y CX,t y CY,t y CZ,t = BX BY BZ } {{ } S y AX,t y AY,t y AZ,t y BX,t y BY,t y BZ,t y CX,t y CY,t y CZ,t }{{} b t CX CY CZ Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 9

21 Hierarchical time series Total A B C y t = AX AY AZ y t y A,t y B,t y C,t y AX,t y AY,t y AZ,t y BX,t y BY,t y BZ,t y CX,t y CY,t y CZ,t = BX BY BZ } {{ } S y AX,t y AY,t y AZ,t y BX,t y BY,t y BZ,t y CX,t y CY,t y CZ,t }{{} b t CX CY CZ Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 9

22 Hierarchical time series Total A B C y t = AX AY AZ y t y A,t y B,t y C,t y AX,t y AY,t y AZ,t y BX,t y BY,t y BZ,t y CX,t y CY,t y CZ,t = BX BY BZ } {{ } S y AX,t y AY,t y AZ,t y BX,t y BY,t y BZ,t y CX,t y CY,t y CZ,t }{{} b t CX CY CZ y t = Sb t Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 9

23 Grouped data AX AY A BX BY B y t = y t y A,t y B,t y X,t y Y,t y AX,t y AY,t y BX,t y BY,t = X Y Total }{{} S y AX,t y AY,t y BX,t y BY,t } {{ } b t Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 10

24 Grouped data AX AY A BX BY B y t = y t y A,t y B,t y X,t y Y,t y AX,t y AY,t y BX,t y BY,t = X Y Total }{{} S y AX,t y AY,t y BX,t y BY,t } {{ } b t Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 10

25 Grouped data AX AY A BX BY B y t = y t y A,t y B,t y X,t y Y,t y AX,t y AY,t y BX,t y BY,t = X Y Total }{{} S y AX,t y AY,t y BX,t y BY,t } {{ } b t y t = Sb t Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 10

26 Hierarchical and grouped time series Every collection of time series with aggregation constraints can be written as where y t = Sb t y t is a vector of all series at time t b t is a vector of the most disaggregated series at time t S is a summing matrix containing the aggregation constraints. Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 11

27 hts package for R hts: Hierarchical and grouped time series Methods for analysing and forecasting hierarchical and grouped time series Version: 4.5 Depends: forecast ( 5.0), SparseM Imports: parallel, utils Published: Author: Rob J Hyndman, Earo Wang and Alan Lee with contributions from Shanika Wickramasuriya Maintainer: Rob J Hyndman <Rob.Hyndman at monash.edu> BugReports: License: GPL ( 2) Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 12

28 Example using R library(hts) # bts is a matrix containing the bottom level time series # nodes describes the hierarchical structure y <- hts(bts, nodes=list(2, c(3,2))) Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 13

29 Example using R library(hts) # bts is a matrix containing the bottom level time series # nodes describes the hierarchical structure y <- hts(bts, nodes=list(2, c(3,2))) Total A B AX AY AZ BX BY Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 13

30 Example using R library(hts) # bts is a matrix containing the bottom level time series # nodes describes the hierarchical structure y <- hts(bts, nodes=list(2, c(3,2))) Total A B AX AY AZ BX BY summary(y) smatrix(y) Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 14

31 Example using R Example 1 illustrating the usage of the "groups" argument abc <- ts(5 + matrix(sort(rnorm(1600)), ncol = 16, nrow = 100)) sex <- rep(c("female", "male"), each = 8) state <- rep(c("nsw","vic","qld","sa","wa","nt","act","tas"), 2) gc <- rbind(sex, state) # a matrix consists of strings. gn <- rbind(rep(1:2, each = 8), rep(1:8, 2)) rownames(gc) <- rownames(gn) <- c("sex", "State") x <- gts(abc, groups = gc) y <- gts(abc, groups = gn) plot(x,level=3) plot(x,level=2) plot(x,level=1) plot(x,level=0) Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 15

32 Example using R Example 2 with two simple hierarchies (geography and product) to show the argument "characters" bnames1 <- c("vicmelba1", "VICMelbA2", "VICGeelA1", "VICGeelA2", "VICMelbB1", "VICMelbB2", "VICGeelB1", "VICGeelB2", "NSWSyndA1", "NSWSyndA2", "NSWWollA1", "NSWWollA2", "NSWSyndB1", "NSWSyndB2", "NSWWollB1", "NSWWollB2") bts1 <- matrix(ts(rnorm(160)), ncol = 16) colnames(bts1) <- bnames1 x1 <- gts(bts1, characters = list(c(3, 4), c(1, 1)), gnames = c("state","state.city","product","product.size", "State.Product","State.Product.Size","State.City.Product")) plot(x1, level=1) plot(x1, level=7) Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 16

33 Example using R Example 3 with a non-hierarchical grouped time series of 3 grouping variables (state, age and sex) bnames2 <- c("vic1f","vic1m","vic2f","vic2m","vic3f","vic3m", "NSW1F","NSW1M","NSW2F","NSW2M","NSW3F","NSW3M") bts2 <- matrix(ts(rnorm(120)), ncol = 12) colnames(bts2) <- bnames2 x2 <- gts(bts2, characters = c(3, 1, 1), gnames=c("state","age","sex", "State.Age","State.Sex","Age.Sex")) plot(x2, level=3) Optimal forecast reconciliation for big time series data Hierarchical and grouped time series 17

34 Outline 1 Hierarchical and grouped time series 2 BLUF: Best Linear Unbiased Forecasts 3 Application: Australian tourism 4 Temporal hierarchies 5 References Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 18

35 Forecasting notation Let ŷ n (h) be vector of initial h-step forecasts, made at time n, stacked in same order as y t. (They may not add up.) Reconciled forecasts must be of the form: ỹ n (h) = SPŷ n (h) for some matrix P. P extracts and combines base forecasts ŷ n (h) to get bottom-level forecasts. S adds them up Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 19

36 Forecasting notation Let ŷ n (h) be vector of initial h-step forecasts, made at time n, stacked in same order as y t. (They may not add up.) Reconciled forecasts must be of the form: ỹ n (h) = SPŷ n (h) for some matrix P. P extracts and combines base forecasts ŷ n (h) to get bottom-level forecasts. S adds them up Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 19

37 Forecasting notation Let ŷ n (h) be vector of initial h-step forecasts, made at time n, stacked in same order as y t. (They may not add up.) Reconciled forecasts must be of the form: ỹ n (h) = SPŷ n (h) for some matrix P. P extracts and combines base forecasts ŷ n (h) to get bottom-level forecasts. S adds them up Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 19

38 Forecasting notation Let ŷ n (h) be vector of initial h-step forecasts, made at time n, stacked in same order as y t. (They may not add up.) Reconciled forecasts must be of the form: ỹ n (h) = SPŷ n (h) for some matrix P. P extracts and combines base forecasts ŷ n (h) to get bottom-level forecasts. S adds them up Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 19

39 Forecasting notation Let ŷ n (h) be vector of initial h-step forecasts, made at time n, stacked in same order as y t. (They may not add up.) Reconciled forecasts must be of the form: ỹ n (h) = SPŷ n (h) for some matrix P. P extracts and combines base forecasts ŷ n (h) to get bottom-level forecasts. S adds them up Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 19

40 General properties: bias ỹ n (h) = SPŷ n (h) Assume: base forecasts ŷ n (h) are unbiased: E[ŷ n (h) y 1,..., y n ] = E[y n+h y 1,..., y n ] Let ˆbn (h) be bottom level base forecasts with β n (h) = E[ˆbn (h) y 1,..., y n ]. Then E[ŷ n (h)] = Sβ n (h). We want the revised forecasts to be unbiased: E[ỹ n (h)] = SPSβ n (h) = Sβ n (h). Revised forecasts are unbiased iff SPS = S. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 20

41 General properties: bias ỹ n (h) = SPŷ n (h) Assume: base forecasts ŷ n (h) are unbiased: E[ŷ n (h) y 1,..., y n ] = E[y n+h y 1,..., y n ] Let ˆbn (h) be bottom level base forecasts with β n (h) = E[ˆbn (h) y 1,..., y n ]. Then E[ŷ n (h)] = Sβ n (h). We want the revised forecasts to be unbiased: E[ỹ n (h)] = SPSβ n (h) = Sβ n (h). Revised forecasts are unbiased iff SPS = S. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 20

42 General properties: bias ỹ n (h) = SPŷ n (h) Assume: base forecasts ŷ n (h) are unbiased: E[ŷ n (h) y 1,..., y n ] = E[y n+h y 1,..., y n ] Let ˆbn (h) be bottom level base forecasts with β n (h) = E[ˆbn (h) y 1,..., y n ]. Then E[ŷ n (h)] = Sβ n (h). We want the revised forecasts to be unbiased: E[ỹ n (h)] = SPSβ n (h) = Sβ n (h). Revised forecasts are unbiased iff SPS = S. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 20

43 General properties: bias ỹ n (h) = SPŷ n (h) Assume: base forecasts ŷ n (h) are unbiased: E[ŷ n (h) y 1,..., y n ] = E[y n+h y 1,..., y n ] Let ˆbn (h) be bottom level base forecasts with β n (h) = E[ˆbn (h) y 1,..., y n ]. Then E[ŷ n (h)] = Sβ n (h). We want the revised forecasts to be unbiased: E[ỹ n (h)] = SPSβ n (h) = Sβ n (h). Revised forecasts are unbiased iff SPS = S. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 20

44 General properties: bias ỹ n (h) = SPŷ n (h) Assume: base forecasts ŷ n (h) are unbiased: E[ŷ n (h) y 1,..., y n ] = E[y n+h y 1,..., y n ] Let ˆbn (h) be bottom level base forecasts with β n (h) = E[ˆbn (h) y 1,..., y n ]. Then E[ŷ n (h)] = Sβ n (h). We want the revised forecasts to be unbiased: E[ỹ n (h)] = SPSβ n (h) = Sβ n (h). Revised forecasts are unbiased iff SPS = S. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 20

45 General properties: bias ỹ n (h) = SPŷ n (h) Assume: base forecasts ŷ n (h) are unbiased: E[ŷ n (h) y 1,..., y n ] = E[y n+h y 1,..., y n ] Let ˆbn (h) be bottom level base forecasts with β n (h) = E[ˆbn (h) y 1,..., y n ]. Then E[ŷ n (h)] = Sβ n (h). We want the revised forecasts to be unbiased: E[ỹ n (h)] = SPSβ n (h) = Sβ n (h). Revised forecasts are unbiased iff SPS = S. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 20

46 General properties: bias ỹ n (h) = SPŷ n (h) Assume: base forecasts ŷ n (h) are unbiased: E[ŷ n (h) y 1,..., y n ] = E[y n+h y 1,..., y n ] Let ˆbn (h) be bottom level base forecasts with β n (h) = E[ˆbn (h) y 1,..., y n ]. Then E[ŷ n (h)] = Sβ n (h). We want the revised forecasts to be unbiased: E[ỹ n (h)] = SPSβ n (h) = Sβ n (h). Revised forecasts are unbiased iff SPS = S. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 20

47 General properties: variance ỹ n (h) = SPŷ n (h) Let error variance of h-step base forecasts ŷ n (h) be W h = Var[y n+h ŷ n (h) y 1,..., y n ] Then the error variance of the corresponding revised forecasts is Var[y n+h ỹ n (h) y 1,..., y n ] = SPW h P S Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 21

48 General properties: variance ỹ n (h) = SPŷ n (h) Let error variance of h-step base forecasts ŷ n (h) be W h = Var[y n+h ŷ n (h) y 1,..., y n ] Then the error variance of the corresponding revised forecasts is Var[y n+h ỹ n (h) y 1,..., y n ] = SPW h P S Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 21

49 BLUF via trace minimization Theorem For any P satisfying SPS = S, then min P = trace[spw h P S ] has solution P = (S W h S) 1 S W h. W h is generalized inverse of W h. Problem: W h hard to estimate, especially for h > 1. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 22

50 BLUF via trace minimization Theorem For any P satisfying SPS = S, then min P = trace[spw h P S ] has solution P = (S W h S) 1 S W h. W h is generalized inverse of W h. Problem: W h hard to estimate, especially for h > 1. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 22

51 BLUF via trace minimization Theorem For any P satisfying SPS = S, then min P = trace[spw h P S ] has solution P = (S W h S) 1 S W h. W h is generalized inverse of W h. Problem: W h hard to estimate, especially for h > 1. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 22

52 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 1: OLS Assume forecast errors have the same aggregation constraints as the data. Then W h = SΩ h S where Ω h is covariance matrix of bottom level errors. If Moore-Penrose generalized inverse used, then (S W h S) 1 S W h = (S S) 1 S. ỹ n (h) = S(S S) 1 S ŷ n (h) Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 23

53 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 1: OLS Assume forecast errors have the same aggregation constraints as the data. Then W h = SΩ h S where Ω h is covariance matrix of bottom level errors. If Moore-Penrose generalized inverse used, then (S W h S) 1 S W h = (S S) 1 S. ỹ n (h) = S(S S) 1 S ŷ n (h) Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 23

54 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 1: OLS Assume forecast errors have the same aggregation constraints as the data. Then W h = SΩ h S where Ω h is covariance matrix of bottom level errors. If Moore-Penrose generalized inverse used, then (S W h S) 1 S W h = (S S) 1 S. ỹ n (h) = S(S S) 1 S ŷ n (h) Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 23

55 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 1: OLS Assume forecast errors have the same aggregation constraints as the data. Then W h = SΩ h S where Ω h is covariance matrix of bottom level errors. If Moore-Penrose generalized inverse used, then (S W h S) 1 S W h = (S S) 1 S. ỹ n (h) = S(S S) 1 S ŷ n (h) Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 23

56 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 1: OLS Assume forecast errors have the same aggregation constraints as the data. Then W h = SΩ h S where Ω h is covariance matrix of bottom level errors. If Moore-Penrose generalized inverse used, then (S W h S) 1 S W h = (S S) 1 S. ỹ n (h) = S(S S) 1 S ŷ n (h) Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 23

57 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 1: OLS Assume forecast errors have the same aggregation constraints as the data. Then W h = SΩ h S where Ω h is covariance matrix of bottom level errors. If Moore-Penrose generalized inverse used, then (S W h S) 1 S W h = (S S) 1 S. ỹ n (h) = S(S S) 1 S ŷ n (h) Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 23

58 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 2: WLS Suppose we approximate W 1 by its diagonal and assume that W h W 1. Easy to estimate, and places weight where we have best forecasts. Empirically, WLS gives better forecasts than OLS. Still need to estimate covariance matrix to produce prediction intervals. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 24

59 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 2: WLS Suppose we approximate W 1 by its diagonal and assume that W h W 1. Easy to estimate, and places weight where we have best forecasts. Empirically, WLS gives better forecasts than OLS. Still need to estimate covariance matrix to produce prediction intervals. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 24

60 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 2: WLS Suppose we approximate W 1 by its diagonal and assume that W h W 1. Easy to estimate, and places weight where we have best forecasts. Empirically, WLS gives better forecasts than OLS. Still need to estimate covariance matrix to produce prediction intervals. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 24

61 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 2: WLS Suppose we approximate W 1 by its diagonal and assume that W h W 1. Easy to estimate, and places weight where we have best forecasts. Empirically, WLS gives better forecasts than OLS. Still need to estimate covariance matrix to produce prediction intervals. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 24

62 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 2: WLS Suppose we approximate W 1 by its diagonal and assume that W h W 1. Easy to estimate, and places weight where we have best forecasts. Empirically, WLS gives better forecasts than OLS. Still need to estimate covariance matrix to produce prediction intervals. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 24

63 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 3: GLS Estimate W 1 using shrinkage to the diagonal and assume that W h W 1. Allows for covariances. Empirically, GLS gives better forecasts than WLS or OLS. Difficult to compute for large numbers of time series. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 25

64 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 3: GLS Estimate W 1 using shrinkage to the diagonal and assume that W h W 1. Allows for covariances. Empirically, GLS gives better forecasts than WLS or OLS. Difficult to compute for large numbers of time series. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 25

65 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 3: GLS Estimate W 1 using shrinkage to the diagonal and assume that W h W 1. Allows for covariances. Empirically, GLS gives better forecasts than WLS or OLS. Difficult to compute for large numbers of time series. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 25

66 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 3: GLS Estimate W 1 using shrinkage to the diagonal and assume that W h W 1. Allows for covariances. Empirically, GLS gives better forecasts than WLS or OLS. Difficult to compute for large numbers of time series. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 25

67 Optimal combination forecasts ỹ n (h) = S(S W h S) 1 S W hŷn(h) Revised forecasts Base forecasts Solution 3: GLS Estimate W 1 using shrinkage to the diagonal and assume that W h W 1. Allows for covariances. Empirically, GLS gives better forecasts than WLS or OLS. Difficult to compute for large numbers of time series. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 25

68 Example using R library(hts) # bts is a matrix containing the bottom level time series # nodes describes the hierarchical structure y <- hts(bts, nodes=list(2, c(3,2))) # Forecast 10-step-ahead using WLS combination method # ETS used for each series by default fc <- forecast(y, h=10) # Select your own methods ally <- aggts(y) allf <- matrix(, nrow=10, ncol=ncol(ally)) for(i in 1:ncol(ally)) allf[,i] <- mymethod(ally[,i], h=10) allf <- ts(allf, start=2004) # Reconcile forecasts so they add up fc2 <- combinef(allf, nodes=y$nodes) Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 26

69 forecast.gts function Usage forecast(object, h, method = c("comb", "bu", "mo", "tdgsf", "tdgsa", "tdfp"), fmethod = c("ets", "rw", "arima"), weights = c("sd", "none", "nseries"), positive = FALSE, parallel = FALSE, num.cores = 2,...) Arguments object Hierarchical time series object of class gts. h Forecast horizon method Method for distributing forecasts within the hierarchy. fmethod Forecasting method to use positive If TRUE, forecasts are forced to be strictly positive weights Weights used for "optimal combination" method. When weights = "sd", it takes account of the standard deviation of forecasts. parallel If TRUE, allow parallel processing num.cores If parallel = TRUE, specify how many cores are going to be used Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 27

70 Exercise Use the infantgts data to: 1 Plot various levels of aggregation. 2 Forecast the series using auto.arima with the default WLS reconciliation method. 3 Plot the reconciled forecasts at various levels of aggregation. Optimal forecast reconciliation for big time series data BLUF: Best Linear Unbiased Forecasts 28

71 Outline 1 Hierarchical and grouped time series 2 BLUF: Best Linear Unbiased Forecasts 3 Application: Australian tourism 4 Temporal hierarchies 5 References Optimal forecast reconciliation for big time series data Application: Australian tourism 29

72 Australian tourism Optimal forecast reconciliation for big time series data Application: Australian tourism 30

73 Australian tourism Hierarchy: States (7) Zones (27) Regions (82) Optimal forecast reconciliation for big time series data Application: Australian tourism 30

74 Australian tourism Hierarchy: States (7) Zones (27) Regions (82) Base forecasts ETS (exponential smoothing) models Optimal forecast reconciliation for big time series data Application: Australian tourism 30

75 Base forecasts Domestic tourism forecasts: Total Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

76 Base forecasts Domestic tourism forecasts: NSW Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

77 Base forecasts Domestic tourism forecasts: VIC Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

78 Base forecasts Domestic tourism forecasts: Nth.Coast.NSW Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

79 Base forecasts Domestic tourism forecasts: Metro.QLD Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

80 Base forecasts Domestic tourism forecasts: Sth.WA Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

81 Base forecasts Domestic tourism forecasts: X201.Melbourne Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

82 Base forecasts Domestic tourism forecasts: X402.Murraylands Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

83 Base forecasts Domestic tourism forecasts: X809.Daly Visitor nights Year Optimal forecast reconciliation for big time series data Application: Australian tourism 31

84 Forecast evaluation Training sets time Optimal forecast reconciliation for big time series data Application: Australian tourism 32

85 Forecast evaluation Training sets Test set h = 1 time Optimal forecast reconciliation for big time series data Application: Australian tourism 32

86 Forecast evaluation Training sets Test set h = 2 time Optimal forecast reconciliation for big time series data Application: Australian tourism 32

87 Forecast evaluation Training sets Test set h = 3 time Optimal forecast reconciliation for big time series data Application: Australian tourism 32

88 Forecast evaluation Training sets Test set h = 4 time Optimal forecast reconciliation for big time series data Application: Australian tourism 32

89 Forecast evaluation Training sets Test set h = 5 time Optimal forecast reconciliation for big time series data Application: Australian tourism 32

90 Forecast evaluation Training sets Test set h = 6 time Optimal forecast reconciliation for big time series data Application: Australian tourism 32

91 Hierarchy: states, zones, regions Forecast horizon RMSE h = 1 h = 2 h = 3 h = 4 h = 5 h = 6 Ave Australia Base Bottom OLS WLS GLS States Base Bottom OLS WLS GLS Regions Base Bottom OLS WLS GLS Optimal forecast reconciliation for big time series data Application: Australian tourism 33

92 Outline 1 Hierarchical and grouped time series 2 BLUF: Best Linear Unbiased Forecasts 3 Application: Australian tourism 4 Temporal hierarchies 5 References Optimal forecast reconciliation for big time series data Temporal hierarchies 34

93 Temporal hierarchies Annual Semi-Annual 1 Semi-Annual 2 Q 1 Q 2 Q 3 Q 4 Basic idea: Forecast series at each available frequency. Optimally combine forecasts within the same year. Optimal forecast reconciliation for big time series data Temporal hierarchies 35

94 Temporal hierarchies Annual Semi-Annual 1 Semi-Annual 2 Q 1 Q 2 Q 3 Q 4 Basic idea: Forecast series at each available frequency. Optimally combine forecasts within the same year. Optimal forecast reconciliation for big time series data Temporal hierarchies 35

95 Monthly series Annual Semi-Annual 1 Semi-Annual 2 Q 1 Q 2 Q 3 Q 4 M 1 M 2 M 3 M 4 M 5 M 6 M 7 M 8 M 9 M 10 M 11 M 12 k = 2, 4, 12 nodes k = 3, 6, 12 nodes Why not k = 2, 3, 4, 6, 12 nodes? Optimal forecast reconciliation for big time series data Temporal hierarchies 36

96 Monthly series Annual FourM 1 FourM 2 FourM 3 BiM 1 BiM 2 BiM 3 BiM 4 BiM 5 BiM 6 M 1 M 2 M 3 M 4 M 5 M 6 M 7 M 8 M 9 M 10 M 11 M 12 k = 2, 4, 12 nodes k = 3, 6, 12 nodes Why not k = 2, 3, 4, 6, 12 nodes? Optimal forecast reconciliation for big time series data Temporal hierarchies 36

97 Monthly series Annual FourM 1 FourM 2 FourM 3 BiM 1 BiM 2 BiM 3 BiM 4 BiM 5 BiM 6 M 1 M 2 M 3 M 4 M 5 M 6 M 7 M 8 M 9 M 10 M 11 M 12 k = 2, 4, 12 nodes k = 3, 6, 12 nodes Why not k = 2, 3, 4, 6, 12 nodes? Optimal forecast reconciliation for big time series data Temporal hierarchies 36

98 Monthly data A SemiA 1 SemiA 2 FourM 1 FourM 2 FourM 3 Q 1. Q 4 BiM 1. BiM 6 M 1. M 12 }{{} (28 1) = I 12 } {{ } S M 1 M 2 M 3 M 4 M 5 M 6 M 7 M 8 M 9 M 10 M 11 M 12 }{{} Optimal forecast reconciliation for big time series data Temporal hierarchies 37 B t

99 In general For a time series y 1,..., y T, observed at frequency m, we generate aggregate series y [k] j = jk t=1+(j 1)k y t, for j = 1,..., T/k k F(m) = {factors of m}. A single unique hierarchy is only possible when there are no coprime pairs in F(m). M k = m/k is seasonal period of aggregated series. Optimal forecast reconciliation for big time series data Temporal hierarchies 38

100 In general For a time series y 1,..., y T, observed at frequency m, we generate aggregate series y [k] j = jk t=1+(j 1)k y t, for j = 1,..., T/k k F(m) = {factors of m}. A single unique hierarchy is only possible when there are no coprime pairs in F(m). M k = m/k is seasonal period of aggregated series. Optimal forecast reconciliation for big time series data Temporal hierarchies 38

101 In general For a time series y 1,..., y T, observed at frequency m, we generate aggregate series y [k] j = jk t=1+(j 1)k y t, for j = 1,..., T/k k F(m) = {factors of m}. A single unique hierarchy is only possible when there are no coprime pairs in F(m). M k = m/k is seasonal period of aggregated series. Optimal forecast reconciliation for big time series data Temporal hierarchies 38

102 WLS weights Hierarchy variance scaling Λ H : diagonal. Series variance scaling Λ V : elements equal within aggregation level. Structural scaling Λ S = diag(s1): elements equal to # nodes at each level. Depends only on seasonal period m. Independent of data and model. Allows forecasts where no errors available. Quarterly example Λ H = (ˆσ ) diag A, 2 ˆσ 2 S 1, ˆσ 2 S 2, ˆσ 2 Q 1, ˆσ 2 Q 2, ˆσ 2 Q 3, ˆσ 2 Q 4 Λ V = (ˆσ ) diag A, 2 ˆσ S, 2 ˆσ S, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ 2 Q Λ S = ( diag 4, 2, 2, 1, 1, 1, ) 1 Optimal forecast reconciliation for big time series data Temporal hierarchies 39

103 WLS weights Hierarchy variance scaling Λ H : diagonal. Series variance scaling Λ V : elements equal within aggregation level. Structural scaling Λ S = diag(s1): elements equal to # nodes at each level. Depends only on seasonal period m. Independent of data and model. Allows forecasts where no errors available. Quarterly example Λ H = (ˆσ ) diag A, 2 ˆσ 2 S 1, ˆσ 2 S 2, ˆσ 2 Q 1, ˆσ 2 Q 2, ˆσ 2 Q 3, ˆσ 2 Q 4 Λ V = (ˆσ ) diag A, 2 ˆσ S, 2 ˆσ S, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ 2 Q Λ S = ( diag 4, 2, 2, 1, 1, 1, ) 1 Optimal forecast reconciliation for big time series data Temporal hierarchies 39

104 WLS weights Hierarchy variance scaling Λ H : diagonal. Series variance scaling Λ V : elements equal within aggregation level. Structural scaling Λ S = diag(s1): elements equal to # nodes at each level. Depends only on seasonal period m. Independent of data and model. Allows forecasts where no errors available. Quarterly example Λ H = (ˆσ ) diag A, 2 ˆσ 2 S 1, ˆσ 2 S 2, ˆσ 2 Q 1, ˆσ 2 Q 2, ˆσ 2 Q 3, ˆσ 2 Q 4 Λ V = (ˆσ ) diag A, 2 ˆσ S, 2 ˆσ S, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ 2 Q Λ S = ( diag 4, 2, 2, 1, 1, 1, ) 1 Optimal forecast reconciliation for big time series data Temporal hierarchies 39

105 WLS weights Hierarchy variance scaling Λ H : diagonal. Series variance scaling Λ V : elements equal within aggregation level. Structural scaling Λ S = diag(s1): elements equal to # nodes at each level. Depends only on seasonal period m. Independent of data and model. Allows forecasts where no errors available. Quarterly example Λ H = (ˆσ ) diag A, 2 ˆσ 2 S 1, ˆσ 2 S 2, ˆσ 2 Q 1, ˆσ 2 Q 2, ˆσ 2 Q 3, ˆσ 2 Q 4 Λ V = (ˆσ ) diag A, 2 ˆσ S, 2 ˆσ S, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ 2 Q Λ S = ( diag 4, 2, 2, 1, 1, 1, ) 1 Optimal forecast reconciliation for big time series data Temporal hierarchies 39

106 WLS weights Hierarchy variance scaling Λ H : diagonal. Series variance scaling Λ V : elements equal within aggregation level. Structural scaling Λ S = diag(s1): elements equal to # nodes at each level. Depends only on seasonal period m. Independent of data and model. Allows forecasts where no errors available. Quarterly example Λ H = (ˆσ ) diag A, 2 ˆσ 2 S 1, ˆσ 2 S 2, ˆσ 2 Q 1, ˆσ 2 Q 2, ˆσ 2 Q 3, ˆσ 2 Q 4 Λ V = (ˆσ ) diag A, 2 ˆσ S, 2 ˆσ S, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ Q, 2 ˆσ 2 Q Λ S = ( diag 4, 2, 2, 1, 1, 1, ) 1 Optimal forecast reconciliation for big time series data Temporal hierarchies 39

107 Experimental setup: M3 forecasting competition (Makridakis and Hibon, 2000, IJF). In total 3003 series. 1,428 monthly series with a test sample of 12 observations each. 756 quarterly series with a test sample of 8 observations each. Forecast each series with ETS models. Optimal forecast reconciliation for big time series data Temporal hierarchies 40

108 Experimental setup: M3 forecasting competition (Makridakis and Hibon, 2000, IJF). In total 3003 series. 1,428 monthly series with a test sample of 12 observations each. 756 quarterly series with a test sample of 8 observations each. Forecast each series with ETS models. Optimal forecast reconciliation for big time series data Temporal hierarchies 40

109 Experimental setup: M3 forecasting competition (Makridakis and Hibon, 2000, IJF). In total 3003 series. 1,428 monthly series with a test sample of 12 observations each. 756 quarterly series with a test sample of 8 observations each. Forecast each series with ETS models. Optimal forecast reconciliation for big time series data Temporal hierarchies 40

110 Experimental setup: M3 forecasting competition (Makridakis and Hibon, 2000, IJF). In total 3003 series. 1,428 monthly series with a test sample of 12 observations each. 756 quarterly series with a test sample of 8 observations each. Forecast each series with ETS models. Optimal forecast reconciliation for big time series data Temporal hierarchies 40

111 Results: Monthly MAE percent difference relative to base max h BU WLS H WLS V WLS S Annual Semi-annual Four-monthly Quarterly Bi-monthly Monthly Optimal forecast reconciliation for big time series data Temporal hierarchies 41

112 Results: Quarterly MAE percent difference relative to base max h BU WLS H WLS V WLS S Annual Semi-annual Quarterly Optimal forecast reconciliation for big time series data Temporal hierarchies 42

113 UK Accidents and Emergency Demand Annual (k=52) Forecast Semi annual (k=26) Forecast Quarterly (k=13) Forecast Monthly (k=4) Bi weekly (k=2) Weekly (k=1) Forecast Forecast base reconciled Forecast Optimal forecast reconciliation for big time series data Temporal hierarchies 43

114 UK Accidents and Emergency Demand 1 Type 1 Departments Major A&E 2 Type 2 Departments Single Specialty 3 Type 3 Departments Other A&E/Minor Injury 4 Total Attendances 5 Type 1 Departments Major A&E > 4 hrs 6 Type 2 Departments Single Specialty > 4 hrs 7 Type 3 Departments Other A&E/Minor Injury > 4 hrs 8 Total Attendances > 4 hrs 9 Emergency Admissions via Type 1 A&E 10 Total Emergency Admissions via A&E 11 Other Emergency Admissions (i.e., not via A&E) 12 Total Emergency Admissions 13 Number of patients spending > 4 hrs from decision to admission Optimal forecast reconciliation for big time series data Temporal hierarchies 44

115 UK Accidents and Emergency Demand Minimum training set: all data except the last year Base forecasts using auto.arima(). Reconciled using WLS V. Mean Absolute Scaled Errors for 1, 4 and 13 weeks ahead using a rolling origin. Aggr. Level h Base Reconciled Change Weekly % Weekly % Weekly % Weekly % Annual % Optimal forecast reconciliation for big time series data Temporal hierarchies 45

116 UK Accidents and Emergency Demand Minimum training set: all data except the last year Base forecasts using auto.arima(). Reconciled using WLS V. Mean Absolute Scaled Errors for 1, 4 and 13 weeks ahead using a rolling origin. Aggr. Level h Base Reconciled Change Weekly % Weekly % Weekly % Weekly % Annual % Optimal forecast reconciliation for big time series data Temporal hierarchies 45

117 UK Accidents and Emergency Demand Minimum training set: all data except the last year Base forecasts using auto.arima(). Reconciled using WLS V. Mean Absolute Scaled Errors for 1, 4 and 13 weeks ahead using a rolling origin. Aggr. Level h Base Reconciled Change Weekly % Weekly % Weekly % Weekly % Annual % Optimal forecast reconciliation for big time series data Temporal hierarchies 45

118 UK Accidents and Emergency Demand Minimum training set: all data except the last year Base forecasts using auto.arima(). Reconciled using WLS V. Mean Absolute Scaled Errors for 1, 4 and 13 weeks ahead using a rolling origin. Aggr. Level h Base Reconciled Change Weekly % Weekly % Weekly % Weekly % Annual % Optimal forecast reconciliation for big time series data Temporal hierarchies 45

119 UK Accidents and Emergency Demand Minimum training set: all data except the last year Base forecasts using auto.arima(). Reconciled using WLS V. Mean Absolute Scaled Errors for 1, 4 and 13 weeks ahead using a rolling origin. Aggr. Level h Base Reconciled Change Weekly % Weekly % Weekly % Weekly % Annual % Optimal forecast reconciliation for big time series data Temporal hierarchies 45

120 UK Accidents and Emergency Demand Minimum training set: all data except the last year Base forecasts using auto.arima(). Reconciled using WLS V. Mean Absolute Scaled Errors for 1, 4 and 13 weeks ahead using a rolling origin. Aggr. Level h Base Reconciled Change Weekly % Weekly % Weekly % Weekly % Annual % Optimal forecast reconciliation for big time series data Temporal hierarchies 45

121 Outline 1 Hierarchical and grouped time series 2 BLUF: Best Linear Unbiased Forecasts 3 Application: Australian tourism 4 Temporal hierarchies 5 References Optimal forecast reconciliation for big time series data References 46

122 References RJ Hyndman, RA Ahmed, G Athanasopoulos, and HL Shang (2011). Optimal combination forecasts for hierarchical time series. Computational statistics & data analysis 55(9), RJ Hyndman, AJ Lee, and E Wang (2014). Fast computation of reconciled forecasts for hierarchical and grouped time series. Working paper 17/14. Monash University SL Wickramasuriya, G Athanasopoulos, and RJ Hyndman (2015). Forecasting hierarchical and grouped time series through trace minimization. Working paper 15/15. Monash University G Athanasopoulos, RJ Hyndman, N Kourentzes, and F Petropoulos (2015). Forecasting with temporal hierarchies. Working paper. Monash University RJ Hyndman, AJ Lee, and E Wang (2015). hts: Hierarchical and grouped time series. R package v4.5 on CRAN. Optimal forecast reconciliation for big time series data References 47

Forecasting: principles and practice. Rob J Hyndman 4 Hierarchical forecasting

Forecasting: principles and practice. Rob J Hyndman 4 Hierarchical forecasting Forecasting: principles and practice Rob J Hyndman 4 Hierarchical forecasting 1 Outline 1 Hierarchical and grouped time series 2 Optimal forecast reconciliation 3 hts package for R 4 Application: Australian

More information

Fast computation of reconciled forecasts in hierarchical and grouped time series

Fast computation of reconciled forecasts in hierarchical and grouped time series Rob J Hyndman Fast computation of reconciled forecasts in hierarchical and grouped time series Total A B C AX AY AZ BX BY BZ CX CY CZ 1 Outline 1 Optimally reconciled forecasts 2 Fast computation 3 hts

More information

Forecasting hierarchical (and grouped) time series

Forecasting hierarchical (and grouped) time series George Athanasopoulos (with Rob J. Hyndman, Nikos Kourentzes and Fotis Petropoulos) Forecasting hierarchical (and grouped) time series Total A B C AA AB AC BA BB BC CA CB CC 1 Outline 1 Hierarchical and

More information

Rob J Hyndman. Visualizing and forecasting. big time series data. Victoria: scaled

Rob J Hyndman. Visualizing and forecasting. big time series data. Victoria: scaled Rob J Hyndman Visualizing and forecasting big time series data Victoria: scaled Outline 1 Examples of biggish time series 2 Time series visualisation 3 BLUF: Best Linear Unbiased Forecasts 4 Application:

More information

Optimal combination forecasts for hierarchical time series

Optimal combination forecasts for hierarchical time series Optimal combination forecasts for hierarchical time series Rob J. Hyndman Roman A. Ahmed Department of Econometrics and Business Statistics Outline 1 Review of hierarchical forecasting 2 3 Simulation study

More information

Fast computation of reconciled forecasts for hierarchical and grouped time series

Fast computation of reconciled forecasts for hierarchical and grouped time series ISSN 1440-771X Department of Econometrics and Business Statistics http://wwwbusecomonasheduau/depts/ebs/pubs/wpapers/ Fast computation of reconciled forecasts for hierarchical and grouped time series Rob

More information

Package hts. June 21, 2017

Package hts. June 21, 2017 Title Hierarchical and Grouped Time Series Version 5.1.4 Package hts June 21, 2017 Methods for analysing and forecasting hierarchical and grouped time series. Depends R (>= 3.0.2), forecast (>= 8.1) Imports

More information

Time series can often be naturally disaggregated

Time series can often be naturally disaggregated Optimally Reconciling Forecasts in a Hierarchy Rob J. Hyndman and George Athanasopoulos Preview We know that when forecasting the product hierarchy, reconciliation is invariably needed to make the sum

More information

Package thief. R topics documented: January 24, Version 0.3 Title Temporal Hierarchical Forecasting

Package thief. R topics documented: January 24, Version 0.3 Title Temporal Hierarchical Forecasting Version 0.3 Title Temporal Hierarchical Forecasting Package thief January 24, 2018 Methods and tools for generating forecasts at different temporal frequencies using a hierarchical time series approach.

More information

Optimal forecast reconciliation for hierarchical and grouped time series through trace minimization

Optimal forecast reconciliation for hierarchical and grouped time series through trace minimization ISSN 1440-771X Department of Econometrics and Business Statistics http://business.monash.edu/econometrics-and-businessstatistics/research/publications Optimal forecast reconciliation for hierarchical and

More information

Optimal forecast reconciliation for hierarchical and grouped time series through trace minimization

Optimal forecast reconciliation for hierarchical and grouped time series through trace minimization Optimal forecast reconciliation for hierarchical and grouped time series through trace minimization Shanika L Wickramasuriya Department of Econometrics and Business Statistics, Monash University, VIC 3800,

More information

Forecasting in Hierarchical Models

Forecasting in Hierarchical Models Forecasting in Hierarchical Models Lucy Morgan Supervisor: Nikolaos Kourentzes 20 th February 2015 Introduction Forecasting is the process of making statements about events whose actual outcomes (typically)

More information

Using Temporal Hierarchies to Predict Tourism Demand

Using Temporal Hierarchies to Predict Tourism Demand Using Temporal Hierarchies to Predict Tourism Demand International Symposium on Forecasting 29 th June 2015 Nikolaos Kourentzes Lancaster University George Athanasopoulos Monash University n.kourentzes@lancaster.ac.uk

More information

Modelling and Forecasting Australian Domestic Tourism

Modelling and Forecasting Australian Domestic Tourism Modelling and Forecasting Australian Domestic Tourism George Athanasopoulos & Rob Hyndman Outline Background 1 Background 2 Data 3 Regression models 4 Exponential smoothing via innovations state space

More information

Forecast UPC-Level FMCG Demand, Part III: Grouped Reconciliation

Forecast UPC-Level FMCG Demand, Part III: Grouped Reconciliation 2016 IEEE International Conference on Big Data (Big Data) Forecast UPC-Level FMCG Demand, Part III: Grouped Reconciliation Dazhi Yang, Gary S. W. Goh, Siwei Jiang and Allan N. Zhang Singapore Institute

More information

Cross-temporal coherent forecasts for Australian tourism

Cross-temporal coherent forecasts for Australian tourism ISSN 1440-771X Department of Econometrics and Business Statistics http://business.monash.edu/econometrics-and-businessstatistics/research/publications Cross-temporal coherent forecasts for Australian tourism

More information

Forecasting hierarchical and grouped time series through trace minimization

Forecasting hierarchical and grouped time series through trace minimization ISSN 1440-771X Department of Econometrics and Business Statistics http://business.monash.edu/econometrics-and-business-statistics/research/publications Forecasting hierarchical and grouped time series

More information

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS ISSN 1440-771X ISBN 0 7326 1085 0 Unmasking the Theta Method Rob J. Hyndman and Baki Billah Working Paper 5/2001 2001 DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS AUSTRALIA Unmasking the Theta method

More information

Lecture 1: Introduction to Forecasting

Lecture 1: Introduction to Forecasting NATCOR: Forecasting & Predictive Analytics Lecture 1: Introduction to Forecasting Professor John Boylan Lancaster Centre for Forecasting Department of Management Science Leading research centre in applied

More information

Grouped time-series forecasting: Application to regional infant mortality counts

Grouped time-series forecasting: Application to regional infant mortality counts Grouped time-series forecasting: Application to regional infant mortality counts Han Lin Shang and Peter W. F. Smith University of Southampton Motivation 1 Multiple time series can be disaggregated by

More information

Regularization in Hierarchical Time Series Forecasting With Application to Electricity Smart Meter Data

Regularization in Hierarchical Time Series Forecasting With Application to Electricity Smart Meter Data Regularization in Hierarchical Time Series Forecasting With Application to Electricity Smart Meter Data Souhaib Ben Taieb Monash Business School Monash University Jiafan Yu Dept. of CEE Stanford University

More information

Forecasting: principles and practice. Rob J Hyndman 1.1 Introduction to Forecasting

Forecasting: principles and practice. Rob J Hyndman 1.1 Introduction to Forecasting Forecasting: principles and practice Rob J Hyndman 1.1 Introduction to Forecasting 1 Outline 1 Background 2 Case studies 3 The statistical forecasting perspective 4 What can we forecast? 2 Resources Slides

More information

Automatic forecasting with a modified exponential smoothing state space framework

Automatic forecasting with a modified exponential smoothing state space framework ISSN 1440-771X Department of Econometrics and Business Statistics http://www.buseco.monash.edu.au/depts/ebs/pubs/wpapers/ Automatic forecasting with a modified exponential smoothing state space framework

More information

The value of feedback in forecasting competitions

The value of feedback in forecasting competitions ISSN 1440-771X Department of Econometrics and Business Statistics http://www.buseco.monash.edu.au/depts/ebs/pubs/wpapers/ The value of feedback in forecasting competitions George Athanasopoulos and Rob

More information

Forecasting with R A practical workshop

Forecasting with R A practical workshop Forecasting with R A practical workshop International Symposium on Forecasting 2016 19 th June 2016 Nikolaos Kourentzes nikolaos@kourentzes.com http://nikolaos.kourentzes.com Fotios Petropoulos fotpetr@gmail.com

More information

Hierarchical Electricity Time Series Forecasting for Integrating Consumption Patterns Analysis and Aggregation Consistency

Hierarchical Electricity Time Series Forecasting for Integrating Consumption Patterns Analysis and Aggregation Consistency Hierarchical Electricity Time Series Forecasting for Integrating Consumption Patterns Analysis and Aggregation Consistency Yue Pang 1, Bo Yao 1, Xiangdong Zhou 1, Yong Zhang 2, Yiming Xu 1 and Zijing Tan

More information

Forecasting using exponential smoothing: the past, the present, the future

Forecasting using exponential smoothing: the past, the present, the future Forecasting using exponential smoothing: the past, the present, the future OR60 13th September 2018 Marketing Analytics and Forecasting Introduction Exponential smoothing (ES) is one of the most popular

More information

A new approach to forecasting based on exponential smoothing with independent regressors

A new approach to forecasting based on exponential smoothing with independent regressors ISSN 1440-771X Australia Department of Econometrics and Business Statistics http://wwwbusecomonasheduau/depts/ebs/pubs/wpapers/ A new approach to forecasting based on exponential smoothing with independent

More information

Package Tcomp. June 6, 2018

Package Tcomp. June 6, 2018 Package Tcomp June 6, 2018 Title Data from the 2010 Tourism Forecasting Competition Version 1.0.1 The 1311 time series from the tourism forecasting competition conducted in 2010 and described in Athanasopoulos

More information

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS ISSN 1440-771X ISBN 0 7326 1091 5 Prediction Intervals for Exponential Smoothing State Space Models Rob J Hyndman, Anne B Koehler, J. Keith Ord and Ralph D Snyder Working Paper 11/2001 2001 DEPARTMENT

More information

Forecasting: Principles and Practice. Rob J Hyndman. 1. Introduction to forecasting OTexts.org/fpp/1/ OTexts.org/fpp/2/3

Forecasting: Principles and Practice. Rob J Hyndman. 1. Introduction to forecasting OTexts.org/fpp/1/ OTexts.org/fpp/2/3 Rob J Hyndman Forecasting: Principles and Practice 1. Introduction to forecasting OTexts.org/fpp/1/ OTexts.org/fpp/2/3 Forecasting: Principles and Practice 1 Resources Slides Exercises Textbook Useful

More information

Modelling and forecasting Australian domestic tourism

Modelling and forecasting Australian domestic tourism * Title page with author details Modelling and forecasting Australian domestic tourism G. Athanasopoulos, R. J. Hyndman Department of Econometrics and Business Statistics, Monash University, Melbourne,

More information

Package MAPA. R topics documented: January 5, Type Package Title Multiple Aggregation Prediction Algorithm Version 2.0.4

Package MAPA. R topics documented: January 5, Type Package Title Multiple Aggregation Prediction Algorithm Version 2.0.4 Type Package Title Multiple Aggregation Prediction Algorithm Version 2.0.4 Package MAPA January 5, 2018 Functions and wrappers for using the Multiple Aggregation Prediction Algorithm (MAPA) for time series

More information

A State Space Framework For Automatic Forecasting Using Exponential Smoothing Methods

A State Space Framework For Automatic Forecasting Using Exponential Smoothing Methods ISSN 1440-771X ISBN 0 7326 1078 8 A State Space Framework For Automatic Forecasting Using Exponential Smoothing s Rob J. Hyndman, Anne B. Koehler, Ralph D. Snyder and Simone Grose Working Paper 9/2000

More information

An Evaluation of Methods for Combining Univariate Time Series Forecasts

An Evaluation of Methods for Combining Univariate Time Series Forecasts An Evaluation of Methods for Combining Univariate Time Series Forecasts Magnus Svensson Bachelor s thesis in Statistics 15 ECTS credits April 2018 Supervised by Jonas Wallin Department of Statistics Lund

More information

Improving forecasting by estimating time series structural components across multiple frequencies

Improving forecasting by estimating time series structural components across multiple frequencies Improving forecasting by estimating time series structural components across multiple frequencies Nikolaos Kourentzes a,, Fotios Petropoulos a, Juan R. Trapero b a Lancaster University Management School

More information

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS ISSN 1440-771X AUSTRALIA DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS Empirical Information Criteria for Time Series Forecasting Model Selection Md B Billah, R.J. Hyndman and A.B. Koehler Working

More information

ARMA (and ARIMA) models are often expressed in backshift notation.

ARMA (and ARIMA) models are often expressed in backshift notation. Backshift Notation ARMA (and ARIMA) models are often expressed in backshift notation. B is the backshift operator (also called the lag operator ). It operates on time series, and means back up by one time

More information

Local linear forecasts using cubic smoothing splines

Local linear forecasts using cubic smoothing splines Local linear forecasts using cubic smoothing splines Rob J. Hyndman, Maxwell L. King, Ivet Pitrun, Baki Billah 13 January 2004 Abstract: We show how cubic smoothing splines fitted to univariate time series

More information

Mathematics I. Exercises with solutions. 1 Linear Algebra. Vectors and Matrices Let , C = , B = A = Determine the following matrices:

Mathematics I. Exercises with solutions. 1 Linear Algebra. Vectors and Matrices Let , C = , B = A = Determine the following matrices: Mathematics I Exercises with solutions Linear Algebra Vectors and Matrices.. Let A = 5, B = Determine the following matrices: 4 5, C = a) A + B; b) A B; c) AB; d) BA; e) (AB)C; f) A(BC) Solution: 4 5 a)

More information

Modified Holt s Linear Trend Method

Modified Holt s Linear Trend Method Modified Holt s Linear Trend Method Guckan Yapar, Sedat Capar, Hanife Taylan Selamlar and Idil Yavuz Abstract Exponential smoothing models are simple, accurate and robust forecasting models and because

More information

Parts Manual. EPIC II Critical Care Bed REF 2031

Parts Manual. EPIC II Critical Care Bed REF 2031 EPIC II Critical Care Bed REF 2031 Parts Manual For parts or technical assistance call: USA: 1-800-327-0770 2013/05 B.0 2031-109-006 REV B www.stryker.com Table of Contents English Product Labels... 4

More information

Package gsarima. February 20, 2015

Package gsarima. February 20, 2015 Package gsarima February 20, 2015 Version 0.1-4 Date 2013-06-17 Title Two functions for Generalized SARIMA time series simulation Author Olivier Briet Maintainer Olivier Briet

More information

S-GSTAR-SUR Model for Seasonal Spatio Temporal Data Forecasting ABSTRACT

S-GSTAR-SUR Model for Seasonal Spatio Temporal Data Forecasting ABSTRACT Malaysian Journal of Mathematical Sciences (S) March : 53-65 (26) Special Issue: The th IMT-GT International Conference on Mathematics, Statistics and its Applications 24 (ICMSA 24) MALAYSIAN JOURNAL OF

More information

EXPONENTIAL SMOOTHING MODELING AND FORECASTING FOR INCIDENCE OF TUBERCULOSIS IN INDIA

EXPONENTIAL SMOOTHING MODELING AND FORECASTING FOR INCIDENCE OF TUBERCULOSIS IN INDIA Global and Stochastic analysis Special Issue: 25th International Conference of Forum for Interdisciplinary Mathematics EXPONENTIAL SMOOTHING MODELING AND FORECASTING FOR INCIDENCE OF TUBERCULOSIS IN INDIA

More information

Final A. Problem Points Score Total 100. Math115A Nadja Hempel 03/23/2017

Final A. Problem Points Score Total 100. Math115A Nadja Hempel 03/23/2017 Final A Math115A Nadja Hempel 03/23/2017 nadja@math.ucla.edu Name: UID: Problem Points Score 1 10 2 20 3 5 4 5 5 9 6 5 7 7 8 13 9 16 10 10 Total 100 1 2 Exercise 1. (10pt) Let T : V V be a linear transformation.

More information

THE TRANSLATION PLANES OF ORDER 49 AND THEIR AUTOMORPHISM GROUPS

THE TRANSLATION PLANES OF ORDER 49 AND THEIR AUTOMORPHISM GROUPS MATHEMATICS OF COMPUTATION Volume 67, Number 223, July 1998, Pages 1207 1224 S 0025-5718(98)00961-2 THE TRANSLATION PLANES OF ORDER 49 AND THEIR AUTOMORPHISM GROUPS C. CHARNES AND U. DEMPWOLFF Abstract.

More information

Nonparametric time series forecasting with dynamic updating

Nonparametric time series forecasting with dynamic updating 18 th World IMAS/MODSIM Congress, Cairns, Australia 13-17 July 2009 http://mssanz.org.au/modsim09 1 Nonparametric time series forecasting with dynamic updating Han Lin Shang and Rob. J. Hyndman Department

More information

Forecasting using R. Rob J Hyndman. 2.4 Non-seasonal ARIMA models. Forecasting using R 1

Forecasting using R. Rob J Hyndman. 2.4 Non-seasonal ARIMA models. Forecasting using R 1 Forecasting using R Rob J Hyndman 2.4 Non-seasonal ARIMA models Forecasting using R 1 Outline 1 Autoregressive models 2 Moving average models 3 Non-seasonal ARIMA models 4 Partial autocorrelations 5 Estimation

More information

Module 09 From s-domain to time-domain From ODEs, TFs to State-Space Modern Control

Module 09 From s-domain to time-domain From ODEs, TFs to State-Space Modern Control Module 09 From s-domain to time-domain From ODEs, TFs to State-Space Modern Control Ahmad F. Taha EE 3413: Analysis and Desgin of Control Systems Email: ahmad.taha@utsa.edu Webpage: http://engineering.utsa.edu/

More information

Based on the original slides from Levine, et. all, First Edition, Prentice Hall, Inc

Based on the original slides from Levine, et. all, First Edition, Prentice Hall, Inc Based on the original slides from Levine, et. all, First Edition, Prentice Hall, Inc Process of predicting a future event Underlying basis of all business decisions Production Inventory Personnel Facilities

More information

Package Metrics. November 3, 2017

Package Metrics. November 3, 2017 Version 0.1.3 Title Evaluation Metrics for Machine Learning Package Metrics November 3, 2017 An implementation of evaluation metrics in R that are commonly used in supervised machine learning. It implements

More information

Exponentially weighted forecasts

Exponentially weighted forecasts FORECASTING USING R Exponentially weighted forecasts Rob Hyndman Author, forecast Simple exponential smoothing Forecasting Notation: ŷ t+h t = y t+h y 1,...,y t point forecast of Forecast Equation: given

More information

peak half-hourly New South Wales

peak half-hourly New South Wales Forecasting long-term peak half-hourly electricity demand for New South Wales Dr Shu Fan B.S., M.S., Ph.D. Professor Rob J Hyndman B.Sc. (Hons), Ph.D., A.Stat. Business & Economic Forecasting Unit Report

More information

peak half-hourly South Australia

peak half-hourly South Australia Forecasting long-term peak half-hourly electricity demand for South Australia Dr Shu Fan B.S., M.S., Ph.D. Professor Rob J Hyndman B.Sc. (Hons), Ph.D., A.Stat. Business & Economic Forecasting Unit Report

More information

A state space model for exponential smoothing with group seasonality

A state space model for exponential smoothing with group seasonality ISSN 1440-771X Department of Econometrics and Business Statistics http://www.buseco.monash.edu.au/depts/ebs/pubs/wpapers/ A state space model for exponential smoothing with group seasonality Pim Ouwehand,

More information

The Art of Forecasting

The Art of Forecasting Time Series The Art of Forecasting Learning Objectives Describe what forecasting is Explain time series & its components Smooth a data series Moving average Exponential smoothing Forecast using trend models

More information

The SAB Medium Term Sales Forecasting System : From Data to Planning Information. Kenneth Carden SAB : Beer Division Planning

The SAB Medium Term Sales Forecasting System : From Data to Planning Information. Kenneth Carden SAB : Beer Division Planning The SAB Medium Term Sales Forecasting System : From Data to Planning Information Kenneth Carden SAB : Beer Division Planning Planning in Beer Division F Operational planning = what, when, where & how F

More information

Meta-learning how to forecast time series

Meta-learning how to forecast time series ISSN 1440-771X Department of Econometrics and Business Statistics http://business.monash.edu/econometrics-and-businessstatistics/research/publications Meta-learning how to forecast time series Thiyanga

More information

Do we need Experts for Time Series Forecasting?

Do we need Experts for Time Series Forecasting? Do we need Experts for Time Series Forecasting? Christiane Lemke and Bogdan Gabrys Bournemouth University - School of Design, Engineering and Computing Poole House, Talbot Campus, Poole, BH12 5BB - United

More information

Department of Econometrics and Business Statistics

Department of Econometrics and Business Statistics ISSN 1440-771X Australia Department of Econometrics and Business Statistics http://www.buseco.monash.edu.au/depts/ebs/pubs/wpapers/ Exponential Smoothing: A Prediction Error Decomposition Principle Ralph

More information

Time Series in R: Forecasting and Visualisation. Forecast evaluation 29 May 2017

Time Series in R: Forecasting and Visualisation. Forecast evaluation 29 May 2017 Time Series in R: Forecasting and Visualisation Forecast evaluation 29 May 2017 1 Outline 1 Forecasting residuals 2 Evaluating forecast accuracy 3 Forecasting benchmark methods 4 Lab session 7 5 Time series

More information

CPSC 313 Introduction to Computability

CPSC 313 Introduction to Computability CPSC 313 Introduction to Computability Grammars in Chomsky Normal Form (Cont d) (Sipser, pages 109-111 (3 rd ed) and 107-109 (2 nd ed)) Renate Scheidler Fall 2018 Chomsky Normal Form A context-free grammar

More information

Hierarchical Probabilistic Forecasting of Electricity Demand with Smart Meter Data

Hierarchical Probabilistic Forecasting of Electricity Demand with Smart Meter Data Hierarchical Probabilistic Forecasting of Electricity Demand with Smart Meter Data Souhaib Ben Taieb Department of Econometrics and Business Statistics, Monash University James W. Taylor Saïd Business

More information

GDP forecast errors Satish Ranchhod

GDP forecast errors Satish Ranchhod GDP forecast errors Satish Ranchhod Editor s note This paper looks more closely at our forecasts of growth in Gross Domestic Product (GDP). It considers two different measures of GDP, production and expenditure,

More information

Comparison between VAR, GSTAR, FFNN-VAR and FFNN-GSTAR Models for Forecasting Oil Production

Comparison between VAR, GSTAR, FFNN-VAR and FFNN-GSTAR Models for Forecasting Oil Production MATEMATIKA, 218, Volume 34, Number 1, 13 111 c Penerbit UTM Press. All rights reserved Comparison between VAR, GSTAR, FFNN-VAR and FFNN-GSTAR Models for Forecasting Oil Production 1 Suhartono, 2 Dedy Dwi

More information

Financial Econometrics

Financial Econometrics Material : solution Class : Teacher(s) : zacharias psaradakis, marian vavra Example 1.1: Consider the linear regression model y Xβ + u, (1) where y is a (n 1) vector of observations on the dependent variable,

More information

Copyright 2010 Pearson Education, Inc. Publishing as Prentice Hall.

Copyright 2010 Pearson Education, Inc. Publishing as Prentice Hall. 13 Forecasting PowerPoint Slides by Jeff Heyl For Operations Management, 9e by Krajewski/Ritzman/Malhotra 2010 Pearson Education 13 1 Forecasting Forecasts are critical inputs to business plans, annual

More information

ECON 427: ECONOMIC FORECASTING. Ch1. Getting started OTexts.org/fpp2/

ECON 427: ECONOMIC FORECASTING. Ch1. Getting started OTexts.org/fpp2/ ECON 427: ECONOMIC FORECASTING Ch1. Getting started OTexts.org/fpp2/ 1 Outline 1 What can we forecast? 2 Time series data 3 Some case studies 4 The statistical forecasting perspective 2 Forecasting is

More information

Forecasting using robust exponential smoothing with damped trend and seasonal components

Forecasting using robust exponential smoothing with damped trend and seasonal components Forecasting using robust exponential smoothing with damped trend and seasonal components Crevits R, Croux C. KBI_1714 Forecasting using Robust Exponential Smoothing with Damped Trend and Seasonal Components

More information

A = 3 B = A 1 1 matrix is the same as a number or scalar, 3 = [3].

A = 3 B = A 1 1 matrix is the same as a number or scalar, 3 = [3]. Appendix : A Very Brief Linear ALgebra Review Introduction Linear Algebra, also known as matrix theory, is an important element of all branches of mathematics Very often in this course we study the shapes

More information

The tourism forecasting competition

The tourism forecasting competition ISSN 1440-771X Department of Econometrics and Business Statistics http://www.buseco.monash.edu.au/depts/ebs/pubs/wpapers/ The tourism forecasting competition George Athanasopoulos, Rob J Hyndman, Haiyan

More information

Modelling Wind Farm Data and the Short Term Prediction of Wind Speeds

Modelling Wind Farm Data and the Short Term Prediction of Wind Speeds Modelling Wind Farm Data and the Short Term Prediction of Wind Speeds An Investigation into Wind Speed Data Sets Erin Mitchell Lancaster University 6th April 2011 Outline 1 Data Considerations Overview

More information

1. Fundamental concepts

1. Fundamental concepts . Fundamental concepts A time series is a sequence of data points, measured typically at successive times spaced at uniform intervals. Time series are used in such fields as statistics, signal processing

More information

A VERY BRIEF LINEAR ALGEBRA REVIEW for MAP 5485 Introduction to Mathematical Biophysics Fall 2010

A VERY BRIEF LINEAR ALGEBRA REVIEW for MAP 5485 Introduction to Mathematical Biophysics Fall 2010 A VERY BRIEF LINEAR ALGEBRA REVIEW for MAP 5485 Introduction to Mathematical Biophysics Fall 00 Introduction Linear Algebra, also known as matrix theory, is an important element of all branches of mathematics

More information

peak half-hourly Tasmania

peak half-hourly Tasmania Forecasting long-term peak half-hourly electricity demand for Tasmania Dr Shu Fan B.S., M.S., Ph.D. Professor Rob J Hyndman B.Sc. (Hons), Ph.D., A.Stat. Business & Economic Forecasting Unit Report for

More information

Transformations for variance stabilization

Transformations for variance stabilization FORECASTING USING R Transformations for variance stabilization Rob Hyndman Author, forecast Variance stabilization If the data show increasing variation as the level of the series increases, then a transformation

More information

Zhaoxing Gao and Ruey S Tsay Booth School of Business, University of Chicago. August 23, 2018

Zhaoxing Gao and Ruey S Tsay Booth School of Business, University of Chicago. August 23, 2018 Supplementary Material for Structural-Factor Modeling of High-Dimensional Time Series: Another Look at Approximate Factor Models with Diverging Eigenvalues Zhaoxing Gao and Ruey S Tsay Booth School of

More information

THE BRAZILIAN QUARTERLY REAL GDP:TEMPORAL DISAGGREGATION AND NOWCASTING.

THE BRAZILIAN QUARTERLY REAL GDP:TEMPORAL DISAGGREGATION AND NOWCASTING. THE BRAZILIAN QUARTERLY REAL GDP:TEMPORAL DISAGGREGATION AND NOWCASTING. André Nunes Maranhão - University of Brasília and Bank of Brazil André Minella - Central Bank of Brazil Cleomar Gomes da Silva -

More information

NATCOR. Forecast Evaluation. Forecasting with ARIMA models. Nikolaos Kourentzes

NATCOR. Forecast Evaluation. Forecasting with ARIMA models. Nikolaos Kourentzes NATCOR Forecast Evaluation Forecasting with ARIMA models Nikolaos Kourentzes n.kourentzes@lancaster.ac.uk O u t l i n e 1. Bias measures 2. Accuracy measures 3. Evaluation schemes 4. Prediction intervals

More information

Package TSPred. April 5, 2017

Package TSPred. April 5, 2017 Type Package Package TSPred April 5, 2017 Title Functions for Benchmarking Time Series Prediction Version 3.0.2 Date 2017-04-05 Author Rebecca Pontes Salles [aut, cre, cph] (CEFET/RJ), Eduardo Ogasawara

More information

Package dlagm. January 17, 2018

Package dlagm. January 17, 2018 Type Package Package dlagm January 17, 2018 Title Time Series Regression Models with Distributed Lag Models Version 1.0.2 Date 2018-01-13 Author Haydar Demirhan Maintainer Haydar Demirhan

More information

VARMA versus VAR for Macroeconomic Forecasting

VARMA versus VAR for Macroeconomic Forecasting VARMA versus VAR for Macroeconomic Forecasting 1 VARMA versus VAR for Macroeconomic Forecasting George Athanasopoulos Department of Econometrics and Business Statistics Monash University Farshid Vahid

More information

Recent Developments in Benchmarking to Annual Totals in X-12-ARIMA and at Statistics Canada

Recent Developments in Benchmarking to Annual Totals in X-12-ARIMA and at Statistics Canada CONFERENCE ON SEASONALITY, SEASONAL ADJUSTMENT AND THEIR IMPLICATIONS FOR SHORT-TERM ANALYSIS AND FORECASTING 10-12 MAY 2006 Recent Developments in Benchmarking to Annual Totals in X-12-ARIMA and at Statistics

More information

page 1 Total ( )

page 1 Total ( ) A B C D E F Costs budget of [Claimant / Defendant] dated [ ] Estimated page 1 Work done / to be done Pre-action Disbs ( ) Time ( ) Disbs ( ) Time ( ) Total ( ) 1 Issue /statements of case 0.00 0.00 CMC

More information

Functional time series

Functional time series Rob J Hyndman Functional time series with applications in demography 4. Connections, extensions and applications Outline 1 Yield curves 2 Electricity prices 3 Dynamic updating with partially observed functions

More information

Time Series Analysis

Time Series Analysis Time Series Analysis hm@imm.dtu.dk Informatics and Mathematical Modelling Technical University of Denmark DK-2800 Kgs. Lyngby 1 Outline of the lecture State space models, 1st part: Model: Sec. 10.1 The

More information

CHAPTER 1: Decomposition Methods

CHAPTER 1: Decomposition Methods CHAPTER 1: Decomposition Methods Prof. Alan Wan 1 / 48 Table of contents 1. Data Types and Causal vs.time Series Models 2 / 48 Types of Data Time series data: a sequence of observations measured over time,

More information

SHORT TERM LOAD FORECASTING

SHORT TERM LOAD FORECASTING Indian Institute of Technology Kanpur (IITK) and Indian Energy Exchange (IEX) are delighted to announce Training Program on "Power Procurement Strategy and Power Exchanges" 28-30 July, 2014 SHORT TERM

More information

MLES & Multivariate Normal Theory

MLES & Multivariate Normal Theory Merlise Clyde September 6, 2016 Outline Expectations of Quadratic Forms Distribution Linear Transformations Distribution of estimates under normality Properties of MLE s Recap Ŷ = ˆµ is an unbiased estimate

More information

Complex exponential Smoothing. Ivan Svetunkov Nikolaos Kourentzes Robert Fildes

Complex exponential Smoothing. Ivan Svetunkov Nikolaos Kourentzes Robert Fildes Complex exponential Smoothing Ivan Svetunkov Nikolaos Kourentzes Robert Fildes 18 March 2015 Introduction Exponential Smoothing methods performed very well in many competitions: M-Competitions in 1982

More information

Complex Exponential Smoothing

Complex Exponential Smoothing Complex Exponential Smoothing Ivan Svetunkov Management School Lancaster University A thesis submitted for the degree of Doctor of Philosophy 2016 Acknowledgements It was one of those evenings during our

More information

Econometric Forecasting

Econometric Forecasting Robert M Kunst robertkunst@univieacat University of Vienna and Institute for Advanced Studies Vienna September 29, 2014 University of Vienna and Institute for Advanced Studies Vienna Outline Introduction

More information

Lecture 1: OLS derivations and inference

Lecture 1: OLS derivations and inference Lecture 1: OLS derivations and inference Econometric Methods Warsaw School of Economics (1) OLS 1 / 43 Outline 1 Introduction Course information Econometrics: a reminder Preliminary data exploration 2

More information

Ehrhart polynomial for lattice squares, cubes, and hypercubes

Ehrhart polynomial for lattice squares, cubes, and hypercubes Ehrhart polynomial for lattice squares, cubes, and hypercubes Eugen J. Ionascu UWG, REU, July 10th, 2015 math@ejionascu.ro, www.ejionascu.ro 1 Abstract We are investigating the problem of constructing

More information

The Multivariate Normal Distribution 1

The Multivariate Normal Distribution 1 The Multivariate Normal Distribution 1 STA 302 Fall 2014 1 See last slide for copyright information. 1 / 37 Overview 1 Moment-generating Functions 2 Definition 3 Properties 4 χ 2 and t distributions 2

More information

Serie temporali: previsioni

Serie temporali: previsioni A. A. 2016-2017 Serie temporali: previsioni prof. ing. Antonio Comi Department of Enterprise Engineering University of Rome Tor Vergata Bibliography Forecasting: principles and practice by Rob J Hyndman

More information

Applied Time Series Topics

Applied Time Series Topics Applied Time Series Topics Ivan Medovikov Brock University April 16, 2013 Ivan Medovikov, Brock University Applied Time Series Topics 1/34 Overview 1. Non-stationary data and consequences 2. Trends and

More information

BALANCING-RELATED MODEL REDUCTION FOR DATA-SPARSE SYSTEMS

BALANCING-RELATED MODEL REDUCTION FOR DATA-SPARSE SYSTEMS BALANCING-RELATED Peter Benner Professur Mathematik in Industrie und Technik Fakultät für Mathematik Technische Universität Chemnitz Computational Methods with Applications Harrachov, 19 25 August 2007

More information

Demographic Data in ArcGIS. Harry J. Moore IV

Demographic Data in ArcGIS. Harry J. Moore IV Demographic Data in ArcGIS Harry J. Moore IV Outline What is demographic data? Esri Demographic data - Real world examples with GIS - Redistricting - Emergency Preparedness - Economic Development Next

More information