A bivariate INAR(1) process with application

Size: px
Start display at page:

Download "A bivariate INAR(1) process with application"

Transcription

1 A bivariate INAR1 process wit application Xanti Pedeli and Dimitris Karlis Department of Statistics Atens University of Economics and Business Abstract Te study of time series models for count data as become a topic of special interest during te last years. However, wile researc on univariate time series for counts now flouris, te literature on multivariate time series models for count data is notably more limited. In te present paper, a bivariate integer-valued autoregressive process of order 1 BINAR1 is introduced. Empasis is placed on model wit bivariate Poisson and bivariate negative binomial innovations. We discuss properties of te BINAR1 model and propose te metod of conditional maximum likeliood for te estimation of its unknown parameters. Issues of diagnostics and forecasting are considered and predictions are produced by means of te conditional forecast distribution. Estimation uncertainty is accommodated by taking advantage of te asymptotic normality of maximum likeliood estimators and constructing appropriate confidence intervals for te -step-aead conditional probability mass function. Te proposed model is applied to a bivariate data series concerning daytime and nigttime road accidents in te Neterlands. Keywords: BINAR; count data; Poisson; negative binomial; bivariate time series. 1 Introduction Multivariate count data occur in several different disciplines like epidemiology, marketing, criminology and engineering just to name a few. In many cases te data are observed across time leading to multivariate time series data as, for example, wen one studies te purcases of different products across time, or te occurrence of different diseases across time. Corresponding autor 1

2 In te literature tere are several models to fit univariate count time series models see Davis et al., A commonly used class of suc models consists of te so-called integer autoregressive time series models, introduced by McKenzie 1985 and Al-Os and Alzaid Te interested reader is referred to McKenzie 2003 and Jung and Tremayne 2006 for a brief but detailed review of suc models. Te literature on multivariate time series models for count data is less developed. Some interesting attempts ave been made during te last decade but most of tem do not arise in te context of INAR processes. Among te models tat ave been built in te aforementioned setting are tose of Latour 1997; Brännäs and Nordström 2000; Heinen and Rengifo 2007; Silva et al and Quoresi Te aim of tis paper is to introduce and examine in detail a bivariate integer-valued autoregressive model of order 1 BINAR1. To motivate te model consider te case of road traffic accidents. Accident analysis assumes tat even if te beavior of crases differs between day and nigt, bot types of accidents sare some common azards. Weater conditions, te road s quality and caracteristics and uman error, i.e. fallible perception, attention and/or memory, count between te factors tat introduce correlation. On te oter and, serial correlation between successive daily cras counts, i.e. autocorrelation, is reported as an important callenge for all accident models Brijs et al., Tus, appropriate time series models are needed to andle te presence of correlation bot between and witin te series of daytime and nigttime cras counts. Te remaining of te paper is structured as follows. A general specification of te BINAR1 process and alternative metods for te estimation of its unknown parameters are given in section 2. In sections 3 we concentrate on te special cases of bivariate Poisson and bivariate negative binomial innovations respectively. In section 4 we give a specification of te model residuals as a diagnostic tool wile issues of forecasting are discussed in section 5. An application to real data concerning daytime and nigttime road accidents follows in section 6. Some concluding remarks are presented in section 7. 2 Te BINAR1 Process 2.1 Model Let X and R be non-negative integer-valued random 2-vectors. Let A be a 2 2 diagonal matrix wit independent elements {α jj } j1,2. Te bivariate 2

3 integer-valued autoregressive process of order 1 can be defined as [ ] [ ] [ ] α1 0 X1,t 1 R1t X t A X t 1 + R t +, t Z α 2 X 2,t 1 R 2t were is te binomial tinning operator defined as α X X i1 Y i Y,were{Y i } X i1 is a sequence of iid Bernoulli random variables suc tat P Y i 1α 1 P Y i 0andα [0, 1] Steutel and van Harn, In te bivariate case, te A operation is a matricial operation wic acts as te usual matrix multiplication keeping in te same time te properties of te binomial tinning operation. One can see tat wit te above definition te jt element, j 1, 2isgivenbyX jt α j X j,t 1 + R jt. Te elements R t wic entered te system in te interval t 1,t] are usually called as innovations. Assuming independence between and witin te tinning operations and {R jt } an iid sequence wit mean λ j and variance σ 2 j υ j λ j, υ j > 0,j 1, 2, te unconditional first and second order moments based on second order stationarity conditions are: EX jt µ Xj λ j 1 α j 2.2 VarX jt σ 2 X j α j + υ j λ j 1 α 2 j 2.3 Cov X jt,x j,t+ γ Xj α j σ 2 X j ; 1, 2, Corr X jt,x j,t+ ρ Xj α j ; 1, 2, Obviously, te mean, variance and autocovariance functions can take only positive values, since λ j,σj 2 and α j are all positive. Depending on weter υ j > 1, υ j 0, 1, or υ j 1, te variance may be larger tan te mean overdispersion, smaller tan te mean underdispersion, or equal to te mean equidispersion respectively. Dependence between te two series tat comprise te BINAR1 process is introduced by allowing for dependence between R 1t and R 2t wile retaining all te previous assumptions fixed. Watever te underlying joint distribution of {R 1t,R 2t } is, it can be sown tat te covariance between te innovations of te two series at time t, totally determines te covariance between te current value of te one process and te innovations of te oter process at te same point in time t and vice versa see Appendix: 3

4 CovX 1t,R 2t CovR 1t,R 2t 2.6 As expected, te covariance between te sequences {X 1t } and {X 2t } at time t is also affected by te corresponding survival parts of te two processes. More specifically it can be sown tat, CovX 1,t+,X 2t α 1 1 α 1 α 2 CovR 1t,R 2t ; 0, 1,... and 2.7 α CorrX 1,t+,X 2t 1 1 α α2 2 1 α 1 α 2 CovR 1t,R 2t ; 0, 1,... α 1 + υ 1 α 2 + υ 2 λ 1 λ Covariances and correlations between X 1t and X 2,t+, 0, 1,...,canbe defined analogously. Note tat 2.7 presumes tat {X t } is a strictly stationary process, i.e. X1t X1,t+ tat te joint distribution of istesameastatof, for X 2t X 2,t+ all. Using te analytical representations [ ] [ ] [ ] X1t α1 0 X1,t 1 R1t X 2t 0 α 2 X 2,t 1 R 2t and [ ] [ ] [ ] X1,t+ α1 0 X1,t+ 1 R1,t X 2,t+ 0 α 2 X 2,t+ 1 R 2,t+ it is easy to see tat strict stationarity does indeed old for {X t } since te variables involved in te rigt-and sides of 2.9 and 2.10 ave identical distributions see also Latour, Estimation As already noted, te structure of te BINAR1 model implies tat te two innovation series {R 1t,R 2t } follow jointly a bivariate distribution. Let G R1,R 2 s 1,s 2 be te joint probability generating function jpgf of{r 1t,R 2t }. Ten, te jpgf of X t {X 1t,X 2t } is given by G Xt s G X1t,X 2t s 1,s 2 G X1,0 1 α1 t + α1s t 1 G X2,0 1 α2 t + α2s t 2 t 1 G R1,R 2 1 α1 i + α1s i 1, 1 α2 i + α2s i

5 wic reduces to G Xt s G X1t,X 2t s 1,s 2 G R1,R 2 1 α1+α i 1s i 1, 1 α2+α i 2s i Te moment generating function M Xt s G Xt e s can ten be used to obtain appropriate sample moments for te estimation of te unknown model parameters. However, wen te definition of a full density function is feasible, maximum-likeliood ML estimation is generally preferable. In te remaining of te present section we describe a general setting for ML estimation of te unknown parameters involved in te conditional mean and covariance functions. Te conditional density for te BINAR1 model can be expressed as te convolution of two binomials, namely X1,t 1 f 1 x 1 α x α 1 X 1,t 1 x f 2 x 2 x 1 X2,t 1 x 2 α x α 2 X 2,t 1 x 2, 2.14 and a bivariate distribution of te form f 3 k, s P R 1t k, R 2t s. Tus te conditional density becomes fx t x t 1, θ f 1 x 1t kf 2 x 2t sf 3 k, s 2.15 k s were θ is te vector of unknown parameters. Te conditional likeliood function is ten given by Lθ x T fx t x t 1, θ 2.16 t1 for some initial value x 0 and ence maximization provides wit te ML estimates. Numerical maximization is straigtforward wit standard statistical packages. 3 Parametric Cases In tis section we discuss two specific BINAR1 models. Te first one comes from te assumption tat te innovations of te two series follow jointly a bivariate Poisson distribution. Te second model assumes a bivariate negative 5

6 binomial distribution for te two innovation processes. Te two representations can be viewed as appropriate tools for modeling equidispersed and overdispersed bivariate time series respectively. Some additional specifications for negative correlation time series data are also briefly considered. 3.1 Te Poisson BINAR1 Process Model Let assume tat te joint probability mass function jpmf of te two innovation processes {R 1t,R 2t } is a bivariate Poisson distribution given by P R 1t x, R 2t y e λ 1+λ 2 φ λ 1 φ x λ 2 φ y x! y! s x y i i i φ i! 3.1 λ 1 φλ 2 φ were s minx, y, λ 1,λ 2 > 0andφ [0, minλ 1,λ 2. We will denote tis distribution as BPλ 1,λ 2,φ. Te bivariate Poisson distribution defined in 3.1 allows for dependence between te two random variables. Marginally eac random variable follows a Poisson distribution wit parameters λ 1 and λ 2 respectively. Parameter φ is te covariance between te two random variables. If φ 0 ten te two variables are independent and te bivariate Poisson distribution reduces to te product of two independent Poisson distributions. For a compreensive treatment of te bivariate Poisson distribution and its multivariate extensions te reader can refer to te books of Kocerlakota and Kocerlakota 1992 and Jonson et al Te above assumption leads to te equidispersion case, i.e. υ j 1,or equivalently assume tat R jt are iid Poisson sequences wit σ 2 j λ j, j 1, 2. Obviously, in tis case te covariance function 2.7 remains unaffected wile te correlation function 2.8 is simplified due to te simplification of te variances of te two processes. Hence, te Poisson BINAR1 model is caracterized by te vector of expectations µ Xt EX t wit elements µ Xjt λ j ; j 1, α j te variance-covariance matrix γ Xt wit diagonal elements CovX j,t+,x jt α j λ j ; 1 α j j 1, 2, 0, 1, and off-diagonal elements 6

7 CovX j,t+,x it α j φ ; j i, 0, 1, α 1 α 2 and te correlation matrix ρ Xt wit diagonal and off-diagonal elements equal to and CorrX j,t+,x jt α j ; j 1, 2, 0, 1, CorrX j,t+,x it α j 1 α1 1 α 2 φ 1 α 1 α 2 ; j i, 0, 1, λ 1 λ 2 respectively. Note also tat conditionally on te previous observations X t 1 {X 1,t 1,X 2,t 1 }, te vector of conditional means µ Xt t 1 EX t t 1 as elements µ Xjt t 1 α j X j,t 1 + λ j, j 1, 2. For 0, te conditional variance-covariance matrix γ Xt t 1 as diagonal and off-diagonal elements equal to CovX j,t+,x jt X j,t 1 α j 1 α j X j,t 1 + λ j and CovX j,t+,x it X j,t 1,X i,t 1 φ respectively, wile oterwise it is te zero matrix Estimation Te conditional density for te Poisson BINAR1 model can be obtained by substituting f 3 k, s e λ 1+λ 2 φ mink,s m0 λ 1 φ k m λ 2 φ s m φ m k m!s m!m! 3.7 in Ten we get fx t x t 1,α 1,α 2,λ 1,λ 2,φe λ 1+λ 2 φ x1,t 1 α x 1t k 1 1 α x 1t k 1 x 1,t 1 x 1t +k g 1 g 2 k0 s0 x2,t 1 x 2t s were g 1 minx 1t,x 1,t 1 andg 2 minx 2t,x 2,t 1. 7 mink,s m0 λ 1 φ k m λ 2 φ s m φ m k m!s m!m! α x 2t s 2 1 α 2 x 2,t 1 x 2t +s 3.8

8 3.2 A BINAR1 Process wit BVNB Innovations Model Assume tat te jpmf of te innovations {R 1t,R 2t } is a bivariate negative binomial distribution of te following form Marsall and Olkin, 1990; Boucer et al., 2008; Ceon et al., 2009: Γβ 1 + x + y P R 1t x, R 2t y Γβ 1 Γx +1Γy +1 x y λ 1 λ 1 + λ 2 + β 1 λ 2 λ 1 + λ 2 + β 1 β 1 λ 1 + λ 2 + β 1 β were λ 1,λ 2,β > 0. We will denote tis distribution as BVNBλ 1,λ 2,β. Note tat te marginal distribution of R jt is univariate negative binomial wit parameters β 1 and p j β 1 /λ j + β 1, j 1, 2 and tat te correlation between te two count variables R 1t and R 2t λ 1 λ 2 β Corrx, y λ 1 β1 + λ 2 β 3.10 must be positive. Tis assumption allows for more flexibility tan te Poisson BINAR1 model does, due to te involvement of te overdispersion parameter β in te model s specification. Recall tat in section 2.1, {R jt } was generally defined as an iid sequence wit mean λ j and variance σ 2 j υ j λ j, υ j > 0, j 1, 2. For te BVNB model, σ 2 j λ j 1 + βλ j implying tat υ j 1+βλ j, λ j,β > 0. Consequently υ j > 1 wic indicates te overdispersion case. However, te resulting model is not a BINAR model wit negative binomial marginals but a model tat effectively accounts for overdispersion. In specific, te statistical properties of te BINAR1 model wit BVNB innovations are encompassed in te vector of expectations µ Xt EX t wit elements µ Xjt λ j ; j 1, α j te variance-covariance matrix γ Xt wit diagonal and off-diagonal elements equal to CovX j,t+,x jt α j λ j 1 + βλ j + α j 1 α 2 j ; j 1, 2, 0, 1,

9 and CovX j,t+,x it α j βλ 1 λ 2 1 α 1 α 2 ; j i, 0, 1, respectively, and te correlation matrix ρ Xt wit diagonal elements CorrX j,t+,x jt αj ; j 1, 2, 0, 1, and off-diagonal elements αj β 1 α CorrX j,t+,x it 11 2 α2λ 2 1 λ 2 ; j i, 0, 1,... 1 α 1 α βλ 1 + α βλ 2 + α Conditionally on te previous observations X t 1 {X 1,t 1,X 2,t 1 },te vector of conditional means µ Xt t 1 EX t t 1 as elements µ Xjt t 1 α j X j,t 1 + λ j, j 1, 2. For 0, te conditional variance-covariance matrix γ Xt t 1 as diagonal and off-diagonal elements equal to CovX j,t+,x jt X j,t 1 α j 1 α j X j,t 1 + λ j 1 + βλ j and CovX j,t+,x it X j,t 1,X i,t 1 βλ 1 λ 2 respectively, wile oterwise it is te zero matrix Estimation For te BINAR1 model wit BVNB innovations it olds tat k s f 3 k, s Γβ 1 + k + s λ 1 λ 2 β 1 Γβ 1 k!s! λ 1 + λ 2 + β 1 λ 1 + λ 2 + β 1 λ 1 + λ 2 + β 1 Tus te conditional density 2.15 becomes β fx t x t 1,α 1,α 2,λ 1,λ 2,β g 1 g 2 Γβ 1 k s + k + s λ 1 λ 2 Γβ 1 k!s! λ k0 s0 1 + λ 2 + β 1 λ 1 + λ 2 + β 1 x1,t 1 α x 1t k 1 1 α x 1t k 1 x 1,t 1 x 1t +k x2,t 1 x 2t s were g 1 minx 1t,x 1,t 1 andg 2 minx 2t,x 2,t 1. 9 β 1 λ 1 + λ 2 + β 1 α x 2t s 2 1 α 2 x 2,t 1 x 2t +s β

10 3.3 Oter distributional coices As mentioned before, te coice of te joint distribution for R 1t and R 2t determines te properties of te underlying process. Wile te bivariate negative binomial provides overdispersion, it is interesting to note tat a selection of a distribution wit negative correlation can also produce negative correlation between te two series see 2.7. Te literature on bivariate count distributions wit negative correlation is limited. One of te reasons is tat negative correlation in bivariate counts occurs rater infrequently. However tere are suc models in te literature, as for example te bivariate Poisson-lognormal model of Aitcinson and Ho 1989 see also Cib and Winkelmann, 2001, te finite mixture model developed in Karlis and Meligkotsidou 2007 and models based on copulas see, e.g. Nikoloulopoulos and Karlis 2009 and te references terein. Finally, noted tat wile we used a certain bivariate negative binomial distribution, tere are certain oter alternatives in te literature wic could ave been used. We ave selected tis one mainly because of its relative simplicity. 4 Diagnostics In tis section we describe diagnostics for assessing te goodness of fit. Usually, in model fitting, tis is accomplised by means of residual analysis. However, due to te structural distinctiveness of INAR-type models, te classical definition of residuals as differences between te observed and fitted values, may prove to be inadequate as a diagnostic tool. We follow Freeland and McCabe 2004a by introducing a definition for residuals for count data tat distinguises between a set of residuals for te continuation process r 1t α X t 1 αx t 1 and anoter for te arrival component r 2t R t λ. In tis section we attempt to extend te ideas of Freeland and McCabe 2004a to te BINAR1 model. For eac one of te two series {X 1t,X 2t }, we define two sets of residuals; one for eac random component. So, for te continuation components we let r j 1t α j X j,t 1 α j X j,t 1 and for te arrival components we let r j 2t R jt λ j, j 1, 2. In order to arrive at a sensible and practical form of te above definitions, te unobservable quantities α j X j,t 1 and R jt sould be replaced wit E t [α j X j,t 1 ]ande t [R jt ] respectively, i.e. wit teir conditional expectations given te observed values of X jt and X j,t 1. PROPOSITION 1. Let E t [ ] denote te conditional expectation to te sigma field, I t σx j0,x j1,..., X jt, j 1, 2. For te BINAR1 model 10

11 wit bivariately distributed innovations te following equalities old: E t [α 1 X 1,t 1 ] α 1X 1,t 1 P x 1t 1 X 1,t 1 1,X 2,t 1 P x 1t X 1,t 1,X 2,t E t [α 2 X 2,t 1 ] α 2X 2,t 1 P x 2t 1 X 1,t 1,X 2,t P x 2t X 1,t 1,X 2,t 1 g 1 g 2 kf 1 x 1t kf 2 x 2t sf 3 k, s x E t [R 1t ] 2t k0 s0 4.3 P x 1t X 1,t 1,X 2,t 1 g 1 g 2 sf 1 x 1t kf 2 x 2t sf 3 k, s x E t [R 2t ] 1t k0 s0 4.4 P x 2t X 1,t 1,X 2,t 1 were te densities f 1 and f 2 are given in 2.13 and 2.14, f 3 k, s P R 1t k, R 2t s, g 1 minx 1t,x 1,t 1 andg 2 minx 2t,x 2,t 1. Using Proposition 1 we can now define te residuals as r j 1t r j 2t E t [r j 1t ]E t [α j X j,t 1 ] α j X j,t 1, and 4.5 E t [r j 2t ]E t [R jt ] λ j, for j 1, Regarding separately eac one of te two series tat comprise te BI- NAR1 model, it is noted tat adding te components of te two new sets of residuals gives te usual definition of residuals, i.e. r j 1t + r j 2t E t [α j X j,t 1 ] α j X j,t 1 + E t [R jt ] λ j E t [α j X j,t 1 + R jt ] α j X j,t 1 λ j X jt α j X j,t 1 λ j r j t. 4.7 Tus, te adequacy of eac component of te model may by assessed by plotting te aformentioned sets of residuals. 5 Forecasting Te usual way to produce forecasts in time series models is via te conditional forecast distribution. Freeland and McCabe 2004b establised te -stepaead conditional distribution of te Poisson INAR1 model, based on te remark of Al-Os and Alzaid 1987 tat 1 X t,x t d α X t + α i R t i,x t

12 were R t is a sequence of uncorrelated non-negative integer-valued random variables wit finite mean and variance. Te above result olds also for te marginal distribution of eac one of te two series X 1t,X 2t tat consist a BINAR1 model. As in te univariate case, αj X j,t X j,t, j 1, 2, as a binomial distribution wit parameters αj,x j,t. Moreover, te joint and marginal distributions of 1 αi 1 R 1,t i and 1 αi 2 R 2,t i are determined by te joint and marginal distributions { of X 1t and X 2t. Tis relation can be described 1 in terms of te jpgf of α1 i R 1,t i, 1 α2 i R 2,t i }. Denote by S j te quantity 1 αj i R 1,t j, j 1, 2. Ten, 1 G S1,S 2 s 1,s 2 G R1,R 2 1 α1 i + α1s i 1, 1 α2 i + α2s i Hence, te joint distribution of {X 1t,X 2t } given {X 1,t,X 2,t } is a convolution of two binomial distributions wit parameters α 1,X 1,t and α 2,X 2,t respectively, and a bivariate distribution wit jpgf of te form 5.2. Obviously, if 5.2 as not a closed-form expression, ten neiter te -step-aead forecast distribution can be specified in closed-form. However, it is straigtforward to evaluate it numerically. For te Poisson BINAR1 model, it can be proved tat [ 1 α G S1,S 2 s 1,s 2 exp 1 1 α 1 1 α + 1 α2 1 α 1 α 2 λ 1 s φs 1 1s α 2 1 α 2 ] λ 2 s wile te corresponding jpgf for te BINAR1 model wit BVNB innovations is given by 1 [ G S1,S 2 s 1,s 2 1 βλ1 α1s i 1 1 βλ 2 α2s i 2 1 ] β wic is not of a convenient form. Note owever tat irrespective of te jpgf of { 1 α1 i R 1,t i, 1 α i 2 R 2,t i }, closed-form expressions are available for teir conditional expectations and 12

13 variances conditional on X 1t,X 2t. More specifically, it can be proved tat 1 1 α E αj i j R j,t i λ j α j and Var 1 αj i R j,t i 1 α 2 j 1 α j υ 1 αj 2 j λ j + 1 α2 j 1 α j 1 αj 2 λ j 5.6 THEOREM 1. Tejpmf of {X 1,T +,X 2,T + } given {x 1T,x 2T } is given by f wit means, P X 1,T + x 1,X 2,T + x 2 x 1T,x 2T minx 1,x 1T minx 2,x 2T x1t x 1 k k0 x2t x 2 s 1 s0 α 2 x 2 s 1 α 2 x 2T x 2 +s α 1 x 1 k 1 α 1 x 1T x 1 +k 1 α1 i R 1,T + i k, α2 i R 2,T + i s x 1T,x 2T 1 α Ex j,t + x 2T,x 2T αj j x jt + ER jt ; j 1, 2, 1, 2,... 1 α j 5.7 and variances, 1 α 2 Varx j,t + x 1T,x 2T αj 1 αj j x jt + VarR 1 αj 2 jt 1 α j + 1 α2 j ER 1 α j 1 αj 2 jt ; j 1, 2, 1, 2, Te corresponding jpgf of {X 1,T +,X 2,T + } given {x 1T,x 2T } is of te form 13

14 G X1,T +,X 2,T + s 1,s 2 x 1T,x 2T 1 α 1+α 1s 1 X 1T 1 α 2+α 2s 2 X 2T G S1,S 2 s 1,s were G S1,S 2 s 1,s 2 is given in 5.2. Corollary 1. For te Poisson BINAR1 model, te jpgf and jpmf of {X 1,T +,X 2,T + } given {x 1T,x 2T } are given by G X1,T +,X 2,T + s 1,s 2 x 1T,x 2T 1 α1 + α1s 1 X 1T 1 α2 + α2s 2 X 2T { } 1 α exp 1 1 α λ 1 s α λ 2 s α2 φs 1 1s α 1 1 α 2 1 α 1 α 2 and P X 1,T + x 1,X 2,T + x 2 x 1T,x 2T minx 1,x 1T minx 2,x 2T x1t x 1 k k0 x2t s0 x 2 s { [ 1 α exp 1 1 α 1 mink,s m0 α 2 x 2 s 1 α 2 x 2T x 2 +s [ 1 α 1 1 α 1 λ 1 respectively, wit means, λ 1 + α 1 x 1 k 1 α 1 x 1T x 1 +k 1 α 2 1 α λ 2 1 α2 1 α 2 1 α 1 α 2 ] k m [ 1 α 2 1 α 2 λ 2 1 α 1 α 2 1 α 1 α 2 φ k m!s m!m! ]} φ 1 α 1 α 2 1 α 1 α 2 φ ] s m [ 1 α 1 α 2 1 α 1 α 2 φ ] m Ex j,t + x 1T,x 2T α j x jt + 1 α j 1 α j λ j ; j 1, 2, 1, 2, variances, Varx j,t + x 1T,x 2T αj 1 αj x jt + 1 α j λ j ; 1 α j j 1, 2, 1, 2, and covariance, 14

15 1 α Covx 1,T +,x 2,T + x 1T,x 2T 1 α2 1 α 1 α 2 φ ; 1, 2, Corollary 2. For te BINAR1 model wit BVNB innovations, te jpgf and te jpmf of {X 1,T +,X 2,T + } given {x 1T,x 2T } are given by and G X1,T +,X 2,T + s 1,s 2 x 1T,x 2T 1 α1 + α1s 1 X 1T 1 α2 + α2s 2 X 2T 1 [ 1 βλ1 α1s i 1 1 βλ 2 α2s i 2 1 ] β 1 f P X 1,T + x 1,X 2,T + x 2 x 1T,x 2T minx 1,x 1T minx 2,x 2T x1t x 1 k k0 x2t x 2 s 1 s0 α 2 x 2 s 1 α 2 x 2T x 2 +s α 1 x 1 k 1 α 1 x 1T x 1 +k 1 α1 i R 1,T + i k, α2 i R 2,T + i s x 1T,x 2T respectively, were f 1 α1 i R 1,T + i k, 1 can be numerically calculated. Te means and variances of tis process are given by, α i 2 R 2,T + i s x 1T,x 2T Ex j,t + x 1T,x 2T αj x jt + 1 α j λ j ; 1 α j j 1, 2, 1, 2, and 1 α 2 Varx j,t + x 1T,x 2T αj 1 αj j x jt βλ 1 αj 2 j λ j 1 α j + 1 α2 j λ 1 α j 1 αj 2 j ; j 1, 2, 1, 2,

16 wereas te covariance function is not of a closed-form. Te marginal probabilities P x 1 x 1T,x 2T andp x 2 x 1T,x 2T canbecalculated directly as, P x 1 x 1T,x 2T x 2 P x 1,x 2 x 1T,x 2T and P x 2 x 1T,x 2T x 1 P x 1,x 2 x 1T,x 2T respectively. Given te fact tat te vector of parameters θ is unknown, in practice we are only able to compute P x 1,x 2 x 1T,x 2T ; ˆθ were ˆθ are typically te maximum likeliood estimators introduced in section 2.2. Lack of knowledge about te true values of te model parameters and te need to estimate tem introduce uncertainty in te estimation of te -step-aead jpmf s. Estimation uncertainty, i.e. te error made in estimating tese probabilities, can be assessed by taking advantage of te asymptotic normality of ML estimators. Under standard regularity conditions, te ML estimator θ, denoted by ˆθ, is asymptotically normally distributed around te true parameter value, i.e. T ˆθ θ0 a N0, i 1, were i 1 is te inverse of te Fiser information matrix Bu and McCabe, Te δ-metod can ten be used for finding te asymptotic distribution of a random variable g ˆθ. An application of te δ-metod to g ˆθ P x x T ; ˆθ provides us wit a confidence interval for te probability associated wit any fixed value of x x 1,x 2 in te forecast distribution. Obviously, tese intervals may be truncated outside [0,1] Freeland and McCabe, 2004b. THEOREM 2 Freeland and McCabe, 2004b: Te quantity P x x T ; ˆθ as an asymptotically normal distribution wit mean P x x T ; θ 0 andvariance { } σx; 2 θ 0 T 1 P ˆθ i 1 P ˆθ 5.15 θθ0 θθ0 It is apparent tat analytical expressions for 5.15 are only available in cases were P x x T as a closed-form expression as is te case for te Poisson BINAR1 model see Appendix. Closing tis section, it is wort noting tat summarizing te forecast distribution by means of conditional expectations, wile ensures a minimum mean square error, it as drawbacks wit respect to data coerency since te integer-valued property of te time series is not taken into account. Freeland and McCabe 2004b suggest instead te use of te median of tis distribution wic always lies in te support of te series and is terefore coerent. Also, Pavlopoulos and Karlis 2008 propose a parametric bootstrap approac wic guarantees bot integer-valued predictions and prediction 16

17 intervals wit integer-valued ends. In our case, since te distributions are discrete, it is relatively easy to find te median to use as prediction instead of te mean, as te median will satisfy te discrete nature of te data. 6 Application Te data used in tis application refer to te joint modelling of daytime and nigttime road accidents in Scipol area, in te Neterlands for te year As nigttime accidents we refer to accidents appened between 10.00am-06.00pm, wile te rest were considered as daytime accidents. In accident analysis tose types of accidents are considered to ave different beavior. During nigttimes te traffic is of different nature e.g. more people travel for entertainment. On te oter and since bot types sare te same environment, like weater conditions, caracteristics of te road, tey are correlated. Te data are daily observations. Data from successive days are typically correlated as tey refer to similar conditions. Ignoring tis time series nature of te data can lead to incorrect inference see Brijs et al., Hence joint modelling of te two time series can be very useful. For suc data typically te autocorrelations are relatively small and tus AR1 models are enoug to capture te time dependence. Te data can be seen in figure 1. Te daytime and nigttime accidents ave mean values variances equal to and respectively implying ovedispersion. Te autocorrelation functions for bot series present a rater exponential decay wit a few exceptions. Te first order autocorrelation coefficient is 0.12 for daytime accidents and 0.13 for nigttime accidents. Finally te correlation between te two time series is revealing a sort of correlation between te series. In order to model te data we considered bot te bivariate Poisson INAR1 model and te INAR1 model wit BVNB innovations. Te results can be seen in Table 1. Comparing te log-likeliood one can see tat bot te time series context and te correlation between te series are needed. Te negative binomial BINAR1 model can also model te overdispersion and tus it provides te better fit. We ave also fitted a model wit bivariate Poisson lognormal innovations wic aving two more parameters offered very small improvement being muc more computationally demanding results are not presented ere. It is clear tat te time series models are better tan te models tat neglect tis. In addition te BVNB INAR1 model is muc better as it captures te overdispersion in te dataset togeter wit te correlation between te two series but also te autocorrelation witin eac series. Te standard 17

18 daytime number of accidents number of accidents days nigttime days Figure 1: Time series plots and acf plots for te series of daytime and nigttime accidents. 18

19 Table 1: Maximum Likeliood Estimates from fitting alternatively a BI- NAR1, two independent INAR1 models and a simple bivariate Poisson model. BINAR1 Independent INAR1 Biv. Poisson Neg.Bin BINAR1 Estimate SE Estimate SE Estimate SE Estimate SE ˆα ˆα ˆλ ˆλ ˆφ ˆβ Log-Lik AIC daytime accidents nigttime accidents errors of te estimates obtained by te two approaces standard errors are derived numerically from te Hessian sow tat fitting a BINAR1 model to te data generally improves te precision of te produced estimates. On te oter and it is apparent tat ignoring any form of te correlation eiter witin or between or te overdispersion leads to incorrect standard errors and ence incorrect inferences. Figures 2-4 and te related inference concern results obtained from te BINAR1 model wit BVNB innovations. Figure 2 includes te plots of te residuals of te two series. Since tese residuals ave not been standardized, te survival and arrival residuals add up to te Pearson residuals. Moreover, a large Pearson residual is comprised by a large survival and arrival residual, wile a small Pearson residual consists of a small survival and arrival residual. Te signs of survival and arrival residuals may also differ in some cases. However, tey still keep teir similarity in pattern. Anoter interesting point is te reflection of te model structure in te correlation between different pairs of residuals. More specifically, te sample correlations between te survival and arrival residuals of eac series are very ig: 0.72 for te daytime series and 0.67 for te nigttime series. Te arrival residuals of te two series are also significantly correlated at 0.16 depicting te structural assumption underlying te BINAR1 model tat te correlation between te two series as been introduced by using correlated innovation terms. Figure 3 sows te one-step-aead marginal predictive distributions P x 1,n+1 x 1n,x 2n andp x 2,n+1 x 1n,x 2n werex 1 corresponds to te daytime series and X 2 corresponds to te nigttime series. Te last observation was equal to 4 for te former series and equal to 1 for te latter series. As 19

20 one can see in Figure 3, bot distributions are skewed to te rigt wic is in accordance wit te sape of te negative binomial distribution. Te most probable one-step-aead predictive value is equal to 7 for te daytime series and equal to 1 for te nigttime series. Te larger dispersion of te series of daytime accidents compared wit te nigttime accidents series is also reflected in te plot of its predictive distribution. Figure 4 sows te observed values of te series of daytime and nigttime accidents wit te corresponding one-step-aead predictions. Te divergence between real data and forecasts is also portrayed. Te orizontal lines correspond to te observed mean values of te two series. Obviously, divergence is larger for observations tat lie far away from te mean. Tis seems to be expected since te one-step-aead predictions ave te same mean but are less dispersed tan te original series. Note also tat te correlation coefficient of te two series of forecasts is equal to te correlation coefficient of te real data series. daytime accidents Arrival Survival Pearson Observation Number nigttime accidents Arrival Survival Pearson Observation Number Figure 2: Non-standardized residuals of te daytime and nigttime accidents series. 20

21 prob prob daytime accidents nigttime accidents Figure 3: Te one-step-aead predictive distributions P x 1,T +1 x 1T,x 2T P x 2,T +1 x 1T,x 2T of te series of daytime and nigttime accidents respectively. Te last observed values n 365 are equal to 4 and 1 accordingly. 21

22 daytime accidents Data Forecasts divergence Observation Number nigttime accidents Data Forecasts divergence Observation Number Figure 4: Observed values of te series of daytime and nigttime accidents and te corresponding one-step-aead predictions.te orizontal lines correspond to te observed mean values of te two series. 22

23 7 Concluding Remarks Te main focus in tis paper is on bivariate time series for count data. Generally, te desired BINAR1 model can be constructed in two different ways: Te first approac prespecifies te form of te marginal distributions and subsequently identifies te required form of te distribution of te innovations in order for stationarity to old. In te second approac it is te coice of te form of te innovations distribution tat leads to te specification of te underlying marginal distributions. Te models proposed in tis paper ave been built following te last approac. In particular, we considered two different BINAR1 models, one wit bivariate Poisson innovations and anoter one wit bivariate negative binomial innovations. Te former specification as te useful property tat te joint distribution of te two series under consideration is also bivariate Poisson. In te latter case, we don t end up wit a bivariate negative binomial INAR1 process but we obtain a BINAR1 model tat effectively accounts for overdispersion. Deviations from te equidispersion restriction could alternatively accounted for by assuming anoter distribution for te innovations, e.g. mixed Poisson, or by te inclusion of appropriate regressors. Results on suc extensions will be reported elsewere. It is of course self-evident tat te proposed model is not a panacea. For example, wen significant correlation between te series under consideration is present at lags iger tan 1, fitting a BINAR1 model proves to be rater inadequate. Tus, extensions of te present model to iger orders would be a useful contribution to te improvement of its flexibility. Moreover, te structure of real-life data frequently implies need for te inclusion of bot autoregressive and moving average components wen for example seasonal patterns are observed in time series of counts. So, extending te bivariate INAR model to a bivariate INARMA model seems to be anoter interesting callenge. Finally, generalization of te proposed process to te multivariate case would provide a great opportunity for modelling more tan two time series of correlated count data. In tis case, te definition of a multivariate discrete distribution for te innovation process is needed. Te existing models ave certain limitations and tey do not lead to models wit well specified marginals. Hence inference can be difficult wit standard metods like maximum likeliood and some alternatives, like composite likeliood, sould be considered. 23

24 References J. Aitcinson and C. Ho. Te multivariate Poisson-log normal distribution. Biometrika, 75: , M.A. Al-Os and A.A. Alzaid. First-Order Integer-Valued Autoregressive Process. Journal of Time Series Analysis, 83: , J.P. Boucer, D. Micel, and G. Montserrat. Models of insurance claim counts wit time dependence based on generalization of poisson and negative binomial distributions. Variance, 21: , K. Brännäs and J. Nordström. A Bivariate Integer Valued Allocation Model for Guest Nigts in Hotels and Cottages. Umeå Economic Studies, 547, T. Brijs, D. Karlis, and G. Weets. Studying te effect of weater conditions on daily cras counts using a discrete time-series model. Accident Analysis and Prevention, 40: , R. Bu and B. McCabe. Model selection, estimation and forecasting in INARp models: A likeliood-based Markov Cain approac. International Journal of Forecasting, 24: , S. Ceon, S.H. Song, and B.C. Jung. Tests for independence in a bivariate negative binomial model. Journal of te Korean Statistical Society, 38: , S. Cib and R. Winkelmann. Markov Cain Monte Carlo Analysis of Correlated Count Data. Journal of Business & Economic Statistics, 194: , R.A. Davis, W.T. Dunsmuir, and Y. Wang. Modelling time series of count data, pages Ed. S. Gos, Marcel Dekker, R.K. Freeland and B.P.M. McCabe. Analysis of low count time series data by Poisson autoregression. Journal of Time Series Analysis, 255: , 2004a. R.K. Freeland and B.P.M. McCabe. Forecasting discrete valued low count time series. International Journal of Forecasting, 20: , 2004b. A. Heinen and E. Rengifo. Multivariate autoregressive modeling of time series count data using copulas. Journal of Empirical Finance, 14: ,

25 N. Jonson, S. Kotz, and N. Balakrisnan. Multivariate Discrete Distributions. Wiley, New York, R.C. Jung and A.R. Tremayne. Binomial tinning models for integer time series. Statistical Modelling, 6:21 96, D. Karlis and L. Meligkotsidou. Finite multivariate Poisson mixtures wit applications. Journal of Statistical Planning and Inference, 137: , S. Kocerlakota and K. Kocerlakota. Bivariate Discrete Distributions, Statistics: textbooks and monograps, volume 132. Markel Dekker, New York, A. Latour. Te Multivariate GINARp Process. Advances in Applied Probability, 291: , A.W. Marsall and I. Olkin. Multivarirate distribution generated from mixtures of convolution and product families, pages Topics in Statistical Dependence, Block,Sampson y Sanits Eds, Institute of Matematical Statistics, E. McKenzie. Some Simple Models for Discrete Variate Time Series. Water Resources Bulletin, 214: , E. McKenzie. Discrete Variate Time Series, volume 21. Sanbag, D. and Rao, C. eds., Elsevier Science, A.K. Nikoloulopoulos and D. Karlis. Finite normal mixture copulas for multivariate discrete data modeling. Journal of Statistical Planning and Inference In Press, H. Pavlopoulos and D. Karlis. INAR1 modeling of overdispersed count series wit an environmental application. Environmetrics, 194: , A.M.M.S. Quoresi. Bivariate Time Series Modeling of Financial Count Data. Communications in Statistics - Teory and Metods, 35: , N. Silva, I. Pereira, and E.S. Siva. Estimation and forecasting in SUINAR1. REVSTAT - Statistical Journal, 63: , F.W. Steutel and K. van Harn. Discrete Analogues of Self - Decomposability and Stability. Te Annals of Probability, 75: ,

26 Appendix I. PROOF OF EQUATION 2.6. CovX 1t,R 2t EX 1t R 2t EX 1t ER 2t EX 1t R 2t µ 1 λ [ 2 ] E α1 i R 1,t i R 2t µ 1 λ 2 α1 i {ER 1,t i R 2t } µ 1 λ 2 ER 1t R 2t + α1 i {ER 1,t 1 R 2t } µ 1 λ 2 i1 CovR 1t,R 2t + CovR 1t,R 2t α1 i {λ 1 λ 2 } µ 1 λ 2 II. ESTIMATION UNCERTAINTY OF THE POISSON BINAR1 MODEL. For te Poisson BINAR1 model, we let ˆθ T ˆα 1, ˆα 2, ˆλ 1, ˆλ 2, ˆφ bete ML estimators of θ α 1,α 2,λ 1,λ 2,φ based on a sample of size T. Ten, 5.15 can be written as σx; 2 θ 0 T 1 P α 1 P + λ 2 θθ0 P P + 2 α 1 λ 1 P P + 2 α 2 λ 1 P P + 2 λ 1 λ 2 θθ0 2 i 1 4,4 + θθ0 θθ0 θθ0 2 i 1 1,1 + P φ i 1 1,3 +2 i 1 2,3 +2 i 1 3,4 +2 P α 2 θθ0 P θθ0 2 i 1 5,5 +2 P α 1 λ 2 P α 2 λ 2 P λ 1 φ P P 2 i 1 2,2 + θθ0 θθ0 θθ0 2 P λ 1 i 1 3,3 θθ0 P P α 1 α 2 i 1 1,2 θθ0 i 1 P P 1,4 +2 α 1 φ i 1 P P 2,4 +2 α 2 φ i 1 P P 3,5 +2 λ 2 φ θθ0 θθ0 θθ0 i 1 1,5 i 1 2,5 i 1 4,5 } 26

27 were i 1 k,j is te k, j-element of te matrix i 1, k, j 1, 2,..., 5and P α 1 { x 1 α1 1T P x 1 1,x 2 x 1T 1,x 2T α1 1 P x 1,x 2 x 1T,x 2T } λ 11 α1 1 1 α1 {P 1 α1 2 x 1,x 2 x 1T,x 2T P x 1 1,x 2 x 1T,x 2T } + α 2φ1 α 1 1 α α 1α 2 1 α 1 α 2 2 {P x 1,x 2 x 1T,x 2T P x 1 1,x 2 x 1T,x 2T P x 1,x 2 1 x 1T,x 2T +P x 1 1,x 2 1 x 1T,x 2T } P α 2 P φ { x 1 α2 2T P x 1,x 2 1 x 1T,x 2T 1 α2 1 P x 1,x 2 x 1T,x 2T } λ 21 α2 1 1 α2 {P 1 α2 2 x 1,x 2 x 1T,x 2T P x 1,x 2 1 x 1T,x 2T } + α 1φ1 α 1 1 α α 1α 2 1 α 1 α 2 2 {P x 1,x 2 x 1T,x 2T P x 1 1,x 2 x 1T,x 2T P x 1,x 2 1 x 1T,x 2T +P x 1 1,x 2 1 x 1T,x 2T } P λ 1 P λ 2 1 α 1 1 α 1 1 α 2 1 α 2 {P x 1,x 2 x 1T,x 2T +P x 1 1,x 2 x 1T,x 2T } {P x 1,x 2 x 1T,x 2T +P x 1,x 2 1 x 1T,x 2T } 1 α 1 α2 {P x 1,x 2 x 1T,x 2T P x 1 1,x 2 x 1T,x 2T 1 α 1 α 2 P x 1,x 2 1 x 1T,x 2T +P x 1 1,x 2 1 x 1T,x 2T } In order to obtain analytical expressions for te elements tat comprise te Fiser information matrix i 1 we follow te notation of Freeland and McCabe 2004b and denote by l θ te second derivatives of te log-likeliood of te Poisson BINAR1 model wit respect to θ [α 1,α 2,λ 1,λ 2,φ] : l θ l α1 α 1 lα1 α 2 lα1 λ 1 lα1 λ 2 lα1 φ l α2 α 2 lα2 λ 1 lα2 λ 2 lα2 φ l λ1 λ 1 lλ1 λ 2 lλ1 φ l λ2 λ 2 lλ2 φ l φφ 27

28 Troug ordinary algebra it can be sown tat l α1 α α 1 2 T { 2x1,t 1 P x 1t 1,x 2t x 1,t 1 1,x 2,t 1 x 1,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 t1 + x 1,t 1x 1,t 1 1P x 1t 2,x 2t x 1,t 1 2,x 2,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 } 2 x1,t 1 P x 1t 1,x 2t x 1,t 1 1,x 2,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 l α2 α α 2 2 T { 2x2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 1 x 2,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 t1 + x 2,t 1x 2,t 1 1P x 1t,x 2t 2 x 1,t 1,x 2,t 1 2 P x 1t,x 2t x 1,t 1,x 2,t 1 } 2 x2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 1 P x 1t,x 2t x 1,t 1,x 2,t 1 l λ1 λ 1 l λ2 λ 2 T t1 T t1 { P x 1t 2,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 { P x 1t,x 2t 2 x 1,t 1,x 2,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 } 2 P x1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 } 2 P x1t,x 2t 1 x 1,t 1,x 2,t 1 P x 1t,x 2t x 1,t 1,x 2,t 1 28

29 l φφ T { 1 P x t1 1t,x 2t x 1,t 1,x 2,t 1 {2P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 2P x 1t 2,x 2t 1 x 1,t 1,x 2,t 1 2P x 1t 1,x 2t 2 x 1,t 1,x 2,t 1 +Px 1t 2,x 2t 2 x 1,t 1,x 2,t 1 + P x 1t 2,x 2t x 1,t 1,x 2,t 1 +Px 1t,x 2t 2 x 1,t 1,x 2,t 1 } + 1 P 2 x 1t,x 2t x 1,t 1,x 2,t 1 {2P x 1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 + 2P x 1t,x 2t 1 x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 2P x 1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 P 2 x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 P 2 x 1t 1,x 2t x 1,t 1,x 2,t 1 P 2 x 1t,x 2t 1 x 1,t 1,x 2,t 1 } } l α1 α 2 T { x 1,t 1 x 2,t 1 1 α t1 1 1 α 2 P 2 x 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t,x 2t x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1 1,x 2,t 1 1 } P x 1t 1,x 2t x 1,t 1 1,x 2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 1} l α1 λ 1 T { x 1,t 1 1 α t1 1 P 2 x 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t,x 2t x 1,t 1,x 2,t 1 P x 1t 2,x 2t x 1,t 1 1,x 2,t 1 } P x 1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t 1,x 2t x 1,t 1 1,x 2,t 1 } l α2 λ 2 T { x 2,t 1 1 α t1 2 P 2 x 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t 2 x 1,t 1,x 2,t 1 1 } P x 1t,x 2t 1 x 1,t 1,x 2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 1} 29

30 l α1 λ 2 T { x 1,t 1 1 α t1 1 P 2 x 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t,x 2t x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1 1,x 2,t 1 } P x 1t,x 2t 1 x 1,t 1,x 2,t 1 P x 1t 1,x 2t x 1,t 1 1,x 2,t 1 } l α2 λ 1 T { x 2,t 1 1 α t1 2 P 2 x 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t,x 2t x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 1 } P x 1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 1} l α1 φ T { { x1,t α t1 1 P x 1t,x 2t x 1,t 1,x 2,t 1 [P x 1t 2,x 2t 1 x 1,t 1 1,x 2,t 1 P x 1t 2,x 2t x 1,t 1 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1 1,x 2,t 1 ] P x 1t 1,x 2t x 1,t 1 1,x 2,t 1 [P x P 2 1t 1,x 2t 1 x 1,t 1,x 2,t 1 x 1t,x 2t x 1,t 1,x 2,t 1 }} P x 1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 ] l α2 φ T { { x2,t α t1 2 P x 1t,x 2t x 1,t 1,x 2,t 1 [P x 1t 1,x 2t 2 x 1,t 1,x 2,t 1 1 P x 1t,x 2t 2 x 1,t 1,x 2,t 1 1 P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 1] P x 1t,x 2t 1 x 1,t 1,x 2,t 1 1 [P x P 2 1t 1,x 2t 1 x 1,t 1,x 2,t 1 x 1t,x 2t x 1,t 1,x 2,t 1 }} P x 1t,x 2t 1 x 1,t 1,x 2,t 1 P x 1t 1,x 2t x 1,t 1,x 2,t 1 ] l λ1 λ 2 T { 1 P 2 x t1 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t,x 2t x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 } P x 1t,x 2t 1 x 1,t 1,x 2,t 1 P x 1t 1,x 2t x 1,t 1,x 2,t 1 } 30

31 l λ1 φ T { 1 P x t1 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t 2,x 2t 1 x 1,t 1,x 2,t 1 P x 1t 2,x 2t x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 } P x 1t 1,x 2t x 1,t 1,x 2,t 1 {P x P 2 1t 1,x 2t 1 x 1,t 1,x 2,t 1 x 1t,x 2t x 1,t 1,x 2,t 1 } P x 1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 } l λ2 φ T { 1 P x t1 1t,x 2t x 1,t 1,x 2,t 1 {P x 1t 1,x 2t 2 x 1,t 1,x 2,t 1 P x 1t,x 2t 2 x 1,t 1,x 2,t 1 P x 1t 1,x 2t 1 x 1,t 1,x 2,t 1 } P x 1t,x 2t 1 x 1,t 1,x 2,t 1 {P x P 2 1t 1,x 2t 1 x 1,t 1,x 2,t 1 x 1t,x 2t x 1,t 1,x 2,t 1 } P x 1t 1,x 2t x 1,t 1,x 2,t 1 P x 1t,x 2t 1 x 1,t 1,x 2,t 1 } Note tat, in contrast to te univariate case, te information as well as te scores of te Poisson BINAR1 model cannot be decomposed into quantities associated wit eac component of te model seperately. Tis barrier is just due to te model s structure, i.e. to its bivariate nature. Te Fiser information matrix i can ten be calculated as usual: [ ] 2 lθ ] i E θ E [ lθ θ θ 2 were lθ is te log-likeliood of te Poisson BINAR1 model. 31

ATHENS UNIVERSITY OF ECONOMICS

ATHENS UNIVERSITY OF ECONOMICS ATHENS UNIVERSITY OF ECONOMICS BIVARIATE INAR1 MODELS Xanthi Pedeli and Dimitris Karlis Department of Statistics Athens University of Economics and Business Technical Report No 247, June, 2009 DEPARTMENT

More information

Advances on time series for multivariate counts

Advances on time series for multivariate counts Advances on time series for multivariate counts Dimitris Karlis Department of Statistics Athens University of Economics Besançon, July 2018 Outline 1 Introduction 2 A bivariate INAR(1) process The BINAR(1)

More information

Financial Econometrics Prof. Massimo Guidolin

Financial Econometrics Prof. Massimo Guidolin CLEFIN A.A. 2010/2011 Financial Econometrics Prof. Massimo Guidolin A Quick Review of Basic Estimation Metods 1. Were te OLS World Ends... Consider two time series 1: = { 1 2 } and 1: = { 1 2 }. At tis

More information

A = h w (1) Error Analysis Physics 141

A = h w (1) Error Analysis Physics 141 Introduction In all brances of pysical science and engineering one deals constantly wit numbers wic results more or less directly from experimental observations. Experimental observations always ave inaccuracies.

More information

4. The slope of the line 2x 7y = 8 is (a) 2/7 (b) 7/2 (c) 2 (d) 2/7 (e) None of these.

4. The slope of the line 2x 7y = 8 is (a) 2/7 (b) 7/2 (c) 2 (d) 2/7 (e) None of these. Mat 11. Test Form N Fall 016 Name. Instructions. Te first eleven problems are wort points eac. Te last six problems are wort 5 points eac. For te last six problems, you must use relevant metods of algebra

More information

EFFICIENCY OF MODEL-ASSISTED REGRESSION ESTIMATORS IN SAMPLE SURVEYS

EFFICIENCY OF MODEL-ASSISTED REGRESSION ESTIMATORS IN SAMPLE SURVEYS Statistica Sinica 24 2014, 395-414 doi:ttp://dx.doi.org/10.5705/ss.2012.064 EFFICIENCY OF MODEL-ASSISTED REGRESSION ESTIMATORS IN SAMPLE SURVEYS Jun Sao 1,2 and Seng Wang 3 1 East Cina Normal University,

More information

Stationary Gaussian Markov Processes As Limits of Stationary Autoregressive Time Series

Stationary Gaussian Markov Processes As Limits of Stationary Autoregressive Time Series Stationary Gaussian Markov Processes As Limits of Stationary Autoregressive Time Series Lawrence D. Brown, Pilip A. Ernst, Larry Sepp, and Robert Wolpert August 27, 2015 Abstract We consider te class,

More information

Polynomial Interpolation

Polynomial Interpolation Capter 4 Polynomial Interpolation In tis capter, we consider te important problem of approximatinga function fx, wose values at a set of distinct points x, x, x,, x n are known, by a polynomial P x suc

More information

Polynomial Interpolation

Polynomial Interpolation Capter 4 Polynomial Interpolation In tis capter, we consider te important problem of approximating a function f(x, wose values at a set of distinct points x, x, x 2,,x n are known, by a polynomial P (x

More information

Katz Family of Distributions and Processes

Katz Family of Distributions and Processes CHAPTER 7 Katz Family of Distributions and Processes 7. Introduction The Poisson distribution and the Negative binomial distribution are the most widely used discrete probability distributions for the

More information

arxiv: v1 [math.pr] 28 Dec 2018

arxiv: v1 [math.pr] 28 Dec 2018 Approximating Sepp s constants for te Slepian process Jack Noonan a, Anatoly Zigljavsky a, a Scool of Matematics, Cardiff University, Cardiff, CF4 4AG, UK arxiv:8.0v [mat.pr] 8 Dec 08 Abstract Slepian

More information

Application of copula-based BINAR models in loan modelling

Application of copula-based BINAR models in loan modelling Application of copula-based BINAR models in loan modelling Andrius Buteikis 1, Remigijus Leipus 1,2 1 Faculty of Mathematics and Informatics, Vilnius University, Naugarduko 24, Vilnius LT-03225, Lithuania

More information

Volume 29, Issue 3. Existence of competitive equilibrium in economies with multi-member households

Volume 29, Issue 3. Existence of competitive equilibrium in economies with multi-member households Volume 29, Issue 3 Existence of competitive equilibrium in economies wit multi-member ouseolds Noriisa Sato Graduate Scool of Economics, Waseda University Abstract Tis paper focuses on te existence of

More information

Numerical Differentiation

Numerical Differentiation Numerical Differentiation Finite Difference Formulas for te first derivative (Using Taylor Expansion tecnique) (section 8.3.) Suppose tat f() = g() is a function of te variable, and tat as 0 te function

More information

New families of estimators and test statistics in log-linear models

New families of estimators and test statistics in log-linear models Journal of Multivariate Analysis 99 008 1590 1609 www.elsevier.com/locate/jmva ew families of estimators and test statistics in log-linear models irian Martín a,, Leandro Pardo b a Department of Statistics

More information

Handling Missing Data on Asymmetric Distribution

Handling Missing Data on Asymmetric Distribution International Matematical Forum, Vol. 8, 03, no. 4, 53-65 Handling Missing Data on Asymmetric Distribution Amad M. H. Al-Kazale Department of Matematics, Faculty of Science Al-albayt University, Al-Mafraq-Jordan

More information

How to Find the Derivative of a Function: Calculus 1

How to Find the Derivative of a Function: Calculus 1 Introduction How to Find te Derivative of a Function: Calculus 1 Calculus is not an easy matematics course Te fact tat you ave enrolled in suc a difficult subject indicates tat you are interested in te

More information

Consider a function f we ll specify which assumptions we need to make about it in a minute. Let us reformulate the integral. 1 f(x) dx.

Consider a function f we ll specify which assumptions we need to make about it in a minute. Let us reformulate the integral. 1 f(x) dx. Capter 2 Integrals as sums and derivatives as differences We now switc to te simplest metods for integrating or differentiating a function from its function samples. A careful study of Taylor expansions

More information

THE ROYAL STATISTICAL SOCIETY GRADUATE DIPLOMA EXAMINATION MODULE 5

THE ROYAL STATISTICAL SOCIETY GRADUATE DIPLOMA EXAMINATION MODULE 5 THE ROYAL STATISTICAL SOCIETY GRADUATE DIPLOMA EXAMINATION NEW MODULAR SCHEME introduced from te examinations in 009 MODULE 5 SOLUTIONS FOR SPECIMEN PAPER B THE QUESTIONS ARE CONTAINED IN A SEPARATE FILE

More information

Continuous Stochastic Processes

Continuous Stochastic Processes Continuous Stocastic Processes Te term stocastic is often applied to penomena tat vary in time, wile te word random is reserved for penomena tat vary in space. Apart from tis distinction, te modelling

More information

Time (hours) Morphine sulfate (mg)

Time (hours) Morphine sulfate (mg) Mat Xa Fall 2002 Review Notes Limits and Definition of Derivative Important Information: 1 According to te most recent information from te Registrar, te Xa final exam will be eld from 9:15 am to 12:15

More information

EFFICIENT REPLICATION VARIANCE ESTIMATION FOR TWO-PHASE SAMPLING

EFFICIENT REPLICATION VARIANCE ESTIMATION FOR TWO-PHASE SAMPLING Statistica Sinica 13(2003), 641-653 EFFICIENT REPLICATION VARIANCE ESTIMATION FOR TWO-PHASE SAMPLING J. K. Kim and R. R. Sitter Hankuk University of Foreign Studies and Simon Fraser University Abstract:

More information

The derivative function

The derivative function Roberto s Notes on Differential Calculus Capter : Definition of derivative Section Te derivative function Wat you need to know already: f is at a point on its grap and ow to compute it. Wat te derivative

More information

First-Order Fractionally Integrated Non-Separable Spatial Autoregressive (FINSSAR(1,1)) Model and Some of its Properties

First-Order Fractionally Integrated Non-Separable Spatial Autoregressive (FINSSAR(1,1)) Model and Some of its Properties First-Order Fractionally Integrated Non-Separable Spatial Autoregressive (FINSSAR(,)) Model and Some of its Properties Alireza Godsi Department of Matematics, Faculty of Science University Putra, Malaysia.

More information

New Distribution Theory for the Estimation of Structural Break Point in Mean

New Distribution Theory for the Estimation of Structural Break Point in Mean New Distribution Teory for te Estimation of Structural Break Point in Mean Liang Jiang Singapore Management University Xiaou Wang Te Cinese University of Hong Kong Jun Yu Singapore Management University

More information

Bootstrap prediction intervals for Markov processes

Bootstrap prediction intervals for Markov processes arxiv: arxiv:0000.0000 Bootstrap prediction intervals for Markov processes Li Pan and Dimitris N. Politis Li Pan Department of Matematics University of California San Diego La Jolla, CA 92093-0112, USA

More information

3.1 Extreme Values of a Function

3.1 Extreme Values of a Function .1 Etreme Values of a Function Section.1 Notes Page 1 One application of te derivative is finding minimum and maimum values off a grap. In precalculus we were only able to do tis wit quadratics by find

More information

Mathematics 105 Calculus I. Exam 1. February 13, Solution Guide

Mathematics 105 Calculus I. Exam 1. February 13, Solution Guide Matematics 05 Calculus I Exam February, 009 Your Name: Solution Guide Tere are 6 total problems in tis exam. On eac problem, you must sow all your work, or oterwise torougly explain your conclusions. Tere

More information

Differentiation in higher dimensions

Differentiation in higher dimensions Capter 2 Differentiation in iger dimensions 2.1 Te Total Derivative Recall tat if f : R R is a 1-variable function, and a R, we say tat f is differentiable at x = a if and only if te ratio f(a+) f(a) tends

More information

The Laplace equation, cylindrically or spherically symmetric case

The Laplace equation, cylindrically or spherically symmetric case Numerisce Metoden II, 7 4, und Übungen, 7 5 Course Notes, Summer Term 7 Some material and exercises Te Laplace equation, cylindrically or sperically symmetric case Electric and gravitational potential,

More information

Reflection Symmetries of q-bernoulli Polynomials

Reflection Symmetries of q-bernoulli Polynomials Journal of Nonlinear Matematical Pysics Volume 1, Supplement 1 005, 41 4 Birtday Issue Reflection Symmetries of q-bernoulli Polynomials Boris A KUPERSHMIDT Te University of Tennessee Space Institute Tullaoma,

More information

Analytic Functions. Differentiable Functions of a Complex Variable

Analytic Functions. Differentiable Functions of a Complex Variable Analytic Functions Differentiable Functions of a Complex Variable In tis capter, we sall generalize te ideas for polynomials power series of a complex variable we developed in te previous capter to general

More information

1. Questions (a) through (e) refer to the graph of the function f given below. (A) 0 (B) 1 (C) 2 (D) 4 (E) does not exist

1. Questions (a) through (e) refer to the graph of the function f given below. (A) 0 (B) 1 (C) 2 (D) 4 (E) does not exist Mat 1120 Calculus Test 2. October 18, 2001 Your name Te multiple coice problems count 4 points eac. In te multiple coice section, circle te correct coice (or coices). You must sow your work on te oter

More information

SIMG-713 Homework 5 Solutions

SIMG-713 Homework 5 Solutions SIMG-73 Homework 5 Solutions Spring 00. Potons strike a detector at an average rate of λ potons per second. Te detector produces an output wit probability β wenever it is struck by a poton. Compute te

More information

Simulation and verification of a plate heat exchanger with a built-in tap water accumulator

Simulation and verification of a plate heat exchanger with a built-in tap water accumulator Simulation and verification of a plate eat excanger wit a built-in tap water accumulator Anders Eriksson Abstract In order to test and verify a compact brazed eat excanger (CBE wit a built-in accumulation

More information

Preface. Here are a couple of warnings to my students who may be here to get a copy of what happened on a day that you missed.

Preface. Here are a couple of warnings to my students who may be here to get a copy of what happened on a day that you missed. Preface Here are my online notes for my course tat I teac ere at Lamar University. Despite te fact tat tese are my class notes, tey sould be accessible to anyone wanting to learn or needing a refreser

More information

Kernel Density Estimation

Kernel Density Estimation Kernel Density Estimation Univariate Density Estimation Suppose tat we ave a random sample of data X 1,..., X n from an unknown continuous distribution wit probability density function (pdf) f(x) and cumulative

More information

7 Semiparametric Methods and Partially Linear Regression

7 Semiparametric Methods and Partially Linear Regression 7 Semiparametric Metods and Partially Linear Regression 7. Overview A model is called semiparametric if it is described by and were is nite-dimensional (e.g. parametric) and is in nite-dimensional (nonparametric).

More information

Section 3.1: Derivatives of Polynomials and Exponential Functions

Section 3.1: Derivatives of Polynomials and Exponential Functions Section 3.1: Derivatives of Polynomials and Exponential Functions In previous sections we developed te concept of te derivative and derivative function. Te only issue wit our definition owever is tat it

More information

lecture 26: Richardson extrapolation

lecture 26: Richardson extrapolation 43 lecture 26: Ricardson extrapolation 35 Ricardson extrapolation, Romberg integration Trougout numerical analysis, one encounters procedures tat apply some simple approximation (eg, linear interpolation)

More information

Click here to see an animation of the derivative

Click here to see an animation of the derivative Differentiation Massoud Malek Derivative Te concept of derivative is at te core of Calculus; It is a very powerful tool for understanding te beavior of matematical functions. It allows us to optimize functions,

More information

Solution. Solution. f (x) = (cos x)2 cos(2x) 2 sin(2x) 2 cos x ( sin x) (cos x) 4. f (π/4) = ( 2/2) ( 2/2) ( 2/2) ( 2/2) 4.

Solution. Solution. f (x) = (cos x)2 cos(2x) 2 sin(2x) 2 cos x ( sin x) (cos x) 4. f (π/4) = ( 2/2) ( 2/2) ( 2/2) ( 2/2) 4. December 09, 20 Calculus PracticeTest s Name: (4 points) Find te absolute extrema of f(x) = x 3 0 on te interval [0, 4] Te derivative of f(x) is f (x) = 3x 2, wic is zero only at x = 0 Tus we only need

More information

Material for Difference Quotient

Material for Difference Quotient Material for Difference Quotient Prepared by Stepanie Quintal, graduate student and Marvin Stick, professor Dept. of Matematical Sciences, UMass Lowell Summer 05 Preface Te following difference quotient

More information

. If lim. x 2 x 1. f(x+h) f(x)

. If lim. x 2 x 1. f(x+h) f(x) Review of Differential Calculus Wen te value of one variable y is uniquely determined by te value of anoter variable x, ten te relationsip between x and y is described by a function f tat assigns a value

More information

3.4 Worksheet: Proof of the Chain Rule NAME

3.4 Worksheet: Proof of the Chain Rule NAME Mat 1170 3.4 Workseet: Proof of te Cain Rule NAME Te Cain Rule So far we are able to differentiate all types of functions. For example: polynomials, rational, root, and trigonometric functions. We are

More information

Exam 1 Review Solutions

Exam 1 Review Solutions Exam Review Solutions Please also review te old quizzes, and be sure tat you understand te omework problems. General notes: () Always give an algebraic reason for your answer (graps are not sufficient),

More information

A MONTE CARLO ANALYSIS OF THE EFFECTS OF COVARIANCE ON PROPAGATED UNCERTAINTIES

A MONTE CARLO ANALYSIS OF THE EFFECTS OF COVARIANCE ON PROPAGATED UNCERTAINTIES A MONTE CARLO ANALYSIS OF THE EFFECTS OF COVARIANCE ON PROPAGATED UNCERTAINTIES Ronald Ainswort Hart Scientific, American Fork UT, USA ABSTRACT Reports of calibration typically provide total combined uncertainties

More information

Te comparison of dierent models M i is based on teir relative probabilities, wic can be expressed, again using Bayes' teorem, in terms of prior probab

Te comparison of dierent models M i is based on teir relative probabilities, wic can be expressed, again using Bayes' teorem, in terms of prior probab To appear in: Advances in Neural Information Processing Systems 9, eds. M. C. Mozer, M. I. Jordan and T. Petsce. MIT Press, 997 Bayesian Model Comparison by Monte Carlo Caining David Barber D.Barber@aston.ac.uk

More information

Order of Accuracy. ũ h u Ch p, (1)

Order of Accuracy. ũ h u Ch p, (1) Order of Accuracy 1 Terminology We consider a numerical approximation of an exact value u. Te approximation depends on a small parameter, wic can be for instance te grid size or time step in a numerical

More information

Derivatives. By: OpenStaxCollege

Derivatives. By: OpenStaxCollege By: OpenStaxCollege Te average teen in te United States opens a refrigerator door an estimated 25 times per day. Supposedly, tis average is up from 10 years ago wen te average teenager opened a refrigerator

More information

= 0 and states ''hence there is a stationary point'' All aspects of the proof dx must be correct (c)

= 0 and states ''hence there is a stationary point'' All aspects of the proof dx must be correct (c) Paper 1: Pure Matematics 1 Mark Sceme 1(a) (i) (ii) d d y 3 1x 4x x M1 A1 d y dx 1.1b 1.1b 36x 48x A1ft 1.1b Substitutes x = into teir dx (3) 3 1 4 Sows d y 0 and states ''ence tere is a stationary point''

More information

More on generalized inverses of partitioned matrices with Banachiewicz-Schur forms

More on generalized inverses of partitioned matrices with Banachiewicz-Schur forms More on generalized inverses of partitioned matrices wit anaciewicz-scur forms Yongge Tian a,, Yosio Takane b a Cina Economics and Management cademy, Central University of Finance and Economics, eijing,

More information

Bootstrap confidence intervals in nonparametric regression without an additive model

Bootstrap confidence intervals in nonparametric regression without an additive model Bootstrap confidence intervals in nonparametric regression witout an additive model Dimitris N. Politis Abstract Te problem of confidence interval construction in nonparametric regression via te bootstrap

More information

Chapter 2 Performance Analysis of Call-Handling Processes in Buffered Cellular Wireless Networks

Chapter 2 Performance Analysis of Call-Handling Processes in Buffered Cellular Wireless Networks Capter 2 Performance Analysis of Call-Handling Processes in Buffered Cellular Wireless Networks In tis capter effective numerical computational procedures to calculate QoS (Quality of Service) metrics

More information

1 The concept of limits (p.217 p.229, p.242 p.249, p.255 p.256) 1.1 Limits Consider the function determined by the formula 3. x since at this point

1 The concept of limits (p.217 p.229, p.242 p.249, p.255 p.256) 1.1 Limits Consider the function determined by the formula 3. x since at this point MA00 Capter 6 Calculus and Basic Linear Algebra I Limits, Continuity and Differentiability Te concept of its (p.7 p.9, p.4 p.49, p.55 p.56). Limits Consider te function determined by te formula f Note

More information

HOMEWORK HELP 2 FOR MATH 151

HOMEWORK HELP 2 FOR MATH 151 HOMEWORK HELP 2 FOR MATH 151 Here we go; te second round of omework elp. If tere are oters you would like to see, let me know! 2.4, 43 and 44 At wat points are te functions f(x) and g(x) = xf(x)continuous,

More information

Tail Conditional Expectations for Extended Exponential Dispersion Models

Tail Conditional Expectations for Extended Exponential Dispersion Models American Researc Journal of Matematics Original Article ISSN 378-704 Volume 1 Issue 4 015 Tail Conditional Expectations for Extended Exponential Dispersion Models Ye (Zoe) Ye Qiang Wu and Don Hong 1 Program

More information

REVIEW LAB ANSWER KEY

REVIEW LAB ANSWER KEY REVIEW LAB ANSWER KEY. Witout using SN, find te derivative of eac of te following (you do not need to simplify your answers): a. f x 3x 3 5x x 6 f x 3 3x 5 x 0 b. g x 4 x x x notice te trick ere! x x g

More information

5 Ordinary Differential Equations: Finite Difference Methods for Boundary Problems

5 Ordinary Differential Equations: Finite Difference Methods for Boundary Problems 5 Ordinary Differential Equations: Finite Difference Metods for Boundary Problems Read sections 10.1, 10.2, 10.4 Review questions 10.1 10.4, 10.8 10.9, 10.13 5.1 Introduction In te previous capters we

More information

Efficient algorithms for for clone items detection

Efficient algorithms for for clone items detection Efficient algoritms for for clone items detection Raoul Medina, Caroline Noyer, and Olivier Raynaud Raoul Medina, Caroline Noyer and Olivier Raynaud LIMOS - Université Blaise Pascal, Campus universitaire

More information

Quantum Numbers and Rules

Quantum Numbers and Rules OpenStax-CNX module: m42614 1 Quantum Numbers and Rules OpenStax College Tis work is produced by OpenStax-CNX and licensed under te Creative Commons Attribution License 3.0 Abstract Dene quantum number.

More information

Chapter 4: Numerical Methods for Common Mathematical Problems

Chapter 4: Numerical Methods for Common Mathematical Problems 1 Capter 4: Numerical Metods for Common Matematical Problems Interpolation Problem: Suppose we ave data defined at a discrete set of points (x i, y i ), i = 0, 1,..., N. Often it is useful to ave a smoot

More information

Discriminate Modelling of Peak and Off-Peak Motorway Capacity

Discriminate Modelling of Peak and Off-Peak Motorway Capacity International Journal of Integrated Engineering - Special Issue on ICONCEES Vol. 4 No. 3 (2012) p. 53-58 Discriminate Modelling of Peak and Off-Peak Motorway Capacity Hasim Moammed Alassan 1,*, Sundara

More information

MA455 Manifolds Solutions 1 May 2008

MA455 Manifolds Solutions 1 May 2008 MA455 Manifolds Solutions 1 May 2008 1. (i) Given real numbers a < b, find a diffeomorpism (a, b) R. Solution: For example first map (a, b) to (0, π/2) and ten map (0, π/2) diffeomorpically to R using

More information

Kernel Density Based Linear Regression Estimate

Kernel Density Based Linear Regression Estimate Kernel Density Based Linear Regression Estimate Weixin Yao and Zibiao Zao Abstract For linear regression models wit non-normally distributed errors, te least squares estimate (LSE will lose some efficiency

More information

Lecture 15. Interpolation II. 2 Piecewise polynomial interpolation Hermite splines

Lecture 15. Interpolation II. 2 Piecewise polynomial interpolation Hermite splines Lecture 5 Interpolation II Introduction In te previous lecture we focused primarily on polynomial interpolation of a set of n points. A difficulty we observed is tat wen n is large, our polynomial as to

More information

1 Calculus. 1.1 Gradients and the Derivative. Q f(x+h) f(x)

1 Calculus. 1.1 Gradients and the Derivative. Q f(x+h) f(x) Calculus. Gradients and te Derivative Q f(x+) δy P T δx R f(x) 0 x x+ Let P (x, f(x)) and Q(x+, f(x+)) denote two points on te curve of te function y = f(x) and let R denote te point of intersection of

More information

Regularized Regression

Regularized Regression Regularized Regression David M. Blei Columbia University December 5, 205 Modern regression problems are ig dimensional, wic means tat te number of covariates p is large. In practice statisticians regularize

More information

Pre-Calculus Review Preemptive Strike

Pre-Calculus Review Preemptive Strike Pre-Calculus Review Preemptive Strike Attaced are some notes and one assignment wit tree parts. Tese are due on te day tat we start te pre-calculus review. I strongly suggest reading troug te notes torougly

More information

Applied Linear Statistical Models. Simultaneous Inference Topics. Simultaneous Estimation of β 0 and β 1 Issues. Simultaneous Inference. Dr.

Applied Linear Statistical Models. Simultaneous Inference Topics. Simultaneous Estimation of β 0 and β 1 Issues. Simultaneous Inference. Dr. Applied Linear Statistical Models Simultaneous Inference Dr. DH Jones Simultaneous Inference Topics Simultaneous estimation of β 0 and β 1 Bonferroni Metod Simultaneous estimation of several mean responses

More information

IEOR 165 Lecture 10 Distribution Estimation

IEOR 165 Lecture 10 Distribution Estimation IEOR 165 Lecture 10 Distribution Estimation 1 Motivating Problem Consider a situation were we ave iid data x i from some unknown distribution. One problem of interest is estimating te distribution tat

More information

Bounds on the Moments for an Ensemble of Random Decision Trees

Bounds on the Moments for an Ensemble of Random Decision Trees Noname manuscript No. (will be inserted by te editor) Bounds on te Moments for an Ensemble of Random Decision Trees Amit Durandar Received: Sep. 17, 2013 / Revised: Mar. 04, 2014 / Accepted: Jun. 30, 2014

More information

Copyright c 2008 Kevin Long

Copyright c 2008 Kevin Long Lecture 4 Numerical solution of initial value problems Te metods you ve learned so far ave obtained closed-form solutions to initial value problems. A closedform solution is an explicit algebriac formula

More information

LIMITATIONS OF EULER S METHOD FOR NUMERICAL INTEGRATION

LIMITATIONS OF EULER S METHOD FOR NUMERICAL INTEGRATION LIMITATIONS OF EULER S METHOD FOR NUMERICAL INTEGRATION LAURA EVANS.. Introduction Not all differential equations can be explicitly solved for y. Tis can be problematic if we need to know te value of y

More information

2.8 The Derivative as a Function

2.8 The Derivative as a Function .8 Te Derivative as a Function Typically, we can find te derivative of a function f at many points of its domain: Definition. Suppose tat f is a function wic is differentiable at every point of an open

More information

1.5 Functions and Their Rates of Change

1.5 Functions and Their Rates of Change 66_cpp-75.qd /6/8 4:8 PM Page 56 56 CHAPTER Introduction to Functions and Graps.5 Functions and Teir Rates of Cange Identif were a function is increasing or decreasing Use interval notation Use and interpret

More information

SECTION 1.10: DIFFERENCE QUOTIENTS LEARNING OBJECTIVES

SECTION 1.10: DIFFERENCE QUOTIENTS LEARNING OBJECTIVES (Section.0: Difference Quotients).0. SECTION.0: DIFFERENCE QUOTIENTS LEARNING OBJECTIVES Define average rate of cange (and average velocity) algebraically and grapically. Be able to identify, construct,

More information

HOW TO DEAL WITH FFT SAMPLING INFLUENCES ON ADEV CALCULATIONS

HOW TO DEAL WITH FFT SAMPLING INFLUENCES ON ADEV CALCULATIONS HOW TO DEAL WITH FFT SAMPLING INFLUENCES ON ADEV CALCULATIONS Po-Ceng Cang National Standard Time & Frequency Lab., TL, Taiwan 1, Lane 551, Min-Tsu Road, Sec. 5, Yang-Mei, Taoyuan, Taiwan 36 Tel: 886 3

More information

(a) At what number x = a does f have a removable discontinuity? What value f(a) should be assigned to f at x = a in order to make f continuous at a?

(a) At what number x = a does f have a removable discontinuity? What value f(a) should be assigned to f at x = a in order to make f continuous at a? Solutions to Test 1 Fall 016 1pt 1. Te grap of a function f(x) is sown at rigt below. Part I. State te value of eac limit. If a limit is infinite, state weter it is or. If a limit does not exist (but is

More information

1 1. Rationalize the denominator and fully simplify the radical expression 3 3. Solution: = 1 = 3 3 = 2

1 1. Rationalize the denominator and fully simplify the radical expression 3 3. Solution: = 1 = 3 3 = 2 MTH - Spring 04 Exam Review (Solutions) Exam : February 5t 6:00-7:0 Tis exam review contains questions similar to tose you sould expect to see on Exam. Te questions included in tis review, owever, are

More information

Optimal Mechanism with Budget Constraint Bidders

Optimal Mechanism with Budget Constraint Bidders Optimal Mecanism wit Budget Constraint Bidders Alexei Bulatov Sergei Severinov Tis draft: November 6 Abstract Te paper deals wit te optimal mecanism design for selling to buyers wo ave commonly known budget

More information

Chapters 19 & 20 Heat and the First Law of Thermodynamics

Chapters 19 & 20 Heat and the First Law of Thermodynamics Capters 19 & 20 Heat and te First Law of Termodynamics Te Zerot Law of Termodynamics Te First Law of Termodynamics Termal Processes Te Second Law of Termodynamics Heat Engines and te Carnot Cycle Refrigerators,

More information

Poisson INAR processes with serial and seasonal correlation

Poisson INAR processes with serial and seasonal correlation Poisson INAR processes with serial and seasonal correlation Márton Ispány University of Debrecen, Faculty of Informatics Joint result with Marcelo Bourguignon, Klaus L. P. Vasconcellos, and Valdério A.

More information

DETERMINATION OF OPTIMAL DESIGN PARAMETERS FOR X CONTROL CHART WITH TRUNCATED WEIBULL IN- CONTROL TIMES

DETERMINATION OF OPTIMAL DESIGN PARAMETERS FOR X CONTROL CHART WITH TRUNCATED WEIBULL IN- CONTROL TIMES International Journal of Production Tecnology and Management (IJPTM) Volume 7, Issue 1, Jan June 016, pp. 01 17, Article ID: IJPTM_07_01_001 Available online at ttp://www.iaeme.com/ijptm/issues.asp?jtype=ijptm&vtype=7&itype=1

More information

Combining functions: algebraic methods

Combining functions: algebraic methods Combining functions: algebraic metods Functions can be added, subtracted, multiplied, divided, and raised to a power, just like numbers or algebra expressions. If f(x) = x 2 and g(x) = x + 2, clearly f(x)

More information

NUMERICAL DIFFERENTIATION. James T. Smith San Francisco State University. In calculus classes, you compute derivatives algebraically: for example,

NUMERICAL DIFFERENTIATION. James T. Smith San Francisco State University. In calculus classes, you compute derivatives algebraically: for example, NUMERICAL DIFFERENTIATION James T Smit San Francisco State University In calculus classes, you compute derivatives algebraically: for example, f( x) = x + x f ( x) = x x Tis tecnique requires your knowing

More information

Conditional Maximum Likelihood Estimation of the First-Order Spatial Non-Negative Integer-Valued Autoregressive (SINAR(1,1)) Model

Conditional Maximum Likelihood Estimation of the First-Order Spatial Non-Negative Integer-Valued Autoregressive (SINAR(1,1)) Model JIRSS (205) Vol. 4, No. 2, pp 5-36 DOI: 0.7508/jirss.205.02.002 Conditional Maximum Likelihood Estimation of the First-Order Spatial Non-Negative Integer-Valued Autoregressive (SINAR(,)) Model Alireza

More information

Section 15.6 Directional Derivatives and the Gradient Vector

Section 15.6 Directional Derivatives and the Gradient Vector Section 15.6 Directional Derivatives and te Gradient Vector Finding rates of cange in different directions Recall tat wen we first started considering derivatives of functions of more tan one variable,

More information

A Goodness-of-fit test for GARCH innovation density. Hira L. Koul 1 and Nao Mimoto Michigan State University. Abstract

A Goodness-of-fit test for GARCH innovation density. Hira L. Koul 1 and Nao Mimoto Michigan State University. Abstract A Goodness-of-fit test for GARCH innovation density Hira L. Koul and Nao Mimoto Micigan State University Abstract We prove asymptotic normality of a suitably standardized integrated square difference between

More information

MAT Calculus for Engineers I EXAM #1

MAT Calculus for Engineers I EXAM #1 MAT 65 - Calculus for Engineers I EXAM # Instructor: Liu, Hao Honor Statement By signing below you conrm tat you ave neiter given nor received any unautorized assistance on tis eam. Tis includes any use

More information

Stochastic Processes

Stochastic Processes Stochastic Processes Stochastic Process Non Formal Definition: Non formal: A stochastic process (random process) is the opposite of a deterministic process such as one defined by a differential equation.

More information

Atm S 547 Boundary Layer Meteorology

Atm S 547 Boundary Layer Meteorology Lecture 9. Nonlocal BL parameterizations for clear unstable boundary layers In tis lecture Nonlocal K-profile parameterization (e. g. WRF-YSU) for dry convective BLs EDMF parameterizations (e. g. ECMWF)

More information

Lines, Conics, Tangents, Limits and the Derivative

Lines, Conics, Tangents, Limits and the Derivative Lines, Conics, Tangents, Limits and te Derivative Te Straigt Line An two points on te (,) plane wen joined form a line segment. If te line segment is etended beond te two points ten it is called a straigt

More information

Nonparametric estimation of the average growth curve with general nonstationary error process

Nonparametric estimation of the average growth curve with general nonstationary error process Nonparametric estimation of te average growt curve wit general nonstationary error process Karim Benenni, Mustapa Racdi To cite tis version: Karim Benenni, Mustapa Racdi. Nonparametric estimation of te

More information

Precalculus Test 2 Practice Questions Page 1. Note: You can expect other types of questions on the test than the ones presented here!

Precalculus Test 2 Practice Questions Page 1. Note: You can expect other types of questions on the test than the ones presented here! Precalculus Test 2 Practice Questions Page Note: You can expect oter types of questions on te test tan te ones presented ere! Questions Example. Find te vertex of te quadratic f(x) = 4x 2 x. Example 2.

More information

New Streamfunction Approach for Magnetohydrodynamics

New Streamfunction Approach for Magnetohydrodynamics New Streamfunction Approac for Magnetoydrodynamics Kab Seo Kang Brooaven National Laboratory, Computational Science Center, Building 63, Room, Upton NY 973, USA. sang@bnl.gov Summary. We apply te finite

More information

For a stochastic process {Y t : t = 0, ±1, ±2, ±3, }, the mean function is defined by (2.2.1) ± 2..., γ t,

For a stochastic process {Y t : t = 0, ±1, ±2, ±3, }, the mean function is defined by (2.2.1) ± 2..., γ t, CHAPTER 2 FUNDAMENTAL CONCEPTS This chapter describes the fundamental concepts in the theory of time series models. In particular, we introduce the concepts of stochastic processes, mean and covariance

More information

University Mathematics 2

University Mathematics 2 University Matematics 2 1 Differentiability In tis section, we discuss te differentiability of functions. Definition 1.1 Differentiable function). Let f) be a function. We say tat f is differentiable at

More information

FUNDAMENTAL ECONOMICS Vol. I - Walrasian and Non-Walrasian Microeconomics - Anjan Mukherji WALRASIAN AND NON-WALRASIAN MICROECONOMICS

FUNDAMENTAL ECONOMICS Vol. I - Walrasian and Non-Walrasian Microeconomics - Anjan Mukherji WALRASIAN AND NON-WALRASIAN MICROECONOMICS FUNDAMENTAL ECONOMICS Vol. I - Walrasian and Non-Walrasian Microeconomics - Anjan Mukerji WALRASIAN AND NON-WALRASIAN MICROECONOMICS Anjan Mukerji Center for Economic Studies and Planning, Jawaarlal Neru

More information

The Krewe of Caesar Problem. David Gurney. Southeastern Louisiana University. SLU 10541, 500 Western Avenue. Hammond, LA

The Krewe of Caesar Problem. David Gurney. Southeastern Louisiana University. SLU 10541, 500 Western Avenue. Hammond, LA Te Krewe of Caesar Problem David Gurney Souteastern Louisiana University SLU 10541, 500 Western Avenue Hammond, LA 7040 June 19, 00 Krewe of Caesar 1 ABSTRACT Tis paper provides an alternative to te usual

More information