Σύγκριση είκτη Αιτιότητας κατά Granger και µεταφορικής εντροπίας και εφαρµογή σε προβλήµατα αγοράς

Μέγεθος: px
Εμφάνιση ξεκινά από τη σελίδα:

Download "Σύγκριση είκτη Αιτιότητας κατά Granger και µεταφορικής εντροπίας και εφαρµογή σε προβλήµατα αγοράς"

Transcript

1 ΑΡΙΣΤΟΤΕΛΕΙΟ ΠΑΝΕΠΙΣΤΗΜΙΟ ΘΕΣΣΑΛΟΝΙΚΗΣ ΣΧΟΛΗ ΘΕΤΙΚΩΝ ΕΠΙΣΤΗΜΩΝ ΤΜΗΜΑ ΜΑΘΗΜΑΤΙΚΩΝ ΜΕΤΑΠΤΥΧΙΑΚΟ ΠΡΟΓΡΑΜΜΑ ΣΠΟΥ ΩΝ ΣΤΑΤΙΣΤΙΚΗ ΚΑΙ ΜΟΝΤΕΛΟΠΟΙΗΣΗ Σύγκριση είκτη Αιτιότητας κατά Granger και µεταφορικής εντροπίας και εφαρµογή σε προβλήµατα αγοράς ΜΕΤΑΠΤΥΧΙΑΚΗ ΙΑΤΡΙΒΗ Κουτσούρη Ελεωνόρα, MSc Οικονοµολόγος Εγκρίθηκε από την τριµελή εξεταστική επιτροπή Κουγιουµτζής ηµήτρης Αν. Καθηγητής ΑΠΘ Μωυσιάδης Πολυχρόνης Καθηγητής ΑΠΘ Ιωαννίδης ηµήτριος Καθηγητής Παν. Μακεδονίας

2 ARISTOTLE UNIVERSITY OF THESSALONIKI FACULTY OF SCIENCES SCHOOL OF MATHEMATICS MASTER OF SCIENCES STATISTICS AND MODELLING A comparaive sudy on Granger Causaliy and Transfer Enropy Indices and an applicaion on consumer goods issues MASTER THESIS Kousouri Eleonora, MSc Economics Supervisor Kugiumzis Dimiris As. Professor AUTH February 203 Thessaloniki

3 Ελεωνόρα Κουτσούρη Οικονοµολόγος Ελεωνόρα Κουτσούρη, 203 Αριστοτέλειο Πανεπιστήµιο Θεσσαλονίκης, 203 Με επιφύλαξη παντός δικαιώµατος. All righs reserved. Απαγορεύεται η αντιγραφή, αποθήκευση και διανοµή της παρούσας εργασίας, εξ ολοκλήρου ή τµήµατος αυτής, για εµπορικό σκοπό. Επιτρέπεται η ανατύπωση, αποθήκευση και διανοµή για σκοπό µη κερδοσκοπικό, εκπαιδευτικής ή ερευνητικής φύσης, υπό την προϋπόθεση να αναφέρεται η πηγή προέλευσης και να διατηρείται το παρόν µήνυµα. Ερωτήµατα που αφορούν τη χρήση της εργασίας για κερδοσκοπικό σκοπό πρέπει να απευθύνονται προς το συγγραφέα. Οι απόψεις και τα συµπεράσµατα που περιέχονται σε αυτό το έγγραφο εκφράζουν το συγγραφέα και δεν πρέπει να ερµηνευτεί ότι εκφράζουν τις επίσηµες θέσεις του Α.Π.Θ. 2

4 Περίληψη Η αιτιότητα αποτελεί ένα ερώτημα που καταλαμβάνει κεντρική θέση στις έρευνες πολλών κλάδων τις τελευταίες δεκατετίες. Πολύ συχνά διατυπώνονται ερωτήματα που αφορούν στη σχέση μεταξύ των διερευνώμενων μεταβλητών, όπως και μια σειρά άλλων υποερωτημάτων σχετικά με την ένταση και τη διεύθυνση αυτής της σχέσης. Σημαντικό είναι να αναφερθεί πως δεν υπάρχει ένας ενιάιος και καθολικός ορισμός της έννοιας της αιτιότητας. Μέχρι και σήμερα οι ερευνητές χρησιμοποιούν την έννοια της αιτιότητας αλλά και διάφορους ορισμούς της όπως αυτοί έχουν κατά καιρούς διατυπωθεί, προκειμένουν να μελετήσουν και να κατανοήσουν την φύση και την συμπεριφορά των παρατηρήσεών τους. Οι διαφορετικοί ορισμοί, είναι εμπνευσμένοι από διαφορετικά πεδία, και διατυπώθηκαν κάθε φορά, ώστε να εξυπηρετήσουν την ακριβή φύση του ερευνητικού αντικειμένου. Στα τέλη της δεκαετίας του 960, οι οικονομολόγοι έθεσαν ερωτήματα για την αιτιότητα και για το κατά πόσο αυτή θα έπρεπε να συνυπολογίζεται κατά τη διαμόρφωση ερευνητικών μοντέλων. Ο πρώτος που ασχολήθηκε αναλυτικά με το θέμα της αιτιότητας και θεμελίωσε μεθοδολογία για την διερεύνηση της αιτιότητας μεταξύ δύο μεταβλητών ήταν ο Clive G. Granger (Granger, 969), αν και είχαν ήδη προηγηθεί μελέτες οι οποίες αναφερόντουσαν σε θεωρήτικό επίπεδο στο ζήτημα της αιτότητας (Wiener, 956). Η αιτιότητα, όπως αυτή ορίζεται από τον Granger, απαντάται όταν η διακύμανση ενός μονομεταβλητού μοντέλου για μια τυχαία μεταβλητή μειώνεται από την εισαγωγή παρελθοντικών τιμών μιας δεύτερης στο επεξηγηματικό κομμάτι του μοντελου της πρώτης. Η δουλειά του Granger έτυχε ευρείας αποδοχής από το ερευνητικό κοινό, και ονομάστηκε αιτιότητα κατά Granger. Μεταγενέστεροι ερευνητές επέκτειναν αυτή τη μεθοδολογία σε πολλαπλές χρονοσειρές, αλλά και σε πολυμεταβλητά συστήματα, αλλά και καθιέρωσαν 3

5 παραμετρικούς ελέγχους ώστε να καταστήσουν εφικτή τη στατιστική συμπερασματολογία στα αποτελέσματα του δείκτη αιτιότητας κατά Granger. Φυσικά υπήρξε και κριτική στο έργο του Granger, η οποία στηρίχθηκε κυρίως στο οτί ο ορισμός που δίνει o Granger στην αιτιότητα είναι πολύ συγκεκριμένος και φέρει αρκετούς περιορισμούς, μεταξύ των οποίων και η παραμετρική διατύπωση ενός μοντέλου, καθώς και η απαραίτητη στασιμότητα των χρονοσειρών που μελετούνται. Παρ όλ αυτά, η μεθοδολογία καθιερώθηκε και χρησιμοποιήθηκε ευρέως σε πολλά ερευνητικά πεδία χάρη στην απλή και εύχρηστή της διατύπωση. Πρόσφατα δε, το ενδιαφέρον της ερευνητικής κοινότητας για την αιτιότητα κατά Granger αναζωπυρώθηκε, καθώς το μέτρο αυτό αξιοποιήθηκε από τον κλάδο της νευροφυσιολογίας, προσφέροντας νέα ευρήματα στην ανάλυση των ηλεκροεγκεφαλογραφημάτων. Παράλληλα με τη χρήση του παραμετρικού ελέγχου, πρόσφατα έχουν προταθεί νέες μεθοδολογίες για τη διερεύνηση της αιτιότητας, οι οποίες έχουν βασιστεί σε μια εναλλακτική διατύπωση της αιτιότητας κι έχουν εμπνευστεί από τη θεωρία της πληροφορίας. Πιο συγκεκριμένα, αφετηρία αυτής της νέας θεώρησης αποτελέι το μέτρο της εντροπίας (Schreiber, 2000) στo οποίο στηρίχτηκε και διαμορφώθηκε το μέτρο της μεταφορικής εντροπίας, η οποία μελετάει την πληροφορία που περιλαμβάνει μια μεταβλητή Υ που χρησιμέυει ώστε να προβλεφτούν μελλοντικές τιμές της μεταβλητής Χ. Η ειδοποιός διαφορά, πέραν του ορισμού αυτού του νέου μέτρου είναι το πλαίσιο στο οποίο αναπτύσσεται, κάθως πρόκειται για ένα μη παραμετρικό μέτρο που δεν επιβάλλει περιορισμούς στις χρονοσειρές που μελετούνται. Ο σκοπός αυτής της εργασίας είναι παρουσιάσει τα κύρια σημεία της βιβλιογραφίας που έχει αναπτυχθεί γύρω από την αιτιότητα κατά Granger αλλά και την μεταφορική εντροπία, σε θεωρητικό πλαίσιο αλλά και σε πρακτικό επίπεδο. Κλασικά πλέον μέτρα, όπως ο δείκτης αιτιότητας κατά Granger, ο κατά συνθήκη δείκτης αιτιότητας κατά Granger, και ο μερικός δείκτης αιτιότητας κατά Granger, παρουσιάζονται μαζί με τις στατιστικές ιδιότητες τους. Σε αντιστοιχία, παρουσιάζεται το μέτρο της μεταφορικής εντροπίας, όπως αυτό έχει εισαχτεί στη 4

6 βιβλιογραφία, αλλά και το μέτρο της μερικής μεταφορικής εντροπίας, ως μια βελτίωση της πρώτης διατύπωσης. Στη συνέχεια τα μέτρα αυτά, παραμετρικά και μη, εφαρμόζονται σε συστήματα με κατασκευασμένα δεδομένα στα οποία ήδη γνωρίζουμε τις σχέσεις μεταξύ των μεταβλητών, προκειμένου να μελετηθεί η συμπεριφορά του κάθε ελέγχου και η επιτυχία διαπίστωσης των αιτιατών σχέσεων του συστήματος. Τα συμπεράσματα στα οποία οδηγηθήκαμε είναι πως και τα τέσσερα μέτρα έχουνε μεγάλο ποσοστό επιτυχίας σε δεδομένα που δημιοργούνται από ένα γραμμικό σύστημα, ή από ένα σύστημα που έχει υποστεί μια μικρή παραμόρφωση. Από την άλλη, το ποσοστό επιτυχίας μειώθηκε δραματικά σε δεδομένα που προήλθαν από μη γραμμικό σύστημα. Επίσης, αξιοσημείωτη ήταν η βελτίωση των αποτελεσμάτων με την χρήση των κατά συνθήκη δεικτών, καθώς τα λανθασμένα συμπεράσματα που προέρχονταν απο έμμεσες αιτιατές σχέσεις απαλείφτηκαν στις περιπτώσεις του γραμμικού και του παραμορφωμένου συστήματος. Τέλος, τα ίδια μέτρα αιτιότητας εφαρμόστηκαν σε πραγματικά δεδομένα που προέρχονται από την αγορά ταχυκίνητων αγαθών, και διερευνήθηκαν οι σχέσεις μεταξύ πέντε βασικών μεταβλητών που έχουν παγιωθεί στις σύγχρονες μεθόδους ανάλυσης μάρκετινγκ δεδομένων. Τα αποτελέσματα δεν είναι κοινά ανάμεσα στους τέσσερις δείκτες που χρησιμοποιήθηκαν, όμως αναδεικνύουν κάποιες βασικές σχέσεις μεταξύ των μεταβλητών που μελετήθηκαν. Λέξεις-Κλειδιά: Χρονοσειρές, Αιτιότητα κατά Granger, ΜεταφορικήΕντροπία, Ταχυκίνητα Αγαθά. 5

7 Absrac This hesis presens an overview of exising lieraure upon conceps of causaliy, boh in a parameric and an informaion heoreic conex. Proposed measures of Granger Causaliy and Transfer Enropy are inroduced and applied on simulaed daa where he causaions beween he generaed variables are a priori known o us, so as o explore he robusness and he fallacies of each measure. The same measures are applied on markeing daa, in an effor o verify cerain common hypoheses ha are commonly used in modern markeing analyics. Keywords: Time Series, Granger Causaliy, Transfer Enropy, FMCG. 6

8 Acknowledgmens I would like o hank my supervisor, Dr. Dimiris Kugiumzis, for his perseverance and valuable help during he wriing of his hesis. 7

9 Conens. Inroducion Time Series: Fundamenal Conceps Time Series Firs Mahemaical Momens and heir Sample Esimaions Populaion Measures... 4 Mean & Variance... 4 Auo Covariance & Auo Correlaion... 4 Cross Covariance & Cross Correlaion... 4 Parial Correlaion Sample Esimaors... 5 Sample Mean & Variance... 5 Sample Auo Covariance & Auo Correlaion... 6 Sample Cross Covariance & Cross Correlaion... 6 Sample Parial Correlaion Time series models... 7 Iid noise... 7 Random walk... 7 Whie noise... 7 Auoregressive process... 8 Moving average process Order Selecion for a model Saionariy The Uni roo problem & Uni Roo Tess Wold Represenaion Theorem Univariae and Mulivariae Time Series

10 Mean & Variance Auo Covariance & Auo Correlaion Cross Covariance & Cross Correlaion Parial Correlaion Sample Mean & Variance Sample Auo Covariance & Auo Correlaion Sample Cross Covariance & Cross Correlaion Sample Parial Correlaion Causaliy Granger Causaliy Condiional Granger Causaliy Parial Granger Causaliy Transfer Enropy Enropy as an informaion meric Transfer Enropy-An Indicaor Parial Transfer Enropy-An Indicaor Saisical Inference on TE and PTE Granger Causaliy and Transfer Enropy on Simulaed Daa Inroducion The esing procedure Linear Sysem Disored linear sysem Non-linear sysem Discussion Applicaion Inroducion

11 6.2. The Applicaion Se Up Uni Sales TPR- Weighed Average Price Reducion Avg Iems/Sore Selling ACV Weighed Disribuion Feaure and Display Price Formulaion of our Expecaions Resuls Overview of Resuls Granger Causaliy Condiional Granger Causaliy Transfer Enropy Parial Transfer Enropy Conclusions Summary References

12 . Inroducion Causaliy has always been a crucial opic for research. Quesions as he exisence of conneciviy beween wo observed variables, he direcion of he link beween hem and he exen o which hey relae, have been a he cener of discussions for many years; unil oday, here is sill heaed debae around his opic, and researchers coninue suggesing new ways of discovering causal links wihin a sochasic environmen. Granger - despie he fac ha his work has been preceded by ohers- has been he firs o horoughly address he issue of causaion wih formal analyical represenaions and mahemaical proof upon ime series. He herefore coined he erm of Granger causaliy, a erm ha describes a very specific form of causaliy beween wo or more variables, and ha iniially was used in economic research. Nowadays, Granger causaliy erm has become a useful and powerful ool in he hands of researchers across many disciplines. In he laes years, Granger causaliy is exensively used in he neurobiology field, in an aemp o decode EEGs (Elecroencephalograms). More recenly, Τransfer Εnropy was inroduced, accouning for causaion in an alernaive way: in a broader sense of informaion direcion. This new meric sems from informaion heory field, which is exensively used in signal analysis. The main purpose of his hesis is o presen main poins from lieraure upon Granger causaliy and Τransfer Εnropy merics, as well as o apply four causaliy ess on acual FMCG (Fas Moving Consumer Goods Daa), in an effor o ouline similariies and differences across all used ess; boh parameric and informaion-heoreic ones. The res of he hesis is organized as follows: Chaper 2 presens some inroducory conceps from Time Series Theory, necessary for he undersanding of he following chapers. Chaper 3 presens he Granger Causaliy concep, ogeher wih parameric ess ha can be used o explore causaliy relaions wihin a parameric sysem. Chaper 4 gives an overview of he lieraure upon Transfer Enropy so far. In Chaper 5, we apply four causaliy ess on simulaed daa, in order o idenify he measure capabiliies as well as

13 some of heir inefficiencies. In Chaper 6, we use he same four causaliy measures on acual markeing daa from he FMCG (Fas Moving Consumer Goods) secor, o discover causal links among five variables. Finally, in Chaper 7 we summarize our work. 2

14 2. Time Series: Fundamenal Conceps 2.. Time Series Time Series is a se of x observaions of a random variable X, each recorded a ime poin, wih aking values,.,n. A ime series can be discree or coninuous as i can consis from eiher T0 discree values or from coninuous recordings wihin a fixed ime period T0 accordingly. The sudy of ime series can prove very helpful as a paern of he underlying process can emerge, enabling he researcher o build a model ha will explain he behavior of he variable X. This is why ime series have been exensively used across many fields, such as engineering, financial forecasing, signal processing ec. A ime series model is a formal represenaion which specifies he oin disribuions (or mos commonly, only up o second order properies) of he x realizaions of a random variable X. Therefore, a valid model specificaion lies on our abiliy o define all oin disribuions of he random vecors (X,X2, Xn) for n=,2, or equivalenly of probabiliies P(X x,x2 x2, Xn xn) for - < x,,xn< and n=,2, Alernaively (and significanly easier) i is sufficien o specify he firs and second order momens (second order properies) of he oin disribuions of he random vecors (X,X2, Xn) for n=,2,, meaning ha we can formulae our expecaions regarding he mean and he variance of he underlying {X} process. 3

15 2.2. Firs Mahemaical Momens and heir Sample Esimaions The mahemaical momens of a random variable X, are measures which are used o describe he properies of is disribuion Populaion Measures Mean & Variance The firs momen is used o describe is populaion mean, E[X ] = µ Χ whereas he second momen is mos usually referred o as variance, σ = var[x ] = E[(X ) ] 2 2 X µ Χ Auo Covariance & Auo Correlaion Furher, we will also define he covariance funcion beween by a ime lag h as γ ( h) = Ε( Χ µ )( Χ µ ), X + h Χ Χ Χ and + h Χ which differ only as well as he correlaion funcion beween Χ and + h Χ as γ X ( h) Ε( Χ+ h µ Χ )( Χ µ Χ ) ρ X ( h) = =. 2 γ (0) σ X Cross Covariance & Cross Correlaion The laer wo measures can also be used o measure he covariance and correlaion beween wo random variables Χ and + h Y, and hen he measures will ake he following form : γ ( h) = Ε[( Χ µ )( Y µ )] XY + h X Y 4

16 γ XY ( h) Ε[( Χ+ h µ X )( Y µ Y )] ρxy ( h) = = γ (0) γ (0) σ σ 2 2 XX YY X Y Parial Correlaion Parial correlaion is a erm used o describe he associaion beween wo variables, while a he same ime aking ino accoun he effec of oher conrolling variables. Le us consider he case where we have hree variables, X, Y, and Z, and we wan o find ou he connecion beween X and Y, while a he same ime we conrol for he any effecs ha migh be driven from Z. ρ XY Ζ ( h) = ρ ( h) ρ ( h) ρ ( h) XY XZ ZY 2 2 ρ XZ (0) ρ ZY (0) Sample Esimaors In mos cases, i is improbable on real applicaions o have access o he whole populaion in order o calculae he firs momens. So, insead of ha, we use a subse of he populaion called sample, and esimae he momen parameers. Therefore, for n observaions in our sample and of course as before, h being he ime lag beween wo ime poins we define: Sample Mean & Variance n x n = x =, he sample variance, 5

17 n 2 sx = ( x x) n = Sample Auo Covariance & Auo Correlaion n h c ( h) = ( x x)( x x) X + h n h = And similarly, he sample auocorrelaion funcion, c rx ( h) = c X X ( h) (0) Sample Cross Covariance & Cross Correlaion We also consider he sample covariance and correlaion beween wo differen random variables X and Y n h c ( h) = ( x x)( y y) XY + h n h = r XY ( h) = c c XY ( h) (0) c (0) 2 2 XX YY Sample Parial Correlaion r XY Ζ ( h) = r ( h) r ( h) r ( h) XY XZ ZY 2 2 r XZ (0) r ZY (0) 6

18 2.3. Time series models In his secion, we presen some ime series models, which are frequenly encounered in ime-series analysis and also used in he following chapers. Iid noise This model is used o describe independen and idenically disribued realizaions of random variables wih zero mean. Since he observaions are independen, i holds ha ( ) ( ) P(X x, X x, X x ) = P(X x ) P X x P X x 2 2 n n 2 2 n n wih E[Χ ] = 0, Random walk Random walk is a sequence S X X 2.. X = of random iid variables { } X = X.. X, where =,2,,n 2 wih E[Χ ] = 0, and var[χ ] = σ Whie noise Whie noise is sequence { ε } of uncorrelaed random variables, wih E [ε ] = 0, and var[ε ] = σ 2 ε 7

19 Auoregressive process An auoregressive process AR(p), is defined is a following: X = ϕ X + ϕ X + + ϕ X + ε, p p where p represens he order of model, showing he number of lags ha explain he behavior of series X a ime, ϕ,..., ϕ pare he coefficiens of he process, ha can be grouped under he noaion ϕ, wih =, 2,..., p deermining he degree o which each X componen affecs he behavior of he process X a he presen. ε represens a whie noise procedure wih zero mean and sandard variance 2 ~ (0, ) ε WN σ ε Wih he help of a back shif operaor B X X =, we can rewrie an AR(p) process as an poluonym expression X = ϕ X + ϕ X ϕ X + ε 2 2 p p p ( ϕ B ϕ B... ϕ B ) X = ε ϕ( B) X 2 2 = ε p where ϕ( B) p = = ϕ B Therefore an auoregressive series of order p can be represened in he following compac formulaion: X p = ϕ X + ε = 8

20 Moving average process A moving average process of order q can be represened in he following way X = θε θ qε q + ε, Where θ,... θ q are real numbers represening he coefficiens deermining he exen o which each pas value of he process deermines is curren values, q is he order of he model, which in accordance o he AR(p) specificaion, represens he number of lags ha explain he behavior of series X a ime, and of course ε represens a whie noise procedure wih zero mean and sandard variance 2 ~ (0, σ ) ε WN ε 9

21 2.4. Order Selecion for a model One of he parameers of he model o be esimaed is he order of he model, meaning he p ime lags ha need o be included in he model so ha he dependen variable is adequaely explained. There exis a number of mehodologies o selec he appropriae order for a model, such as he AIC crierion, which selecs an appropriae order for our model by accouning for he radeoff beween he number of variables which add explanaory power in our model, bu ha also add variance ino i. AIC=-2ln(LLN)+2g, where g is he number of explanaory variables insered ino he model, and LLN is he maximized value of he Log Likelihood Funcion. Alernaively, one can use he Bayesian Informaion Crierion (BIC), which penalizes harder he number of independen variables insered ino he model. One can also chose he appropriae order by employing empirical mehods, and examining he auocorrelaion and parial auocorrelaion graphs, because he parial auocorrelaion funcion urns o zero for lags greaer han he order of he model. 20

22 2.5. Saionariy We can call a ime series saionary, when is properies do no change over ime. The main inuiion behind his is ha he sysem will give us back he same informaion a any given ime-poin, allowing us hus o apply several generalizing assumpions for he underlying process. There exis wo forms of saionariy, he sric and he weak saionariy, however weak saionariy is usually sufficien o allow he sudy of he respecive ime series. A formal definiion of weak saionariy: Le X be a sochasic ime series wih E[ X ] = µ, a consan, ime-invarian mean, 2 2 var[ ] [( ) ] X = E X µ = σ, a consan variance, var[ X ] = E[( X µ )( X µ ] = γ, a seady covariance, + h h Then, X can be considered covariance saionary process, for all ime poins. There are more han one ways o es wheher a series is saionary. An experienced eye can recognize a mean revering procedure only by is graphical represenaion. There are also more formal ways o es wheher a series is saionary, called uni roos ess. Broadly speaking, a sochasic process is said o be saionary if is mean and variance are consan over ime and he value of he covariance beween he wo ime periods depends only on he disance or gap or lag beween he wo ime periods and no he acual ime a which he covariance is compued. In he ime series lieraure, such a sochasic process is known as a weakly saionary, or covariance saionary, or second-order saionary, or wide sense, sochasic process, (Guarai D., 2004) 2

23 2.6. The Uni roo problem & Uni Roo Tess In lieraure, a uni roo process is a non-saionary process. In he example below, we explain how a uni roo process is non-saionary. Le us consider he following AR () process: X = ϕ X + ε, wih In he case where he coefficien φ akes he value of, hen our AR process becomes a random walk process, where all he new elemens sem from he innovaion erm, rendering us a non-predicable series. This case is broadly known in he lieraure as uni roo problem. If, however, φ < and he series has a consan mean μ for all ime poins, hen we can show ha he series fulfills weak saionariy condiions, as is mean will ake he following form: E( X ) = ϕe( X ) + E( ε ) µ = ϕµ + 0 µ = ϕ I is obvious now ha if φ = hen we canno have a mean. The same holds for he series variance provided we have a seady variance over ime. var( X ) = ϕ var( X ) + var( ε ) var( X ) = ϕ var( X ) + σ σε var( X ) = ϕ ε 22

24 There have been developed many ess o check he sabiliy of a process, esing wheher he esimaed parameer for φ is saisically differen han zero. Such a es, ha is widely used, is he Dickey Fuller es: Consider an AR(p) process ha includes an inercepα and a ime rend η, where η is he seady rae a which he series evolves hrough ime. X = α + η + ϕ X + ε The firs differences of his procedure will hen be X X = α + η + ( ϕ ) X + ε X = α + η + β X + ε, where β = ( ϕ ) And he es saisic will be ˆ ADF = β, which follows a non-sandard disribuion 2 s ( ˆ β) compued by Dickey & Fuller, wih H ˆ 0 : β = 0 and alernaive hypohesis H ˆ : β < 0, i.e. esing wheher φ is significanly differen han. An exension of his es is also he augmened Dickey Fuller es, which resores he disorion ha he model migh possibly include due o auocorrelaions in he error erm, wih he inclusion of lagged values of X in he model: Then he es on he same parameer β = ( ϕ ) is realized on he firs differences of he augmened version of he model: X = α + η + ϕ X + ϕ X + ε p i p i= X X X... p X p = α + η + β + ϕ + + ϕ + ε 23

25 A poin o consider while using he ADF es saisic is he selecion of he order p of he model. There exiss a radeoff beween he size and he power of he es, as a high order is beneficial for he es size, bu decreases our possibiliy of reecing he null hypohesis. We can use a crierion, such as he AIC or BIC crierion in order o choose he opimal order for he model. 24

26 2.7. Wold Represenaion Theorem The Wold s decomposiion heorem shows ha a weakly saionary series can be wrien in he form of an infinie moving average process. Consider he following process AR() process: X = ϕ X + ε, wih 2 ~ (0, ) ε WN σ ε Then he process can be re-wrien in an MA( ) represenaion, by replacing he righ hand side of he expression wih heir own AR represenaion: X = ϕ( ϕx 2 + ε ) + ε =... = ϕ ε = 0 Where he coefficiens ϕ are square summable ϕ < o secure sabiliy, hey are causal in he sense ha no <0 and hey are sable hrough ime. = 0 25

27 2.8. Univariae and Mulivariae Time Series So far, we have been referring o univariae ime series; however, i is more usual ha a process X is comprised by many such univariae componens which have some degree of dependence beween hem. The mulivariae case is a naural expansion of he univariae case, so we will quickly refer o basic properies, such as he firs order momens and heir esimaion, as well as he vecor represenaion (mulivariae case) of an AR(p) and of an MA(q) process. Please noe, ha from now on we will use vecor noaion o denoe he mulivariae cases (bold case leers). Le us consider a mulivariae ime series X, Χ =..., comprising of m Χ univariae ime series wih =,2, ime poins. X, m Consider also he AR represenaion of = = 0 Χ φ ε. Then, he parameers φ shoud fulfill he crieria for causaliy which is = 0 de[ φ ( z)] 0, z <. Then heir firs order momens and heir corresponding sample esimaors will be he following: 26

28 Mean & Variance µ X, µ X = E[ Χ ] =..., µ X, m σ = var[ X ] = E[( X µ )( X µ ) '] 2 Χ X X Auo Covariance & Auo Correlaion Γ ( h ) = Ε [( Χ - µ )( Χ - µ )'] X + h X X ρ Γ ( h) Ε[( Χ µ )( Χ µ ) '] X + h X X X ( h ) = = 2 Γ X (0) σ X Cross Covariance & Cross Correlaion γ( + h, ) K γm ( + h, ) ΓΧΥ ( h) = Ε[( Χ + h µ X )( Y µ Y ) '] = M O M γ m( + h, ) γ mm ( + h, ) L ρ ΧΥ ( h) ρ ( + h, ) K ρ ( + h, ) m Ε[( Χ + h µ X )( Y µ Y ) '] = = 2 2 M O M ΓX (0) ΓY (0) ρm( + h, ) L ρmm ( + h, ) Parial Correlaion ρxy ( h) ρxz ( h) ρzy ( h) ρxy Ζ ( h) = 2 2 ρ (0) ρ (0) XZ ZY In accordance o he univariae esimaors we have already defined, he sample esimaions of he above menioned heoreical momens will be he following for he mulivariae case: 27

29 Sample Mean & Variance n n = x = x, and n 2 s = ( x x ) X n = Sample Auo Covariance & Auo Correlaion n h C ( h) = ( x+ h x)( x x ), X n h C ( h) r ( h ) = X X C (0) X = Sample Cross Covariance & Cross Correlaion We also consider he sample covariance and correlaion beween wo differen random variables X and Y n h C ( h) = ( x+ h x)( y y) XY n h = r XY ( h ) = C C XY ( h) C 2 2 X Y Sample Parial Correlaion rxy ( h) rxz ( h) rzy ( h) rxy Ζ ( h) = 2 2 r (0) r (0) XZ ZY 28

30 29

31 3. Causaliy In his chaper we presen he classical definiion Granger has given o causaliy (Granger, 969). We also explore expansions of Granger-Causaliy, such as Condiional and Parial Granger-Causaliy. Noe ha in all cases, we will only visi he case of unidirecional causaliy from one variable owards anoher. Measures for accouning causaliy will be given, as well as mehods for drawing saisical inference upon hese measures. 3.. Granger Causaliy As briefly discussed in he inroducory par, Granger causaliy is a concep ha explores he exen o which one variable causes anoher. More precisely, Granger defines i (Granger 969, Seh 200) as a siuaion where he pas values of a variable Y improve he linear predicion of X. He also gives he following formulaion, σ 2 ( X U)<( X U Y) 3. which shows ha if he variance of he random variable X given he informaion se U ha we have available a ime (including pas values of Y), is smaller han he variance of X given all informaion available, excluding pas values of Y, hen we have indicaion ha Y causes X. To his end, Granger has consruced wo models, one called resriced (excluding all pas values of Y and one named unresriced (conaining informaion upon pas values of Y ), in order o compare he variances of he error erms and draw inference upon poenial causaliy from Y o X (Geweke, 984). In he following secions we will presen in deail he formal represenaion of he wo bivariae auoregressive models and a es saisic for Granger causaliy. 30

32 Le X be a saionary random vecor ha can be represened by an auoregressive model of order p, which represens he maximal lagged observaions included in he model (Seh 200). X = ϕ X + ϕ X ϕ X + ε r r r r 2 p p X where p r ϕ <, =, 2,..., = p r and he parameers of he equaion lie inside he uni circle < ϕ <, and he innovaion erm (sochasic par) belongs o a whie noise process, ε σ 2 ~ WN (0, r ) ε X We also inroduce a second variable Y ha has he same properies Y = ξ Y + ξ Y ξ Y + ε r r r r 2 p p Y wih p r ξ <, =, 2,..., = p r and < ξ <, ' 2 and ε ~ WN(0, σ ) r Y ε As we have shown in chaper 2, he wo processes can be re-wrien shorly in he following compac form X Y p r ϕ X = r = + ε p r r = ξ Y + εy = Χ 3.2 3

33 The noaion r on he parameers of he models presened so far, is an abbreviaion for he word resriced, as we will call model 3.2 as resriced model (for from now on. X and Y accordingly) Now, we will inroduce he unresriced model, where besides he auo-lagged values of each variable, he pas values of he oher variable also ake par in he predicion of and Y accordingly. Formally, X p p u u u = ϕ + ζ + ε Χ = = X X Y where p u ϕ <, = u and he parameers of he equaion lie inside he uni circle < ϕ <, =, 2,..., p u 2 and he innovaion erm (sochasic par) belongs o a whie noise process, ε ~ WN (0, σ ) X ε u X p p u u u = ξ + ϕ + εy = = Y Y X wih p u ξ <, = u and < ξ <, =, 2,..., p u 2 and ε ~ WN(0, σ ) Y u Y ε In accordance wih he resriced models, indicaor u sands for unresriced, and (3.3) will be called unresriced model from now on. 32

34 p p u u u = ϕ + ζ + ε Χ = = X X Y p p u u u = ξ + ϕ + εy = = Y Y X 3.3 The heoreical measure ha Geweke gave o measure linear dependence from Y o X was = σ ln( ) σ 3.4 F Y X 2 r ε X 2 u ε X Under he null hypohesis H : 0 0 F =, meaning ha here exiss no linear dependence Y X from Y o X, ha is, pas values of Y canno improve he predicion of X. (Inuiively, he saisic FY X would ake he value of 0 in he case where σ = σ 2 2 r u ε X ε X By performing Ordinary Leas Squares or Maximum Likelihood Esimaion on equaions (3.2 ) and (3.3 ), o rerieve he opimal linear esimaor for X we also ge he residual sum of squares RRSS, URSS of he resriced and he unresriced model respecively. Then, we can consruc an F es o es he significance of our es, under he null hypohesis ha here exiss no unidirecional causaliy from Y o X Fˆ Y X ( RRSS URSS) / q = URSS / ( n k) 3.5 H : 0 0 FY X = where n is our sample size, k he number of parameers esimaed from he unresriced model and q represens he number of resricions. 33

35 3.2. Condiional Granger Causaliy Real applicaion daa do no confine o bivariae cases. A variable migh be simulaneously affeced by wo or more variables. I is also possible and happens quie ofen ha he researcher derives false causaions beween wo ses of variables, due o a spurious correlaion wih a hird unexamined variable. As Granger firs poined ou early in his work (Granger, 969), a variable migh falsely seem o be caused by a second one, even hough boh variables are caused by a hird one, bu a differen ime lags. For example, we migh be able o derive a false causal connecion from our naional Gross Domesic Produc (GDP) and he monhly rae of car sales. Bu if one observes closely, boh variables migh be driven by a hird variable, e.g. consumpion, which is no as frequenly recorded as he monhly car sales. This example demonsraes how he naure of our daa affecs our resuls as well as ha he repeiive realizaion of bivariae analysis beween our ses of variables can someimes reurn misleading resuls. I herefore underlines he necessiy of an MVAR represenaion of he series ha are possibly relaed. Geweke se he ground for Condiional Granger Causaliy (Geweke, 984) as a means o conrol for indirec causaliy beween wo variables ha migh a firs insance appear direc, bu is in ruh mediaed by a hird se of series (Guo e al, 2008). Suppose ha we wan o explore he causaliy from he ime series Y owards X bu a he same ime, we wan o condiion on a hird se of random variables called Z. Le Z be a mulivariae ime series consised of m random saionary Zi ime series, wih i=,,m ha has he following represenaion and properies: Z = ψ Z + ψ Z ψ Z + ε Z r r r r 2 p p p where ψ = u <, =,2,.., p 34

36 and he parameers of he equaion lie inside he uni circle r < ψ <, 2 ε ~ (0, r ) and he innovaion erm (sochasic par) belongs o a whie noise process, WN Z. σ ε Then, he resriced model ha we inroduced in he previous secions will become p p r r r = + + Χ = = X φ X δ Z ε p p r r r = + + Y = = Y ξ Y ν Ζ ε 3.6 Γ = r r r var(ε ) cov(ε,ε ) Χ Χ Z r r r cov(ε,ε ) var(ε ) Z Χ Z 3.7 Whereas he unresriced will ake he following form: p p p u u u u = Χ = = = X φ X ζ Y δ Z ε p p p u u u u = Y = = = Y ξ Y φ X ν Ζ ε p p p u u u u = π + ρ + υ + Ζ = = = Ζ X Y Ζ ε 3.8 The noise covariance marix of his new sysem will urn o: u u u u u var(ε ) cov(ε,ε ) cov(ε,ε ) Χ Χ Y Χ Z u u u u u Γ 2 = cov(ε,ε ) var(ε ) cov(ε,ε ) 3.9 Y Χ Y Y Z u u u u u cov(ε,ε ) cov(ε,ε ) var(ε ) Z Χ Z Y Z 35

37 Clearly, he Granger causaliy from Y o X condiioned on Z will be: c F = ln( Γ / Γ ) Y X Inuiively, his measure will show us how he unexplained par of he X ime series has been reduced by he inclusion of he Y series as an explanaory facor, bu condiional on he Z ime series. Therefore, under a FY->X ha has a posiive value, we can conclude ha here exiss some direc influence from Y o X, no mediaed by he hird variable Z. Bu in he case where he causal influence measure akes he value of 0, we have subsanial evidence ha Y is no direcly causing X, bu he causaliy is being driven by he hird facor, Z. By linearly esimaing (OLS) models 3.6 and 3.8, we rerieve he esimaed residual marices C2 and C3 accordingly, which boh follow a chi-square disribuion, so by using he appropriae ransformaion using he degrees of freedom of he unresriced model and he number of resricion, we can consruc an F-es saisic, comparable o he Granger Causaliy Index. Then he condiional granger causaliy Index will become Fˆ H c Y X 0 ( RRSS URSS) / q = URSS / ( n k) c : F = 0 Y X 3. wih RRSS, URSS he residual sum of squares of he resriced and he unresriced model respecively, and n our sample size, k he number of parameers esimaed from he unresriced model, and q he number of resricions imposed. 36

38 3.3. Parial Granger Causaliy Parial Granger Causaliy is a furher improvemen upon he measure of condiional G- Causaliy consruced o allow he researcher o conrol for exogenous inpus and laen variables which migh no be included in he VAR model bu migh be criical for he resuls (Guo e al., 2008). The main inuiion is ha he effec of exogenous or laen facors will be refleced in he covariance marices (Seh 200). The model seup is very similar o ha of condiional G-causaliy. The sole difference lies in he error erms ha are now defined as he sum of he innovaion erm, he noise due o exogenous inpus and laen facors. Thus, he resriced model becomes: p p r r r ' r ' r E L =, where ( ) + + = + + Bx L Χ Χ Χ Χ Χ = = X φ X δ Z ε ε ε ε ε p p r r r ' r ' r E L =, where ( ) + + Y Y = B Y + + Y x L Y = = Y ξ Y ν Ζ ε ε ε ε ε 3.2 Whereas he unresriced akes he form p p p u u u u ' u ' u E L =, where ( ) = + + B L Χ Χ Χ Χ Χ Χ = = = X φ X ζ Y δ Z ε ε ε ε ε p p p u u u u' u ' u E L =, where ( ) Y = + + B Y Y Y Y L Y = = = Y ξ Y φ X ν Ζ ε ε ε ε ε p p p u u u π ρ υ = = = Ζ = X + Y + Ζ + ε ', where ' Ζ ε = ε + ε + BΖ ( L) ε u u u E L Ζ Ζ Ζ Ζ 3.3 Wih heir corresponding noise covariance marices Γ 3 = r ' r ' r ' var(ε ) cov(ε,ε ) Χ Χ Z r ' r ' r ' cov(ε,ε ) var(ε ) Z Χ Z 37

39 u ' u' u' u' u ' var(ε ) cov(ε,ε ) cov(ε,ε ) Χ Χ Y Χ Z u ' u' u ' u ' u' Γ 4 = cov(ε,ε ) var(ε ) cov(ε,ε ) Y Χ Y Y Z u' u' u' u' u' cov(ε,ε ) cov(ε,ε ) var(ε ) Z Χ Z Y Z By pariioning he noise variance marices, o manage o conrol for he effec ha he exogenous inpu have, we rerieve r r ' r ' r ' r ' r ' Γ 3 = var(ε ) cov(ε,ε )var(ε ) cov(ε,ε ) Χ Χ Z Z Z Χ u u ' u ' u ' u ' u ' Γ 4 = var(ε ) cov(ε,ε )var(ε ) cov(ε,ε ) Χ Χ Z Z Z Χ 3.4 The Parial Causaliy Index is he log raio of he wo pariioned noise covariance marices p F = ln( Γ / Γ ) 3.5 Y X 3 4 Unlike sandard and condiional Granger Causaliy indices, he saisical disribuion of he Parial Causaliy Index is unknown o us, as we are unaware of he saisical properies of he exogenous facors (Seh, 200), so we have o resor o randomizaion ess in order o be able o draw some conclusion upon he saisical significance of our indicaors. 38

40 4. Transfer Enropy 4.. Enropy as an informaion meric Transfer enropy is a concep inroduced by Schreiber o describe a measure commonly used in informaion heory ha would ake ino accoun muual informaion bu would also share he dynamics of informaion ranspor ino accoun (Schreiber, 2000). Though difficul o read, enropy in informaion heory has an one-on-one relaionship o ha of causaliy, as enropy is a erm used o describe he degree a which a bi of informaion reveals he rue message. As Schreiber described in his paper, his informaion meric quanifies he saisical coherence beween sysems evolving in ime. Shannon was he firs one o borrow he classic Bolzman enropy and inser i ino informaion heory as a measure of uncerainy. I is imporan o noe ha informaion heory analyzes sochasic processes which generae discree bis of informaion. As Shannon menions in his work (Shannon, 948), coninuous processes can be regarded as discree sources, provided we can apply some discreizaion process. The Shannon enropy is mahemaically formulaed as following: Le p i wih i=,2,,n be he probabiliy of a se of n possible evens. I is more inuiive o hink of a sae space comprising of n cells, and being a he cell i. p i represens he probabiliy of a sysem Then is enropy will be given by (Shannon, 948): n H = pi log( pi ) 4. i= 39

41 Now consider wo evens x and y, where p( i), p( ) are he probabiliy ses for x and y respecively and p( i, ) represens he oin occurrence of i for x and for y. Then he enropy for each even will form as following, H ( x) = p( i, ) log p( i, ) i, H ( y) = p( i, ) log p( i, ) i, i and he oin occurrence of he evens is given by H( x, y) = p( i, ) log p( i, ) i, whereas if he wo evens are independen, H ( x, y) = p( i, ) log( p( i) p( )) I i, Then he muual informaion beween X and Y evens will be given by he difference of he wo evens (acual relaionship minus he case hey where independen) MI = H ( x, y) H ( x, y) xy I MI = p ( i, ) log( p ( i) p ( )) + p ( i, ) log( p ( i, )) xy xy x y xy xy i, i, MI = p ( i, ) [ log( p ( i, )) log( p ( i) p ( ))] xy xy xy x y i pxy ( i, ) MI xy = pxy ( i, ) log i px ( i) py ( )

42 4.2. Transfer Enropy-An Indicaor Schreiber was he firs one o coin he erm ransfer enropy, ha would isolae he dynamics of muual informaion (due o a common hisory) revealing hus he direcion of informaion exchange beween wo sysems. Transfer Enropy can be represened in wo equivalen ways, one using he expressions of Shannon enropy, and one resembling he muual informaion crierion. Below we presen boh formulaions for consideraion. Consider wo sochasic variables X and Y. The informaion ha variable Y conains on variable X a h ime seps forward, based also on he curren saus of X is called ransfer enropy from Y o X. As already menioned, in order o be able o measure ransfer enropy, hen we have o be able o discreize he sochasic processes of X and Y, wih he binning of heir coninuous ime span ino ( ) = m τ +,..., n h ime seps, wih m being he embedding dimension used for he phase space reconsrucion, τ represening he delay dimension, and h he ime horizon ahead, ha we wish o examine. The vecors x and x = [ x, x,..., x ]' y denoe he phase sae represenaion of X and Y respecively, wih ( ) τ m τ y = [ y, y,..., y ]' ( ) τ m τ The expression of ransfer enropy from Y o X is given by T = H ( x x ) H ( x x, y ) 4.3 Y X + h + h and can be expressed hus as he difference beween x + h, condiioned on is own pas and on he pas of y (Barne, Seh, 2009). 4

43 The represenaion of TY X wih a formulaion resembling ha of muual informaion is given by he following ype (Schreiber, 2000). T = p( x, x, y ) log Y X + h p( x+ h x, y ) p( x x ) + h 42

44 4.3. Parial Transfer Enropy-An Indicaor The parial ransfer enropy, is an exension of he measure of ransfer enropy. The main feaure ha renders Parial Transfer Enropy as an imporan improvemen over he sandard Transfer Enropy measure is he fac ha i is able o conrol for he effec of a hird variable Z, hrough is inclusion of he oin probabiliy densiy funcions in he expression of he enropies (Papana e al., 202). Consider he wo vecors we inroduced in he previous chaper, wih idenical paramers (embedding dimension, ime delay). x = [ x, x,..., x ]' ( -) τ m y = [ y, y,..., y ]' ( -) τ m τ τ Le Z be a hird sochasic variable and z = [ z, z τ,..., z ( -) ]' m τ be is vecor phase sae represenaion. Then he Transfer Enropy from y o PTE = H ( x x, z ) H ( x x, y, z ) 4.4 Y X + h + h x conrolling for he effec of z will be given by The PTEY X can also be expressed in he muual informaion formulaion, by insering he condiioning on z ino he log raio (Schreiber, 2000). 43

45 4.4. Saisical Inference on TE and PTE As he saisical disribuions of he Transfer Enropy and Parial Transfer Enropy measures are no known o us, we are no able o perform any parameric ess o es heir saisical significance. In he lieraure here can be found some surrogae daa mehods, such as boosrapping and random permuaion (Seh, 200). We will also resor o a surrogae daa mehod (Kugiumzis, 2000) in order o consruc p-values on he measures and herefore be able o es he null hypohesis ha he ime series ha we examine for causaliy, do no bear any links. The inuiion behind he surrogae daa is ha afer reshifing he causal variable, hen he wo ime series bear no correlaion. Wih random resampling, we creae a disribuion of lack of correlaion, herefore we can check he null hypohesis wheher he value ha our measure has aken belongs o his disribuion of no correlaion. 44

46 5. Granger Causaliy and Transfer Enropy on Simulaed Daa 5.. Inroducion In he previous chapers we have presened ways one can accoun for causaliy beween variables. In his secion we will demonsrae four causaliy measures ha we have previously inroduced; Granger Causaliy, Condiional Granger Causaliy, Transfer Enropy and Parial Transfer Enropy. Thus we can compare he resuls ha he parameric ess deliver, in conuncion o he corresponding ones from he informaion-heoreic ess. To his end, we have applied he measures on hree variables which hide causal links. The hree variables ha we esed were consruced so as o follow cerain assumpions. Our goal was o explore he dynamics of each measure under differen models; herefore, we creaed hree ses of variables, each one generaed under differen assumpions in each sysem: a linear, a slighly disored linear one, as well as a non-linear. For each one of hese sysems, we check wheher he four causaliy merics give us back subsanial and saisically significan resuls upon he causaliy relaions of he sysem. The firs se of variables is based on a rivial linear sysem. The second sysem generaing our variables is very similar o he iniial linear sysem, wih a sligh differeniaion; by applying absolue values on our variables, we explore he behavior of our ess under a sligh deviaion from normaliy. Finally, we have also used a non-linear sysem o furher explore he possibiliies of he four mehods. 45

47 5.2. The esing procedure The seps from our esing procedure are described in he following. Firs we define he sysem which should generae he ime series we would like o es. The ime series creaed are being fied agains an AR model, and he residuals are being esed wheher hey belong o a whie noise process. This is o safeguard ha hey fulfill he crieria in order o be insered in he analysis. The four ess are applied on he hree generaed ime series, and we receive he values of he respecive causaliy measures. Because we have no sandardized way of measuring he significance of he informaionheoreic measures (ransfer enropy and parial ransfer enropy), we follow a surrogae procedure ha was shorly inroduced in previous chaper: A hundred of new ime series are generaed by randomly shifing he originally generaed series (he causal one). The purpose of his process is o creae new ses of series ha are uncorrelaed and have no causal links. Having creaed he 00 pairs, we hen calculae again he causaliy measures for each one of hese 00 ses. Thus, we rerieve a disribuion of causaliy values in cases where we have no causaliy. In he final sep, we compare he value ha he causaliy measures delivered us for he original ime series o he disribuion of values ha he non-causal surrogaes delivered. Obviously, if he original value falls ino he disribuion of he non-correlaed surrogaes, hen his would mean ha we do no have subsanial proof of causaliy for he original ime series as well. If however, he value is significanly differen from he values ha he surrogaes delivered, or lies in he ails of he non-correlaed disribuion (5% of each ail), hen we have subsanial evidence ha he wo ime series we are examining bear causal links. 46

48 Figure 5- If he value of he causaliy measure lies on he ales of he disribuion, hen we reec he null hypohesis of non- causaliy This procedure is repeaed 00 imes, o reveal he exen o which he reecions of our null hypohesis are randomly or seadily replicaed. I is imporan o noe ha we have no used any parameric ess for he Granger and Condiional Granger Causaliy. As we waned o creae a common benchmark for comparing he four measures, we have used he same procedure of surrogaes for saisical inference upon all four of hem, despie he fac ha he disribuion of he parameric ess is known, and ha confidence inervals can be consruced for hem. The specificaions of he sysems generaing he ime series, as well of he ess are shorly described below: In all hree sysems we have generaed hree ime series of lengh equal o 024 observaions. Therefore, for each measure, here were evaluaed 3x3 causaliy measures (of course he diagonal elemens, esing he causaliy of one variable owards iself resuled o zero). We chose a low order for our model, as we do no wan o decrease dramaically our es power. The embedding delay was se o 2, whereas he ime delay was se o, for compuaional reasons. We also se embedding delay and ime delay parameers for each pair of variables esed equal, as i is common use (Schreiber, 2000). Mehod of neares neighbors (Kraskov, 2004) was used o esimae he enropies, for he 47

49 calculaion of ransfer enropy; he arbirary parameer of number of neighbors for he densiy esimaion was se o 5. 48

50 5.3. Linear Sysem The linear sysem we used o creae he firs se of variables is he following: X = 0.4X 0.5X + e X,,, 2, = 0.4X 0.3X + 0.6X + e 5. 2, 2, 2, 2, 2, X = 0.5X 0.7X 0.3X + e 3, 3, 3, 2 2, 3, I is eviden in our VAR formulaion ha while he firs variable X of our sysem is no impaced by any oher variables excep for is own pas, i however influences he fuure values of X2. X2 on he oher hand direcly impacs on X3 whereas X3 does no influence any oher variable. X X2 X3 Figure 5-2 Represenaion of he rue causaliy links in he linear sysem Following, we will explore he causal links wihin he sysem hrough he four causaliy measures we have presened, and we will see how hese heoreical measures respond in praxis. The four measures we will be using are: Granger Causaliy Index, Condiional Granger causaliy Index, Transfer Enropy and Condiional Transfer Enropy. In he following diagram we see he ime series ha our sysem generaed, 49

51 Figure 5-3 The ime series ha he linear sysem has generaed, n=024 and i is obvious even from he char ha he series are saionary and able o be used in our analysis. The following able gives us an overview of he resuls we rerieved. In deail, he figures show us he percenage of successful causal links, ha is, of he imes we managed o reec he null hypohesis of no causaliy. Parameric Tess Informaion Theoreic Tess Condiional Granger Causaliy Transfer Enropy Table 5- Overview of empirical resuls on a linear sysem: Parial Transfer Enropy Var Granger Causaliy X X2 00% 00% 00% 00% X X3 00% 3% 84% 6% X2 X % 2% 6% 9% X2 X3 00% 00% 00% 88% X3 X 7% 7% 0% 9% X3 X2 6% % 2% 5% 50

52 The able presens he percenage of significan causaliies (reeced null hypoheses of non causaliy) found in 00 repeiions. As we can see in he able above, he Granger Causaliy es deeced causal links from X owards X2 and X3, as well as a link from X2 o X3 in all 00 repeiions, failing hus, o recognize he indirec flow of informaion from X o X3 hrough X2. The measure of Condiional Granger Causaliy improved our findings as he null hypohesis was reeced in only 3 ou of 00 repeiions. As for he remaining, rue causal relaions of he sysem, he measure of Condiional Granger Causaliy had a 00% success in idenifying he in all repeiions. Transfer Enropy exposed he same links wih Granger Causaliy Index, as i idenified he indirec link beween X and X3. I is herefore sill vulnerable o indirec causaliy, while parial ransfer enropy has resored his false link. The conclusions driving his comparaive sudy in a simple linear sysem are quie sraigh forward and in line wih he bibliography. All four measures have revealed he fundamenal links hroughou our sysem, while condiioning on he hird variable has refined our findings, in boh parameric and informaion-heoreic measures, by eliminaing cases of indirec causaions. The comparison has proven ha he parameric ess gave us back more robus resuls, as he same significan causaions were replicaed in all 00 replicaions, whereas he informaion heoreic measures, did no deliver he same success rae in all cases. 5

53 Figure 5-4 The figures depic he values ha he four causaliy measures ook during 00 realizaions for he linear sysem. The figure above represens graphically he values ha he causaliy indexes ook for he original ime series esed hroughou he 00 repeiions of he esing procedure. For he parameric es we can have a sraighforward inerpreaion of his graphic, because he furher away from zero he value of our causaliy index, he sronger he causaion i reflecs (provided of course, i is also found saisically significan). So we can also see graphically how he indirec link from X owards X3 is already weaker han he remaining wo causal relaions, and ha by condiioning on X2, his link vanishes. 52

54 5.4. adisored linear sysem In he previous secion, we esed causal relaions of a simple linear sysem using four differen echniques. Bu wha if we were o slighly aler he sysem we are examining? Would his significanly aler he oucome of he iniial paradigm? We will inroduce a sligh disorion o he sysem used in chaper 5.3; ha is inroducing an absolue value o he auoregressive par of he deerminisic equaion of each variable comprising our VAR sysem. X = 0.4 X 0.5X + e X,,, 2, = 0.4 X 0.3X + 0.6X + e 5.2 2, 2, 2, 2, X = 0.5 X 0.7X 0. 3X + e 3, 3, 3, 2 2, 3, The underlying connecions of our sysem are he same as he ones of he linear sysem, graphically shown in he below flow char: X X2 X3 Figure 5-3 Represenaion of he rue causaliy linkes in a disored linear sysem The series ha he sysem generaed are depiced in he following graph: 53

55 Figure 5-5 The ime series ha he disored linear sysem has generaed, n=024 Again he original se of series is underlined by a mean revering procedure, being able o be furher analyzed. Parameric Tess Informaion Theoreic Tess Parial Var Granger Causaliy Condiional Granger Causaliy Transfer Enropy Transfer Enropy X X2 00% 00% 00% 00% X X3 00% 0% 65% 20% X2 X 5% 5% 20% 0% X2 X3 00% 00% 00% 00% X3 X 5% 0% 5% 5% X3 X2 5% 0% 5% 5% Table 5-2 Overview of empirical resuls on a disored linear sysem: The able presens he percenage of significan causaliies (reeced null hypoheses of no causaliy) found in 00 repeiions. As we can see in he able above, he Granger Causaliy es has exacly he same behavior as in he linear sysem; i has deeced causal links from X owards X2 and X3, as well as a 54

56 link from X2 o X3 in all 00 repeiions. Thus, he sandard Granger Causaliy Index failed again o recognize he indirec flow of informaion from X o X3 hrough X2. The remaining hree ess were more successful in he sense ha rue links were found ou wih a 00% rae of success. The conclusions ha he disored linear sysem has led us o are very similar, almos equivalen o hose of he linear sysem. All four measures have revealed he fundamenal links hroughou our sysem, whereas condiioning on he hird variable has refined our findings, in boh parameric and informaion-heoreic measures, by eliminaing cases of indirec causaions. An ineresing fac when comparing Granger Causaliy and Transfer Enropy resuls, is ha he Transfer Enropy indicaor performed beer in comparison o is corresponding parameric es, as i failed o reec he hypohesis of non-causaliy in 35% of he cases, in conradicion o G-causaliy ha reeced he null in all insances. In oher words, ransfer enropy managed in some cases o reveal he indirec flow of informaion from X o X3. So, he examinaion of he disored linear sysem allowed us o see ha he informaion heoreic es have a sligh advanage in environmens where variables migh deviae from normaliy, or exhibi peculiar behavior, or even when he researcher is unaware of possible indirec links wihin he sysem. 55

57 Figure 5-62 The figures depic he values ha he four causaliy measures ook during 00 realizaions for he disored linear sysem. 56

58 5.5. Non-linear sysem Challenging he robusness of he four measures, we ook one sep furher and esed all four of hem on a purely non-linear sysem. We chose no o disurb he iniial assumpions and o keep he same causal links hroughou he sysem. So our sysem becomes he following: X = 0.4X 0.5X + e,,,, X = 0.4X 0.3 X (0.4 X ) + e 2, 2, 2,, 2, X = 0.5X 0.3X + e 3, 3, 2, 3, (0.7 X 3, ) 5.3 The pair of original series generaed by he sysem is depiced in he below graph. Figure 5-7 The ime series ha he non-linear sysem has generaed, n=024 57

Χρονοσειρές Μάθημα 3

Χρονοσειρές Μάθημα 3 Χρονοσειρές Μάθημα 3 Ασυσχέτιστες (λευκός θόρυβος) και ανεξάρτητες (iid) παρατηρήσεις Chafield C., The Analysis of Time Series, An Inroducion, 6 h ediion,. 38 (Chaer 3): Some auhors refer o make he weaker

Διαβάστε περισσότερα

ΕΡΓΑΣΙΑ ΜΑΘΗΜΑΤΟΣ: ΘΕΩΡΙΑ ΒΕΛΤΙΣΤΟΥ ΕΛΕΓΧΟΥ ΦΙΛΤΡΟ KALMAN ΜΩΥΣΗΣ ΛΑΖΑΡΟΣ

ΕΡΓΑΣΙΑ ΜΑΘΗΜΑΤΟΣ: ΘΕΩΡΙΑ ΒΕΛΤΙΣΤΟΥ ΕΛΕΓΧΟΥ ΦΙΛΤΡΟ KALMAN ΜΩΥΣΗΣ ΛΑΖΑΡΟΣ ΑΡΙΣΤΟΤΕΛΕΙΟ ΠΑΝΕΠΙΣΤΗΜΙΟ ΘΕΣΣΑΛΟΝΙΚΗΣ ΤΜΗΜΑ ΜΑΘΗΜΑΤΙΚΩΝ ΜΕΤΑΠΤΥΧΙΑΚΟ ΠΡΟΓΡΑΜΜΑ ΣΠΟΥΔΩΝ ΘΕΩΡΗΤΙΚΗ ΠΛΗΡΟΦΟΡΙΚΗ ΚΑΙ ΘΕΩΡΙΑ ΣΥΣΤΗΜΑΤΩΝ & ΕΛΕΓΧΟΥ ΕΡΓΑΣΙΑ ΜΑΘΗΜΑΤΟΣ: ΘΕΩΡΙΑ ΒΕΛΤΙΣΤΟΥ ΕΛΕΓΧΟΥ ΦΙΛΤΡΟ KALMAN ΜΩΥΣΗΣ

Διαβάστε περισσότερα

9.1 Introduction 9.2 Lags in the Error Term: Autocorrelation 9.3 Estimating an AR(1) Error Model 9.4 Testing for Autocorrelation 9.

9.1 Introduction 9.2 Lags in the Error Term: Autocorrelation 9.3 Estimating an AR(1) Error Model 9.4 Testing for Autocorrelation 9. 9.1 Inroducion 9.2 Lags in he Error Term: Auocorrelaion 9.3 Esimaing an AR(1) Error Model 9.4 Tesing for Auocorrelaion 9.5 An Inroducion o Forecasing: Auoregressive Models 9.6 Finie Disribued Lags 9.7

Διαβάστε περισσότερα

The Student s t and F Distributions Page 1

The Student s t and F Distributions Page 1 The Suden s and F Disribuions Page The Fundamenal Transformaion formula for wo random variables: Consider wo random variables wih join probabiliy disribuion funcion f (, ) simulaneously ake on values in

Διαβάστε περισσότερα

4.6 Autoregressive Moving Average Model ARMA(1,1)

4.6 Autoregressive Moving Average Model ARMA(1,1) 84 CHAPTER 4. STATIONARY TS MODELS 4.6 Autoregressive Moving Average Model ARMA(,) This section is an introduction to a wide class of models ARMA(p,q) which we will consider in more detail later in this

Διαβάστε περισσότερα

Appendix. The solution begins with Eq. (2.15) from the text, which we repeat here for 1, (A.1)

Appendix. The solution begins with Eq. (2.15) from the text, which we repeat here for 1, (A.1) Aenix Aenix A: The equaion o he sock rice. The soluion egins wih Eq..5 rom he ex, which we reea here or convenience as Eq.A.: [ [ E E X, A. c α where X u ε, α γ, an c α y AR. Take execaions o Eq. A. as

Διαβάστε περισσότερα

( ) ( ) ( ) Fourier series. ; m is an integer. r(t) is periodic (T>0), r(t+t) = r(t), t Fundamental period T 0 = smallest T. Fundamental frequency ω

( ) ( ) ( ) Fourier series. ; m is an integer. r(t) is periodic (T>0), r(t+t) = r(t), t Fundamental period T 0 = smallest T. Fundamental frequency ω Fourier series e jm when m d when m ; m is an ineger. jm jm jm jm e d e e e jm jm jm jm r( is periodi (>, r(+ r(, Fundamenal period smalles Fundamenal frequeny r ( + r ( is periodi hen M M e j M, e j,

Διαβάστε περισσότερα

Key Formulas From Larson/Farber Elementary Statistics: Picturing the World, Second Edition 2002 Prentice Hall

Key Formulas From Larson/Farber Elementary Statistics: Picturing the World, Second Edition 2002 Prentice Hall 64_INS.qxd /6/0 :56 AM Page Key Formulas From Larson/Farber Elemenary Saisics: Picuring he World, Second Ediion 00 Prenice Hall CHAPTER Class Widh = round up o nex convenien number Maximum daa enry - Minimum

Διαβάστε περισσότερα

Reservoir modeling. Reservoir modelling Linear reservoirs. The linear reservoir, no input. Starting up reservoir modeling

Reservoir modeling. Reservoir modelling Linear reservoirs. The linear reservoir, no input. Starting up reservoir modeling Reservoir modeling Reservoir modelling Linear reservoirs Paul Torfs Basic equaion for one reservoir:) change in sorage = sum of inflows minus ouflows = Q in,n Q ou,n n n jus an ordinary differenial equaion

Διαβάστε περισσότερα

University of Washington Department of Chemistry Chemistry 553 Spring Quarter 2010 Homework Assignment 3 Due 04/26/10

University of Washington Department of Chemistry Chemistry 553 Spring Quarter 2010 Homework Assignment 3 Due 04/26/10 Universiy of Washingon Deparmen of Chemisry Chemisry 553 Spring Quarer 1 Homework Assignmen 3 Due 4/6/1 v e v e A s ds: a) Show ha for large 1 and, (i.e. 1 >> and >>) he velociy auocorrelaion funcion 1)

Διαβάστε περισσότερα

HOMEWORK 4 = G. In order to plot the stress versus the stretch we define a normalized stretch:

HOMEWORK 4 = G. In order to plot the stress versus the stretch we define a normalized stretch: HOMEWORK 4 Problem a For the fast loading case, we want to derive the relationship between P zz and λ z. We know that the nominal stress is expressed as: P zz = ψ λ z where λ z = λ λ z. Therefore, applying

Διαβάστε περισσότερα

Lecture 12 Modulation and Sampling

Lecture 12 Modulation and Sampling EE 2 spring 2-22 Handou #25 Lecure 2 Modulaion and Sampling The Fourier ransform of he produc of wo signals Modulaion of a signal wih a sinusoid Sampling wih an impulse rain The sampling heorem 2 Convoluion

Διαβάστε περισσότερα

3 Frequency Domain Representation of Continuous Signals and Systems

3 Frequency Domain Representation of Continuous Signals and Systems 3 Frequency Domain Represenaion of Coninuous Signals and Sysems 3. Fourier Series Represenaion of Periodic Signals............. 2 3.. Exponenial Fourier Series.................... 2 3..2 Discree Fourier

Διαβάστε περισσότερα

2 Composition. Invertible Mappings

2 Composition. Invertible Mappings Arkansas Tech University MATH 4033: Elementary Modern Algebra Dr. Marcel B. Finan Composition. Invertible Mappings In this section we discuss two procedures for creating new mappings from old ones, namely,

Διαβάστε περισσότερα

16. 17. r t te 2t i t 1. 18 19 Find the derivative of the vector function. 19. r t e t cos t i e t sin t j ln t k. 31 33 Evaluate the integral.

16. 17. r t te 2t i t 1. 18 19 Find the derivative of the vector function. 19. r t e t cos t i e t sin t j ln t k. 31 33 Evaluate the integral. SECTION.7 VECTOR FUNCTIONS AND SPACE CURVES.7 VECTOR FUNCTIONS AND SPACE CURVES A Click here for answers. S Click here for soluions. Copyrigh Cengage Learning. All righs reserved.. Find he domain of he

Διαβάστε περισσότερα

= e 6t. = t 1 = t. 5 t 8L 1[ 1 = 3L 1 [ 1. L 1 [ π. = 3 π. = L 1 3s = L. = 3L 1 s t. = 3 cos(5t) sin(5t).

= e 6t. = t 1 = t. 5 t 8L 1[ 1 = 3L 1 [ 1. L 1 [ π. = 3 π. = L 1 3s = L. = 3L 1 s t. = 3 cos(5t) sin(5t). Worked Soluion 95 Chaper 25: The Invere Laplace Tranform 25 a From he able: L ] e 6 6 25 c L 2 ] ] L! + 25 e L 5 2 + 25] ] L 5 2 + 5 2 in(5) 252 a L 6 + 2] L 6 ( 2)] 6L ( 2)] 6e 2 252 c L 3 8 4] 3L ] 8L

Διαβάστε περισσότερα

6.003: Signals and Systems. Modulation

6.003: Signals and Systems. Modulation 6.3: Signals and Sysems Modulaion December 6, 2 Subjec Evaluaions Your feedback is imporan o us! Please give feedback o he saff and fuure 6.3 sudens: hp://web.mi.edu/subjecevaluaion Evaluaions are open

Διαβάστε περισσότερα

CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS

CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS CHAPTER 5 SOLVING EQUATIONS BY ITERATIVE METHODS EXERCISE 104 Page 8 1. Find the positive root of the equation x + 3x 5 = 0, correct to 3 significant figures, using the method of bisection. Let f(x) =

Διαβάστε περισσότερα

Other Test Constructions: Likelihood Ratio & Bayes Tests

Other Test Constructions: Likelihood Ratio & Bayes Tests Other Test Constructions: Likelihood Ratio & Bayes Tests Side-Note: So far we have seen a few approaches for creating tests such as Neyman-Pearson Lemma ( most powerful tests of H 0 : θ = θ 0 vs H 1 :

Διαβάστε περισσότερα

ω = radians per sec, t = 3 sec

ω = radians per sec, t = 3 sec Secion. Linear and Angular Speed 7. From exercise, =. A= r A = ( 00 ) (. ) = 7,00 in 7. Since 7 is in quadran IV, he reference 7 8 7 angle is = =. In quadran IV, he cosine is posiive. Thus, 7 cos = cos

Διαβάστε περισσότερα

Math221: HW# 1 solutions

Math221: HW# 1 solutions Math: HW# solutions Andy Royston October, 5 7.5.7, 3 rd Ed. We have a n = b n = a = fxdx = xdx =, x cos nxdx = x sin nx n sin nxdx n = cos nx n = n n, x sin nxdx = x cos nx n + cos nxdx n cos n = + sin

Διαβάστε περισσότερα

( ) ( t) ( 0) ( ) dw w. = = β. Then the solution of (1.1) is easily found to. wt = t+ t. We generalize this to the following nonlinear differential

( ) ( t) ( 0) ( ) dw w. = = β. Then the solution of (1.1) is easily found to. wt = t+ t. We generalize this to the following nonlinear differential Periodic oluion of van der Pol differenial equaion. by A. Arimoo Deparmen of Mahemaic Muahi Iniue of Technology Tokyo Japan in Seminar a Kiami Iniue of Technology January 8 9. Inroducion Le u conider a

Διαβάστε περισσότερα

6.003: Signals and Systems

6.003: Signals and Systems 6.3: Signals and Sysems Modulaion December 6, 2 Communicaions Sysems Signals are no always well mached o he media hrough which we wish o ransmi hem. signal audio video inerne applicaions elephone, radio,

Διαβάστε περισσότερα

The Euler Equations! λ 1. λ 2. λ 3. ρ ρu. E = e + u 2 /2. E + p ρ. = de /dt. = dh / dt; h = h( T ); c p. / c v. ; γ = c p. p = ( γ 1)ρe. c v.

The Euler Equations! λ 1. λ 2. λ 3. ρ ρu. E = e + u 2 /2. E + p ρ. = de /dt. = dh / dt; h = h( T ); c p. / c v. ; γ = c p. p = ( γ 1)ρe. c v. hp://www.nd.ed/~gryggva/cfd-corse/ The Eler Eqaions The Eler Eqaions The Eler eqaions for D flow: + + p = x E E + p where Define E = e + / H = h + /; h = e + p/ Gréar Tryggvason Spring 3 Ideal Gas: p =

Διαβάστε περισσότερα

1. Ευθύγραμμη ομαλή κίνηση 2. Εξίσωση κίνησης 3. Μετατόπιση & διάστημα 4. ιάγραμμα ταχύτητας χρόνου 5. Στρατηγική λύσης προβλημάτων.

1. Ευθύγραμμη ομαλή κίνηση 2. Εξίσωση κίνησης 3. Μετατόπιση & διάστημα 4. ιάγραμμα ταχύτητας χρόνου 5. Στρατηγική λύσης προβλημάτων. 24/9/214 Γενική Φσική Κωνσταντίνος Χ. Παύλο Φσικός Ραδιοηλεκτρολόγος (MSc) Καστοριά, Σεπτέμβριος 14 1. 2. Εξίσωση κίνησης 3. Μετατόπιση & διάστημα 4. ιάγραμμα ταχύτητας χρόνο 5. ονομάζεται η κίνηση πο

Διαβάστε περισσότερα

Problem Set 3: Solutions

Problem Set 3: Solutions CMPSCI 69GG Applied Information Theory Fall 006 Problem Set 3: Solutions. [Cover and Thomas 7.] a Define the following notation, C I p xx; Y max X; Y C I p xx; Ỹ max I X; Ỹ We would like to show that C

Διαβάστε περισσότερα

Solution Series 9. i=1 x i and i=1 x i.

Solution Series 9. i=1 x i and i=1 x i. Lecturer: Prof. Dr. Mete SONER Coordinator: Yilin WANG Solution Series 9 Q1. Let α, β >, the p.d.f. of a beta distribution with parameters α and β is { Γ(α+β) Γ(α)Γ(β) f(x α, β) xα 1 (1 x) β 1 for < x

Διαβάστε περισσότερα

Homework 3 Solutions

Homework 3 Solutions Homework 3 Solutions Igor Yanovsky (Math 151A TA) Problem 1: Compute the absolute error and relative error in approximations of p by p. (Use calculator!) a) p π, p 22/7; b) p π, p 3.141. Solution: For

Διαβάστε περισσότερα

Lecture 2. Soundness and completeness of propositional logic

Lecture 2. Soundness and completeness of propositional logic Lecture 2 Soundness and completeness of propositional logic February 9, 2004 1 Overview Review of natural deduction. Soundness and completeness. Semantics of propositional formulas. Soundness proof. Completeness

Διαβάστε περισσότερα

6.3 Forecasting ARMA processes

6.3 Forecasting ARMA processes 122 CHAPTER 6. ARMA MODELS 6.3 Forecasting ARMA processes The purpose of forecasting is to predict future values of a TS based on the data collected to the present. In this section we will discuss a linear

Διαβάστε περισσότερα

EE512: Error Control Coding

EE512: Error Control Coding EE512: Error Control Coding Solution for Assignment on Finite Fields February 16, 2007 1. (a) Addition and Multiplication tables for GF (5) and GF (7) are shown in Tables 1 and 2. + 0 1 2 3 4 0 0 1 2 3

Διαβάστε περισσότερα

On Strong Product of Two Fuzzy Graphs

On Strong Product of Two Fuzzy Graphs Inernaional Journal of Scienific and Research Publicaions, Volume 4, Issue 10, Ocober 014 1 ISSN 50-3153 On Srong Produc of Two Fuzzy Graphs Dr. K. Radha* Mr.S. Arumugam** * P.G & Research Deparmen of

Διαβάστε περισσότερα

Concrete Mathematics Exercises from 30 September 2016

Concrete Mathematics Exercises from 30 September 2016 Concrete Mathematics Exercises from 30 September 2016 Silvio Capobianco Exercise 1.7 Let H(n) = J(n + 1) J(n). Equation (1.8) tells us that H(2n) = 2, and H(2n+1) = J(2n+2) J(2n+1) = (2J(n+1) 1) (2J(n)+1)

Διαβάστε περισσότερα

Μηχανική Μάθηση Hypothesis Testing

Μηχανική Μάθηση Hypothesis Testing ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ Μηχανική Μάθηση Hypothesis Testing Γιώργος Μπορμπουδάκης Τμήμα Επιστήμης Υπολογιστών Procedure 1. Form the null (H 0 ) and alternative (H 1 ) hypothesis 2. Consider

Διαβάστε περισσότερα

Statistical Inference I Locally most powerful tests

Statistical Inference I Locally most powerful tests Statistical Inference I Locally most powerful tests Shirsendu Mukherjee Department of Statistics, Asutosh College, Kolkata, India. shirsendu st@yahoo.co.in So far we have treated the testing of one-sided

Διαβάστε περισσότερα

Example Sheet 3 Solutions

Example Sheet 3 Solutions Example Sheet 3 Solutions. i Regular Sturm-Liouville. ii Singular Sturm-Liouville mixed boundary conditions. iii Not Sturm-Liouville ODE is not in Sturm-Liouville form. iv Regular Sturm-Liouville note

Διαβάστε περισσότερα

C.S. 430 Assignment 6, Sample Solutions

C.S. 430 Assignment 6, Sample Solutions C.S. 430 Assignment 6, Sample Solutions Paul Liu November 15, 2007 Note that these are sample solutions only; in many cases there were many acceptable answers. 1 Reynolds Problem 10.1 1.1 Normal-order

Διαβάστε περισσότερα

The conditional CAPM does not explain assetpricing. Jonathan Lewellen & Stefan Nagel. HEC School of Management, March 17, 2005

The conditional CAPM does not explain assetpricing. Jonathan Lewellen & Stefan Nagel. HEC School of Management, March 17, 2005 The condiional CAPM does no explain assepricing anomalies Jonahan Lewellen & Sefan Nagel HEC School of Managemen, March 17, 005 Background Size, B/M, and momenum porfolios, 1964 001 Monhly reurns (%) Avg.

Διαβάστε περισσότερα

Section 8.3 Trigonometric Equations

Section 8.3 Trigonometric Equations 99 Section 8. Trigonometric Equations Objective 1: Solve Equations Involving One Trigonometric Function. In this section and the next, we will exple how to solving equations involving trigonometric functions.

Διαβάστε περισσότερα

derivation of the Laplacian from rectangular to spherical coordinates

derivation of the Laplacian from rectangular to spherical coordinates derivation of the Laplacian from rectangular to spherical coordinates swapnizzle 03-03- :5:43 We begin by recognizing the familiar conversion from rectangular to spherical coordinates (note that φ is used

Διαβάστε περισσότερα

TRM +4!5"2# 6!#!-!2&'!5$27!842//22&'9&2:1*;832<

TRM +4!52# 6!#!-!2&'!5$27!842//22&'9&2:1*;832< TRM!"#$%& ' *,-./ *!#!!%!&!3,&!$-!$./!!"#$%&'*" 4!5"# 6!#!-!&'!5$7!84//&'9&:*;83< #:4

Διαβάστε περισσότερα

Μάθημα 5-6: Στάσιμες πολυμεταβλητές χρονοσειρές και μοντέλα Διασυσχέτιση Διανυσματικά αυτοπαλίνδρομα μοντέλα Δίκτυα από πολυμεταβλητές χρονοσειρές

Μάθημα 5-6: Στάσιμες πολυμεταβλητές χρονοσειρές και μοντέλα Διασυσχέτιση Διανυσματικά αυτοπαλίνδρομα μοντέλα Δίκτυα από πολυμεταβλητές χρονοσειρές Μάθημα 5-6: Στάσιμες πολυμεταβλητές χρονοσειρές και μοντέλα Διασυσχέτιση Διανυσματικά αυτοπαλίνδρομα μοντέλα Δίκτυα από πολυμεταβλητές χρονοσειρές Αιτιότητα κατά Granger Ασκήσεις Ανάλυση μονομεταβλητής

Διαβάστε περισσότερα

Riemann Hypothesis: a GGC representation

Riemann Hypothesis: a GGC representation Riemann Hypohesis: a GGC represenaion Nicholas G. Polson Universiy of Chicago Augus 8, 8 Absrac A GGC Generalized Gamma Convoluion represenaion for Riemann s reciprocal ξ-funcion is consruced. This provides

Διαβάστε περισσότερα

Matrices and Determinants

Matrices and Determinants Matrices and Determinants SUBJECTIVE PROBLEMS: Q 1. For what value of k do the following system of equations possess a non-trivial (i.e., not all zero) solution over the set of rationals Q? x + ky + 3z

Διαβάστε περισσότερα

Anti-aliasing Prefilter (6B) Young Won Lim 6/8/12

Anti-aliasing Prefilter (6B) Young Won Lim 6/8/12 ni-aliasing Prefiler (6B) Copyrigh (c) Young W. Lim. Permission is graned o copy, disribue and/or modify his documen under he erms of he GNU Free Documenaion License, Version. or any laer version published

Διαβάστε περισσότερα

5.4 The Poisson Distribution.

5.4 The Poisson Distribution. The worst thing you can do about a situation is nothing. Sr. O Shea Jackson 5.4 The Poisson Distribution. Description of the Poisson Distribution Discrete probability distribution. The random variable

Διαβάστε περισσότερα

Predictability and Model Selection in the Context of ARCH Models

Predictability and Model Selection in the Context of ARCH Models Predicabiliy and Model Selecion in he Conex of ARCH Models Savros Degiannakis and Evdokia Xekalaki Deparmen of Saisics Ahens Universiy of Economics and Business 76 Paission Sree 434 Ahens Greece echnical

Διαβάστε περισσότερα

Every set of first-order formulas is equivalent to an independent set

Every set of first-order formulas is equivalent to an independent set Every set of first-order formulas is equivalent to an independent set May 6, 2008 Abstract A set of first-order formulas, whatever the cardinality of the set of symbols, is equivalent to an independent

Διαβάστε περισσότερα

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8 questions or comments to Dan Fetter 1

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8  questions or comments to Dan Fetter 1 Eon : Fall 8 Suggested Solutions to Problem Set 8 Email questions or omments to Dan Fetter Problem. Let X be a salar with density f(x, θ) (θx + θ) [ x ] with θ. (a) Find the most powerful level α test

Διαβάστε περισσότερα

Fractional Colorings and Zykov Products of graphs

Fractional Colorings and Zykov Products of graphs Fractional Colorings and Zykov Products of graphs Who? Nichole Schimanski When? July 27, 2011 Graphs A graph, G, consists of a vertex set, V (G), and an edge set, E(G). V (G) is any finite set E(G) is

Διαβάστε περισσότερα

d dt S = (t)si d dt R = (t)i d dt I = (t)si (t)i

d dt S = (t)si d dt R = (t)i d dt I = (t)si (t)i d d S = ()SI d d I = ()SI ()I d d R = ()I d d S = ()SI μs + fi + hr d d I = + ()SI (μ + + f + ())I d d R = ()I (μ + h)r d d P(S,I,) = ()(S +1)(I 1)P(S +1, I 1, ) +()(I +1)P(S,I +1, ) (()SI + ()I)P(S,I,)

Διαβάστε περισσότερα

ΟΙΚΟΝΟΜΟΤΕΧΝΙΚΗ ΑΝΑΛΥΣΗ ΕΝΟΣ ΕΝΕΡΓΕΙΑΚΑ ΑΥΤΟΝΟΜΟΥ ΝΗΣΙΟΥ ΜΕ Α.Π.Ε

ΟΙΚΟΝΟΜΟΤΕΧΝΙΚΗ ΑΝΑΛΥΣΗ ΕΝΟΣ ΕΝΕΡΓΕΙΑΚΑ ΑΥΤΟΝΟΜΟΥ ΝΗΣΙΟΥ ΜΕ Α.Π.Ε Τμήμα Ηλεκτρονικών Μηχανικών Τ.Ε. ΟΙΚΟΝΟΜΟΤΕΧΝΙΚΗ ΑΝΑΛΥΣΗ ΕΝΟΣ ΕΝΕΡΓΕΙΑΚΑ ΑΥΤΟΝΟΜΟΥ ΝΗΣΙΟΥ ΜΕ Α.Π.Ε Πτυχιακή Εργασία Φοιτητής: Γεμενής Κωνσταντίνος ΑΜ: 30931 Επιβλέπων Καθηγητής Κοκκόσης Απόστολος Λέκτορας

Διαβάστε περισσότερα

6.1. Dirac Equation. Hamiltonian. Dirac Eq.

6.1. Dirac Equation. Hamiltonian. Dirac Eq. 6.1. Dirac Equation Ref: M.Kaku, Quantum Field Theory, Oxford Univ Press (1993) η μν = η μν = diag(1, -1, -1, -1) p 0 = p 0 p = p i = -p i p μ p μ = p 0 p 0 + p i p i = E c 2 - p 2 = (m c) 2 H = c p 2

Διαβάστε περισσότερα

Phys460.nb Solution for the t-dependent Schrodinger s equation How did we find the solution? (not required)

Phys460.nb Solution for the t-dependent Schrodinger s equation How did we find the solution? (not required) Phys460.nb 81 ψ n (t) is still the (same) eigenstate of H But for tdependent H. The answer is NO. 5.5.5. Solution for the tdependent Schrodinger s equation If we assume that at time t 0, the electron starts

Διαβάστε περισσότερα

Managing Production-Inventory Systems with Scarce Resources

Managing Production-Inventory Systems with Scarce Resources Managing Producion-Invenory Sysems wih Scarce Resources Online Supplemen Proof of Lemma 1: Consider he following dynamic program: where ḡ (x, z) = max { cy + E f (y, z, D)}, (7) x y min(x+u,z) f (y, z,

Διαβάστε περισσότερα

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013 Notes on Average Scattering imes and Hall Factors Jesse Maassen and Mar Lundstrom Purdue University November 5, 13 I. Introduction 1 II. Solution of the BE 1 III. Exercises: Woring out average scattering

Διαβάστε περισσότερα

3.4 SUM AND DIFFERENCE FORMULAS. NOTE: cos(α+β) cos α + cos β cos(α-β) cos α -cos β

3.4 SUM AND DIFFERENCE FORMULAS. NOTE: cos(α+β) cos α + cos β cos(α-β) cos α -cos β 3.4 SUM AND DIFFERENCE FORMULAS Page Theorem cos(αβ cos α cos β -sin α cos(α-β cos α cos β sin α NOTE: cos(αβ cos α cos β cos(α-β cos α -cos β Proof of cos(α-β cos α cos β sin α Let s use a unit circle

Διαβάστε περισσότερα

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 19/5/2007

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 19/5/2007 Οδηγίες: Να απαντηθούν όλες οι ερωτήσεις. Αν κάπου κάνετε κάποιες υποθέσεις να αναφερθούν στη σχετική ερώτηση. Όλα τα αρχεία που αναφέρονται στα προβλήματα βρίσκονται στον ίδιο φάκελο με το εκτελέσιμο

Διαβάστε περισσότερα

Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit

Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit Ting Zhang Stanford May 11, 2001 Stanford, 5/11/2001 1 Outline Ordinal Classification Ordinal Addition Ordinal Multiplication Ordinal

Διαβάστε περισσότερα

Econ Spring 2004 Instructor: Prof. Kiefer Solution to Problem set # 5. γ (0)

Econ Spring 2004 Instructor: Prof. Kiefer Solution to Problem set # 5. γ (0) Cornell University Department of Economics Econ 60 - Spring 004 Instructor: Prof. Kiefer Solution to Problem set # 5. Autocorrelation function is defined as ρ h = γ h γ 0 where γ h =Cov X t,x t h =E[X

Διαβάστε περισσότερα

Nonlinear Analysis: Modelling and Control, 2013, Vol. 18, No. 4,

Nonlinear Analysis: Modelling and Control, 2013, Vol. 18, No. 4, Nonlinear Analysis: Modelling and Conrol, 23, Vol. 8, No. 4, 493 58 493 Exisence and uniqueness of soluions for a singular sysem of higher-order nonlinear fracional differenial equaions wih inegral boundary

Διαβάστε περισσότερα

Congruence Classes of Invertible Matrices of Order 3 over F 2

Congruence Classes of Invertible Matrices of Order 3 over F 2 International Journal of Algebra, Vol. 8, 24, no. 5, 239-246 HIKARI Ltd, www.m-hikari.com http://dx.doi.org/.2988/ija.24.422 Congruence Classes of Invertible Matrices of Order 3 over F 2 Ligong An and

Διαβάστε περισσότερα

INDIRECT ADAPTIVE CONTROL

INDIRECT ADAPTIVE CONTROL INDIREC ADAPIVE CONROL OULINE. Inroducion a. Main properies b. Running example. Adapive parameer esimaion a. Parameerized sysem model b. Linear parameric model c. Normalized gradien algorihm d. Normalized

Διαβάστε περισσότερα

Approximation of distance between locations on earth given by latitude and longitude

Approximation of distance between locations on earth given by latitude and longitude Approximation of distance between locations on earth given by latitude and longitude Jan Behrens 2012-12-31 In this paper we shall provide a method to approximate distances between two points on earth

Διαβάστε περισσότερα

The Simply Typed Lambda Calculus

The Simply Typed Lambda Calculus Type Inference Instead of writing type annotations, can we use an algorithm to infer what the type annotations should be? That depends on the type system. For simple type systems the answer is yes, and

Διαβάστε περισσότερα

ST5224: Advanced Statistical Theory II

ST5224: Advanced Statistical Theory II ST5224: Advanced Statistical Theory II 2014/2015: Semester II Tutorial 7 1. Let X be a sample from a population P and consider testing hypotheses H 0 : P = P 0 versus H 1 : P = P 1, where P j is a known

Διαβάστε περισσότερα

ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ. Ψηφιακή Οικονομία. Διάλεξη 7η: Consumer Behavior Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών

ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ. Ψηφιακή Οικονομία. Διάλεξη 7η: Consumer Behavior Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών ΕΛΛΗΝΙΚΗ ΔΗΜΟΚΡΑΤΙΑ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΡΗΤΗΣ Ψηφιακή Οικονομία Διάλεξη 7η: Consumer Behavior Mαρίνα Μπιτσάκη Τμήμα Επιστήμης Υπολογιστών Τέλος Ενότητας Χρηματοδότηση Το παρόν εκπαιδευτικό υλικό έχει αναπτυχθεί

Διαβάστε περισσότερα

Durbin-Levinson recursive method

Durbin-Levinson recursive method Durbin-Levinson recursive method A recursive method for computing ϕ n is useful because it avoids inverting large matrices; when new data are acquired, one can update predictions, instead of starting again

Διαβάστε περισσότερα

is the home less foreign interest rate differential (expressed as it

is the home less foreign interest rate differential (expressed as it The model is solved algebraically, excep for a cubic roo which is solved numerically The mehod of soluion is undeermined coefficiens The noaion in his noe corresponds o he noaion in he program The model

Διαβάστε περισσότερα

The choice of an optimal LCSCR contract involves the choice of an x L. such that the supplier chooses the LCS option when x xl

The choice of an optimal LCSCR contract involves the choice of an x L. such that the supplier chooses the LCS option when x xl EHNIA APPENDIX AMPANY SIMPE S SHARIN NRAS Proof of emma. he choice of an opimal SR conrac involves he choice of an such ha he supplier chooses he S opion hen and he R opion hen >. When he selecs he S opion

Διαβάστε περισσότερα

ΕΚΠΑΙΔΕΥΣΗ, ΑΜΥΝΤΙΚΕΣ ΔΑΠΑΝΕΣ ΚΑΙ ΟΙΚΟΝΟΜΙΚΗ ΑΝΑΠΤΥΞΗ: ΜΙΑ ΕΜΠΕΙΡΙΚΗ ΈΡΕΥΝΑ ΓΙΑ ΤΗΝ ΚΥΠΡΟ

ΕΚΠΑΙΔΕΥΣΗ, ΑΜΥΝΤΙΚΕΣ ΔΑΠΑΝΕΣ ΚΑΙ ΟΙΚΟΝΟΜΙΚΗ ΑΝΑΠΤΥΞΗ: ΜΙΑ ΕΜΠΕΙΡΙΚΗ ΈΡΕΥΝΑ ΓΙΑ ΤΗΝ ΚΥΠΡΟ ΕΚΠΑΙΔΕΥΣΗ, ΑΜΥΝΤΙΚΕΣ ΔΑΠΑΝΕΣ ΚΑΙ ΟΙΚΟΝΟΜΙΚΗ ΑΝΑΠΤΥΞΗ: ΜΙΑ ΕΜΠΕΙΡΙΚΗ ΈΡΕΥΝΑ ΓΙΑ ΤΗΝ ΚΥΠΡΟ ΜΕ ΤΗΝ ΑΝΑΛΥΣΗ ΤΗΣ ΑΙΤΙΟΤΗΤΑΣ Νίκος Δριτσάκης - Τάσος Στυλιανού Τμήμα Εφαρμοσμένης Πληροφορικής, Πανεπιστήμιο Μακεδονίας

Διαβάστε περισσότερα

Notes on the Open Economy

Notes on the Open Economy Notes on the Open Econom Ben J. Heijdra Universit of Groningen April 24 Introduction In this note we stud the two-countr model of Table.4 in more detail. restated here for convenience. The model is Table.4.

Διαβάστε περισσότερα

A Simple Version of the Lucas Model

A Simple Version of the Lucas Model Aricle non publié May 11, 2007 A Simple Version of he Lucas Model Mazamba Tédie Absrac This discree-ime version of he Lucas model do no include he physical capial. We inregrae in he uiliy funcion he leisure

Διαβάστε περισσότερα

Levin Lin(1992) Oh(1996),Wu(1996) Papell(1997) Im, Pesaran Shin(1996) Canzoneri, Cumby Diba(1999) Lee, Pesaran Smith(1997) FGLS SUR

Levin Lin(1992) Oh(1996),Wu(1996) Papell(1997) Im, Pesaran Shin(1996) Canzoneri, Cumby Diba(1999) Lee, Pesaran Smith(1997) FGLS SUR EVA M, SWEEEY R 3,. ;. McDonough ; 3., 3006 ; ; F4.0 A Levin Lin(99) Im, Pesaran Shin(996) Levin Lin(99) Oh(996),Wu(996) Paell(997) Im, Pesaran Shin(996) Canzoner Cumby Diba(999) Levin Lin(99) Coe Helman(995)

Διαβάστε περισσότερα

Practice Exam 2. Conceptual Questions. 1. State a Basic identity and then verify it. (a) Identity: Solution: One identity is csc(θ) = 1

Practice Exam 2. Conceptual Questions. 1. State a Basic identity and then verify it. (a) Identity: Solution: One identity is csc(θ) = 1 Conceptual Questions. State a Basic identity and then verify it. a) Identity: Solution: One identity is cscθ) = sinθ) Practice Exam b) Verification: Solution: Given the point of intersection x, y) of the

Διαβάστε περισσότερα

The canonical 2nd order transfer function is expressed as. (ω n

The canonical 2nd order transfer function is expressed as. (ω n Second order ransfer funcions nd Order ransfer funcion - Summary of resuls The canonical nd order ransfer funcion is expressed as H(s) s + ζ s + is he naural frequency; ζ is he damping coefficien. The

Διαβάστε περισσότερα

Lecture 2: Dirac notation and a review of linear algebra Read Sakurai chapter 1, Baym chatper 3

Lecture 2: Dirac notation and a review of linear algebra Read Sakurai chapter 1, Baym chatper 3 Lecture 2: Dirac notation and a review of linear algebra Read Sakurai chapter 1, Baym chatper 3 1 State vector space and the dual space Space of wavefunctions The space of wavefunctions is the set of all

Διαβάστε περισσότερα

THEORETICAL PROPERTIES OF THE WEIGHTED FELLER-PARETO AND RELATED DISTRIBUTIONS

THEORETICAL PROPERTIES OF THE WEIGHTED FELLER-PARETO AND RELATED DISTRIBUTIONS ASIAN JOURNAL OF MATHEMATICS AND APPLICATIONS Volume 204 Aricle ID ama073 2 pages ISSN 2307-7743 hp://scienceasia.asia THEORETICAL PROPERTIES OF THE WEIGHTED FELLER-PARETO AND RELATED DISTRIBUTIONS OLUSEYI

Διαβάστε περισσότερα

The Probabilistic Method - Probabilistic Techniques. Lecture 7: The Janson Inequality

The Probabilistic Method - Probabilistic Techniques. Lecture 7: The Janson Inequality The Probabilistic Method - Probabilistic Techniques Lecture 7: The Janson Inequality Sotiris Nikoletseas Associate Professor Computer Engineering and Informatics Department 2014-2015 Sotiris Nikoletseas,

Διαβάστε περισσότερα

Generating Set of the Complete Semigroups of Binary Relations

Generating Set of the Complete Semigroups of Binary Relations Applied Mathematics 06 7 98-07 Published Online January 06 in SciRes http://wwwscirporg/journal/am http://dxdoiorg/036/am067009 Generating Set of the Complete Semigroups of Binary Relations Yasha iasamidze

Διαβάστε περισσότερα

ΣΧΕΣΕΙΣ ΑΛΛΗΛΕΞΑΡΤΗΣΗΣ ΚΑΙ ΑΠΟΤΕΛΕΣΜΑΤΙΚΟΤΗΤΑ ΣΤΟ ΧΡΗΜΑΤΙΣΤΗΡΙΟ ΑΞΙΩΝ ΑΘΗΝΩΝ

ΣΧΕΣΕΙΣ ΑΛΛΗΛΕΞΑΡΤΗΣΗΣ ΚΑΙ ΑΠΟΤΕΛΕΣΜΑΤΙΚΟΤΗΤΑ ΣΤΟ ΧΡΗΜΑΤΙΣΤΗΡΙΟ ΑΞΙΩΝ ΑΘΗΝΩΝ Ελληνικό Στατιστικό Ινστιτούτο Πρακτικά 20 ου Πανελληνίου Συνεδρίου Στατιστικής (2007), σελ 373-382 ΣΧΕΣΕΙΣ ΑΛΛΗΛΕΞΑΡΤΗΣΗΣ ΚΑΙ ΑΠΟΤΕΛΕΣΜΑΤΙΚΟΤΗΤΑ ΣΤΟ ΧΡΗΜΑΤΙΣΤΗΡΙΟ ΑΞΙΩΝ ΑΘΗΝΩΝ Μαριέττα Σιταρά Τμήμα Επιστήμης

Διαβάστε περισσότερα

ANSWERSHEET (TOPIC = DIFFERENTIAL CALCULUS) COLLECTION #2. h 0 h h 0 h h 0 ( ) g k = g 0 + g 1 + g g 2009 =?

ANSWERSHEET (TOPIC = DIFFERENTIAL CALCULUS) COLLECTION #2. h 0 h h 0 h h 0 ( ) g k = g 0 + g 1 + g g 2009 =? Teko Classes IITJEE/AIEEE Maths by SUHAAG SIR, Bhopal, Ph (0755) 3 00 000 www.tekoclasses.com ANSWERSHEET (TOPIC DIFFERENTIAL CALCULUS) COLLECTION # Question Type A.Single Correct Type Q. (A) Sol least

Διαβάστε περισσότερα

Strain gauge and rosettes

Strain gauge and rosettes Strain gauge and rosettes Introduction A strain gauge is a device which is used to measure strain (deformation) on an object subjected to forces. Strain can be measured using various types of devices classified

Διαβάστε περισσότερα

SCHOOL OF MATHEMATICAL SCIENCES G11LMA Linear Mathematics Examination Solutions

SCHOOL OF MATHEMATICAL SCIENCES G11LMA Linear Mathematics Examination Solutions SCHOOL OF MATHEMATICAL SCIENCES GLMA Linear Mathematics 00- Examination Solutions. (a) i. ( + 5i)( i) = (6 + 5) + (5 )i = + i. Real part is, imaginary part is. (b) ii. + 5i i ( + 5i)( + i) = ( i)( + i)

Διαβάστε περισσότερα

CRASH COURSE IN PRECALCULUS

CRASH COURSE IN PRECALCULUS CRASH COURSE IN PRECALCULUS Shiah-Sen Wang The graphs are prepared by Chien-Lun Lai Based on : Precalculus: Mathematics for Calculus by J. Stuwart, L. Redin & S. Watson, 6th edition, 01, Brooks/Cole Chapter

Διαβάστε περισσότερα

Necessary and sufficient conditions for oscillation of first order nonlinear neutral differential equations

Necessary and sufficient conditions for oscillation of first order nonlinear neutral differential equations J. Mah. Anal. Appl. 321 (2006) 553 568 www.elsevier.com/locae/jmaa Necessary sufficien condiions for oscillaion of firs order nonlinear neural differenial equaions X.H. ang a,, Xiaoyan Lin b a School of

Διαβάστε περισσότερα

HW 3 Solutions 1. a) I use the auto.arima R function to search over models using AIC and decide on an ARMA(3,1)

HW 3 Solutions 1. a) I use the auto.arima R function to search over models using AIC and decide on an ARMA(3,1) HW 3 Solutions a) I use the autoarima R function to search over models using AIC and decide on an ARMA3,) b) I compare the ARMA3,) to ARMA,0) ARMA3,) does better in all three criteria c) The plot of the

Διαβάστε περισσότερα

Estimation for ARMA Processes with Stable Noise. Matt Calder & Richard A. Davis Colorado State University

Estimation for ARMA Processes with Stable Noise. Matt Calder & Richard A. Davis Colorado State University Estimation for ARMA Processes with Stable Noise Matt Calder & Richard A. Davis Colorado State University rdavis@stat.colostate.edu 1 ARMA processes with stable noise Review of M-estimation Examples of

Διαβάστε περισσότερα

APPENDIX A DERIVATION OF JOINT FAILURE DENSITIES

APPENDIX A DERIVATION OF JOINT FAILURE DENSITIES APPENDIX A DERIVAION OF JOIN FAILRE DENSIIES I his Appedi we prese he derivaio o he eample ailre models as show i Chaper 3. Assme ha he ime ad se o ailre are relaed by he cio g ad he sochasic are o his

Διαβάστε περισσότερα

6. MAXIMUM LIKELIHOOD ESTIMATION

6. MAXIMUM LIKELIHOOD ESTIMATION 6 MAXIMUM LIKELIHOOD ESIMAION [1] Maximum Likelihood Estimator (1) Cases in which θ (unknown parameter) is scalar Notational Clarification: From now on, we denote the true value of θ as θ o hen, view θ

Διαβάστε περισσότερα

k A = [k, k]( )[a 1, a 2 ] = [ka 1,ka 2 ] 4For the division of two intervals of confidence in R +

k A = [k, k]( )[a 1, a 2 ] = [ka 1,ka 2 ] 4For the division of two intervals of confidence in R + Chapter 3. Fuzzy Arithmetic 3- Fuzzy arithmetic: ~Addition(+) and subtraction (-): Let A = [a and B = [b, b in R If x [a and y [b, b than x+y [a +b +b Symbolically,we write A(+)B = [a (+)[b, b = [a +b

Διαβάστε περισσότερα

Approximation of the Lerch zeta-function

Approximation of the Lerch zeta-function Approximaion of he Lerch zea-funcion Ramūna Garunkši Deparmen of Mahemaic and Informaic Vilniu Univeriy Naugarduko 4 035 Vilniu Lihuania ramunagarunki@mafvul Abrac We conider uniform in parameer approximaion

Διαβάστε περισσότερα

the total number of electrons passing through the lamp.

the total number of electrons passing through the lamp. 1. A 12 V 36 W lamp is lit to normal brightness using a 12 V car battery of negligible internal resistance. The lamp is switched on for one hour (3600 s). For the time of 1 hour, calculate (i) the energy

Διαβάστε περισσότερα

& Risk Management , A.T.E.I.

& Risk Management , A.T.E.I. Μεταβλητότητα & Risk Managemen Οικονοµικό Επιµελητήριο της Ελλάδας Επιµορφωτικά Σεµινάρια Σταύρος. Ντεγιαννάκης, Οικονοµικό Πανεπιστήµιο Αθηνών Χρήστος Φλώρος, A.T.E.I. Κρήτης Volailiy - Μεταβλητότητα

Διαβάστε περισσότερα

Probability and Random Processes (Part II)

Probability and Random Processes (Part II) Probability and Random Processes (Part II) 1. If the variance σ x of d(n) = x(n) x(n 1) is one-tenth the variance σ x of a stationary zero-mean discrete-time signal x(n), then the normalized autocorrelation

Διαβάστε περισσότερα

Modbus basic setup notes for IO-Link AL1xxx Master Block

Modbus basic setup notes for IO-Link AL1xxx Master Block n Modbus has four tables/registers where data is stored along with their associated addresses. We will be using the holding registers from address 40001 to 49999 that are R/W 16 bit/word. Two tables that

Διαβάστε περισσότερα

Homework for 1/27 Due 2/5

Homework for 1/27 Due 2/5 Name: ID: Homework for /7 Due /5. [ 8-3] I Example D of Sectio 8.4, the pdf of the populatio distributio is + αx x f(x α) =, α, otherwise ad the method of momets estimate was foud to be ˆα = 3X (where

Διαβάστε περισσότερα

Second Order Partial Differential Equations

Second Order Partial Differential Equations Chapter 7 Second Order Partial Differential Equations 7.1 Introduction A second order linear PDE in two independent variables (x, y Ω can be written as A(x, y u x + B(x, y u xy + C(x, y u u u + D(x, y

Διαβάστε περισσότερα

Quadratic Expressions

Quadratic Expressions Quadratic Expressions. The standard form of a quadratic equation is ax + bx + c = 0 where a, b, c R and a 0. The roots of ax + bx + c = 0 are b ± b a 4ac. 3. For the equation ax +bx+c = 0, sum of the roots

Διαβάστε περισσότερα

HISTOGRAMS AND PERCENTILES What is the 25 th percentile of a histogram? What is the 50 th percentile for the cigarette histogram?

HISTOGRAMS AND PERCENTILES What is the 25 th percentile of a histogram? What is the 50 th percentile for the cigarette histogram? HISTOGRAMS AND PERCENTILES What is the 25 th percentile of a histogram? The point on the horizontal axis such that of the area under the histogram lies to the left of that point (and to the right) What

Διαβάστε περισσότερα