Bounds for the asymptotic normality of the maximum likelihood estimator using the Delta method

Size: px
Start display at page:

Download "Bounds for the asymptotic normality of the maximum likelihood estimator using the Delta method"

Transcription

1 ALEA, Lat. Am. J. Probab. Math. Stat. 14, Bous for the asymtotic normality of the maximum likelihoo estimator using the Delta metho Areas Anastasiou a Christohe Ley Loon School of Economics, Deartment of Statistics Columbia House, Houghton Street, WC2A 2AE Loon, Unite Kingom. aress: a.anastasiou@lse.ac.uk Ghent University, Deartment of Alie Mathematics, Comuter Science a Statistics Krijgslaan 281, S9, 9000 Gent, Belgium. aress: christohe.ley@ugent.be Abstract. The asymtotic normality of the Maximum Likelihoo Estimator MLE is a cornerstone of statistical theory. In the resent aer, we rovie shar exlicit uer bous on Zolotarev-tye istances between the exact, unknown istribution of the MLE a its limiting normal istribution. Our aroach to this fuamental issue is base on a sou combination of the Delta metho, Stein s metho, Taylor exansions a coitional exectations, for the classical situations where the MLE can be exresse as a function of a sum of ieeent a ientically istribute terms. This result is tailore for the broa class of one-arameter exonential family istributions. 1. Introuction The asymtotic normality of maximum likelihoo estimators MLEs is one of the best-known a most fuamental results in mathematical statistics. Uer certain regularity coitions given later in this section, we have the following classical theorem, first iscusse in Fisher Receive by the eitors Aril 4th, 2016; accete February 2, Mathematics Subject Classification. 62F12, 62E17. Key wors a hrases. Delta metho, Maximum likelihoo estimator, Normal aroximation, Stein s metho. A greatartof thiswork has been comlete whileareas Anastasiou wasworking at the University of Oxfor. Areas Anastasiou was suorte by a Teaching Assistantshi Bursary from the Deartment of Statistics, University of Oxfor, a EPSRC grant EP/K503113/1. Christohe Ley thanks the Fos National e la Recherche Scientifique, Communauté Française e Belgique, for financial suort via a Maat e Chargé e Recherche FNRS.. 153

2 154 Areas Anastasiou a Christohe Ley Theorem 1.1 Asymtotic Normality of the MLE. Let X 1,...,X n be i.i.. raom variables with robability ensity or mass function fx i θ, where θ is a scalar arameter. Its true value is enote by θ 0. Assume that the MLE exists a it is unique a coitions R1-R4, see below, are satisfie. Then niθ0 ˆθn X θ 0 N0,1, n where iθ 0 is the execte Fisher information quantity a means convergence in istribution. Theaimoftheresentaeristocomlementthisqualitativeresultwithaquantitative statement, in other wors, to fi the best ossible aroximation for the istance, at finite samle size n, between the istribution of niθ 0 ˆθn X θ 0 on the one ha a N0,1 on the other ha. In mathematical terms, for Z N0,1, we are intereste in the quantity H niθ0 ˆθn X θ 0,Z ] = su E h niθ0 ˆθn X θ 0 EhZ] 1.1 with h H H = {h : R R, absolutely continuous a boue}. 1.2 Distances of this tye are calle Zolotarev-tye istances. Our focus will lie on the classical situations where the MLE can be exresse as a function of a sum of ieeent a ientically istribute terms. Consier an i.i.. samle of observations X = X 1,...,X n. Writing ˆθ n X the MLE of the scalar arameter of interest θ Θ R, we are intereste in settings where there exists a one-to-one twice ifferentiable maing q : Θ R with q θ 0 θ Θ such that qˆθn X = 1 n n gx i 1.3 for some g : R R. Situations of this ki are all but rare; with fx θ the robability ensity or mass function, classical examles inclue thenormalistributionwithensityfx µ,σ 2 = 1 σ ex 1 2π 2σ x µ 2, 2 x R, for which µ R is our unknown arameter, whereas σ > 0 is consiere to be known. The MLE for θ = µ is ˆθ n X = 1 n X i ; n the normal istribution, where now the mean µ is known a θ = σ 2 reresents the unknown arameter, with ˆθ n X = 1 n X i µ 2 ; n the Weibull istribution with ensity fx α,σ = α x α 1ex σ σ x α σ, x 0, where σ is the unknown scale arameter a α > 0 is fixe. The

3 Delta metho for bous on MLE 155 MLE for θ = σ is efine through α 1 n ˆθn X = Xi α ; n the Lalace scale moel with ensity fx σ = 1 2σ ex x /σ,θ = σ > 0, over R, for which ˆθ n X = 1 n X i. n The broa one-arameter exonential families o satisfy coition 1.3; actually, it is ossible to show that MLEs of the form 1.3 are characteristic of these families, see Proosition 3.1 for etails. Hence, our results o aly to most of the wellknown istributions. We now resent in etail the notation use throughout the aer. We write E θ ] the exectation uer the secific value θ of the arameter. In line with the notation use above, the joint ensity or robability mass function of X is written fx θ. The true, unknown value of the arameter is θ 0 a Θ enotes the arameter sace. For X i = x i some observe values, the likelihoo function is enote by Lθ; x = fx θ a we enote its natural logarithm, calle the log-likelihoo function, by lθ; x. The erivatives of the log-likelihoo function with resect to θ are l θ;x,l θ;x,...,l j θ;x, for j any integer greater than 2, a iθ enotes the execte Fisher information number for one raom variable. Whenever the MLE exists a is also unique, we will write it as before uer the form ˆθ n X. For Θ being an oen interval, we use the results in Mäkeläinen et al to secure the existence a uniqueness of the MLE. Thus, it suffices to assume that: A1 The log-likelihoo function lθ;x is a twice continuously ifferentiable function with resect to θ a the arameter varies in an oen interval a,b, where a,b R {, } a a < b; A2 lim θ a,b lθ;x = ; A3 l θ;x < 0 at every oint θ a,b for which l θ;x = 0. Note that we tacitly assume those coitions in Theorem 1.1 when requiring existence a uniqueness of the MLE. Asymtotic normality further requires the following sufficient regularity coitions: R1 the arameteris ientifiable, which means that if θ θ, then x : fx θ fx θ ; R2 the ensity fx θ is three times ifferentiable with resect to θ, the thir erivative is continuous in θ a fx θx can be ifferentiate three times uer the integral sign; R3 for any θ 0 Θ a for X enoting the suort of fx θ, there exists a ositive number ǫ a a function Mx both of which may ee on θ 0 such that 3 θ 3 logfx θ Mx x X, θ 0 ǫ < θ < θ 0 +ǫ, with E θ0 MX] < ; R4 iθ > 0, θ Θ. These coitions, in articular R2, ensure that, rovie the resective exressions exist, E θ l θ;x] = 0 a Var θ l θ;x] = niθ. These coitions form the

4 156 Areas Anastasiou a Christohe Ley basis of Theorem 1.1 above; see age 472 of Casella a Berger 1990 for a basic sketch of the roof. The first aer to aress the roblem of assessing the accuracy of the normal aroximation for the MLE is Anastasiou a Reinert After eriving general bous on Zolotarev-tye istances, they use the boue Wasserstein istance bw, whichisalsoknownasfortet-mourieristancesee, e.g., NourinaPeccati, 2012 a is linke to the Kolmogorov istance H is the class of iicator functions of half-saces via K, 2 bw,. We state in Theorem 2.4 of Section 2 the bou obtaine in that aer. For the broa class of istributions satisfying 1.3, our bou is better than, or at least as goo as, the Anastasiou a Reinert 2017 bou hereafter referre to as AR-bou both in terms of sharness a simlicity. The tools we use to reach this result are the Delta metho, Stein s metho for normal aroximation, Taylor exansions a coitional exectations. The aer is organise as follows. Our new uer bou is escribe, rove a comare to the AR-bou in Section 2, a ractical statistical alications are exhibite. In Section 3 we then aly our results to the class of one-arameter exonential family istributions a treat some secific examles in etail. 2. New bous on the istance to the normal istribution for the MLE In orer to obtain bous on the aforementione istance, we artly emloy the following lemma. From now on, unless otherwise state, enotes the infinity norm. Also, for the sake of resentation we ro the subscrit θ 0 from the exectation a variance. Lemma 2.1 Reinert, Let Y 1,...,Y n be ieeent raom variables with EY i = 0, VarY i = σ 2 > 0 a E Y i 3] <. Let W = 1 n n Y i, with EW = 0, VarW = σ 2 a let K N0,σ 2. Then for any function h H, with H given in 1.2, one has EhW] EhK] h 2+ 1 Y n σ 3E 1 3]. As we shall see below, our strategy consists in benefiting from the secial form of qˆθ n X, which is a sum of raom variables a thus allows us to use the shar bou of this lemma. It is recisely at this oint that the Delta metho comes into lay: abusing notations a language, instea of comaring ˆθ n X to Z N0,1 we rather comare qˆθ n X to Z, a then bou the istance between ˆθ n X a qˆθ n X. The outcome of this aroach is the next theorem, the main result of the resent aer. Theorem 2.2. Let X 1,...,X n be i.i.. raom variables with robability ensity or mass function fx i θ a let Z N0,1. Assume that R1-R4 are satisfie. Furthermore, assume that the MLE exists a 1.3 is satisfie. In aition, the maing g : R R is such that E gx 1 qθ 0 3] < for θ 0 Θ the true value of the arameter. Then, for any h H as in 1.2,

5 Delta metho for bous on MLE if q is not uniformly boue but for any θ 0 Θ there exists 0 < ǫ = ǫθ 0 such that q θ <, we have E h niθ0 h 2+ iθ 0] 3 2 n su θ: θ θ 0 <ǫ ] ˆθn X θ 0 EhZ] q θ 0 3E gx 1 qθ 0 3] niθ0 ] 2 +E ˆθn X θ 0 2 h ǫ 2 ½{ θ Θ : qθ θ}+ h 2 q θ 0 su q θ ; θ: θ θ 0 ǫ if q is uniformly boue, with q θ B θ Θ for some B > 0, we have ] E h niθ0 ˆθn X θ 0 EhZ] h 2+ iθ 0] 3 2 n q θ 0 3E gx 1 qθ 0 3] + h B niθ0 2 q θ 0 E ˆθn X θ 0 2 ] ½{ θ Θ : qθ θ}. 2.2 Proof: The asymtotic normality of the MLE is exlicitly state in Theorem 1.1. Alying the wiely known Delta metho to this result in combination with the requirement q θ 0 0 yiels niθ0 qˆθn X qθ 0 q θ 0 with qˆθn X = 1 n n gx i. Using the triangle inequality, N0,1 2.3 n ] E h niθ0 ˆθn X θ 0 EhZ] E niθ0 h q qˆθn X qθ 0 ] EhZ] 2.4 θ 0 + E niθ0 h niθ0 ˆθn X θ 0 h q qˆθn X qθ 0 ]. θ We first obtain an uer bou for 2.4 using iirectly Stein s metho via Lemma 2.1. Some simle rewriting yiels niθ0 niθ0 1 n q qˆθn X qθ 0 = θ 0 q gx i qθ 0 θ 0 n { } = 1 n iθ0 n q θ 0 gx i qθ 0 = 1 n Y i, n

6 158 Areas Anastasiou a Christohe Ley iθ0 where Y i = q θ gx 0 i qθ 0,i = 1,2,...,n a, obviously, the Y i s are ieeent a ientically istribute raom variables. The Central Limit Theorem alie to 1 n n Y 1 i imlies that n n Y i EY 1 N0,VarY 1. n From 2.3 we know however that 1 n n Y i N0,1; comaring the two n asymtotic results reveals that, necessarilysince two normal istributions can only be equal if their exectations a variances are the same, we have EgX 1 ] = qθ 0 a VargX 1 ] = q θ 0 2, iθ 0 where coition R4 allows to ivie by iθ 0. Hence, as Lemma 2.1 requires, EY i ] = 0, VarY i ] = 1, a E Y i 3 ] < thanks to the aitional coition on g. Alying the result of the lemma we get E niθ0 h q qˆθn X qθ 0 ] EhZ] θ 0 h 2+ iθ 0] 3 2 n q θ 0 3E gx 1 qθ 0 3]. 2.6 Now we are searching for an uer bou on 2.5. Since the case qθ = θ θ Θ is obvious, we from here on assume that qθ θ. We enote A := Aq,θ 0,X niθ0 := h niθ0 ˆθn X θ 0 h q θ 0 qˆθn X qθ 0 a our scoe is to fi an uer bou for EA]. Case 1: θ Θ there exists 0 < ǫ = ǫθ 0 such that su q θ <. Then, θ: θ θ 0 <ǫ using the law of total exectation relate to coitioning on ˆθ n X θ 0 ǫ or ˆθ n X θ 0 < ǫ yiels EA] E A ] = E A ˆθ n X θ 0 ǫ ]P ˆθn X θ 0 ǫ +E A ˆθ n X θ 0 < ǫ ]P ˆθn X θ 0 < ǫ. Markov s inequality a the elementary results of P ˆθn X θ 0 < ǫ 1 a A 2 h further yiel E EA 2 h ˆθn X θ 0 2 ] ǫ 2 +E A ] ˆθ n X θ 0 < ǫ. 2.7 We now focus on the coitional exectation on the right-ha sie of 2.7. A seco-orer Taylor exansion of qˆθn x about θ 0 gives qˆθn x = qθ 0 + ˆθn x θ 0 q θ q ˆθn x θ 0 θ, 2.8 2

7 Delta metho for bous on MLE 159 for θ between ˆθ n x a θ 0. Since q θ 0 θ Θ is assume, we can multily niθ0 both sies in 2.8 with q θ 0. Rearranging the terms, niθ0 qˆθn x qθ 0 q θ 0 = niθ0 2. niθ 0 ˆθn x θ 0 + 2q θ 0 q θ ˆθn x θ 0 The above result along with another first-orer Taylor exansion recall that n ˆθn X θ 0 2 = op 1 as n gives niθ0 h niθ0 ˆθn x θ 0 h q θ 0 = qˆθn x qθ 0 niθ0 2h 2q θ 0 q θ ˆθn x θ 0 tx, 2.9 ˆθn x θ 0 where tx is between niθ 0 a niθ0 q θ 0 qˆθn x qθ 0. Equality 2.9 combine with Lemma 2.1 in Anastasiou a Reinert 2017 relate to coitional exectations yiel ] E A ˆθ n X θ 0 < ǫ ] = E niθ0 2h 2q θ 0 q θ ˆθn X θ 0 tx ˆθ n X θ 0 < ǫ h niθ 0 2 q E θ 0 h niθ 0 2 q θ 0 h niθ 0 2 q θ 0 2 ] q θ ˆθn X θ 0 ˆθn X θ 0 < ǫ 2 ] su q θ E ˆθn X θ 0 ˆθn X θ 0 < ǫ θ: θ θ 0 <ǫ su q θ E θ: θ θ 0 <ǫ ˆθn X θ 0 2 ] Combining the bous in 2.6, 2.7 a 2.10 we get the result in 2.1. Case 2: There exists B > 0 such that q is uniformly boue with q θ B θ Θ. In this case, we o not nee to take coitional exectations. The result in 2.9 gives E A ] h niθ 0 ] 2 E q θ ˆθn X θ 0 2 q θ 0 B h niθ 0 ] 2 2 q E ˆθn X θ θ 0 The above result a the bou in 2.6 give the bou in 2.2 for the case where q is uniformly boue in θ. Remark Theconvergenceofthe termsaartfromthe firstone isgoverne ] 2 by the asymtotic behaviour of the mean square error E ˆθn X θ 0, whose

8 160 Areas Anastasiou a Christohe Ley rate of convergence is O 1 n. This result is obtaine using the ecomosition E ˆθ n X θ 0 2] ] ] = Varˆθn X +bias 2ˆθn X Uer the staar asymtoticsfrom the regularity coitions the MLE is asymtotically efficient, meaning that ] nvarˆθn X iθ 0] 1, n a hence the variance of the MLE is of orer 1 n. In aition, from Theorem 1.1 the bias of the MLE is of orer 1 n ; see also Cox a Snell 1968, where no exlicit coitions are given. Combining these two results a using 2.12 shows that the mean square error of the MLE is of orer 1 n. 2 In the simlest ossible situation where ˆθ n X is alreay a sum of i.i.. terms, qx = x a hence our uer bou simlifies to ] E h niθ0 ˆθn X θ 0 EhZ] h 2+iθ 0 ] 3 2 E gx 1 θ 0 3], n which is equivalent to Lemma One may woer whether asking R1-R4 to hol is not a too strong assumtion given that we assume the MLE to be of the secial form 1.3. Iee, one can easily get from the Central Limit Theorem that n qˆθx EgX 1 ] n N0,VargX 1] without neeing the asymtotic normalityof ˆθX through R1-R4. However, without this asymtotic normality of ˆθX wecannot obtain 2.3, a without the latterexlicit asymtoticresult we cannotlink EgX 1 ] a VargX 1 ] to their corresoingq-basequantities. The sequel of the roof woul boil own. This is the erhas not so obvious reason why from the beginning of the roof we nee the regularity coitions R1-R4. In orer to areciate the sharness a simlicity of our new bous, we comare our result to the AR-bou. For ease of resentation the comarison is concerne with the case where q is not uniformly boue, but only boue in a neightbourhoo of θ 0. Therefore, we emloy the result in 2.1. Similar results hol when q is uniformly boue. We now state the main result of Anastasiou a Reinert Theorem 2.4 AnastasiouaReinert,2017. Let X 1,X 2,...,X n be i.i.. raom variables with ensity or frequency function fx i θ such that the regularity coitions R1-R4 are satisfie a that the MLE, ˆθ n X, exists a it is unique. Assume that E θ logfx 1 θ ] 3 ] 4 θ=θ 0 < a that E ˆθn X θ 0 <. Let 0 < ǫ = ǫθ 0 be such that θ 0 ǫ,θ 0 +ǫ Θ as in R3 a let Z N0,1.

9 Delta metho for bous on MLE 161 Then for any function h that is absolutely continuous a boue, ] E h niθ0 ˆθn X θ 0 EhZ] h 1 2+ E 3 n iθ 0 ] 3 2 θ logfx 1 θ θ=θ 0 ] 2 E ˆθn X θ 0 +2 h ǫ 2 + h ] niθ0 E R 2 θ 0 ;X ˆθ n X θ 0 ǫ ]]1 E su l 3 4 θ;x 2 θ: θ θ 0 ǫ ˆθ 2 n X θ 0 ǫ E ˆθn X θ 0, where R 2 θ 0 ;x = ˆθ n x θ 0 l θ 0 ;x+niθ Obvious observations are that the AR-bou requires finiteness of the fourth moment of ˆθ n X θ 0 a that this bou is more comlicate than ours. Let us now comment on the bous term by term. In the first term of the bous, the ifferent ositioning of the execte Fisher information number is exlaine by the fact that we aly Lemma 2.1 to the staariseversionof gx 1,gX 2,...,gX n, which havevariance q θ 0] 2 iθ 0, while Anastasiou a Reinert 2017 obtain the result by alying the lemma after staarising θ logfx 1 θ θ=θ0, θ logfx 2 θ θ=θ0,..., θ logfx n θ θ=θ0, which have variance equal to iθ 0. The seco a thir terms vanish in our bou when qθ = θ θ Θ, while the AR-bou oes not take this simlification into account. In aition, when qθ θ the seco term is the same in both bous, whereas the thir term in our bou reas ] 2 h E ˆθn niθ 0 X θ 0 2 q su q θ 2.14 θ 0 θ: θ θ 0 ǫ a is to be comare to E h 2 niθ 0 + h su θ: θ θ 0 ǫ l 3 θ;x 2 ˆθn X θ 0 ǫ] 1 2 ]]1 4 2 E ˆθn X θ 0 ] R E 2 θ 0 ;X ˆθ n X θ 0 ǫ n iθ where R 2 θ 0 ;x = ˆθ n x θ 0 l θ 0 ;x+niθ 0. The seco erivative, q θ, lays in our bou the role of l 3 θ;x, u to an imortant ifference: l 3 θ;x is a sum. Consequently, the first term in 2.15 has n in its numerator, exactly as in The istinct ositioning of the information quantity iθ 0 has the same reason as exlaine above. Besies the

10 162 Areas Anastasiou a Christohe Ley obvious aitional term in the AR bou the seco term in 2.15, our bou is also clearly sharer at the level of moments of ˆθ n X θ 0 since ] ]] E ˆθn X θ 0 E ˆθn X θ 0 by the Cauchy-Schwarz inequality. From this comarison one sees that our new bou is simler a, moreover, has one term less. This is articularly striking in the simlest ossible setting where ˆθ n X is a sum of i.i.. terms, where our bou clearly imroves on the AR-bou. An avantage of the AR-bou is its wier alicability as it works for all MLE settings, even when an analytic exression of the MLE is not known. We conclue this section by iicating ractical statistical alications of our bous in 2.1 a 2.2. This will be shown through their usefulness in the constructionofconfienceintervalsforθ 0. TheKolmogorovistancerelatesirectly to exact conservative confience intervals a the next roosition links our results to the Kolmogorov istance. Slightly ifferent results for conservative confience intervals were also given in Anastasiou a Reinert Proosition 2.5. Let W be any real-value raom variable a Z N0, 1. Assume that there exist δ 1 > 0 a δ 2 0 such that, for any function h with h a h boue, Then K W,Z 2 δ 1 +δ 2. EhW] EhZ] δ 1 h +δ 2 h Proof: The roof of this roosition follows the aroach exlaine in the roof of Theorem 3.3 on age 48 of Chen et al. 2011, where a similar result is shown that links the Wasserstein a the Kolmogorov istance. Let z R a for α = δ1 2π 1 4 let 1, if w z, h α w = 1+ z w α, if z < w z +α, 0, if w > z +α, so that h α is boue Lischitz with h α 1 a h α 1 α. The efinition of h α combine with 2.16 leas to P W z P Z z Eh α W] Eh α Z]+Eh α Z] P Z z Eh α W] Eh α Z] +Eh α Z] P Z z δ 1 h α +δ 2 h α +P z < Z z +α δ 1 α +δ 2 + α 2π Similarly, using now δ1 = 2 +δ 2π δ 1 +δ , if w z α, h α w = w z α α, if z α < w z, 1, if w > z,

11 Delta metho for bous on MLE 163 we can show that P W z P Z z 2 δ 1 +δ 2, which comletes the roof. Our bous are of the form 2.16, where in our case W = niθ 0 ˆθn X θ 0 1 a δ 1 a δ 2 are exlicitly given with δ 1 = O n a δ 2 = O 1 n in 2.1, while for 2.2 take δ 2 = 0. Using Proosition 2.5 we get K niθ0 ˆθ n X θ 0,Z 2 δ 1 +δ 2 B K, whereb K isasharbounoteeingonθ 0 ortheata. InviewofTheorem2.2, this bou can be obtaine by uer bouing 1/q θ 0 a su θ: θ θ0 ǫ q θ which is one on a case-by-case basis as well as the mean square error ] 2 E ˆθn X θ 0 e.g., through Theorem 5.1 in Anastasiou a Reinert, The resence of the Fisher information quantity is in many settings not a roblem, as it often oes not ee on θ 0 ; this is tyically the case for location moels. Therefore, for y R we have P niθ0 ˆθn X θ 0 y P Z y B K B K P niθ0 ˆθn X θ 0 y P Z y B K With Φ 1 the quantile function of the staar normal istribution, α 0,1] a α 2 B K > 0, alying 2.17 to y = Φ 1 α 2 B K a to y = Φ 1 1 α 2 +B K yiels P Φ 1 α 2 B K niθ 0 ˆθn X θ 0 Φ 1 1 α 2 +B K 1 α. Hence, if the execte Fisher information number for one raom variable, iθ 0, is known a oes not ee on θ 0, then ˆθ n X Φ 1 1 α 2 +B K, ˆθ n X Φ 1 α 2 B K niθ0 niθ0 is a conservative 1001 α% confience interval for θ Calculation of the bou in ifferent scenarios In this section we shall consier examles for which we exlicitly calculate our uer bous from Theorem 2.2 a comare them to the AR-bou. In articular, we shall show that our results are tailor-mae for one-arameter exonential families. To further assess their accuracy, we simulate ata from exonential istributions a comare our corresoing bou to the actual istance between the unknown exact law of the MLE a its asymtotic normal law, for istinct values of the samle size n Bous for one-arameter exonential families. The robability ensity or mass function for one-arameter exonential families is given by fx θ = ex{kθtx Aθ+Sx} ½ {x B}, 3.1 where the set B = {x : fx θ > 0} is the suort of the ensity a oes not ee on θ; kθ a Aθ are functions of the arameter; Tx a Sx are

12 164 Areas Anastasiou a Christohe Ley functions only of the ata. Whenever kθ = θ we have the so-calle canonical case, where θ a TX are calle the natural arameter a natural observation Casella a Berger, The ientifiability constraint in R1 entails that k θ 0 Geyer, 2013, an imortant etail for the following investigation. Proosition 3.1. Suose X 1,...,X n are i.i.. with robability ensity or mass function that can be exresse in the form of 3.1. Assume that the regularity coitions R1-R4 are satisfie a the MLE ˆθ n X exists a is unique. Let θ Dθ = A θ k θ be invertible. Then q = D, with q : Θ R as in Theorem 2.2. Moreover, uer the same regularity coitions, no istributions other than 3.1 amit an MLE of the form 1.3. Proof: Using 3.1, we have that a hence Lθ;x = { n n fx i θ = ex kθ Tx i naθ+ lθ;x = kθ n Tx i naθ+ l θ;x = k θ n Sx i, n Tx i na θ = 0 Dθ = 1 n } n Sx i, n Tx i, whichmeansthat ˆθ n X = D 1 1 n n TX i uerthe invertibilityassumtion for Dθ. It follows that q = D. For the seco statement, let us start from qˆθn X = 1 n n gx i. Such an MLE form necessarily comes from a log-likelihoo equation of the form qθ = 1 n n gx i n gx i nqθ = 0. Let us write the invertible function qθ as some ratio α θ/β θ for two ifferentiable functions α a β with β θ 0 θ Θ. Fromthisweeucethatlθ;x = βθ n gx i nαθ+fx for some function f of the observations. Since lθ;x is a log-likelihoo built from ieeent observations, the function f necessarily is a sum which we write n Sx iwhichcoulalsocontainconstants. Theclaimnowreailyfollows. This result hence shows that, as announce in the Introuction, the broa onearameter exonential families o satisfy 1.3 a there is no istribution outsie these families that can be of that form uer the given regularity coitions. This remarkable result reresents a new characterization of the exonential families through their MLE. With this in ha, Theorem 2.2 can be alie to 3.1, resulting in Corollary 3.2. Let X 1,...,X n be i.i.. raom variables with the robability ensity or mass function of a one-arameter exonential family. Assume that R1-R4 are satisfie a the MLE ˆθ n X exists a is unique. Then, with Z N0,1 a h H as efine in 1.2,

13 Delta metho for bous on MLE if D is not uniformly boue but for any θ 0 Θ there exists 0 < ǫ = ǫθ 0 such that D θ <, we have su θ: θ θ 0 <ǫ ] E h niθ0 ˆθn X θ 0 EhZ] h k θ 0 3 E TX 1 Dθ 0 3] 2+ n A θ 0 k θ 0 Dθ ] 2 +E ˆθn X θ 0 2 h ½{ θ Θ : Dθ θ} ǫ2 h n k θ A θ 0 k θ 0 Dθ 0 su θ: θ θ 0 ǫ D θ ; if D is uniformly boue, with D θ B θ Θ for some B > 0, we have ] E h niθ0 ˆθn X θ 0 EhZ] h k θ 0 3 E TX 1 Dθ 0 3] 2+ n A θ 0 k θ 0 Dθ h B n k θ 0 2 A θ 0 k θ 0 Dθ 0 E Proof: We reaily have ˆθn X θ 0 2 ] ½{ θ Θ : Dθ θ}. 3.3 iθ 0 = E l θ 0 ;X 1 ] = A θ 0 k θ 0 ETX 1 ] = A θ 0 k θ 0 k θ 0 A θ 0 k θ 0 a q θ 0 = A θ 0k θ 0 k θ 0A θ 0 k θ 0] 2. Combining these two results yiels iθ0 q θ 0 = k θ A θ 0 k θ 0 k θ 0 A θ 0 = k θ 0 A θ 0 k θ 0 Dθ 0. This result, along with the fact that gx = Tx a qθ = Dθ by Proosition 3.1, allows to euce the announce uer bou from Theorem 2.2. Remark 3.3. It is articularly interesting to sell out this bou in the canonical case kθ = θ. Since then k θ = 1, Dθ = A θ, we fi that if A is not uniformly boue but for any θ 0 Θ there exists 0 < ǫ = ǫθ 0 such that

14 166 Areas Anastasiou a Christohe Ley su A θ <, then θ: θ θ 0 <ǫ ] E h niθ0 ˆθn X θ 0 EhZ] h E TX 1 A θ 0 3] 2+ n A θ E ˆθn X θ 0 2 ] 2 h ǫ 2 ½{ θ Θ : A θ θ} + h n 2 su A θ A θ 0 θ: θ θ 0 ǫ Asimilar resultas the onein 3.3 hols when A is uniformly boue. Now, as iθ = A θ a l θ;x = na θ, R 2 θ;x = 0 a straightforwarmaniulations show that all terms in the AR-bou coincie with those in our bou 3.2, ]]1 4 2 excet for E ˆθn X θ 0, making the AR-bou less shar than ours. However, Anastasiou a Reinert 2017 have shown that, in the canonical exonential setting, their bou can actually have an E ] 2 ˆθn X θ 0 factor, imlying that both bous are exactly the same in the canonical case. In orer to get an iea of how our bou imroves on the AR-bou in non-canonical cases, we treat the exonential istribution uer a non-canonical arametrisation in subsection Bous for the Generalise Gamma istribution. Let us consier X 1,...,X n i.i.. raom variables from the Generalize Gamma GGθ,, istribution, where the shae arameters, > 0 are consiere to be known a the scale arameter θ is the unknown arameter of interest. The Generalise Gamma istribution inclues many other known istributions as secial cases: the Weibull for =, the Gamma when = 1, a the negative exonential when = = 1. Iee, with Γ enoting the Gamma function, the robability ensity function for x > 0 is { x } θ fx θ = x 1 θ ex Γ = ex { x +log logθ + 1logx log θ where, in the terminology of one-arameter exonential families, B = 0,, Θ = Γ. } 0,, Tx = x, kθ = 1 θ, Aθ = logθ a Sx = log + 1logx log Γ. Simle stes yiel lθ 0 ;x = 1 n θ x i +nlog logθ n 0 + 1log x i nlog Γ 0 1 l θ 0 ;x = n θ +1 x i n = 0 θ ˆθ n n x = x i. 0 0

15 Delta metho for bous on MLE 167 It is easy to verify that iee l ˆθ n x;x = n < 0, which shows that the ˆθ nx] 2 MLE exists a is unique. The regularity coitions R1-R4 are also satisfie a Dθ 0 = qθ 0 = θ 0 with its seco erivative not being uniformly boue but boue in a neighbourhoo of θ 0. Using therefore the first result of Corollary 3.2 for ǫ = θ0 2 we obtain ] E h niθ0 ˆθn X θ 0 Γ Γ + 8 h + h EhZ] h n Γ +2 ½{{ 1} { 1}} Γ 1 ½{ < 2} ½{ 2}] Let us briefly show how to obtain this bou. As alreay iicate, for the Generalise Gamma istribution, Dθ 0 = θ 0 TX Dθ a thus E 0 3] = X E 3]. This thir absolute moment is very comlicate to calculate. θ 0 Therefore, we use Höler s inequality a the fact that X GGθ 0,, a thus X Gamma, 1 to get θ 0 E X 3] θ 0 E X ]] θ 0 = = E X 4] θ 4 θ E X 2] θ 4 θ ]3 4 = θ 3 0 Simler calculations yiel ] 3 EX ] 4 θ 0 E X 3] k θ 0 A θ 0 k θ 0 Dθ 0 = θ 0 Using that X i Gamma, 1 n θ X i Gamma n 0, 1 θ 0 ] 2 E ˆθn X θ 0 = = θ n E 2 Γ Γ X i θ θ 2 0 2θ Γ. 3.6 n θ 0 E Γ +1, we get X i 1

16 168 Areas Anastasiou a Christohe Ley 1 = θ Γ Γ Γ Γ The seco erivative of the function Dθ is not uniformly boue a thus we use the result in 3.2. Regaring D θ, one has to be careful as the suremum ees on the value of : su θ: θ θ 0 ǫ su θ: θ θ 0 ǫ 1θ 2 = 1 su θ: θ θ 0 ǫ θ 2 θ 0 ǫ 2, if 0 < < 2 = θ 0 +ǫ 2, if 2. Thus, alying now the results of 3.5, 3.6, 3.7 a 3.8 on the general exression for the first uer bou in Corollary 3.2, we obtain the result in Remark The bou in 3.4 is O n. This is not obvious as we nee to comment on the orer of the term Γ 2 Γ+2. Using Γ + Γ the Taylor exansion for a ratio of Gamma functions see Tricomi a Erélyi, 1951 Γz +a Γz +b = za b 1+ a ba+b 1 +O z 2 2z for large z here, / a boue a a b, we can see that this term is of orer 1 n, leaing to the overall orer of n 1. 2 The iicator function in the last term of the bou in 3.4 comes from the fact that qθ = θ θ Θ, = Bous for the negative exonential istribution. In this subsection, we consier the most famous secial case of the Generalise Gamma istribution: the negative exonential istribution. First we will treat the canonical form of the istribution a then we will change the arametrisation to iscuss the more interesting non-canonical setting. In the canonical case D is not uniformly boue but it is boue in a neighbourhoo of θ 0 ; therefore, 3.2 of Corollary 3.2 is use. As for the non-canonical setting, we will show that Dθ = θ θ Θ which means that only the same first term of the bous in 3.2 a 3.3 survives The canonical case: Exθ. We start with X 1,...,X n exonentially istribute i.i.. raom variables with scale arameter θ > 0 a robability ensity function fx θ = θex{ θx} = ex{logθ θx} for x > 0, which we write Exθ. In terms of 3.1, this means B = 0,, Θ = 0,, Tx = x, kθ = θ, Aθ = logθ a Sx = 0. Further we have that l θ;x = n n θ x i, l θ;x = n θ 2, the unique MLE is given by ˆθ n X = 1/ X with X = 1 n n X i a R1-R4 are satisfie.

17 Delta metho for bous on MLE 169 With this in ha, we can easily see that Dθ 0 = qθ 0 = A θ 0 k θ 0 = 1 θ 0 a k θ 0 A θ 0 k θ 0 Dθ 0 = θ Simle calculations allow us here to byass the Höler inequality use for the Generalize Gamma case a to obtain E TX Dθ 0 3] 1 = E θ 0 X 3] Since X i Exθ i {1,2,...,n} then X Gamn,nθ, with Gamα,β being the Gamma istribution with shae arameter α a rate arameter β. Consequently, Eˆθ n X θ 0 2 ] = nθ 0 2 n 1n 2 2nθ2 0 n 1 +θ2 0 = n+2θ2 0 n 1n 2. Moreover, for ǫ > 0 such that 0 < ǫ < θ 0, we obtain Choosing ǫ = θ0 2, we get su θ: θ θ 0 ǫ Corollary 3.21 gives D θ = 16 θ 3 0 θ 3 0 su D 2 θ = θ 0 ǫ. 3 θ: θ θ 0 ǫ. Using this result a 3.9, ] E h niθ0 ˆθn X θ 0 EhZ] h n+2 +8 h n n 1n 2 nn+2 +8 h n 1n This bou is of orer O n a coincies, as iscusse in Remark 3.3, with the AR-bou The non-canonical case: Ex 1 θ. We now rocee to examine the more interesting case where X 1,...,X n are i.i.. raom variables from Ex 1 θ. The robability ensity function is fx θ = 1 { θ ex 1 } θ x = ex { logθ 1θ } x corresoing to B = 0,, Θ = 0,, Tx = x, kθ = 1 θ, Aθ = logθ a Sx = 0. As before, simle stes give that the MLE exists, it is unique a equal to ˆθ n X = X. The regularity coitions are satisfie a we obtain using will give the same result that E h niθ0 ] ˆθn X θ 0 EhZ] h n Iee, Dθ 0 = qθ 0 = θ 0, making the last two terms of both bous in Corollary 3.2 vanish. The result then follows from E TX Dθ 0 3] = E X θ 0 3] θ 3 0 a k θ 0 A θ 0 k θ 0 Dθ 0 = 1 θ 0. Remark The orer of the bou in terms of the samle size is, as execte, 1 n, corresoing to the orer obtaine for the Generalize Gamma istribution. The constant here is better than the one inherite from 3.4 for = = 1, thanks

18 170 Areas Anastasiou a Christohe Ley to a sharer bou for E TX Dθ 0 3]. 2 The AR-bou is given by h n +8 h n +2 h n +80 h n 6 n +3 1/2, showing that our new bou is an imrovement Emirical results. For a more comlete icture, we also assess the accuracy of our results using simulation-base ata. The rocess we follow is quite simle. We generate10000trials ofn = 10,100,1000,10000a raom i.i.. observationsfrom the exonential istribution Ex 1 2 non-canonicalcase. As function h we choose hx = 1 x 2 +2 with h H, h = 0.5 a h = With ẼhZ] being the aroximation of EhZ] u to three ecimal laces, simle calculations yiel ẼhZ] = Each trial gives an MLE ˆθ n X, hence we have em- ˆθn X θ 0 to comare to Taking the aver- ] ˆθn X θ 0 EhZ], niθ0 irical values of h age rovies a simulate estimation ofe h niθ0 which we comare to the uer bou given in Our bou rovies a very strong imrovement on the AR-bou see Table 3.1. Of course, this estimate istance is only a lower bou to the true istance, as we have chosen a articular function hinstea ofthe suremumoverallfunctions h H, but itscalculationstill rovies an iea of the accuracy of our bous. This closeness logically increases with the samle size a becomes quite shar for n 100. Table 3.1. Simulation results for the Ex 1 2 istribution treate as a non-canonical exonential family niθ0 ] n Ê h ˆθ n X θ 0 ẼhZ] New Bou AR-bou ACKNOWLEDGMENTS The authors woul like to thank the Eitor, an Associate Eitor a an anonymous referee for helful comments that le to an imrovement of the aer. They further thank Gesine Reinert for various insightful comments a suggestions. References A. Anastasiou a G. Reinert. Bous for the normal aroximation of the maximum likelihoo estimator. Bernoulli 23 1, MR G. Casella a R. L. Berger. Statistical inference. The Wasworth & Brooks/Cole Statistics/Probability Series. Wasworth& Brooks/Cole Avance Books& Software, Pacific Grove, CA ISBN MR

19 Delta metho for bous on MLE 171 L. H. Y. Chen, L. Golstein a Q.-M. Shao. Normal aroximation by Stein s metho. Probability a its Alications New York. Sringer, Heielberg ISBN MR D. R. Cox a E. J. Snell. A general efinition of resiuals. J. Roy. Statist. Soc. Ser. B 30, MR R. A. Fisher. Theory of Statistical Estimation. Mathematical Proceeings of the Cambrige Philosohical Society 22, DOI: /S C. J. Geyer. Asymtotics of maximum likelihoo without the LLN or CLT or samle size going to infinity. In Avances in Moern Statistical Theory a Alications: A Festschrift in honor of Morris L. Eaton, ages Es.: G. Jones a X. Shen, Beachwoo, OH: Institute of Mathematical Statistics DOI: /12-IMSCOLL1001. T. Mäkeläinen, K. Schmit a G. P. H. Styan. On the existence a uniqueness of the maximum likelihoo estimate of a vector-value arameter in fixe-size samles. Ann. Statist. 9 4, MR I. Nourin a G. Peccati. Normal aroximations with Malliavin calculus, volume 192 of Cambrige Tracts in Mathematics. Cambrige University Press, Cambrige ISBN MR G. Reinert. Coulings for normal aroximations with Stein s metho. In Microsurveys in iscrete robability Princeton, NJ, 1997, volume 41 of DIMACS Ser. Discrete Math. Theoret. Comut. Sci., ages Amer. Math. Soc., Provience, RI MR F. G. Tricomi a A. Erélyi. The asymtotic exansion of a ratio of gamma functions. Pacific J. Math. 1, MR

Submitted to the Brazilian Journal of Probability and Statistics

Submitted to the Brazilian Journal of Probability and Statistics Submitted to the Brazilian Journal of Probability and Statistics Multivariate normal approximation of the maximum likelihood estimator via the delta method Andreas Anastasiou a and Robert E. Gaunt b a

More information

Consistency and asymptotic normality

Consistency and asymptotic normality Consistency an asymtotic normality Class notes for Econ 842 Robert e Jong March 2006 1 Stochastic convergence The asymtotic theory of minimization estimators relies on various theorems from mathematical

More information

Bivariate distributions characterized by one family of conditionals and conditional percentile or mode functions

Bivariate distributions characterized by one family of conditionals and conditional percentile or mode functions Journal of Multivariate Analysis 99 2008) 1383 1392 www.elsevier.com/locate/jmva Bivariate istributions characterize by one family of conitionals an conitional ercentile or moe functions Barry C. Arnol

More information

Consistency and asymptotic normality

Consistency and asymptotic normality Consistency an ymtotic normality Cls notes for Econ 842 Robert e Jong Aril 2007 1 Stochtic convergence The ymtotic theory of minimization estimators relies on various theorems from mathematical statistics.

More information

Lenny Jones Department of Mathematics, Shippensburg University, Shippensburg, Pennsylvania Daniel White

Lenny Jones Department of Mathematics, Shippensburg University, Shippensburg, Pennsylvania Daniel White #A10 INTEGERS 1A (01): John Selfrige Memorial Issue SIERPIŃSKI NUMBERS IN IMAGINARY QUADRATIC FIELDS Lenny Jones Deartment of Mathematics, Shiensburg University, Shiensburg, Pennsylvania lkjone@shi.eu

More information

Elementary Analysis in Q p

Elementary Analysis in Q p Elementary Analysis in Q Hannah Hutter, May Szedlák, Phili Wirth November 17, 2011 This reort follows very closely the book of Svetlana Katok 1. 1 Sequences and Series In this section we will see some

More information

Colin Cameron: Asymptotic Theory for OLS

Colin Cameron: Asymptotic Theory for OLS Colin Cameron: Asymtotic Theory for OLS. OLS Estimator Proerties an Samling Schemes.. A Roama Consier the OLS moel with just one regressor y i = βx i + u i. The OLS estimator b β = ³ P P i= x iy i canbewrittenas

More information

Econometrics I. September, Part I. Department of Economics Stanford University

Econometrics I. September, Part I. Department of Economics Stanford University Econometrics I Deartment of Economics Stanfor University Setember, 2008 Part I Samling an Data Poulation an Samle. ineenent an ientical samling. (i.i..) Samling with relacement. aroximates samling without

More information

Lecture 6 : Dimensionality Reduction

Lecture 6 : Dimensionality Reduction CPS290: Algorithmic Founations of Data Science February 3, 207 Lecture 6 : Dimensionality Reuction Lecturer: Kamesh Munagala Scribe: Kamesh Munagala In this lecture, we will consier the roblem of maing

More information

Colin Cameron: Brief Asymptotic Theory for 240A

Colin Cameron: Brief Asymptotic Theory for 240A Colin Cameron: Brief Asymtotic Theory for 240A For 240A we o not go in to great etail. Key OLS results are in Section an 4. The theorems cite in sections 2 an 3 are those from Aenix A of Cameron an Trivei

More information

MATH 2710: NOTES FOR ANALYSIS

MATH 2710: NOTES FOR ANALYSIS MATH 270: NOTES FOR ANALYSIS The main ideas we will learn from analysis center around the idea of a limit. Limits occurs in several settings. We will start with finite limits of sequences, then cover infinite

More information

NOTES. The Primes that Euclid Forgot

NOTES. The Primes that Euclid Forgot NOTES Eite by Sergei Tabachnikov The Primes that Eucli Forgot Paul Pollack an Enrique Treviño Abstract. Let q 2. Suosing that we have efine q j for all ale j ale k, let q k+ be a rime factor of + Q k j

More information

An Estimate For Heilbronn s Exponential Sum

An Estimate For Heilbronn s Exponential Sum An Estimate For Heilbronn s Exonential Sum D.R. Heath-Brown Magdalen College, Oxford For Heini Halberstam, on his retirement Let be a rime, and set e(x) = ex(2πix). Heilbronn s exonential sum is defined

More information

Improved Bounds on Bell Numbers and on Moments of Sums of Random Variables

Improved Bounds on Bell Numbers and on Moments of Sums of Random Variables Imroved Bounds on Bell Numbers and on Moments of Sums of Random Variables Daniel Berend Tamir Tassa Abstract We rovide bounds for moments of sums of sequences of indeendent random variables. Concentrating

More information

LECTURE 7 NOTES. x n. d x if. E [g(x n )] E [g(x)]

LECTURE 7 NOTES. x n. d x if. E [g(x n )] E [g(x)] LECTURE 7 NOTES 1. Convergence of random variables. Before delving into the large samle roerties of the MLE, we review some concets from large samle theory. 1. Convergence in robability: x n x if, for

More information

Elements of Asymptotic Theory. James L. Powell Department of Economics University of California, Berkeley

Elements of Asymptotic Theory. James L. Powell Department of Economics University of California, Berkeley Elements of Asymtotic Theory James L. Powell Deartment of Economics University of California, Berkeley Objectives of Asymtotic Theory While exact results are available for, say, the distribution of the

More information

ON THE AVERAGE NUMBER OF DIVISORS OF REDUCIBLE QUADRATIC POLYNOMIALS

ON THE AVERAGE NUMBER OF DIVISORS OF REDUCIBLE QUADRATIC POLYNOMIALS ON THE AVERAGE NUMBER OF DIVISORS OF REDUCIBLE QUADRATIC POLYNOMIALS KOSTADINKA LAPKOVA Abstract. We give an asymtotic formula for the ivisor sum c

More information

On Wald-Type Optimal Stopping for Brownian Motion

On Wald-Type Optimal Stopping for Brownian Motion J Al Probab Vol 34, No 1, 1997, (66-73) Prerint Ser No 1, 1994, Math Inst Aarhus On Wald-Tye Otimal Stoing for Brownian Motion S RAVRSN and PSKIR The solution is resented to all otimal stoing roblems of

More information

#A47 INTEGERS 15 (2015) QUADRATIC DIOPHANTINE EQUATIONS WITH INFINITELY MANY SOLUTIONS IN POSITIVE INTEGERS

#A47 INTEGERS 15 (2015) QUADRATIC DIOPHANTINE EQUATIONS WITH INFINITELY MANY SOLUTIONS IN POSITIVE INTEGERS #A47 INTEGERS 15 (015) QUADRATIC DIOPHANTINE EQUATIONS WITH INFINITELY MANY SOLUTIONS IN POSITIVE INTEGERS Mihai Ciu Simion Stoilow Institute of Mathematics of the Romanian Academy, Research Unit No. 5,

More information

Chapter 7: Special Distributions

Chapter 7: Special Distributions This chater first resents some imortant distributions, and then develos the largesamle distribution theory which is crucial in estimation and statistical inference Discrete distributions The Bernoulli

More information

The Effect of a Finite Measurement Volume on Power Spectra from a Burst Type LDA

The Effect of a Finite Measurement Volume on Power Spectra from a Burst Type LDA The Effect of a Finite Measurement Volume on Power Sectra from a Burst Tye LDA Preben Buchhave 1,*, Clara M. Velte, an William K. George 3 1. Intarsia Otics, Birkerø, Denmark. Technical University of Denmark,

More information

Various Proofs for the Decrease Monotonicity of the Schatten s Power Norm, Various Families of R n Norms and Some Open Problems

Various Proofs for the Decrease Monotonicity of the Schatten s Power Norm, Various Families of R n Norms and Some Open Problems Int. J. Oen Problems Comt. Math., Vol. 3, No. 2, June 2010 ISSN 1998-6262; Coyright c ICSRS Publication, 2010 www.i-csrs.org Various Proofs for the Decrease Monotonicity of the Schatten s Power Norm, Various

More information

Maximum Likelihood Asymptotic Theory. Eduardo Rossi University of Pavia

Maximum Likelihood Asymptotic Theory. Eduardo Rossi University of Pavia Maximum Likelihood Asymtotic Theory Eduardo Rossi University of Pavia Slutsky s Theorem, Cramer s Theorem Slutsky s Theorem Let {X N } be a random sequence converging in robability to a constant a, and

More information

Mod p 3 analogues of theorems of Gauss and Jacobi on binomial coefficients

Mod p 3 analogues of theorems of Gauss and Jacobi on binomial coefficients ACTA ARITHMETICA 2.2 (200 Mo 3 analogues of theorems of Gauss an Jacobi on binomial coefficients by John B. Cosgrave (Dublin an Karl Dilcher (Halifax. Introuction. One of the most remarkable congruences

More information

IMPROVED BOUNDS IN THE SCALED ENFLO TYPE INEQUALITY FOR BANACH SPACES

IMPROVED BOUNDS IN THE SCALED ENFLO TYPE INEQUALITY FOR BANACH SPACES IMPROVED BOUNDS IN THE SCALED ENFLO TYPE INEQUALITY FOR BANACH SPACES OHAD GILADI AND ASSAF NAOR Abstract. It is shown that if (, ) is a Banach sace with Rademacher tye 1 then for every n N there exists

More information

7. Introduction to Large Sample Theory

7. Introduction to Large Sample Theory 7. Introuction to Large Samle Theory Hayashi. 88-97/109-133 Avance Econometrics I, Autumn 2010, Large-Samle Theory 1 Introuction We looke at finite-samle roerties of the OLS estimator an its associate

More information

Lecture 4: Law of Large Number and Central Limit Theorem

Lecture 4: Law of Large Number and Central Limit Theorem ECE 645: Estimation Theory Sring 2015 Instructor: Prof. Stanley H. Chan Lecture 4: Law of Large Number and Central Limit Theorem (LaTeX reared by Jing Li) March 31, 2015 This lecture note is based on ECE

More information

1 Riesz Potential and Enbeddings Theorems

1 Riesz Potential and Enbeddings Theorems Riesz Potential and Enbeddings Theorems Given 0 < < and a function u L loc R, the Riesz otential of u is defined by u y I u x := R x y dy, x R We begin by finding an exonent such that I u L R c u L R for

More information

Topic 7: Convergence of Random Variables

Topic 7: Convergence of Random Variables Topic 7: Convergence of Ranom Variables Course 003, 2016 Page 0 The Inference Problem So far, our starting point has been a given probability space (S, F, P). We now look at how to generate information

More information

Elements of Asymptotic Theory. James L. Powell Department of Economics University of California, Berkeley

Elements of Asymptotic Theory. James L. Powell Department of Economics University of California, Berkeley Elements of Asymtotic Theory James L. Powell Deartment of Economics University of California, Berkeley Objectives of Asymtotic Theory While exact results are available for, say, the distribution of the

More information

Digitally delicate primes

Digitally delicate primes Digitally elicate rimes Jackson Hoer Paul Pollack Deartment of Mathematics University of Georgia Athens, Georgia 30602 Tao has shown that in any fixe base, a ositive roortion of rime numbers cannot have

More information

Towards understanding the Lorenz curve using the Uniform distribution. Chris J. Stephens. Newcastle City Council, Newcastle upon Tyne, UK

Towards understanding the Lorenz curve using the Uniform distribution. Chris J. Stephens. Newcastle City Council, Newcastle upon Tyne, UK Towards understanding the Lorenz curve using the Uniform distribution Chris J. Stehens Newcastle City Council, Newcastle uon Tyne, UK (For the Gini-Lorenz Conference, University of Siena, Italy, May 2005)

More information

1 Probability Spaces and Random Variables

1 Probability Spaces and Random Variables 1 Probability Saces and Random Variables 1.1 Probability saces Ω: samle sace consisting of elementary events (or samle oints). F : the set of events P: robability 1.2 Kolmogorov s axioms Definition 1.2.1

More information

The inverse Goldbach problem

The inverse Goldbach problem 1 The inverse Goldbach roblem by Christian Elsholtz Submission Setember 7, 2000 (this version includes galley corrections). Aeared in Mathematika 2001. Abstract We imrove the uer and lower bounds of the

More information

Probabilistic Learning

Probabilistic Learning Statistical Machine Learning Notes 11 Instructor: Justin Domke Probabilistic Learning Contents 1 Introuction 2 2 Maximum Likelihoo 2 3 Examles of Maximum Likelihoo 3 3.1 Binomial......................................

More information

Elementary theory of L p spaces

Elementary theory of L p spaces CHAPTER 3 Elementary theory of L saces 3.1 Convexity. Jensen, Hölder, Minkowski inequality. We begin with two definitions. A set A R d is said to be convex if, for any x 0, x 1 2 A x = x 0 + (x 1 x 0 )

More information

Convergence of random variables, and the Borel-Cantelli lemmas

Convergence of random variables, and the Borel-Cantelli lemmas Stat 205A Setember, 12, 2002 Convergence of ranom variables, an the Borel-Cantelli lemmas Lecturer: James W. Pitman Scribes: Jin Kim (jin@eecs) 1 Convergence of ranom variables Recall that, given a sequence

More information

#A37 INTEGERS 15 (2015) NOTE ON A RESULT OF CHUNG ON WEIL TYPE SUMS

#A37 INTEGERS 15 (2015) NOTE ON A RESULT OF CHUNG ON WEIL TYPE SUMS #A37 INTEGERS 15 (2015) NOTE ON A RESULT OF CHUNG ON WEIL TYPE SUMS Norbert Hegyvári ELTE TTK, Eötvös University, Institute of Mathematics, Budaest, Hungary hegyvari@elte.hu François Hennecart Université

More information

ON THE NORM OF AN IDEMPOTENT SCHUR MULTIPLIER ON THE SCHATTEN CLASS

ON THE NORM OF AN IDEMPOTENT SCHUR MULTIPLIER ON THE SCHATTEN CLASS PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 00, Number 0, Pages 000 000 S 000-9939XX)0000-0 ON THE NORM OF AN IDEMPOTENT SCHUR MULTIPLIER ON THE SCHATTEN CLASS WILLIAM D. BANKS AND ASMA HARCHARRAS

More information

Survey Sampling. 1 Design-based Inference. Kosuke Imai Department of Politics, Princeton University. February 19, 2013

Survey Sampling. 1 Design-based Inference. Kosuke Imai Department of Politics, Princeton University. February 19, 2013 Survey Sampling Kosuke Imai Department of Politics, Princeton University February 19, 2013 Survey sampling is one of the most commonly use ata collection methos for social scientists. We begin by escribing

More information

Lecture 3 January 16

Lecture 3 January 16 Stats 3b: Theory of Statistics Winter 28 Lecture 3 January 6 Lecturer: Yu Bai/John Duchi Scribe: Shuangning Li, Theodor Misiakiewicz Warning: these notes may contain factual errors Reading: VDV Chater

More information

Probabilistic Learning

Probabilistic Learning Statistical Machine Learning Notes 10 Instructor: Justin Domke Probabilistic Learning Contents 1 Introuction 2 2 Maximum Likelihoo 2 3 Examles of Maximum Likelihoo 3 3.1 Binomial......................................

More information

Estimation of the large covariance matrix with two-step monotone missing data

Estimation of the large covariance matrix with two-step monotone missing data Estimation of the large covariance matrix with two-ste monotone missing data Masashi Hyodo, Nobumichi Shutoh 2, Takashi Seo, and Tatjana Pavlenko 3 Deartment of Mathematical Information Science, Tokyo

More information

Journal of Mathematical Analysis and Applications

Journal of Mathematical Analysis and Applications J. Math. Anal. Al. 44 (3) 3 38 Contents lists available at SciVerse ScienceDirect Journal of Mathematical Analysis and Alications journal homeage: www.elsevier.com/locate/jmaa Maximal surface area of a

More information

He s Homotopy Perturbation Method for solving Linear and Non-Linear Fredholm Integro-Differential Equations

He s Homotopy Perturbation Method for solving Linear and Non-Linear Fredholm Integro-Differential Equations nternational Journal of Theoretical an Alie Mathematics 2017; 3(6): 174-181 htt://www.scienceublishinggrou.com/j/ijtam oi: 10.11648/j.ijtam.20170306.11 SSN: 2575-5072 (Print); SSN: 2575-5080 (Online) He

More information

On Z p -norms of random vectors

On Z p -norms of random vectors On Z -norms of random vectors Rafa l Lata la Abstract To any n-dimensional random vector X we may associate its L -centroid body Z X and the corresonding norm. We formulate a conjecture concerning the

More information

4. Score normalization technical details We now discuss the technical details of the score normalization method.

4. Score normalization technical details We now discuss the technical details of the score normalization method. SMT SCORING SYSTEM This document describes the scoring system for the Stanford Math Tournament We begin by giving an overview of the changes to scoring and a non-technical descrition of the scoring rules

More information

1 Extremum Estimators

1 Extremum Estimators FINC 9311-21 Financial Econometrics Handout Jialin Yu 1 Extremum Estimators Let θ 0 be a vector of k 1 unknown arameters. Extremum estimators: estimators obtained by maximizing or minimizing some objective

More information

Robustness of classifiers to uniform l p and Gaussian noise Supplementary material

Robustness of classifiers to uniform l p and Gaussian noise Supplementary material Robustness of classifiers to uniform l and Gaussian noise Sulementary material Jean-Yves Franceschi Ecole Normale Suérieure de Lyon LIP UMR 5668 Omar Fawzi Ecole Normale Suérieure de Lyon LIP UMR 5668

More information

On Doob s Maximal Inequality for Brownian Motion

On Doob s Maximal Inequality for Brownian Motion Stochastic Process. Al. Vol. 69, No., 997, (-5) Research Reort No. 337, 995, Det. Theoret. Statist. Aarhus On Doob s Maximal Inequality for Brownian Motion S. E. GRAVERSEN and G. PESKIR If B = (B t ) t

More information

On a class of Rellich inequalities

On a class of Rellich inequalities On a class of Rellich inequalities G. Barbatis A. Tertikas Dedicated to Professor E.B. Davies on the occasion of his 60th birthday Abstract We rove Rellich and imroved Rellich inequalities that involve

More information

Convergence Analysis of Terminal ILC in the z Domain

Convergence Analysis of Terminal ILC in the z Domain 25 American Control Conference June 8-, 25 Portlan, OR, USA WeA63 Convergence Analysis of erminal LC in the Domain Guy Gauthier, an Benoit Boulet, Member, EEE Abstract his aer shows how we can aly -transform

More information

THE ZEROS OF A QUADRATIC FORM AT SQUARE-FREE POINTS

THE ZEROS OF A QUADRATIC FORM AT SQUARE-FREE POINTS THE ZEROS OF A QUADRATIC FORM AT SQUARE-FREE POINTS R. C. BAKER Abstract. Let F(x 1,..., x n be a nonsingular inefinite quaratic form, n = 3 or 4. Results are obtaine on the number of solutions of F(x

More information

HENSEL S LEMMA KEITH CONRAD

HENSEL S LEMMA KEITH CONRAD HENSEL S LEMMA KEITH CONRAD 1. Introduction In the -adic integers, congruences are aroximations: for a and b in Z, a b mod n is the same as a b 1/ n. Turning information modulo one ower of into similar

More information

An Improved Generalized Estimation Procedure of Current Population Mean in Two-Occasion Successive Sampling

An Improved Generalized Estimation Procedure of Current Population Mean in Two-Occasion Successive Sampling Journal of Modern Alied Statistical Methods Volume 15 Issue Article 14 11-1-016 An Imroved Generalized Estimation Procedure of Current Poulation Mean in Two-Occasion Successive Samling G. N. Singh Indian

More information

RAMANUJAN-NAGELL CUBICS

RAMANUJAN-NAGELL CUBICS RAMANUJAN-NAGELL CUBICS MARK BAUER AND MICHAEL A. BENNETT ABSTRACT. A well-nown result of Beuers [3] on the generalized Ramanujan-Nagell equation has, at its heart, a lower bound on the quantity x 2 2

More information

Positivity, local smoothing and Harnack inequalities for very fast diffusion equations

Positivity, local smoothing and Harnack inequalities for very fast diffusion equations Positivity, local smoothing and Harnack inequalities for very fast diffusion equations Dedicated to Luis Caffarelli for his ucoming 60 th birthday Matteo Bonforte a, b and Juan Luis Vázquez a, c Abstract

More information

t 0 Xt sup X t p c p inf t 0

t 0 Xt sup X t p c p inf t 0 SHARP MAXIMAL L -ESTIMATES FOR MARTINGALES RODRIGO BAÑUELOS AND ADAM OSȨKOWSKI ABSTRACT. Let X be a suermartingale starting from 0 which has only nonnegative jums. For each 0 < < we determine the best

More information

Asymptotically Optimal Simulation Allocation under Dependent Sampling

Asymptotically Optimal Simulation Allocation under Dependent Sampling Asymtotically Otimal Simulation Allocation under Deendent Samling Xiaoing Xiong The Robert H. Smith School of Business, University of Maryland, College Park, MD 20742-1815, USA, xiaoingx@yahoo.com Sandee

More information

ON THE LEAST SIGNIFICANT p ADIC DIGITS OF CERTAIN LUCAS NUMBERS

ON THE LEAST SIGNIFICANT p ADIC DIGITS OF CERTAIN LUCAS NUMBERS #A13 INTEGERS 14 (014) ON THE LEAST SIGNIFICANT ADIC DIGITS OF CERTAIN LUCAS NUMBERS Tamás Lengyel Deartment of Mathematics, Occidental College, Los Angeles, California lengyel@oxy.edu Received: 6/13/13,

More information

Sums of independent random variables

Sums of independent random variables 3 Sums of indeendent random variables This lecture collects a number of estimates for sums of indeendent random variables with values in a Banach sace E. We concentrate on sums of the form N γ nx n, where

More information

Convex Optimization methods for Computing Channel Capacity

Convex Optimization methods for Computing Channel Capacity Convex Otimization methods for Comuting Channel Caacity Abhishek Sinha Laboratory for Information and Decision Systems (LIDS), MIT sinhaa@mit.edu May 15, 2014 We consider a classical comutational roblem

More information

#A64 INTEGERS 18 (2018) APPLYING MODULAR ARITHMETIC TO DIOPHANTINE EQUATIONS

#A64 INTEGERS 18 (2018) APPLYING MODULAR ARITHMETIC TO DIOPHANTINE EQUATIONS #A64 INTEGERS 18 (2018) APPLYING MODULAR ARITHMETIC TO DIOPHANTINE EQUATIONS Ramy F. Taki ElDin Physics and Engineering Mathematics Deartment, Faculty of Engineering, Ain Shams University, Cairo, Egyt

More information

Review of Probability Theory II

Review of Probability Theory II Review of Probability Theory II January 9-3, 008 Exectation If the samle sace Ω = {ω, ω,...} is countable and g is a real-valued function, then we define the exected value or the exectation of a function

More information

p-adic Measures and Bernoulli Numbers

p-adic Measures and Bernoulli Numbers -Adic Measures and Bernoulli Numbers Adam Bowers Introduction The constants B k in the Taylor series exansion t e t = t k B k k! k=0 are known as the Bernoulli numbers. The first few are,, 6, 0, 30, 0,

More information

arxiv:cond-mat/ v2 25 Sep 2002

arxiv:cond-mat/ v2 25 Sep 2002 Energy fluctuations at the multicritical oint in two-dimensional sin glasses arxiv:cond-mat/0207694 v2 25 Se 2002 1. Introduction Hidetoshi Nishimori, Cyril Falvo and Yukiyasu Ozeki Deartment of Physics,

More information

Additive results for the generalized Drazin inverse in a Banach algebra

Additive results for the generalized Drazin inverse in a Banach algebra Additive results for the generalized Drazin inverse in a Banach algebra Dragana S. Cvetković-Ilić Dragan S. Djordjević and Yimin Wei* Abstract In this aer we investigate additive roerties of the generalized

More information

Khinchine inequality for slightly dependent random variables

Khinchine inequality for slightly dependent random variables arxiv:170808095v1 [mathpr] 7 Aug 017 Khinchine inequality for slightly deendent random variables Susanna Sektor Abstract We rove a Khintchine tye inequality under the assumtion that the sum of Rademacher

More information

Submitted to the Journal of Hydraulic Engineering, ASCE, January, 2006 NOTE ON THE ANALYSIS OF PLUNGING OF DENSITY FLOWS

Submitted to the Journal of Hydraulic Engineering, ASCE, January, 2006 NOTE ON THE ANALYSIS OF PLUNGING OF DENSITY FLOWS Submitte to the Journal of Hyraulic Engineering, ASCE, January, 006 NOTE ON THE ANALYSIS OF PLUNGING OF DENSITY FLOWS Gary Parker 1, Member, ASCE an Horacio Toniolo ABSTRACT This note is evote to the correction

More information

System Reliability Estimation and Confidence Regions from Subsystem and Full System Tests

System Reliability Estimation and Confidence Regions from Subsystem and Full System Tests 009 American Control Conference Hyatt Regency Riverfront, St. Louis, MO, USA June 0-, 009 FrB4. System Reliability Estimation and Confidence Regions from Subsystem and Full System Tests James C. Sall Abstract

More information

ECON 4130 Supplementary Exercises 1-4

ECON 4130 Supplementary Exercises 1-4 HG Set. 0 ECON 430 Sulementary Exercises - 4 Exercise Quantiles (ercentiles). Let X be a continuous random variable (rv.) with df f( x ) and cdf F( x ). For 0< < we define -th quantile (or 00-th ercentile),

More information

ON MINKOWSKI MEASURABILITY

ON MINKOWSKI MEASURABILITY ON MINKOWSKI MEASURABILITY F. MENDIVIL AND J. C. SAUNDERS DEPARTMENT OF MATHEMATICS AND STATISTICS ACADIA UNIVERSITY WOLFVILLE, NS CANADA B4P 2R6 Abstract. Two athological roerties of Minkowski content

More information

Lower Confidence Bound for Process-Yield Index S pk with Autocorrelated Process Data

Lower Confidence Bound for Process-Yield Index S pk with Autocorrelated Process Data Quality Technology & Quantitative Management Vol. 1, No.,. 51-65, 15 QTQM IAQM 15 Lower onfidence Bound for Process-Yield Index with Autocorrelated Process Data Fu-Kwun Wang * and Yeneneh Tamirat Deartment

More information

MATHEMATICAL MODELLING OF THE WIRELESS COMMUNICATION NETWORK

MATHEMATICAL MODELLING OF THE WIRELESS COMMUNICATION NETWORK Comuter Modelling and ew Technologies, 5, Vol.9, o., 3-39 Transort and Telecommunication Institute, Lomonosov, LV-9, Riga, Latvia MATHEMATICAL MODELLIG OF THE WIRELESS COMMUICATIO ETWORK M. KOPEETSK Deartment

More information

General Linear Model Introduction, Classes of Linear models and Estimation

General Linear Model Introduction, Classes of Linear models and Estimation Stat 740 General Linear Model Introduction, Classes of Linear models and Estimation An aim of scientific enquiry: To describe or to discover relationshis among events (variables) in the controlled (laboratory)

More information

PETER J. GRABNER AND ARNOLD KNOPFMACHER

PETER J. GRABNER AND ARNOLD KNOPFMACHER ARITHMETIC AND METRIC PROPERTIES OF -ADIC ENGEL SERIES EXPANSIONS PETER J. GRABNER AND ARNOLD KNOPFMACHER Abstract. We derive a characterization of rational numbers in terms of their unique -adic Engel

More information

Lecture 6. 2 Recurrence/transience, harmonic functions and martingales

Lecture 6. 2 Recurrence/transience, harmonic functions and martingales Lecture 6 Classification of states We have shown that all states of an irreducible countable state Markov chain must of the same tye. This gives rise to the following classification. Definition. [Classification

More information

A Social Welfare Optimal Sequential Allocation Procedure

A Social Welfare Optimal Sequential Allocation Procedure A Social Welfare Otimal Sequential Allocation Procedure Thomas Kalinowsi Universität Rostoc, Germany Nina Narodytsa and Toby Walsh NICTA and UNSW, Australia May 2, 201 Abstract We consider a simle sequential

More information

Estimating function analysis for a class of Tweedie regression models

Estimating function analysis for a class of Tweedie regression models Title Estimating function analysis for a class of Tweedie regression models Author Wagner Hugo Bonat Deartamento de Estatística - DEST, Laboratório de Estatística e Geoinformação - LEG, Universidade Federal

More information

Lecture 2: Consistency of M-estimators

Lecture 2: Consistency of M-estimators Lecture 2: Instructor: Deartment of Economics Stanford University Preared by Wenbo Zhou, Renmin University References Takeshi Amemiya, 1985, Advanced Econometrics, Harvard University Press Newey and McFadden,

More information

CONVOLVED SUBSAMPLING ESTIMATION WITH APPLICATIONS TO BLOCK BOOTSTRAP

CONVOLVED SUBSAMPLING ESTIMATION WITH APPLICATIONS TO BLOCK BOOTSTRAP Submitted to the Annals of Statistics arxiv: arxiv:1706.07237 CONVOLVED SUBSAMPLING ESTIMATION WITH APPLICATIONS TO BLOCK BOOTSTRAP By Johannes Tewes, Dimitris N. Politis and Daniel J. Nordman Ruhr-Universität

More information

A MIXED CONTROL CHART ADAPTED TO THE TRUNCATED LIFE TEST BASED ON THE WEIBULL DISTRIBUTION

A MIXED CONTROL CHART ADAPTED TO THE TRUNCATED LIFE TEST BASED ON THE WEIBULL DISTRIBUTION O P E R A T I O N S R E S E A R C H A N D D E C I S I O N S No. 27 DOI:.5277/ord73 Nasrullah KHAN Muhammad ASLAM 2 Kyung-Jun KIM 3 Chi-Hyuck JUN 4 A MIXED CONTROL CHART ADAPTED TO THE TRUNCATED LIFE TEST

More information

Application of Measurement System R&R Analysis in Ultrasonic Testing

Application of Measurement System R&R Analysis in Ultrasonic Testing 17th Worl Conference on Nonestructive Testing, 5-8 Oct 8, Shanghai, China Alication of Measurement System & Analysis in Ultrasonic Testing iao-hai ZHANG, Bing-ya CHEN, Yi ZHU Deartment of Testing an Control

More information

On the Toppling of a Sand Pile

On the Toppling of a Sand Pile Discrete Mathematics and Theoretical Comuter Science Proceedings AA (DM-CCG), 2001, 275 286 On the Toling of a Sand Pile Jean-Christohe Novelli 1 and Dominique Rossin 2 1 CNRS, LIFL, Bâtiment M3, Université

More information

Department of Mathematics

Department of Mathematics Deartment of Mathematics Ma 3/03 KC Border Introduction to Probability and Statistics Winter 209 Sulement : Series fun, or some sums Comuting the mean and variance of discrete distributions often involves

More information

COMMUNICATION BETWEEN SHAREHOLDERS 1

COMMUNICATION BETWEEN SHAREHOLDERS 1 COMMUNICATION BTWN SHARHOLDRS 1 A B. O A : A D Lemma B.1. U to µ Z r 2 σ2 Z + σ2 X 2r ω 2 an additive constant that does not deend on a or θ, the agents ayoffs can be written as: 2r rθa ω2 + θ µ Y rcov

More information

arxiv:math/ v2 [math.nt] 21 Oct 2004

arxiv:math/ v2 [math.nt] 21 Oct 2004 SUMS OF THE FORM 1/x k 1 + +1/x k n MODULO A PRIME arxiv:math/0403360v2 [math.nt] 21 Oct 2004 Ernie Croot 1 Deartment of Mathematics, Georgia Institute of Technology, Atlanta, GA 30332 ecroot@math.gatech.edu

More information

New Information Measures for the Generalized Normal Distribution

New Information Measures for the Generalized Normal Distribution Information,, 3-7; doi:.339/info3 OPEN ACCESS information ISSN 75-7 www.mdi.com/journal/information Article New Information Measures for the Generalized Normal Distribution Christos P. Kitsos * and Thomas

More information

MODELING THE RELIABILITY OF C4ISR SYSTEMS HARDWARE/SOFTWARE COMPONENTS USING AN IMPROVED MARKOV MODEL

MODELING THE RELIABILITY OF C4ISR SYSTEMS HARDWARE/SOFTWARE COMPONENTS USING AN IMPROVED MARKOV MODEL Technical Sciences and Alied Mathematics MODELING THE RELIABILITY OF CISR SYSTEMS HARDWARE/SOFTWARE COMPONENTS USING AN IMPROVED MARKOV MODEL Cezar VASILESCU Regional Deartment of Defense Resources Management

More information

ON FREIMAN S 2.4-THEOREM

ON FREIMAN S 2.4-THEOREM ON FREIMAN S 2.4-THEOREM ØYSTEIN J. RØDSETH Abstract. Gregory Freiman s celebrated 2.4-Theorem says that if A is a set of residue classes modulo a rime satisfying 2A 2.4 A 3 and A < /35, then A is contained

More information

Multiplicity of weak solutions for a class of nonuniformly elliptic equations of p-laplacian type

Multiplicity of weak solutions for a class of nonuniformly elliptic equations of p-laplacian type Nonlinear Analysis 7 29 536 546 www.elsevier.com/locate/na Multilicity of weak solutions for a class of nonuniformly ellitic equations of -Lalacian tye Hoang Quoc Toan, Quô c-anh Ngô Deartment of Mathematics,

More information

ON UNIFORM BOUNDEDNESS OF DYADIC AVERAGING OPERATORS IN SPACES OF HARDY-SOBOLEV TYPE. 1. Introduction

ON UNIFORM BOUNDEDNESS OF DYADIC AVERAGING OPERATORS IN SPACES OF HARDY-SOBOLEV TYPE. 1. Introduction ON UNIFORM BOUNDEDNESS OF DYADIC AVERAGING OPERATORS IN SPACES OF HARDY-SOBOLEV TYPE GUSTAVO GARRIGÓS ANDREAS SEEGER TINO ULLRICH Abstract We give an alternative roof and a wavelet analog of recent results

More information

The non-stochastic multi-armed bandit problem

The non-stochastic multi-armed bandit problem Submitted for journal ublication. The non-stochastic multi-armed bandit roblem Peter Auer Institute for Theoretical Comuter Science Graz University of Technology A-8010 Graz (Austria) auer@igi.tu-graz.ac.at

More information

JUST THE MATHS UNIT NUMBER DIFFERENTIATION 2 (Rates of change) A.J.Hobson

JUST THE MATHS UNIT NUMBER DIFFERENTIATION 2 (Rates of change) A.J.Hobson JUST THE MATHS UNIT NUMBER 10.2 DIFFERENTIATION 2 (Rates of change) by A.J.Hobson 10.2.1 Introuction 10.2.2 Average rates of change 10.2.3 Instantaneous rates of change 10.2.4 Derivatives 10.2.5 Exercises

More information

#A6 INTEGERS 15A (2015) ON REDUCIBLE AND PRIMITIVE SUBSETS OF F P, I. Katalin Gyarmati 1.

#A6 INTEGERS 15A (2015) ON REDUCIBLE AND PRIMITIVE SUBSETS OF F P, I. Katalin Gyarmati 1. #A6 INTEGERS 15A (015) ON REDUCIBLE AND PRIMITIVE SUBSETS OF F P, I Katalin Gyarmati 1 Deartment of Algebra and Number Theory, Eötvös Loránd University and MTA-ELTE Geometric and Algebraic Combinatorics

More information

LORENZO BRANDOLESE AND MARIA E. SCHONBEK

LORENZO BRANDOLESE AND MARIA E. SCHONBEK LARGE TIME DECAY AND GROWTH FOR SOLUTIONS OF A VISCOUS BOUSSINESQ SYSTEM LORENZO BRANDOLESE AND MARIA E. SCHONBEK Abstract. In this aer we analyze the decay and the growth for large time of weak and strong

More information

Notes on Instrumental Variables Methods

Notes on Instrumental Variables Methods Notes on Instrumental Variables Methods Michele Pellizzari IGIER-Bocconi, IZA and frdb 1 The Instrumental Variable Estimator Instrumental variable estimation is the classical solution to the roblem of

More information

The thermal wind 1. v g

The thermal wind 1. v g The thermal win The thermal win Introuction The geostrohic win is etermine by the graient of the isobars (on a horizontal surface) or isohyses (on a ressure surface). On a ressure surface the graient of

More information

Asymptotic theory for linear regression and IV estimation

Asymptotic theory for linear regression and IV estimation Asymtotic theory for linear regression and IV estimation Jean-Marie Dufour McGill University First version: November 20 Revised: December 20 his version: December 20 Comiled: December 3, 20, : his work

More information

Research Article An iterative Algorithm for Hemicontractive Mappings in Banach Spaces

Research Article An iterative Algorithm for Hemicontractive Mappings in Banach Spaces Abstract and Alied Analysis Volume 2012, Article ID 264103, 11 ages doi:10.1155/2012/264103 Research Article An iterative Algorithm for Hemicontractive Maings in Banach Saces Youli Yu, 1 Zhitao Wu, 2 and

More information