A note on strategies for bandit problems with infinitely many arms

A bandit problem consisting of a sequence of n choices (n) from a number of infinitely many Bernoulli arms is considered. The parameters of Bernoulli arms are independent and identically distributed random variables from a common distribution F on the interval [0,1] and F is continuous with F(0)=0 and F(1)=1. The goal is to investigate the asymptotic expected failure rates of k-failure strategies, and obtain a lower bound for the expected failure proportion over all strategies presented in Berry et al. (1997). We show that the asymptotic expected failure rates of k-failure strategies when 0<b1 and a lower bound can be evaluated if the limit of the ratio F(1)–F(t) versus (1–t)^b exists as t1^– for some b>0.     

Continuous representations of interval orders based on induced preorders

Conditions are given for the existence of a pair of continuous real valued functions (u, v) representing an interval order on a topological space (X, ), in the sense that [xy if and only ifu(x)].I am deeply indepted to Professor Alain Chateauneuf for many valuable suggestions.     

Minimal efficiency of designs under the class of orthogonally invariant information criteria

Consider the linear regression model with uncorrelated errors and an experimental design . In the article, we address the problem of calculating the minimal efficiency of with respect to the class of orthogonally invariant information criteria, containing all Kiefers criteria of _p-optimality, among others. We show that the -minimal efficiency of is equal to the minimal efficiency of with respect to a finite class of criteria which generalize the criterion of E-optimality. We also formulate conditions under which a design is maximin efficient, i.e. the most efficiency-stable for criteria from . To illustrate the results, we calculated the -minimal efficiency of _p (in particular D, A and E) optimal designs for polynomial regression on [–1,1] up to degree 4. Moreover, for the quadratic model we explicitly constructed the -maximin efficient design.Acknowledgement. The author would like to thank prof. Pázman as well as an anonymous referee for useful and inspiring comments on earlier versions of this article.Supported by: VEGA grant of the Slovak Agency No. 1/0264/03     

On characterizations of distributions by regression of adjacent generalized order statistics

Let , r ≥ 1, denote generalized order statistics, with arbitrary parameters , based on distribution function F. In this paper we characterize continuous distributions F by the regression of adjacent generalized order statistics, i.e. where are continuous and increasing functions and ψ is strictly increasing. Further we investigate in detail the case when ψ(x) = x and g is a linear function of the form g(x) = cx + d for some .     

$${\phi_p}$$-optimal designs for a linear log contrast model for experiments with mixtures

A mixture experiment is an experiment in which the k ingredients are nonnegative and subject to the simplex restriction on the (k − 1)-dimensional probability simplex S ^k-1. In this work, an essentially complete class of designs under the Kiefer ordering for a linear log contrast model with a mixture experiment is presented. Based on the completeness result, -optimal designs for all p,−∞ ≤ p ≤ 1 including D- and A-optimal are obtained, where the eigenvalues of the design moment matrix are used. By using the approach presented here, we gain insight on how these -optimal designs behave. Mong-Na Lo Huang was supported in part by the National Science Council of Taiwan, ROC under grant NSC 93-2118-M-110-001.     

Some new mathematical methods in dynamic programming over infinite horizon

Two properties of the potentialsU(x, y) are studied. They are the acyclicity and the strong convexity. By mean of them, information on the dynamic behavior of the optimal choice functions is obtained. The previous results are then applied,via Dynamic Programming, to the models of optimization over an infinite horizon (discounted Ramsey models). Qualitative information on the dynamics in such models is derived and some new stability results are given.

---

Riassunto Nel presente lavoro vengono proposte due proprietà dei potenzialiU(x, y): l'aciclicità e la convessità forte. Entrambe permettono di ottenere informazioni sul comportamento dinamico delle funzioni di scelta ottimale .La precedente teoria viene poi applicata, utilizzando la programmazione dinamica, al problema di dedurre informazioni qualitative sulle dinamiche nei modelli di ottimizzazione ad orizzonte infinito (modelli di Ramsey con utilità scontate). Si ottengono in questo modo alcuni nuovi risultati di stabilità delle soluzioni in questo modello.

---

---

The research of the author was partially supported by a grant from the «Ministero della Pubblica Istruzione». A first version of this paper was delivered at the «VIII Convegno A.M.A. S.E.S., Modena 26–29 September 1984». In am grateful to E. Castagnoli and P. Mazzoleni for helpful suggestions. For any remaining errors, I am entirely responsible.     

On exact D-optimal designs with 2 two-level factors and n autocorrelated observations

In this paper we consider the exact D-optimal designs for estimation of the unknown parameters in the two factors, each at only two-level, main effects model with autocorrelated errors. The vector of the n random errors in the observed responses is assumed to follow a first-order autoregressive model (AR(1)). The exact D-optimal designs seek the optimal combinations of the design levels as well as the optimal run orders, so that the determinant of the information matrix of BLUEs for the unknown parameters is maximized. Bora-Senta and Moyssiadis (1999) gave some conjectures about the exact D-optimal designs based on their experience of several exhaustive searches. In this paper their conjectures are partially proved to be true.Received: January 2003 / Accepted: October 2003Partially supported by the National Science Council of Taiwan, R.O.C. under grant NSC 91-2115-M-008-013.Supported in part by the National Science Council of Taiwan, R.O.C. under grant NSC 89-2118-M-110-003.     

A reexamination of additivity of power in randomized social preference

We show that Barberá and Sonnenscheins (1978) power function becomes additive if we replace the Paretian condition by nonimposition and monotonicity. Since these conditions are very much in the spirit of Arrow (1951), our result sharpens the analogy to Arrows theorem.Received: 29 November 2001, Accepted: 4 May 2003, JEL Classification: D71Shasikanta Nandeibam: I would like to thank Professor Prasanta Pattanaik for very helpful discussions. I have also benefitted from the comments and suggestions of an associate editor. The usual disclaimer applies.     

A remark on the recursive estimation of ARMA order

Summary It is well known how, for an ARMA process of order (p ₀,q ₀), max (p ₀,q ₀) may be recursively estimatedHannan/Rissanen. Assuming max (p ₀,q ₀) to be known and, in addition,p ₀q ₀, a simple procedure for the recursive estimation of (p ₀,q ₀) is presented.     

Characterizations of distributions via linearity of regression of generalized order statistics

Let X (r, n, m, k), 1 r n, denote generalized order statistics based on an absolutely continuous distribution function F. We characterize all distribution functions F for which the following linearity of regression holds E(X(r+l,n,m,k) | X(r,n,m,k))=aX(r,n,m,k)+b.We show that only exponential, Pareto and power distributions satisfy this equation. Using this result one can obtain characterizations of exponential, Pareto and power distributions in terms of sequential order statistics, Pfeifers records and progressive type II censored order statistics. Received July 2001/Revised August 2002     

Who consults investment analysts?

This paper shows that the questions: Who consults investment analysts? and Who goes to the El Farol bar in Santa Fe? are similar. Thus, investors use a mixed optimal strategy. The demand for consulting services is also characterized.Received: 25 September 2002, Accepted: 26 May 2003, JEL Classification: D81, C70Jacob Paroush: The author is grateful to Yaw Nyarko, Yigal Milchtaich and an anonymous referee for helpful comments and suggestions.     

Über die Existenz überall trennscharfer Tests im nicht-dominierten Fall

Einleitung Es sei (,A) ein Meßraum undP eine Familie von Wahrscheinlichkeitsmaßen auf . IstP durch ein -endliches Maß dominiert, so ist nachPfanzagl [1960] für die Existenz eines überall trennscharfen Tests zu jedem Niveau für die HypotheseP=P _o (P, P _o P) gegen die AlternativePP ₀ notwendig und hinreichend, daßP/{P ₀} isotonen Likelihood-Quotienten bzgl.P ₀ besitzt. Für den Fall, daßP total geordnet und dominiert ist, gilt nach [Pfanzagl, 1963] eine entsprechende Aussage: Genau dann existiert zu jedem Niveau ein überall trennscharfer Test, wennP isotonen Likelihood-Quotienten besitzt.Die vorliegende Arbeit zeigt, daß auf die Annahme der Dominiertheit verzichtet werden kann, und liefert darüber hinaus einen einheitlichen Beweis für die beiden oben zitierten Sätze vonPfanzagl.

---

Summary Two theorems ofPfanzagl [1960, 1963] about necessary and sufficient conditions for the existence of uniformly most powerful tests are generalized to the undominated case. Moreover a unified proof for the two theorems ofPfanzagl is given.

     

Der diskontierte Einarmige Bandit

Summary A sequential stochastic process is observed in discrete time intervals; the process is generated by Bernoulli trials with unknown meanp. Given an a-priori distribution forp, regarded as a random variable, and discounting future payoffs with a factor , O<<1, optimal and suboptimal stopping rules (depending on ) are constructed. this leads to the connection of the process under consideration with another process of the same structure, but with known meanP _o for the Bernoulli trials, thus finally resulting in an One-Armed-Bandit problem.     

Scharfe untere und obere Schranken für die Wahrscheinlichkeit der Realisation vonk untern Ereignissen

Summary LetA ₁,...,A _n be events in a probability space (,A,W). We denote byL _k the event, that at leastk events among then eventsA ₁,...A _n occur, and byK _k the event, that exactlyk events occur. If only the inequalities _i W(A _i ) _i ,i=1,...,n, are known, we calculate sharp lower and upper bounds forW(L _k ) andW(K _k ). These bounds only depend onn, k and _i , _i ,i=1,...,n. They are relevant, when treating combined tests or confidence procedures.     

An alternative direct proof of Gibbard’s random dictatorship theorem

We present an alternative proof of the Gibbards random dictatorship theorem with ex post Pareto optimality. Gibbard(1977) showed that when the number of alternatives is finite and larger than two, and individual preferences are linear (strict), a strategy-proof decision scheme (a probabilistic analogue of a social choice function or a voting rule) is a convex combination of decision schemes which are, in his terms, either unilateral or duple. As a corollary of this theorem (credited to H. Sonnenschein) he showed that a decision scheme which is strategy-proof and satisfies ex post Pareto optimality is randomly dictatorial. We call this corollary the Gibbards random dictatorship theorem. We present a proof of this theorem which is direct and follows closely the original Gibbards approach. Focusing attention to the case with ex post Pareto optimality our proof is more simple and intuitive than the original Gibbards proof.Received: 15 October 2001, Accepted: 23 May 2003, JEL Classification: D71, D72Yasuhito Tanaka: The author is grateful to an anonymous referee and the Associate editor of this journal for very helpful comments and suggestions. And this research has been supported by a grant from the Zengin Foundation for Studies on Economics and Finance in Japan.     

Charakterisierung der zweiseitigen Exponentialverteilung

LetF () be a family of distribution functions with a translation parameter such thatF (0) has a densityf. It is well known that each sample median is a maximum likelihood estimate of , iff belongs to the classE of all bilateral exponential densities which are symmetric about 0. Here it is shown that, conversely,fE holds, either if there is an evenm such that for every sample of sizem each median is an MLE of , or if there is an infinite setM such that for every sample of any sizemM at least one median is an MLE of .     

Goodness-of-fit test with nuisance regression and scale

In the linear model Y _i = x _i + e _i, i=1,,n, with unknown (, ), {\open R}^p, >0, and with i.i.d. errors e ₁,,e _n having a continuous distribution F, we test for the goodness-of-fit hypothesis H ₀:F(e)F ₀(e/), for a specified symmetric distribution F ₀, not necessarily normal. Even the finite sample null distribution of the proposed test criterion is independent of unknown (,), and the asymptotic null distribution is normal, as well as the distribution under local (contiguous) alternatives. The proposed tests are consistent against a general class of (nonparametric) alternatives, including the case of F having heavier (or lighter) tails than F ₀. A simulation study illustrates a good performance of the tests. Received July 2001     

Minimax estimators andΓ-minimax estimators for a bounded normal mean under the lossl p (θ, d)=|θ-d|p

Summary Let the random variableX be normal distributed with known varianceσ ²>0. It is supposed that the unknown meanθ is an element of a bounded intervalΘ. The problem of estimatingθ under the loss functionl _p (θ, d)=|θ-d| ^p p≥2 is considered. In case the length of the intervalθ is sufficiently small the minimax estimator and theΓ(β, τ)-minimax estimator, whereΓ(β, τ) represents special vague prior information, are given.