Parametric links for binary choice models: A Fisherian–Bayesian colloquy

The familiar logit and probit models provide convenient settings for many binary response applications, but a larger class of link functions may be occasionally desirable. Two parametric families of link functions are investigated: the Gosset link based on the Student t latent variable model with the degrees of freedom parameter controlling the tail behavior, and the Pregibon link based on the (generalized) Tukey λ family, with two shape parameters controlling skewness and tail behavior. Both Bayesian and maximum likelihood methods for estimation and inference are explored, compared and contrasted. In applications, like the propensity score matching problem discussed below, where it is critical to have accurate estimates of the conditional probabilities, we find that misspecification of the link function can create serious bias. Bayesian point estimation via MCMC performs quite competitively with MLE methods; however nominal coverage of Bayes credible regions is somewhat more problematic.     

Geographical concentration and the dynamics of countries' specialization in technologies

This paper examines the empirical dynamics of countries' technological specialization in six technology fields using distribution dynamics. In all technology fields innovation activities are performed by relatively few countries and the degree of concentration is fairly stable in time. Intra-distribution dynamics is characterized by persistence of within field countries' specialization levels around or below the mean, while high specialization levels revert towards lower values. This strengthens the case for absorptive capacity. Electronics show some distinctive properties: they have the highest degree of geographical concentration and numerous small countries among those specialized; they also are the least mobile technology field. In a Schumpeterian perspective, this is in line with "creative accumulation".     

Cost-effectiveness analysis of disease modifiying drugs (interferons and glatiramer acetate) as first line treatments in remitting-relapsing multiple sclerosis patients

Abstract

Objective:

The aim of this study was to assess cost-effectiveness of the different Disease Modifying Drugs (DMD) used as first-line treatments (interferons IM IFNβ-1a, SC IFNβ-1a, SC IFNβ-1b, and glatiramer acetate, GA) in Remitting-Relapsing Multiple Sclerosis (RRMS) in Spain.

Methods:

A Markov model was developed to simulate the progression of a cohort of patients with RRMS, during a period of 10 years. Seven health states, defined by the Expanded Disability Status Scale (EDSS), were considered in the model. Patients with an EDSS score less than 6.0 were assumed to be treated with one of the DMD. In addition, all patients were assumed to receive symptomatic treatment. The monthly transition probabilities of the model were obtained from the literature. The analysis was performed from the societal perspective, in which both direct and indirect (losses in productivity) healthcare costs (€, 2010) were included. A discount rate of 3% was applied to both costs and efficacy results.

Results:

GA was the less costly strategy (€322,510), followed by IM IFNβ-1a (€329,595), SC IFNβ-1b (€ 333,925), and SC IFNβ-1a (€348,208). IM IFNβ-1a has shown the best efficacy results, with 4.176 quality-adjusted life years (QALY), followed by SC IFNβ-1a (4.158 QALY), SC IFNβ-1b (4.157 QALY), and GA (4.117 QALY). Incremental costs per QALY gained with IM IFNβ-1a were €?1,005,194/QALY, €?223,397/QALY, and €117,914/QALY in comparison to SC IFNβ-1a, SC IFNβ-1b, and GA, respectively.

Conclusions:

First-line treatment with GA is the less costly strategy for the treatment of patients with RRMS. Treatment with IM IFNβ-1a is a dominant strategy (lower cost and higher QALY) compared with SC IFNβ-1a and SC IFNβ-1b. However, IM IFNβ-1a is not a cost-effective strategy vs GA, because incremental cost per QALY gained with IM IFNβ-1a exceeds the €30,000 per QALY threshold commonly used in Spain.

Limitations:

The highly-restrictive inclusion criteria of clinical trials limits generalization of the results on efficacy to all patients with multiple sclerosis. Availability of data for head-to-head comparisons is associated with the use of information from clinical trials.     

Cost utility of allogeneic stem cell transplantation with matched unrelated donor versus treatment with imatinib for adult patients with newly diagnosed chronic myeloid leukaemia

Objective: To estimate, from the perspective of the German statutory health insurance, the cost utility of allogeneic stem cell transplantation with matched unrelated donor (MUD-SCT) in newly diagnosed, chronic-phase chronic myeloid leukaemia (CML) patients aged 40 years or younger, relative to the treatment with imatinib.

Methods: The incremental cost-effectiveness ratio (ICER) of the additional cost of imatinib versus MUD-SCT per quality-adjusted life year (QALY) gained was chosen as a target assessment. ICER was quantified using a Markov cohort modelling approach. The evaluation encompassed 5 years of treatment with either approach, and only direct medical costs (in €, year 2005) were considered.

Results: There were incremental costs of €77,410 for imatinib therapy per QALY gained versus MUD-SCT. No strategy was clearly dominant; on average, during 5 years, cost savings of €63,433 were obtained and 0.82 QALY lost by SCT compared to treatment with imatinib. QALYs gained in CML patients with either treatment resulted in considerable cost to the third-party payer in Germany. The results were particularly sensitive to the price of imatinib.

Conclusions: The analysis finds that imatinib is more costly but more effective (as measured in QALYs) over a 5-year time horizon. The resulting ICER of €77,410 per QALY is higher than commonly cited thresholds. The cost utility of MUD-SCT to treat CML in patients with a European Group for Blood and Marrow Transplantation score ≤ to 2 compares with that of the imatinib strategy.     

Cost-effectiveness achieved through changing the composition of renal replacement therapy in Japan

Abstract

Objective:

The cost-effectiveness of renal replacement therapy (RRT) is affected by the composition of treatment. This study aimed to estimate the costs and outcomes associated with changing the composition of RRT modality over time.

Methods:

By using clinical and cost data from a systematic review, a Markov model was developed to assess the costs and benefits of the four main treatments available for RRT in Japan. The model included direct health service costs and quality-adjusted life years (QALY). Sensitivity analyses were performed to assess the robustness of the results.

Results:

Over the 15-year period of the model, the current composition of RRT (i.e., the base composition of RRT) was $84,008/QALY. The most cost-effective treatment was when the likelihood of a living donor transplant was increased by 2.4-times ($70,581/QALY). Compared with the base composition of RRT, dominant treatments with respect to cost-effectiveness were when the likelihood of a deceased donor transplant was increased by 22-times and when the likelihood of a pre-emptive living donor transplant was increased by 2.4-times. Little difference was found between these two treatments. One-way sensitivity analysis did not change the cost effectiveness except for costs of chronic hemodialysis and a living donor transplant in subsequent years.

Limitations:

It is difficult to increase the rate of transplant overall in the shorter term nationally and internationally.

Conclusions:

Appropriate distribution of all transplant options and hemodialysis is necessary to achieve the most cost-effective solution.     

Emergence of macro-variables by evaluation and clustering of micro- activities

Recent finance and economic forecasting and risk calculation failures made obvious that macro-modelling without micro-foundation may be treacherous. Reliable macro-modelling requires the consistent bundling of individual actions into intermediate and macro-variables exploiting the individual actions’ coordination and its dynamics. The degree of coordination may range from chaos – absence of coordination – to determined situations caused by macro-level equilibrium dictating any agent's actions and inhibiting interactions. Coordination clusters individual actions into real decision units such as companies, political parties and unions. It structures the emergent intermediate and macro-level situations vitally.The paper presents first a centennial history of prominent scholars’ quotes questioning the equilibrium paradigm, a short survey of prevailing paradigm's deficiencies laid bare once again by the latest financial crises.It proposes second discrete choice (DC) – successfully applied in different fields – to model the individual agent's decision. DCs innovative integration into a Markov process provides a steady foundation to model interactions of individual agents consistently.The final section justifies the actions’ proposed interactive bundling by referring to recent advances in data processing and network topology. The dynamic modelling of the actions’ and interactions’ coordination breaks fresh grounds both with regards to mathematical, computational and economic modeling requirements. The combination of latest developments in data processing like Big Data and the recently (re)discovered network topology capabilities may cope with these challenges.     

Positive return premia in Japan

This paper examines Jensen's [J. Finance, 1968, 23, 389–416] alphas and the time-varying return premia unexplained by standard risk factors in Japan and presents several new findings. First, in contrast to the US experience, positive alphas remain after Fama and French's three factors are applied to excess stock returns in Japan. Second, positive alphas remain in Japan, even if the Fama–French three factors combined with momentum and reversal factors are applied to excess stock returns. Third, the positive return premia unexplained by these five factors bear little relation to the dynamics of the Japanese macroeconomy. Fourth, the time series evolution of the positive return premia indicates autonomous dynamics with at least three regimes. Fifth, we can predict or time the acquisition of the positive return premia for small-size portfolios in Japan by observing the direction and effect of the return premia of large-size portfolios and high-book equity to market equity (BE/ME) portfolios. Finally, application of the self-exciting threshold autoregressive (SETAR) model shows that the size effects are stronger than the BE/ME effects in Japan, given that the return premia from small-size portfolios in the SETAR model are bounded by positive thresholds, while the return premia from high-BE/ME portfolios are bounded by negative thresholds.     

Some results on the compound Markov binomial model

Abstract

Some years ago, in the course of an analysis of upper and lower limits for incomplete moments of statistical distributions I established an elementary summation formula¹ which proved rather useful for the purpose I had in view. Subsequently the formula was generalized by professor Steffensen, who showed² that the formula in question could be looked upon as giving the first term of an expansion in a certain type of series. Professor Steffensen established recurrence formulae for the coefficients of the series and computed the second, third and fourth term and the corresponding remainders¹, but did not arrive at a general, explicite expression for the coefficient of the n-th term and the corresponding remainder. A year later I found these expressions accidentally while I was working on some other problem. I also discovered the real nature of the procedure in question which proved to be a certain kind of least square fitted polynomial approximation. I did not, however, at the time publish the result. Taking the question up again later I found that the whole problem could be considerably generalized. The type of generalization in question is analogous to the generalization from polynomials to arbitrary functions.     

Ruin probabilities in models with a Markov chain dependence structure

In this paper we derive explicit expressions for the probability of ruin in a renewal risk model with dependence among the increments (Z _k ) _k>0. We study the case where the dependence structure among (Z _k ) _k>0 is driven by a Markov chain with a transition kernel that can be described via ordinary differential equations with constant coefficients.     

Stochastic better-reply dynamics in finite games

In this paper, we analyze a model where individuals from finite populations are repeatedly drawn to play a finite game and in every period choose a weakly better reply to a sample distribution from a finite history of past play. For all finite games and sufficiently incomplete information, we prove convergence to minimal sets closed under better replies. This result complements previous findings in a deterministic continuous-time framework and implies convergence to strict Nash equilibria in many well-known classes of games.