Speaker:

Professor Dan Lovallo

Speaker:

Associate Professor Richard Potter, Department of Information and Decision Sciences, The University of Illinois at Chicago

Title:

Adventures in Interdisciplinary Business Research: Solo Walkabout and Strategic Campaigns

Abstract:

Interdisciplinary research can take many forms and involve varying quantities of people, from the solo practitioner to teams of complimentary experts spanning specializations within a limited area (e.g., a business school) to teams that have expertise drawn from disparate fields (e.g., medicine, economics, sociology). The latter can often by considered strategic campaigns, typically formed to address complex problems not easily understood or remedied from a single disciplinary perspective. Dr. Richard Potter will provide an overview of his interdisciplinary research as a solo practitioner, spanning behavioral decision theory, cognitive psychology, cross-cultural management, organizational development, leadership, health care management and policy, and information technology development and management. The discussion will provide highlights of two studies involving the application of cognitive psychology theory and decision theory to information technology design and use, and organizational development theory and methodology applied to virtual team assessment and management. Potter will provide commentary on the challenges of interdisciplinary research, with respect to publication, career management, and the advancement of knowledge. Management and challenges of interdisciplinary strategic campaigns will also be discussed.

Speaker:

Professor Chihwa Kao, Centre for Policy Research and Department of Economics, Syracuse University, Syracuse, NY

Title:

Panel Cointegration with Global Stochastic Trends

Abstract:

This paper studies estimation of panel cointegration models with cross-sectional dependence generated by unobserved global stochastic trends. The standard least squares estimator is, in general, inconsistent owing to the spuriousness induced by the unobservable (1) trends. We propose two iterative procedures that jointly estimate the slope parameters and the stochastic trends. The resulting estimators are referred to respectively as CupBC (continuously-updated and bias-corrected) and the CupFM (continuously-updated and fully-modi.ed) estimators. We establish their consistency and derive their limiting distributions. Both are asymptotically unbiased and asymptotically normal and permit inference to be conducted using standard test statistics. The estimators are also valid when there are mixed stationary and non-stationary factors, as well as when the factors are all stationary.

Speaker:

Associate Professor Anurag Banerjee

Title:

One Step Towards a Less Sensitive Approach

Abstract:

Econometric estimates or decisions are applicable if they are not sensitive to small changes of 'nuisance' parameters. The statistic which measures these violations is called a sensitivity measure. In this article, we give a sufficient condition for a statistic to be sensitive, leading to the definition of the sensitivity measure. The measure suggests when the sensitivity statistic is 'small', we need not bother about the nuisance parameters. The main contribution of this article is to propose a correction factor for the statistic of interest if we find 'large' sensitivity. We define such an estimator and analyse its properties. As an application we analyse the sensitivity of the slope and variance estimates of the linear model with respect to the presence of associated nuisance parameters in the variance covariance matrix.

Speaker:

Professor Geert Dhaene, Department of Economics, Katholieke Universiteit Leuve

Title:

Split-panel jackknife estimation of fixed effects models

Abstract:

Split-panel jackknife estimators are proposed for reducing the bias of the maximum likelihood estimator (MLE) of dynamic panel data models with fixed effects. The bias is reduced from O(T−1) to O(T−2) or smaller, where T is the number of periods observed. The split-panel jackknife combines the MLE computed from the full panel, θ, with the MLEs computed from shorter subpanels. For example, the half-panel jackknife uses θ and the MLEs corresponding to two non-overlapping half-panels, each using T/2 observations and all N cross-sections units. The half-panel jackknife estimator has bias O(T−2). The bias is further reduced to O(T−3) (or smaller) if two (or more) partitions of the panel are used, for example two half-panels and three 1/3-panels, and the MLEs corresponding to the subpanels. The asymptotic distribution of the jackknife estimators is normal, correctly centered at the true value, has variance equal to that of the MLE, and allows T to grow only slowly with N. The split-panel jackknife can also be employed to correct the profile likelihood function to any order. Maximising the jackknife-corrected likelihood yields estimators with essentially the same properties as the jackknife-corrected MLE. The large N, fixed T asymptotic variance of the split-panel jackknife estimators can be estimated consistently by the bootstrap or by the delete-one jackknife in the cross-section dimension. Simulation results for the probit and logit binary AR(1) models and for the linear AR(1) model show that even in small, short panels such as N = 25 and T = 9, the split-panel jackknife is very effective in reducing the bias of the MLE, has smaller mean squared error, and yields confidence intervals with much better coverage.

Speaker:

Corrado Di Guilmi, Department of Economics, Universit`a Politecnica delle Marche

Title:

Financial Fragility, Mean-field Interaction and Macroeconomic Dynamics: A Stochastic Model

Abstract:

The links between aggregate financial indicators and business fluctuations have been widely addressed in literature while the same interest has not been devoted to the role of microeconomic financial variables in determining macroeconomic results. One of the causes may be individuated in the lack of suitable analytical tools. Firms are different each other as regards, at least, financial structure and size. Therefore, their responses to shocks come out to be different and asymmetric. Moreover, firms are reciprocally linked, and their diverse reactions influence the whole system at micro, macro and meso level. The uncertainty about the final outcome is amplified by feedback effects from aggregate level to firms. In this work we model an economic system populated by heterogeneous firms that, reacting to stochastic shocks to maximize their profit, modify the ratio among liabilities and equities. These variations influence the financial environment and the demography of firms population, endogenously generating business fluctuations. Using a stochastic aggregation method we define a system of coupled dynamic equations that describe long run path and cycles of aggregate output.

Speaker:

Neville Weber, School of Mathematics and Statistics, University of Sydney

Title:

Saddlepoint Approximations and some Applications

Abstract:

Saddlepoint approximations are powerful tools for estimating tail probabilities. They have found application in many different contexts. This seminar will give an overview of the background to saddlepoint approximations and, in particular, look at the formula developed by Lugannani and Rice (Advances in Applied Probability (1980), 12, 475-490). We will then illustrate the use of this approximation by considering two applications. The first deals with finding p-values for a test for serial correlation in panel time series data and the second involves estimating tail probabilities for heavy tailed distributions. Cox and Solomon (Biometrika (1988), 75, 145-148) developed a statistic for testing serial correlation in large numbers of small samples from multivariate normal populations. They established that their statistic has an asymptotic normal distribution. The Lugannani-Rice formula provides improved approximations for the p-values of their statistic. Further we indicate how the ideas can be extended to develop tests when the normal assumption is relaxed. The saddlepoint approximation technique uses the cumulant generating function of the underlying distribution and so is not a procedure that one would normally consider when dealing with heavy tailed data where higher moments are not necessarily finite. We will explain how the Lugannani- Rice formula can still be of benefit in the heavy tailed senario by utilizing the empirical moment generating function.

Speaker:

Andrey Vasnev, Discipline of Econometrics and Business Statistics, University of Sydney

Title:

Locan Sensitivity and Diagnostic Tests

Abstract:

In this paper, we confront sensitivity analysis with diagnostic testing. Every model is misspecified (in the sense that no model coincides with the data-generating process), but a model is useful if the parameters of interest (the focus) are not sensitive to small perturbations in the underlying assumptions. The study of the effect of these violations on the focus is called sensitivity analysis. Diagnostic testing, on the other hand, attempts to find out whether a nuisance parameter is (statistically) 'large' or 'small'. Both aspects are important, but traditional applied econometrics tends to use only diagnostics and forget about sensitivity analysis. We develop a theory of sensitivity in a maximum likelihood framework, give conditions under which the diagnostic and the sensitivity are asymptotically independent, and demonstrate with three core examples that this independence is the rule rather than the exception, thus underlying the importance of sensitivity analysis.

Speaker:

Felix Chan, School of Economics and Finance, Curtin University of Technology

Title:

Modelling Time-Varying Higher Moments with Maximum Entropy Density

Abstract:

Since the introduction of the Autoregressive Conditional Heteroscedasticity (ARCH) model, the literature of modelling the conditional second moment has become increasingly popular in the last two decades. Many extensions and alternate models of the original ARCH have been proposed in the literature aiming to capture the dynamics of volatility more accurately. Interestingly, the Quasi Maximum Likelihood Estimator (QMLE) with normal density is typically used to estimate the parameters in these models. As such, the higher moments of the underlying distribution are assumed to be the same as the normal distribution. However, various studies reveal that the higher moments, such as skewness and kurtosis of the distribution of financial returns are not likely to be the same as the normal distribution, and in some cases, they are not even constant over time. This has significant implications in risk management, especially in the calculation of Value-at-Risk (VaR), which focuses on the negative quantile of the return distribution. Failure to accurately capture the shape of the negative quantile would produce inaccurate measures of risk, and subsequently lead to misleading decision in risk management. This paper proposes a general framework to model the distribution of financial returns using Maximum Entropy Density (MED). The main advantage of MED is that it provides a general framework to estimate the distribution function directly based on a given set of data, and it provides a convenient framework to model higher order moments up to any arbitrary finite order k. However this flexibility comes with a high cost in computational time as k increases, therefore this paper proposes an alternative model that would reduce computation time substantially. Moreover, the sensitivity of the parameters in the MED with respect to the dynamic changes of moments is derived analytically. This result is important as it relates the dynamic structure of the moments to the parameters in the MED. The usefulness of this approach will be demonstrated using 5 minutes intra-daily returns of the Euro/USD exchange rate.

Speaker:

Moshe Sniedovich, Department of Mathematics and Statistics, The University of Melbourne

Title:

Responsible Decision-Making in the Face of Severe Uncertainty

Abstract:

Severe uncertainty is an ever-present phenomenon that lurks in the background, and yet we manage to carry on with our daily life without giving it too much thought. Indeed, we do not allow it to paralyze us, perhaps because there is not much that we can do about it. However, what about the experts? What do the experts say? For obvious reasons the paradigms offered by classical decision theory for tackling severe uncertainty are extremely simple, so much so that prima facie they seem to be naive. This is a reflection of the fact that conditions of severe uncertainty leaves one precious little to work with.

We must therefore expect models designed for decision-making under severe uncertainty to be austere. Consequently, the state of the art is such that no fully satisfactory recipes exist for dealing with problems involving decision-making under severe uncertainty. In this presentation we examine the conceptual, methodological and practical aspects of this important area of decision theory. The discussion is based on the speaker's interaction over the past four years with academics, practitioners and research centers in Australia regarding methods available for dealing with practical problems in the areas of conservation biology, homeland security, financial analysis, and others.

Speaker

Dick van Dijk, Econometric Institute, Erasmus University Rotterdam

Title:

Getting the Most Out of Macroeconomic Information for Predicting Stock Returns and Volatility

Abstract:

This paper considers the application of approximate dynamic factors for predicting stock returns and volatility. This approach enables us to incorporate a large amount of macroeconomic information into the forecasting models, while the model size remains reasonable. Our findings demonstrate that factor-based approaches provide substantial gains when predicting monthly S&P500 excess returns and volatility, compared to benchmark models that use a small number of specific financial and macroeconomic variables. At the same time we find that pre-selecting macroeconomic variables is of crucial importance for the performance of the factor models. A striking result is that the performance of this procedure does not deteriorate during the 1990s, which typically is thought of as a period of declining predictability. We evaluate the economic value of incorporating a large amount of macroeconomic information by simulating a mean-variance investor who uses return and volatility forecasts to determine optimal portfolio weights. We find that the investor would be willing to pay several hundreds of basis points per annum to switch from passive and dynamic strategies based on benchmark models to the dynamic strategy that employs the factor-based approach.

Speaker:

Alan Woodland, Discipline of Econometrics and Business Statistics, University of Sydney

Title:

Steepest Ascent Tariff Reforms

Abstract:

This paper introduces the concept of a steepest ascent tariff reform for a small open economy. By construction, it is locally optimal in that it yields the highest gain in utility of any feasible tariff reform vector. Accordingly, it provides a convenient benchmark for the evaluation of the welfare effectiveness of other well known tariff reform rules, such as the proportional and the concertina rules. We develop the properties of this tariff reform in detail and provide geometric illustrations of our method. Overall, the paper's contribution lies in developing a theoretical concept where the focus is upon the size of welfare gains accruing from tariff reforms rather than simply with the direction of welfare effects that has been the concern of the existing literature.

Speaker:

Giovanni Forchini, Department of Econometrics and Business Statistics, Monash University

Title:

The effect of partial identification of the asymptotic distributions of estimators and test statistics

Abstract:

In a partially identified structural equation model only some of the structural parameters are identified. Phillips (1989) and Choi and Phillips (1992) have shown that the distribution of the TSLS estimator of the identified parameters has a nonstandard asymptotic distribution that is a mean- and covariance-matrix-mixed normal and that the TSLS estimator of the unidentified parameters has nondegenerate asymptotic distribution. This talk reviews and extends some of these results to include the LIML and some other estimators. The asymptotic distributions of the Sargan/Byron/Wegge and Basmann overidentification test statistics under partial identification are also discussed.

Speaker:

Eddie Anderson, Discipline of Econometrics and Business Statistics, University of Sydney

Using supply function models for electricity markets

Abstract:

In wholesale electricity markets generators offer a schedule of prices and quantities, rather than competing on price or quantity alone. We say that generators offer a supply function: if demand is high then generators will be paid a higher price for the power they generate.

The talk will be in two parts. First we will consider the optimization problem faced by the generator in deciding on an optimal supply function to bid. We assume that both the demand for electricity and the behaviour of competing generators is unknown, but can be represented by a probability distribution. Optimality conditions are derived and illustrated on some examples.

In the second part of the talk we analyse the form of equilibria that can occur in a market of this sort, and also show how these equilibria can be effectively calculated numerically. Previous analyses have been restricted to the symmetric case where each firm is the same, but we are interested in asymmetric equilibria.

The talk will give an overview of the work carried out in a series of papers that have appeared since 2002, together with some recent work that has not yet been published. Our aim is to demonstrate the usefulness of the supply function approach; so that it can be considered as a genuine alternative to the more tractable Cournot model in which firms can only choose their output quantities.

Speaker:

George Steiner, DeGroote School of Business, McMaster University

Scheduling and the Travelling Salesman Problem on Permuted Monge Matrices

Abstract:

A large variety of scheduling problems has recently been shown to be solvable as a special case of the Travelling Salesman Problem (TSP) on permuted Monge matrices. Although the TSP on permuted Monge matrices is known to be strongly NP-hard, we present polynomial-time solutions for many of the special cases generated by the scheduling problems. We also show that a simple subtour-patching heuristic is asymptotically optimal for the TSP on permuted Monge matrices under some mild technical conditions.

Speaker:

Donggyu Sul, Department of Economics, University of Auckland

Estimating and Testing Idiosyncratic Equations Using Cross-Section Dependent Panel Data: Application to the Feldstein-Horioka Puzzle

Abstract:

Econometricians must often control for unobservable factors when estimating equations using cross-section dependent panel data. Based on a common factor structure of cross section dependence, we suggest a new method for estimating idiosyncratic equations and demonstrate that the proposed method works under more general conditions than those required by existing estimators, such as Pesaran?s (2006) common correlated effects estimator and Bai?s (2005) iterated principal component least-squares estimator. A Monte-Carlo study shows that the proposed estimation method works well in the finite sample. Applying the method to the Feldstein-Horioka puzzle, we find that the correlation between the log investment ratio and the log saving ratio is insignificant for a panel of 29 developed countries over the 1980 to 2004 period.

Speaker:

Abstract:

Speaker:

Mike So, Department of Information and Systems Management, The Hong Kong University of Science & Technology

A threshold factor multivariate stochastic volatility model

Speaker:

Pavel Chigansky, Department of Statistics, The Hebrew University

Martingale convergence and stability of the nonlinear filtering equation in Hidden Markov Models

Abstract:

Speaker:

Boris Choy, Discipline of Econometrics and Business Statistics, Faculty of Economics and Business, The University of Sydney

Bayesian Student-t Stochastic Volatility Models via Scale Mixtures

Abstract:

Speaker:

Jacek Krawczyk, School of Economics and Finance, Victoria University of Wellington

The invisible polluter: Can regulators save consumer surplus?

Abstract:

Consider an electricity market populated by competitive agents using thermal generating units. Such generation involves the emission of pollutants, on which a regulator might impose constraints. Transmission capacities for sending energy may naturally be restricted by the grid facilities. Both pollution standards and transmission capacities can impose several constraints upon the joint strategy space of the agents. We propose a coupled constraints equilibrium as a solution to the regulator?s problem of avoiding both congestion and excessive pollution. Using the coupled constraints? Lagrange multipliers as taxation coefficients the regulator can compel the agents to obey the multiple constraints. However, for this modification of the players? payoffs to induce the required behaviour a coupled constraints equilibrium needs to exist and must also be unique. A three-node market example with a dcmodel of the transmission line constraints described in [8] and [2] possesses these properties. We extend it here to utilise a two-period load duration curve and, in result, obtain a two-period game. The implications of the game solutions obtained for several weights, which the regulator can use to vary the level of generators? responsibilities for the constraints? satisfaction, for consumer and producer surpluses will be discussed.

Speaker:

Hiroaki Suenaga, School of Economics and Finance, Curtin University of Technology

Volatility dynamics of NYMEX energy complex: seasonality and its implications for crack spread hedging

Abstract:

We examine the volatility dynamics of three petroleum commodities traded on the NYMEX; crude oil, unleaded gasoline, and heating oil. We extend the partially overlapping time-series (POTS) model of Smith (2005, Journal of Applied Econometrics) into three commodity, six factor setting and decompose daily returns on futures contracts into common factors affecting prices of simultaneously traded contracts and contract-specific terms. Factor loadings and variance of contract‐specific terms are specified by flexible non‐parametric functions so that the model accommodates nonlinear volatility dynamics resulting from peculiarities of the commodities. The model reveals several important features about the volatility and co-variability of three commodity prices: (1) three commodities exhibit time-to-maturity effects, (2) volatility exhibits substantial seasonality for two refined commodities?it peaks in late summer for unleaded gasoline and in winter for heating oil, (3) the identified long-term factors exhibit high cross-market correlation whereas the short-term factors exhibit only moderate correlation, (4) for both factors, cross-commodity correlation is highly persistent yet exhibits substantial variation over the period between 1984 and 2006, and (5) the conditional variance of latent factors is highly persistent and correlated across three markets.

The depicted volatility dynamics implies that a trader in need of hedging spot price risk should cross hedge with futures contracts of at least three months to maturity to avoid high contract-specific volatility of contracts closer to maturity. More distant contract is used when spot price is subject to less short-term shocks. Contract near maturity is used when spot price is exposed to greater short-term shocks. The suggested strategy, on average, reduces the variance of portfolio returns by 8 to 14 percent, relative to the strategy utilizing the second-position contracts. It reduces only moderately the variance of returns to crack spreads, however.

Speaker:

David Johnstone, Discipline of Accounting, Faculty of Economics and Business, The University of Sydney

Economic Interpretation of Probabilities Estimated by Maximum Likelihood

Abstract:

From the perspective of a decision maker with log utility, maximum likelihood estimation [MLE] yields the best set of probabilities available from the forecaster?s model, given the data in use. Of all the possible probability estimates that the chosen model might have generated, the MLE estimates are the probabilities that would have produced maximum financial return to a log utility investor, had they been known and acted upon before the events in question (rather than being estimated afterwards). Decision makers with other utility functions may not be so well accounted for. A bootstrap experiment based on a representative set of corporate bankruptcy data suggests that there is something to be gained by customizing the estimation criterion to suit the user?s risk aversion level. The principle of MLE is widely justified by the proven asymptotic frequentist properties of its estimates. That MLE and other abstract statistical estimation criteria can be interpreted as inherently subjective (personal) is not generally understood.

Speaker:

Robert Kohn, School of Economics, UNSW

Regression Density Estimation Using Smooth Adaptive Gaussian Mixtures

Abstract:

We model a regression density flexibly so that at each value of the covariates the density is a mixture of normals with the means, variances and mixture probabilities of the components changing smoothly as a function of the covariates. The model extends existing models in two important ways. First, the components are allowed to be heteroscedastic regressions as the standard model with homoscedastic regressions can give a poor fit to heteroscedastic data, especially when the number of covariates is large. Furthermore, we typically need a lot fewer heteroscedastic components, which makes it easier to interpret the model and speeds up the computation. The second main extension is to introduce a novel variable selection prior into all the components of the model. The variable selection prior acts as a self-adjusting mechanism that prevents overfitting and makes it feasible to fit flexible high-dimensional surfaces. We use Bayesian inference and Markov Chain Monte Carlo methods to estimate the model. Simulated and real examples are used to show that the full generality of our model is required to fit a large class of densities, but also that special cases of the general model are interesting models for economic data.

Speaker:

George Athanasopoulos, Department of Econometrics and Business Statistics, Monash University

VARMA models for macroeconomic modelling and forecasting

Abstract:

I will present an overview of my research on the identification and estimation of VARMA models. I will argue that given the recent advances in VARMA modelling methodology and the improvements in computing power, there is no compelling reason for restricting the class of multivariate models considered for macroeconomic modelling and forecasting to VARs. To support this claim I will present a complete scalar component methodology for identifying and estimating canonical VARMA models and demonstrate its application to a well known multivariate data set. The methodology will then be used in an extensive forecasting competition where I will show that VARMA models forecast macroeconomic variables more accurately than VARs. Through Monte-Carlo simulations I will highlight the key reasons for this important result. Finally if time permits I will briefly present the alternative Echelon form VARMA methodology, its relationship to scalar components, and discuss our work in progress to fully automate this process which I believe is the key for popularising VARMA models.

Speaker:

Professor Philippe Baptiste

Mixed Integer Programming, Branch-and-Cut and Constraint Programming: A comparison of three different approaches to solve the runway sequencing problem

Abstract:

In this talk we compare the efficiency of three different exact optimization techniques on a simple problem coming from Air Traffic Control: When aircraft reach the final descent in the "terminal radar approach control'', a set of disjoint time windows in which the landing is possible, can be automatically assigned to each aircraft. Our goal is to compute landing times (within these time windows) that are spaced as much as possible to avoid wake vortex effect.

We first study the complexity of the problem and we identify some special cases that are polynomially solvable. We then provide a brief overview of discrete optimization techniques such as Mixed Integer Programming (MIP) or Constraint Programming (CP) that are used to model and to solve our problem. Within the CP framework, we rely on a global constraint, the "inter-distance constraint" and we describe the first algorithm that achieves Arc-B-Consistency on this constraint. We also introduce a hybrid branch and cut relying both on MIP and CP.

Speaker:

Professor Wieslaw Kubiak

Just-in-Time Smoothing Through Batching

Abstract:

We present two methods to solve the production smoothing problem in mixed-model just-in-time (JIT) systems with large setup and processing time variability between different models the systems produce. The problem is motivated by production planning at a leading U.S. automotive pressure hose manufacturer. One method finds all Pareto-optimal solutions that minimize total production rate variation of models and work in process, and maximize system utilization and responsiveness. These Pareto-optimal solutions are found efficiently in polynomial time with respect to total demand by an algorithm proposed in the presentation. The other method relies on Daniel Webster?s method of apportionment for production smoothing, which produces periodic, uniform, and reflective production sequences that can improve operations management of the JIT systems. Finally, we present the results of a computational experiment with the two methods.