On the Estimation of the Parameters of ECONOMIC Models: Dealing With

PROFESSOR BERTILLE ANTOINEPROFESSOR Understanding econometrics

Passionate about making sense of the world through mathematics and statistics, Professor Bertille Antoine discusses the importance of her research into economic modelling and how it is shaping the future of econometrics

components: first, my aim is to develop is weak? Or, should she use some complicated relevant economic models that can best alternative method that is always valid, but capture the complexity of the behaviour may lead to conservative inference when the of economic entities such as countries, identification is not that weak? In practice, companies or individuals; second, my goal is the level of identification is always unknown, to develop the associated reliable econometric and choosing the ‘right’ method is not an procedures that can easily be used by applied obvious task. From a practical point of view, it is economists to estimate the parameters of the therefore important to disentangle two classes model. More specifically, I have been looking of identification strength (eg. weak vs not at frameworks where standard models and too weak). methods yield inaccurate results. My current research focuses on two issues that often How is the stability of an economic arise in practice: the estimation of weakly model important? identified models, that is, models where some correlations in the data are so weak that Another issue that has received a lot of attention standard econometric methods fail; and the in econometrics is related to the stability of estimation of time-varying parameters, that the economic relationships described by the is, parameters in the model that are changing chosen model. Indeed, the stability of the over time in unknown ways. economic model (and its associated parameters) What is your background and how did you over time is often questionable, especially for become involved in research on econometrics? Why are user-friendly and reliable macroeconomic models covering extended time econometrics vital to the accuracy of the work periods, but also for microeconomic models. In I grew up in France. After graduating from high of applied economists? many cases, it is important to account for such school, and spending two years in Classe Prepa, instabilities in order to develop an economic I joined ENSAI, a graduate school affiliated Applied economists often design economic model that is relevant to address the issue with the French National Institute for Statistics models to address one of the following goals: at stake. and Economic Studies because I wanted to do either to forecast key variables such as inflation something less theoretical, more orientated or growth rates; or to evaluate the impact of In what ways do you see your research into towards data and real-world questions. This a programme or policy, such as the lunch box economic models developing in the future? is where I discovered economics, and more programmes or the reduction of class sizes on specifically econometrics, which applies students’ performance. The quality of these To develop economic models that can best mathematical and statistical methods to address forecasts and programme evaluations hinges capture the complexity of the behaviour of economic questions. I then spent one year at the critically on the quality of the estimated economic entities, richer datasets are needed. University of Montreal and joined the MSc in econometric model. This is the great advantage offered by panel mathematical and computational finance. After data (also known as longitudinal data), a graduating, I worked as a research associate How will novel tests that you have dataset in which the behaviour of the same under Professor Eric Renault’s supervision, a created aid economists’ everyday work entities is observed across time. The worthiness leader in financial econometrics, at a local developing models? of panel data is now well established in many research center, CIRANO. I really enjoyed my areas of empirical research both in macro and experience and decided I wanted to ‘do more As mentioned, an economic model can only be micro-economics. It has become increasingly research’ in econometrics. A few months later, useful to forecast or evaluate a given policy if available and popular among applied I was starting my PhD in Economics under its parameters are estimated as accurately as researchers because it allows identification and Professor Renault’s supervision. possible. When several estimation methods are estimation of more sophisticated economic available, this also means that the economist models which could not even be considered Could you discuss your current research focus has to choose one of them: should she use when using either time series or cross-sectional and your aims and objectives? the usual one that is easy-to-use and efficient data alone. In the future, I plan to tackle under standard conditions, but may not some of the new issues related to the use of My long-term research interests have two produce reliable estimators if the identification larger datasets.

WWW.INTERNATIONALINNOVATION.COM 83 PROFESSOR BERTILLE ANTOINE

Economic enigmas

Predicting economic trends is challenging. A researcher at Simon Fraser University, Canada, along with collaborators in the USA, France and The Netherlands, is using new statistical and mathematical techniques to tackle these difficulties and increase the reliability of current economic models

EVERYDAY, IMPORTANT DECISIONS are University in the USA, Toulouse School of the data (as done by the Maximum Likelihood made based on information from economic Economics in France and Tilburg University in Estimation), which is often unknown in practice. forecasting models. Whether it is the rate of the Netherlands to alleviate this uncertainty by Such moment conditions naturally appear in inflation or the success of a government scheme, providing more tailored, and subsequently more many economic models as first-order conditions the ability to reliably predict the outcome reliable, models. of an underlying optimisation problem (such as is vital to economic success. Subsequently, Euler equations). “Standard inference methods econometric models are used in a diverse array (such as GMM) may not be reliable when the ESTABLISHING THE CONTRIBUTORS of environments to ensure financial profitability identification [of certain parameters] is too and stability. The complexities of creating a To realise their ambition, the researchers must weak”. Consequently, Antoine and her colleague model are increased by the vast number of redefine some of the fundamentals of forecast Professor Eric Renault have designed a system factors that must be incorporated and the scale model building by designing new methods of whereby this problem can be quantitatively over which it can be applicable, from individuals assessing how effectively existing techniques assessed. This system involves a statistical test to whole countries. function. This task involves constructing unique that gauges the strength of the identification ways of identifying parameters which may have of an economic model. In turn, this determines The purpose of an econometric model is to a weaker, or more uncertain contribution to whether identification is sufficient that GMM hypothetically simulate a real world event so the statistical outcome. This step is commonly may be used in its standard form. This test that the physical outcome can be anticipated neglected in standard procedures that do not uses the rate of convergence of a moment and the necessary actions taken. While common account for the relative influence the parameter condition to assess the quality of the statistical econometric techniques can go part way to may exert on the final model. Professor Bertille input, and subsequently whether the input is make these predictions, some inaccuracy is Antoine from Simon Fraser University highlights sufficiently reliable. prevalent. Consequently, to improve the success the importance of this exercise: “An economic of such a model, two factors must be considered; model can only be useful to forecast or evaluate WHAT TIME? how accurately the model reflects reality, and a given policy if its parameters are estimated as how correctly the individual parameters that accurately as possible”. She elaborates on this The researchers’ understanding of the individual comprise the model are estimated. The latter of further using the example of the Generalized influence of certain parameters has stretched these factors is something which is, historically, Method of Moments (GMM, Hansen 1982), to other areas of econometrics, including the the source of some uncertainty in economic one of the most popular method to estimate influence of the timescale on the stability forecasting. A researcher at the Department parameters in statistical models that relies of economic relationships and subsequent of Economics at Simon Fraser University in on ‘moment conditions’ rather than the full economic modelling. In some economic Canada is working with economists from Brown specification of the distribution function of frameworks, parameters are assumed to be fixed

84 INTERNATIONAL INNOVATION INTELLIGENCE ON THE ESTIMATION OF THE PARAMETERS OF ECONOMIC MODELS: DEALING WITH STABILITY AND IDENTIFICATION ISSUES OBJECTIVES • To develop a new theoretical framework that refl ects more accurately what is observed in the real world • To develop econometric procedures to detect and precisely estimate changes in parameters and in their identifi cation properties

Boldea from Tilburg University, The Netherlands. • To demonstrate how the proposed She explains the diffi culties of addressing the framework and econometric methods problem and why their joint research is at the improve the accuracy of estimated With increased computational forefront of econometrics: “Dealing with one of parameters using simulation studies and empirical studies power and a vast spectrum of data these issues is already a challenging task, mainly because it relies on state-of-the-art statistical • To demonstrate how the proposed available for incorporation into and mathematical tools in order to develop framework and methods improve the the associated asymptotic theory required for quality of forecasting exercises and the adapted models, the study of valid inference. This is why teaming up with Dr programme evaluations Boldea was key to this research project”. Using econometrics holds promise of real-world data, the team has found evidence KEY COLLABORATORS to suggest that economic processes often Associate Professor Otilia Boldea, Tilburg heightened model complexity and display poor identifi cation while also changing University in unexpected ways over time. subsequent improved accuracy ADDITIONAL COLLABORATORS Consequently, developing economic models to account for both these issues is crucial for Professor Eric Renault, Brown University in time, something which is unlikely in reality creating accurate forecasts. Antoine points to Professor Pascal Lavergne, Toulouse School where contributing factors can be variable and the New Keynesian Phillips Curve (NKPC) – a of Economics inconstant. Researchers at the university are leading macroeconomic policy model, which striving to create models that allow for such describes the dynamics of infl ation in terms CONTACT variation. Antoine gives an example of why of unemployment, production costs and Associate Professor Bertille Antoine this is so crucial: “A well-known stylised fact in economic output – as a commonly used model Department of Economics labour economics is that the gap between the which experiences this problem. “Infl ation earnings of college and high-school graduates prediction via the NKPC has recently become Simon Fraser University grew sharply during 1970s and 1980s. Therefore, more challenging, suggesting instabilities in the 8888 University Drive allowing parameters to change over time seems predictive power of the NKPC,” she explains. “In Burnaby important if one aims at developing a model addition, many empirical studies also document British Columbia that refl ects what happens in practice”. the weak identifi cation and the instabilities Canada issues related to its estimation.” Antoine is V5A 1S6 working with her colleague Boldea to address IDENTIFICATION AND STABILITY T +1 778 782 4514 this subject and is hopeful that they can design E [email protected] While variation through time is an important a framework that will account for both issues factor to incorporate, there are still many arising simultaneously. In the future, they hope www.sfu.ca/~baa7 further variations and correlations which to clarify this matter further by establishing the must be included in a successful econometric effect that this lack of integration may have on PROFESSOR BERTILLE ANTOINE model. The unpredictable nature of data the validity of existing models. graduated from Ecole Nationale de la produced from real world events means that Statistique et de l’Analyse de l’Information this is a complex task, but nonetheless, one in With the advances already made, the next step (ENSAI), a French Grande Ecole affi liated which Antoine and her colleagues are making is to integrate these refi ned and more accurate with the French National Institute of considerable headway. When considering the procedures with a larger range of datasets Statistics. She obtained a MSc in Statistics current econometric techniques available, to create more comprehensive models with from University of Carolina, USA in 2005, one area that is lacking involves the design greater real-world relevance. With increased and a PhD in Economics from University of of frameworks that can handle several issues computational power and a vast spectrum Montreal, Canada in 2007. After being hired simultaneously, particularly in relation to of data available for incorporation into the on a tenure-track position in the department stability and identifi cation. Although some adapted models, the study of econometrics of Economics at Simon Fraser University in econometric frameworks allow the parameters holds promise of heightened model complexity 2007, she is now an Associate Professor with of the model to be weakly identifi ed, they hold and subsequent improved accuracy. In times of tenure at SFU. the assumption that these parameters remain economic turbulence, the process of developing stable over time, while other models may and redeveloping models to become as reliable account for their instability, but not for their as possible is essential for making sensible poor identifi cation. economic decisions. With fi nancial markets implementing substantial international To fi nd a solution to the issue, Antoine has infl uence, decisions made using econometrics teamed up with Associate Professor Otilia are likely to have global impact.