Workshop: Automatic Model Selection with Applications

Speaker:  Dr. Jurgen DOORNIK (University of Oxford)

Date:       26 June 2019 (Wed)

Time:       11:00am – 1:00 pm

Venue:     RLG402, Research Complex


Automatic model selection is a powerful tool for the empirical modeller. This workshop will introduce Autometrics, which successfully implements the general-to-specific approach. Foundations of the algorithm will be described, together with interesting extensions, including applications that have more variables than observations. Hands-on computer illustrations will be used throughout.

A modeller is confronted with:

  • many possible inter-related variables that matter,
  • subject to intermittent breaks (earthquakes, financial crises, etc.),
  • possible outliers (oil shocks, tsunamies, etc.)
  • changes to the measurement system
  • and dramatic changes over longer periods (recessions,technological progress, climate change)

Big data may help, but can bring further problems:

  • data overload,
  • too much heterogeneity,
  • hidden stratification and unknown dependence.
  • Information overload and complexity

Automatic modelling essential to cope with these challenges. Autometrics is the implementation of automatic model selection, designed to handle these challenges:

  • Automatic: computer is powerful modelling aid,
  • General to specific: can maintain econometric properties,
  • Extensive search: to handle correlated data,
  • Efficient search: need to estimate many models,
  • Statistical congruence: maintained as a search constraint,
  • Statistical properties: extensively researched,
  • Not maximizing goodness-of-fit: avoids overfitting,
  • Controlled by gauge: expected number of falsely selected variables,
  • Flexible: more variables than observations, different model classes (logit, SEM), …
XlModeler brings regression and volatility models of OxMetrics to Excel

I am impressed by the power of XIModeler to estimate 300 models and select the best model within 1 sec automatically.



An introduction to XImodeler:

Oxmetrics: is a family of of software packages providing an integrated solution for the econometric analysis of time series, forecasting, financial econometric modelling, or statistical analysis of cross-section and panel data. OxMetrics consists of a front-end program called OxMetrics, and individual application modules such as PcGive, STAMP, etc. Oxmetrics

Workshop PPT

Seminar: Automatic Selection of Multivariate Dynamic Econometric Models

Speaker:  Dr. Jurgen DOORNIK (University of Oxford)

Date: 26 June 2019 (Wed)

Time: 3:30pm – 5:30pm

Venue: RLB303, Research Complex


Automatic general-to-specific selection of univariate econometric models is now well established and available in software. Extensions include saturation estimators, e.g. adding an impulse dummy for every observation to handle outliers. This seminar will provide an overview of the approach, and then consider extension of these procedures to the multivariate setting. The starting point is a vector autoregression, and the final stage can be a simultaneous equations model where the role of identification is considered. The aim is to obtain procedures that are relevant for empirical modelling.

The need for machine-assisted learning in econometrics:

  • Developing good models is difficult.
  • Working with economic data is difficult:
  • approximate measurements subject to revisions on a system that is huge,
  • evolving, intercorrelated, maybe nonlinear, and prone to abrupt shifts.
  • Need models for policy as well as forecasting :
  • Black-box models insufficient: need to understand,
  • Nonlinearities of secondary importance.
  • Proliferation of data: Big data:
  • But is there a proliferation of insight?

General-to-specific model selection (Gets, ‘Hendry’ or ‘LSE’ methodology) largely driven by David Hendry (DHSY, PcGive, Alchemy, Dynamic Econometrics, …)

General-to-specific automatic model selection, developed methodology and algorithms to handle these challenges


Doornik, J. A. (2009). Autometrics.
In J. L. Castle and N. Shephard (Eds.), The Methodology and Practice of Econometrics: Festschrift in Honour of David F. Hendry. Oxford: Oxford University Press.

Empirical Model Discovery and Theory Evaluation Automatic Selection Methods in Econometrics By David F. Hendry and Jurgen A. Doornik
Published by the MIT Press

Doornik, J. A. and K. Juselius (2018).
CATS 3: Cointegration Analysis of Time Series in OxMetrics.
London: Timberlake Consultants Press.

Seminar PPT

Nonlinear econometric methods

I would like to express our appreciation to Professor WONG Wing-Keung for providing a seminar on “Do both demand-following and supply-leading theories hold true in developing countries?” and a workshop on “nonlinear co-integration and causality tests”.

The illustration of the sample code ( R programming ) covering time series analysis techniques including unit root test, cointegration test, VAR estimation, linear Granger causality test and non-linear Granger causality test are essential for improving the research skill of our colleagues and students. The guidance of necessary software (Rstudio and Code:Block) environment setup and installation are extremely informative.

Once again, thank Professor WONG Wing-Keung for his contribution to the “IIDS Project – Recent Developments in Theoretical and Applied Econometrics Analysis”. We are looking forward to our research collaborations in the future.

PI Website:

Workshop: Practice in nonlinear co-integration and causality tests

Speaker:  Professor Wong, Wing Keung (Asia University, Taiwan)

Date:  10 June 2019 (Mon)
Time:  14:00 – 16:00
Venue:    RLG402, Research Complex


This workshop will discuss the econometric programs for linear and nonlinear co-integration and causality. Sample computer codes will be illustrated during the workshop. It is designed to familiarize participants with the common computer codes for tackling financial and economic research questions.


R :

Rstudio :

code::Blocks :

Seminar: Do both demand-following and supply-leading theories hold true in developing countries?

Speaker:  Professor Wong, Wing Keung (Asia University, Taiwan)

Date:   3 June 2019 (Mon)

Time:   15:30 – 17:30
Venue:   RLB 303, Research Complex


In this seminar, the speaker recommends using both multivariate linear and nonlinear causality tests to analyze the relationship between financial development and economic growth. In particular, multivariate nonlinear causality test allows us to consider dependent and joint effects among financial variables, and detect a multivariate nonlinear deterministic process. By the end of the seminar, the recent applications of multivariate nonlinear co-integration and causality tests will be discussed.


  • First to use cointegration and (non)linear causality to study financial development and economic growth.
  • Financial development and economic growth are moving together in some developing countries.
  • Both demand-following and supply-leading theories hold for all of the countries studied in our paper.
  • Including nonlinear test allow us to detect causality in five more countries.
  • Our finding helps in the decision making in the development of the countries and reducing poverty.

To overcome the limitations of the traditional approach which uses linear causality to examine whether the supply-leading and demand-following theories hold. As certain countries will be found not to follow the theory by using the traditional approach, this paper first suggests using all the proxies of financial development and economic growth as well as both multivariate and bivariate linear and nonlinear causality tests to analyze the relationship between financial development and economic growth. The multivariate nonlinear test not only takes into consideration both dependent and joint effects among variables, but is also able to detect a multivariate nonlinear deterministic process that cannot be detected by using any linear causality test. We find five more countries in which the supply-leading hypothesis and/or demand-following hypothesis hold true than with the traditional approach. However, there is still one country, Pakistan, for which no linear or nonlinear causality is found between its financial development and economic growth.

To overcome this limitation, this paper suggests including cointegration in the analysis. This leads us to conclude that either supply-leading or demand-following hypotheses or both hold for all countries without any exception. There will be some types of relationships between economic growth and financial development in any country such that either they move together or economic growth causes financial development or financial development causes economic growth without any exception. The finding in our paper is may be useful for governments, politicians, and other international institutions in their decision making process for the development of the countries and reducing poverty.


Dynamic Panel Data Analysis

I would like to express my appreciation to Prof Cheng HSIAO (Department of Economics, University of Southern California) for providing a seminar on “Panel Parametric, Semi-parametric and Nonparametric Construction of Counterfactuals” and a workshop on “Important Considerations in Working Panel Dynamic Models” in Hong Kong Shue Yan University on 16 April, 2019. It is extremely informative, and the active participation illustrated the importance of these topics to our colleagues and students.

Thanks Prof HSIAO for his valuable and essential contribution to the project “IIDS – Recent Developments in Theoretical and Applied Econometrics Analysis” .

PI website:

Seminar: Advancements in panel data analysis

Speaker: Professor HSIAO Cheng (University of Southern California)

Presentation: Panel Parametric, Semi-parametric and Nonparametric Construction of Counterfactuals

Date: 16 April, 2019 (TUE)

Time: 15:00-17:00

Venue: C/F, Library Complex, Hong Kong Shue Yan University

We consider panel parametric, semiparametric and nonparametric methods of constructing counterfactuals. We show through extensive simulations that no method is able to dominate other methods in all circumstances, since the true data‐generating process is typically unknown. We therefore also suggest a model‐averaging method as a robust method to generate counterfactuals.

The advantage of the parametric model approach is that it allows efficient estimation of the unknown parameters; hence, it is possible to identify the impact of each covariate on the outcome. The disadvantage is that, if the model is misspecified, then the resulting inference could be misleading. The advantage of the panel nonparametric approach is that there is no need to consider the conditional mean or the error distribution. The disadvantage is that there is no structural interpretation of the casual effects, only the measurement of the treatment effects. The semiparametric approach is somewhere in between. It assumes that the conditional mean of the observed covariates is specified correctly, but it lets the data control the impact of unobserved factors. Each method has its advantages and disadvantages. Our simulation results show that, if the observed data are stationary, the panel semi-parametric method appears capable of generating counterfactuals close to the (true) data generating process in a wide array of situations. If the data are nonstationary, then the panel nonparametric method appears to dominate the parametric and semiparametric approaches. However, no method appears capable of dominating all other methods under all different data generating processes and different sample configurations of cross-sectional dimension N and pre-treatment time dimension T0: Since the true data generating process is usually unknown and the statistical findings could be very different for different situations, we have also suggested a model averaging method as a robust method for generating counterfactuals.

Advantage and disadvantage of Non-parametric, Parametiric and Semi-parametric approach to Dynamic Panel data


Wiley Online Library:



PPT Download

Workshop: Applications of Dynamic and nonlinear Panel data modelling

Speaker: Professor HSIAO Cheng (University of Southern California)

Date: 16 April 2019 (TUE)

Presentation: Important Considerations in Working Panel Dynamic Model

Time: 19:00-21:00

Venue: Room 205, Main Academic Building, Hong Kong Shue Yan University

“All interesting economic behavior is inherently dynamic, dynamic panel models are the only relevant models; what might superficially appear to be a static model only conceals underlying dynamics, since any state variables presumed to influence present behavior is likely to depend in some way on past behavior.” M. Nerlove (2002, p.46)

Three major issues arise in the analysis of linear dynamic panel data models:

  1. initial value distribution;
  2. controlling the impact of incidental parameters to obtain valid inference on structural parameters; and
  3. relative sample size between cross-sectional dimension N and time series dimension T

For nonlinear  dynamic models, these issues become even more difficult to handle.  The derivation of initial value distribution is much more difficult to handle (e.g.  binary outcomes model). Nor is there any linear transformation to remove the incidental parameters. The identification for the method of moments estimator becomes much harder to derive (e.g. Honore (1993)).


Hsiao, C. (2014). Analysis of Panel Data (Econometric Society Monographs). Cambridge: Cambridge University Press. doi:10.1017/CBO9781139839327

Library of Congress Cataloging in Publication Data :

Test (2007) 16: 1–22. DOI 10.1007/s11749-007-0046-x. INVITED PAPER. Panel data analysis—advantages and challenges. Cheng Hsiao.:

PPT Download