Skip to main content

Additional November CIC seminar

Date: November 30, 2016

Time: 1pm to 2pm, followed by afternoon tea

Location: CLT Learning Space, Building 105, Room 107







Speaker: Prof Laszlo Matyas

Title: Big Data and Econometrics – Event Count Estimation

RSVP: Please email Linda Lilly by COB Monday 28 November, 2016

About the speaker:

Professor Laszlo Matyas is a university professor at the Central European University and a former Provost, and Head of the Department of Economics at the same university. He previously held academic positons at Universite de Paris XII and Monash University, and also worked as Assistant deputy-state secretary at the Ministry of Economy in Hungary. He was the founding director of the Institute for Economic Analysis in Hungary. He also served as the Managing Director of Nortel Networks Financial Services Ltd and the Managing Director of Moore International Hungary. He has published over 50 journal papers and 10 books. His best known paper on gravity models has been cited over 800 times. His main research interests include Panel Data Econometrics, and Modelling International Trade.


In classical econometrics and statistics efficient estimation has had a central role. It has been quite important to squeeze out every possible bit of information from the data. To be able to do so one had to rely one many assumptions and exact “metric”-type measures. The cost has been estimation methods which frequently are not robust against the underlying assumptions, outliers, etc. In times of big data, when almost unbounded information is available, efficiency becomes much less relevant, instead robustness and flexibility become the most desirable properties for “Big Data” estimation methods. This paper proposes a conceptually simple new technique, the so-called Event Count Estimator (hereafter ECE), which turns a blind eye towards optimality and efficiency, but is robust against several assumptions and data related problems. While optimal estimation methods (like Least Squares, Maximum Likelihood, etc.) are more efficient than the ECE when all their assumptions are satisfied, the gained (slight) precision is quickly offset by the biases optimal techniques suffer from when any (or some) of their assumptions are violated.