Speaker: Dr. Aman Ullah, University of California, Riverside, USA
Date & Hour: 12 December 2019, 4:00 pm
Venue: Seminar Hall 1
Title: A Journey in Search of the Unknown “True” Model: Data–Based Nonparametric Econometrics and Empirical Economics
Abstract: In the early 20th century, the legendary statistician Sir R.A. Fisher and others set in motion what is known today as the classical parametric approach to Statistics — estimation of a finite number of population parameters using sample data. Thus began the practice of statistical inference (estimation and testing, and their properties) and this laid the foundation of econometrics inference. In 1930, the international Econometric Society was established in the USA, with the purpose of advancing the study of Econometrics, formally described as the data analysis (measurement) of mathematical economic models using statistical-inference methods. These models which relate economic variables–– called regression or conditional mean models –– were used to answer questions regarding marginal effects (elasticities) and predictions/forecasting for policy purposes. For example, economists are regularly asked to advise on: the effects of imposing or lifting a tax on employment or sales in an industry, effect of education on earning, and forecasts of S&P 500 stock returns or the US-Canadian exchange rates. Traditionally, these efforts are dealt with by assuming a parametric model, usually linear. However, the true functional form of a model is rarely, if ever, known and any misspecification in this regard may lead to misleading estimates and policy conclusions. Consequently Ullah (1985, 1988, —-, 2019) moved to develop data-based nonparametric kernel econometrics (theory and practice) suitable for the estimation of unknown models for prediction and marginal effects. Along with this, other complex nonparametric models were developed and studied such as conditional variance (volatility), conditional correlation, quantile, probability of decision, panel data, and models with endogenous variables, which are useful in Economics, Finance, and other applied subjects. The paper traces the path of these developments with a link to recent work on machine learning (Ho (1995) and Breiman (2001)).
