 So, as the economics department is famous for its teaching and research in heterodox economics, and here one of my major teaching subjects is economics. Over the years I try to teach this subject in a very applied and practical manner, and here especially Cata for our students who are interested in international economics and development economics. Especially textbook econometrics and especially its heavy mathematical contents are largely irrelevant for applied works. And today I'm going to show this argument with a couple of slides. Now what I think that textbook econometrics suffer from a serious methodological flaws and I'm going to borrow Steven Sun's diagram from his paper, Mathematics, Governance or Handmaids for this argument, and here he made a critique of some statistical research which ignore the link with real problem and application, but focus more or less on the elegant mathematical solutions for idealized problem and actually textbook econometrics fall into the same line. When you think about it, especially for those students who have done economics before, you realize that in the textbook econometrics we assume that a structural model or alternatively is known as the target function from machine learning is assumed to be given and known before any data analysis, and then textbook econometrics move on from this assumption and plus it with some new assumptions on probability distribution so that they come up with a consistent estimator for the given structural model. So that is the core of textbook econometrics, and that means that in order to get those consistent estimator, the very concept of consistency assume those structural models without being tested by data are actually globally correct and well known, and that implies that textbook econometrics give no role for formal statistical learning. One of the best known examples of this consistency is the endogenetic bias. The basic idea of endogenetic bias is to prove that open-release-ware method known as OLS is biased, and it's not a consistent estimator for the assumed structural model be the reason as simultaneity or selection bias or omitted variable bias, but the end result is OLS is biased, and therefore the proof is very nicely clean, but a strange thing is that all these proofs are actually independent of data analysis, it's all done by algebraic proof or analytical proofs independent of any data experiments. The reason that they can do these proofs is because the bias is proved not within the sample, but out of sample, and therefore the bias is proved on the out-of-sample model predictive errors. Now what are these model predictive errors in practice? Actually you can find a very simple analysis of these errors from machine learning textbooks. Actually the out-sample errors can be decomposed into two components. One is model bias, and the other is estimation error. And in order to prove OLS bias and all these bias to cater for a consistent estimator, one needs a very strong assumption that means that a priori structural model is bias free, and therefore whatever those analytical proofs of OLS bias is based on these models have no bias. Now you can see this is such a strong assumption which cannot be held in any applied circumstances. And therefore when we teach applied econometrics we have to remove this assumption, we have to acknowledge the fact that in reality the target function is incomplete or actually unknown. So once we removed this assumption it's obvious so all the proof of consistent estimator become irrelevant in practice. And that's why we can get by teaching applied econometrics without actually using those textbooks heavy, mathematical based consistent estimators. So I've moved the essential teaching target in my course to try to search for the best and workable models through a combination of learning from theory and data. Fortunately this missing part of learning from data is readily available from machine learning textbooks so I've tried to combine econometrics with machine learning to my students and to make the whole teaching more motivating and stimulating so the students can learn how to work with real economic data and also critically evaluate previous empirical research.