Amol Agrawal
The application of machine learning in economics is the cool thing. Machine learning uses algorithms and statistical models that perform tasks based on patterns and inferences. It is not really a new idea, but its application in economics is going to increase significantly as seen by the recent working paper by a team of RBI economists, which looks at machine learning for economic forecasting.
To get an idea on machine learning in economics, one should go back to the Jean Monnet lecture by Susan Athey, Professor, Economics of Technology, Stanford University. Her lecture 'Machine learning in economics’ was delivered at the 4th ECB Annual Research Conference on September 5-6.
Athey is no ordinary economist as she is the first female recipient of John Bates Clark Medal in 2007, an award given to “that American economist under the age of 40 who is judged to have made the most significant contribution to economic thought and knowledge”. Athey’s prior education in computers, along with economics, helps her understand the field of machine learning much better than others.
In fact, earlier, people with maths or physics background did well in economics, and in today’s world it will be those who have a background in computer science.
Athey’s lecture started with a quote from Marc Anderssen, the cofounder of Mosaic and Netscape. Andressen once said “we are in the middle of a dramatic and broad technological and economic shift in which software companies are poised to take over large swathes of the economy”. Increasingly, all businesses are becoming online services, from food to movies to national defence.
According to Athey, there are several applications of machine learning in economics.
First, is text and image recognition which we see all the time when a website wants to check whether you are a human being. This is irritating for sure, but helps websites ward off threats from hackers.
Second, the combination of so-called Big Data, artificial intelligence (AI) and machine learning (ML) provides many more opportunities. The machine algorithms can sift through tonnes of data and look at pointers such as which customer is likely to leave shopping at, say a departmental store company. This knowhow can then be used to encourage the customer to continue shopping with the company.
Then, take the case of banks and financial institutions which generate a lot of data about all kinds of behaviour of depositors and loan-seekers. Athey pointed out that banks and FIs have multiple AI/ML projects, but a majority of them deal in image or text recognition. The banks have automated some of these processes such as call centres where humans handle complaints to an app. However, much of this is automation of certain steps and mainly an efficiency enhancer.
The banks could go bigger on ML/AI and even allow things like small quick loans being handled by similar algorithms. The loan officer does give loans based on some data and information, but it has limitations. The machine can do a much better job in the given time as it can scan endless amount of data and information and help make better decisions.
Another example is how cameras in manufacturing plants can stream information about workers to a machine and tell quickly whether they are following the safety procedures and rules. This can then be modelled and applied across similar plants in the world and help improve safety of workers.
A similar application is for Uber drivers where most of them get a higher rating irrespective of the quality of driving. How often we are told by the driver to give higher ratings! Instead of these passive ratings, one could build a model capturing the data of the entire ride and see how driver navigated speed breakers and so on. This gives a more reliable rating of the driver who could then be given safety tips, especially in cases where the driving has been rash and yet the driver gets 5 star ratings!
She added that statistical discrimination plays a major role in human decisions. Say you met Athey over a recruitment dinner, the human brain can only decide on her as a person based on what she says over the conversation and written in the CV. A machine which sifts through her CV can give more ideas about her candidature.
This idea of trying to minimise statistical discrimination is perhaps a good “two word” explanation to build better macroeconomic models. In the earlier models, the economists wrote economics models and then ran those models via computers/machines to get a result. In ML, machines will play a more central role, doing both modelling and analysis.
The recent paper by RBI economists Bhanu Pratap and Shovon Sengupta is in similar spirit. They ask whether machine learning can improve macroeconomic forecasting. They compare the ML models with traditional ones and not surprisingly, their results show ML models produce better results.
The question is not if, but when ML will take centre stage in most aspects of economics, especially forecasting macroeconomics and financial markets. It is a good time for those with a computer science background to think of a career in economics and finance. The cover page of the latest edition of The Economist (October 5) is also on 'How machines are taking over Wall Street', showing the idea has already arrived. This does not mean those with a pure economics background will not be in demand. Just that the cutting edge will lie with those who can mix these two streams.
The end of Cold War led to several physicists getting absorbed in financial market predictions and modelling. The economists will hope that there is no such event that leads to freeing up several computer scientists to take on their jobs.
Amol Agrawal is faculty at Ahmedabad University. Views expressed are personal.
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!
