Welcome, I'm John Mulvey, a professor at Princeton University, and I'm here today to talk about the course I've developed with my colleague Linda Martin in Python and Machine Learning for asset management. The course is a bit different than the ones you've taken previously. We are not going to be teaching programming per se, instead, we're going to look at notebooks and I'm going to show how one can apply these methods, these notebooks in practice. We will provide these notebooks, we'll ask you to run them and look at the results and try to understand how they're used. The area of Machine Learning has become a great interest throughout the world, not only in finance, but also health care, many other domains. Part of the reason has to do with the availability of data at the micro level, at the individual level. This data can be used to supplement the macro-level indexes. We get the macroeconomic indicators, also market data, and now, we have a brand new source of data in many ways. We can look at sentiment, we can look at what humans are doing, activities of individuals. This gives us a whole new, richer set of data that we can use. This data can be used in conjunction with the new techniques that are developed in Machine Learning. These techniques are very easily accessible through Python Machine Learning techniques. You'll see the Python notebooks and the Jupyter Notebooks are easy to use and they gave us great abilities to try out different methods with different kinds of data. Of course, the great incentives in the investment area because even a small improvement in forecasting have a major impact on results. Now, there are challenges, of course, along with this, some of the same reasons for the motivation. Large amounts of data that are available, there's so much data that one has to choose. Machine Learning can help us in some of those ways, but it doesn't provide the only answer when you need to know something about the problem. At the same time, we have all these methods, so which methods are used for what problems? Again, there are some challenges that we face with that. Importantly, there are challenges with regard to the assumptions of Machine Learning, the stability assumptions, the IID processes that you see in Machine Learning. These assumptions are not always valid. Humans are prone to many different, hard to replicate, hard to forecast situations. Take the pandemic in 2020, this was totally unexpected, so we've got have to be very careful about applying these methods. The structure of the course is really different, as I said than the previous courses. We're not teaching programming, rather, we're going to introduce a financial problem, then, we will show the traditional statistical methods, and then, finally, we will show how one can use Machine Learning for those examples. The examples are curated and really, are based on the problems that we've worked on. We can't possibly cover all areas of financial and asset management. Rather, we pick problems that fit our backgrounds. For example, Linda and I are consulted for CalPERS, the largest pension finding United States to California employees pensions. We worked on factor investing, we worked on asset liability management, and these are the techniques that one sees in the course. I've had over 35 years experience as a professor, actually longer than 45 years teaching, but 35 years working with some of the largest institutional investors in the world. Large pension plans, endowments, hedge funds, multi strategy, reinsurance companies, and individual investors, and family offices. Each of these applications you'll see, has lessons that we can learn. You may not be interested necessarily in pension plans, but there seems we can learn there that we can apply in other cases. As in the side, the area of Machine Learning has commonality with decision-making under uncertainty, the stochastic optimization algorithms that we use in models for decision-making in finance and other areas also apply to statistical decision theory. That is to say, Machine Learning techniques are trying to minimize a loss function and that's the stochastic optimization problem. So some are the same concepts. The techniques, the software that you see in the course can be applied in both areas. In summary, or we're trying to show you how one can use these techniques, we assume some background, some fundamental background that you've had in [inaudible] theory, and in elementary level, we also assume that you have some basic background in statistics. We are not, again, trying to cover all materials, we're looking at specific applications that fit our interests. These had been applied. I can guarantee you that these same techniques can be applied broadly across many different domains. At the same time, there are any many advanced methods we have. My doctoral students and many others are working on natural language processing using more advanced methods in the Data Sciences, deep known networks to solve multi-period optimization problems of many new things exciting coming along. Our goal here though, is to teach you the fundamentals and to show how these methods are used. Thanks again for your interest. Hope you'll participate and we look forward to your feedback and work to improve the course and hope that we hear from you if so. Bye-bye.
