Welcome back. Today, we're going to discuss applications of machine learning in finance. We're going to give some examples, and we're going to discuss some issues that come up when we apply these techniques that we introduced last time into the finance area. First topic I want to mention is one that I've seen over in China. I have a project with Alibaba, and Ant Financial which is their division that manages Alipay, the system that's used by 600 million Chinese to manage their affairs. Here, you can see on Alipay, the ability to get a loan almost instantaneously. You sign into Alipay, you ask for a lone, 50,000 renminbi, and you can get a result almost immediately. There is no loan officers, it's simply machine-learning in an action here. Second example has to do with robo-advisors. This is an area where instead of having a human advisor, we have a computer that's helping us make decisions about our investments. How much should we put into stock? How much should we put into bonds? Should we save a certain amount of money every year? How do we manage our affairs? The idea of robo-advisor is somehow developing that computers can be low cost, and they could manage people's financial affairs in a way that would be much more efficient than we would get through paying a fee to a human advisor. The third area that has become of interest is the notion that we can look inside of micro data, and use that to make macro impacts. For example, instead of waiting to see if the GDP is high or low for this quarter, having it revise, this quarter revised again, we can in fact have instantaneous feedback by tracking individuals, and seeing where they actually manage their affairs. Another example would be in the area of sentiment. By tracking what people were saying, what they're buying, we track what they feel, and in sense, we can understand how markets are changing over time. We can get an a notion of the tracking of that information in a way that's real time. We're seeing hedge funds, other investors trying to track that information in a much more systematic way. A fourth area that we're going to look at in the course involves factor investing. The second module, we're going to go into the notion of factor investing, and we're going to see that inside of investment, we have risk that are called factors or features. Now, in this context of identifying, we go back to our facial recognition problem. The computer's job here is to understand what features lend themselves to identifying that object. We're going to do a similar kind of exercise by looking at what are the features that lend themselves to crash or to a normal regime. So those features can be of various kinds. We can have macro-factors, we're going to micro factors, those features are the ones that really are the essence of this machine learning. Can we identify the features? Secondly, since this area has some large databases, we have the ability to then look at individuals, and there's many issues in the context of ethical concerns about whether individuals have access to that data, the data is being used in other areas, and so we have privacy concerns, and ethical issues involved with that. Once we start collecting the data, who has access to the data and how is it used? In particular, since we have so much data, we can start perhaps using it for good purposes but also for purposes which are not so good, this notion of fake news that we've seen more recently. Another issue that comes up when we apply these techniques to financial problems has to do with the question of interpretability of the results. In most cases, first, especially for strategic decisions, it's going to be difficult for the top managers to make important decisions without understanding what are the characteristics of the model. How can we interpret it? Some techniques are easy to interpret in machine learning, and others are very difficult to interpret. When we get to deep neural networks or we get to the reinforcement learning in the AlphaGo system, extremely hard to understand what it's telling us, and this is going to be a barrier that we're going to have to worry about in the area of a financial applications interpretability. Sometimes, the most successful methods are the ones that are least interpretable, so we're going to have to find ways to bridge that gap. So let's think about some take home points from this section here. The first one is that we've seen successes in classification, and we're trying to see some of that taking place into financial applications. However, finance is more complicated because we have an evolving system, and we have to worry about not just how it occurred 10 years ago or five years ago, what's happening today? We need to be able to do the adaptability of the data over time. Secondly, as we start to collect information at the micro level, and we use it for identifying, let's say how loans are being taken out in certain parts of the country, different parts of the cities, how do we use that information? There's going to be ethical and privacy concerns? Once we have that information, how do we use it for purposes which are consistent with our values? Thirdly, it's difficult to understand how some of these systems change over time, so we have to deal with the temporal aspects of it. How's the dynamics occurring and changing over time? Finally, we're going to look at the question of interpretability. If we can interpret the data algorithms, so much the better. However, in many of these examples in machine learning, it's very difficult to understand how to interpret it, so we have to think about ways to explain these techniques to executives and other decision makers, so we can make the best decisions possible while at the same time, being understand what is driving the results. Of course, the feature selection is going to be one of the most important characteristics that we look at.