Now, after we learned how to run a neural network intensive flow, we are finally ready to start trying our new tools on real world financial data. And as long as this is a specialization about machine learning in finance, we will, of course, start with stocks. More seriously, while the field of finance covers many asset classes, including equities, fixed income, commodities, foreign exchange and so on, in this specialization we will mostly focus on only two asset classes, equities and fixed income. The main reason for this choice is popularity of these asset classes among both financial professionals and amateur investors. The other related reason is availability of data. Data for equities and fixed income, such as consumer credit, is both more widely available and more commonly used in financial applications of machine learning methods. So by sticking to these asset classes, we will have some benchmarks from the literature. Let's start with a short overview of common tools used for investment decisions in the world of equities, where investors decide to buy or sell a stock of a particular company. There are four major types of analysis in equity space, fundamental analysis, technical analysis, quantitative analysis, and alternative data analysis. The most important difference between them lies in the data used in these approaches. Let's take a look at these approaches from the perspective of the data used. Let's start with the fundamental analysis. Fundamental analysis deals with valuation of securities using accounting information, such as the balance sheet and income statement information. The main idea there is that the accounting information determines the true value of a firm. If you know this true firm value, you can compare it with the market value for the same firm as implied by the current stock price and the number of shares outstanding. You then buy the most undervalued stocks and sell the most overpriced stocks. This is the essence of the so called value investment. The second type of analysis of equities is called technical analysis. This analysis focuses on the market pricing data only. In other words, it looks at the prices of stocks and nothing else. Then it tries to formulate rules-based strategies that produce signals to buy or sell based on patterns of the past stock prices' behavior. And if you think at this point that this sounds already a bit like a machine learning approach, this is because it does. This approach can indeed be formulated as a data driven predictive modeling of stock returns, or stock price changes, where only the pricing data is used in the analysis. The third approach is what is traditionally referred to as quantitative analysis. Similar to technical analysis, it looks at the market data, sometimes in combination with macro-economic data. Unlike technical analysis, it relies on probabilistic modeling framework, where quantities that are modeled or optimized have a probabilistic interpretation. As an example, we can look at the most classical financial problem of optimal portfolios, as formulated by Harry Markowitz in a 1957 work that earned him a Nobel Prize in economics in 1990. In Markowitz' Portfolio Theory, we choose a portfolio that maximizes the expected return at the given level of portfolio variance. Other portfolio models optimize some other portfolio measures. For example, they can minimize the tail risk, that is the probability of large portfolio losses. Other quantitative models are encountered in systematic stock trading strategies. For example, in classical statistical arbitrage strategies that look for tradable signals by analyzing time series of residual returns that are obtained from stock returns by subtracting a market mode contribution from all of them. If you're interested in more details of this approach, additional references are provided in this week's reading list. Again, quantitative analysis can be rephrased as a probabilistic machine learning framework. Once you formulate it in this manner, it opens a way to extend or enhance traditional quantitative approaches using other machine learning methods. For example, non-parametric models implemented via neural networks. Finally, equity analysts often look at alternative data. This is currently a trending topic in finance. In general, alternative data is any data complementary to traditional data used in equity analysis, such as fundamental data, market data and macro-economic data. This may include sentiment indicators extracted from using geolocation data, satellite image data, and other data that helps better predict such things, as for example, companies' earnings or sales. For example, traffic data collected from mobile phones of customers that visit retail stores can be used to improve predictions of sales. In its turn, this might help to improve predictions of earnings and therefore may also be useful for designing trading strategies. Now, let's recap. We introduced four major types of analysis in equity research, fundamental analysis, technical analysis, quantitative analysis, and alternative data analysis. These methods differ most importantly in data they use, as well as in the methods that they employ. The most important point, however, is that all of them are amenable to machine learning formulations. This allows us to extend and enrich these methods and make them more data driven and less model dependent. Another important point to take away is that most of these problems are problems of regression. We use data to predict such quantities as companies' earnings or stock returns. Sometimes, such problems are reduced to classification problems, where one tries to predict the direction of the price change instead of the amount of the price change. This means, among other things, that nearly all regression and classification methods can be used alongside more traditional financial models to address different problems in the equity research space. And moreover, machine learning approaches let us build models that integrate different sources of data or different approaches. For example, the so called quantamental approach in trading that combines the fundamental and quantitative approaches can be formulated in the machine learning framework. In the next video, we will take a close look at how fundamental analysis can be formulated as a machine learning task.
