Now, after we've discussed four major types of equity analysis, let's take a closer look at fundamental analysis. First let's start with the accurate definition of fundamental analysis. Citing from a classical 1992 paper by Penman, we can say that fundamental analysis focuses on evaluation of securities from available information with a particular emphasis on financial accounting information. Now, financial accounting information is information contained in financial reports filed by companies. For example in the US all corporations that either are listed on NASDAQ, or have more than 10 millions in assets and more than 500 shareholders have to file quarterly reports 10Q to the Securities and Exchange Commission, or SEC. 10Q reports contain both numerical data and unstructured data that is the texts of the report. Fundamental analysis focuses on two components of numerical data presented in corporate finance, income statement data and balance sheet data. Following a 97 paper by Abarbanell and Bushee, we will refer to these variables as fundamental accounting variables. Starting from the income statement data, we can mention such numbers as sales revenues, cost of goods sold or COGs, general and administrative expenses, operating income, net income, capital expenditures, dividends, common shares outstanding, and price per share. On the Balance sheet sides, we can mention cash & cash equivalents, accounts receivables, current and total assets, current and total liabilities, long-term debts, preferred stocks, and shareholder equity. Now how we can use this quantitative information in income statements and balance sheet variables. The answer depends on what type of analysis you want to do with this data. First let's start with what you want to predict. One way to use corporate finance is to predict either future fundamental quantities such as net income or earnings per share, or EPS. This is a task that is often solved both by outside investors as well as the firms themselves as a part of internal forecasting and corporate planning. Another potential use of accounting data is for prediction of stock insurance for medium to long term horizons that are usually measured in months or years. Now this is about the output variables, but what about input variables? Well, features that you construct out of raw income statement and balance sheet variable depend on the setting of your regression. If you want to do the analysis for a single firm, then one set of variables could be provided by standardization of the raw input features. Now this, in this case we would simply subtract the historical mean of each feature and divide it by its standard deviation. For example, one can build linear and non-linear models where the set of predictors is chosen to be inventory, accounts receivable, capital expenditures, gross margin, which is sales less cost, selling and administrative expenses, and effective tax rate, which is in income tax divided by p tax income. You might want to look at implementations of linear regression models for individual firms with such predictors, along with references to original publications in a demo Jupyter notebook for this week. As a part of your homework, you will be asked to extend this notebook for other types of EPS forecasts, including a cross sectional analysis. Now if you want to a cross-sectional analysis that looks at several or many firms fundamental simultaneously, then your input features should be constructed differently. The reason is that firms of different size might have a very different values of fundamental account variables. They also might have different number of shares outstanding to make predictions of EPS comparable across different firms. We have to account for these differences, and one way to do this is to divide all raw input fundamental variables by the market capitalization of the firm. That would be equal to the product of a share price and the number of shares outstanding. Another useful feature to include instead of predictors for a cross-sectional analysis would be a categorical variable describing the industrial sector of the firm. You will be able to try such cross-sectional analysis of EPS or stock returns forecast in your homework for this week. Beyond the use for prediction of EPS or future values of other accounting variable, yet another use of these variables has to do with value investing. As what we've discussed for example by French and Fama in 1992 paper, under-valued stocks typical have a High-Book-to-Market Equity or B/M ratio, a High-Earnings-to-Price or E/P ratio, and a High-Flow-to-Price or C/P ratio. Related set of features for value investing was suggested by Piotroski in 2000 who suggested particular features shown on the right of this diagram as representatives of three broad categories of features. These categories include profitability, leverage and liquidity and operating efficiency. By picking nine specific predictors from these categories as shown here, the model then tries to get a representative view of a firm for investment purposes. Let's pause here for a moment to see what we have learned here, and then summarize this video right after that. Let's summarize what we covered in this video. We first spoke about the analysis of companies fundamentals, that is, the accounting information to gain insights about future performance of a company. As a part of this complex problem, we analyzed models that focus on predicting the next quarter earning per share, or EPS from the past accounting information. We also spoke about value investing and how machine learning methods can help for this task. In your homework for this week, you will have a few assignments where you will analyze the fundamental data for Dow Jones and S&P stocks using Python and TensorFlow. And this will end our first encounter with regression machine learning tasks in finance. We will have more to say about regression models in our next course in this specialization, but for now I would like to move on with our next topics.
