Machine Learning For Trading Exam 1 Flashcards
What is Machine Learning according to Tom Mitchel ?
Tom Mitchel defines Machine Learning as the study of systems that learn from experience (E) given some set of Tasks (T) evaluated on some performance measure (P). He asserts that such systems are said to learn from experience if their performance on said Tasks, improves with experience (E).
Why is the probablistic model adopted for studying and designing ML systems ?
- Because it provides an optimal way to make decisions under uncertainty
- It is a shared domain language that cuts across the sciences
What is Classification ?
Classification is a supervised learning technique where, predictions are made over an unordered set of mutually exclusive labels called classes. These classes represent datapoints grouped together based on some characteristics.
Why is it necessary to carry out an Exploratory Data Analysis before designing a machine learning model and what methods are available for doing such ?
- Because it helps reveal patterns in data which informs what machine learning method to use
- It is useful for spotting potential problems with the data like (label bias or outliers).
Explain Supervised and Unsupervised learning and how they differ
Supervised learning is concern with learning a mapping f from a set of inputs x ∈ X to a set of outputs y ∈ Y where x is called the feature, covariate or predictor and y is called the response, label or target. It assumes that for every input featuture x there is a corresponding target value y that has predictive value.
Unsupervised learning on the other hand is concerned with the learning patterns in data without any labelled outputs. This makes is less time consuming and effective since it skips the overhead of labelling targets. On the other hand it is memory inefficient when compared to supervised learning because of the need to store large amounts of training data from which it makes predictions from.
What is Regression and explain how it is different from Classification
Regression is a statisitical learning technique used to predict quantitative values. Classification is a technique that makes qualitiative or categorical predictions. Data is grouped into cateogries or classes based on common characteristics.
What is Clustering ?
Clustering is the process of grouping/partitioning datapoints into groups called clusters.
Explain what the Sharpe Ratio is and what it is used for
The Sharpe Ratio is a performance metric that compares a portfolios return with its risk. Its also called the risk adjusted return. It is calculated as the expected value (mean) of the portfolio’s return minus the risk-free rate divided by the standard deviation of the portfolio’s return.
What are Beta and Alpha in the CAPM formula
Both alpha and beta are performance measures used ing CAPM. Beta measures the volatility of a portfolio relative to the market.
Alpha on the other hand measures the excess returns on a portfolio after adjusting for market risks and other fluctuations. It tells how better or worse a portfolio performed relative to its beta predictions.
What are the assumptions of the Efficient Market Hypothesis
- There are large number of investors who are incentivized to seek profit. That is to say they are constantly seeking opportunities where the price of stock is not reflective of its true value.
- Information arrives randomly.
- Prices adjust quickly
- The current price of a stock reflects all available information
In the context of EMH information comes from a variety of sources. Can you name some sources
- Price/Volume
- Fundamentals (earnings, assets, liabilities etc)
- Exogenuous (having an external cause or origin)
- Company Insiders
What are the forms of the EMH and what do they prohibit ?
- Weak form: Future asset prices cannot be predicted from historical price and volume data. This suggests that it if the weak form is correct, Technical analysis isn’t possible because technical anaysis relies on price and volume data.
- Semi-Strong form: Asset prices adjust immediately to all public information. Including that which may be available through a compay’s fundamentals like (financial disclosure)
- Strong EMH: Asset prices adjust to refelect all relevant information include that available to insiders.
The weak form if correct prohibits technical analysis
The semi-strong form if correct prohibits both technical analysis and fundamental analysis
The strong form if correct prohibits technical analysis, fundamental analysis and insider information
What is an efficient market ?
An efficient market is one where information that can affect the price of a stock travels rapidly and causes the prices to adjust accordingly.
What is the adjusted price of a stock ?
The adjusted price of a stock is the price after factoring corporate actions like stock splits and dividends issued.
What is a stock split ?
A stock split is when a company increases the amout of shares by spliting existing shares in a bid to reduce the price per share. The total value of all shares remains the same at a reduced price per share but more shares are made available as a result of the split.
What are dividends ?
Income earned on a stock without selling it. Divideds are paid from the company’s earnings. When a company pays dividends, the price of the stock is reduced, this is because the company makes these payments from its holdings and by doing so reduces its assets.
What is the divided yield and how is it calculated ?
The income generating potential of a stock. It is measured as the amout of dividends per share divided by the share price.
Briefly explain reinforcement learning and how it differs from supervised learning
Reinforcement learning solves problems where an agent learns from its environment. The learning is encoded in the form of a policy which specifies what action an agent should take for every possible input from its environment.
It differs from supervised learning in that the agent is not provided any ground truth (labelled data) but an occasional reward or penalty signal for each action the agent takes.
Key downsides to this approach are
* The agent isn’t told what is the best action to take, It just gets occasional signals on if the action it took was good or bad.
* It is difficult to tell which inputs let to an action
Explain what a portfolio is
A portfolio is a weighted sum of assets. The sum of all weights must equal 1 or 100%. However for leveraged portoflios, this sum can exceed 1.
What is the Market Capitalization of a company ?
The market capitalization of a company is the number of shares outstanding multiplied by the share price.
Allocations in a portoflio are usually cap weighted, How are these weights calculated ?
Weights are calculated by dividing the market capitalization of the asset by the total capitalization of all assets in the portfolio
What does CAPM say about α ?
The expected value of α is always zero. CAPM stipulates that the return on a stock is dependent on the market alone.
