Exam 1 Flashcards
Equation for daily return
daily return = (today’s price / yesterday’s price) - 1
Equation for cumulative return
cumulative return = (today’s price / first day’s price) - 1
Kurtosis
Describes tails of a distribution.
Fat tails = positive kurtosis
Skinny tails = negative kurtosis
Equation for present value
PV = FV/(1 + IR)^i
and
PV = FV/DR where FV can be dividend, DR is discount rate
Sharpe Ratio equation
SR = mean(daily returns - daily risk factor) / std(daily returns)
Book value equation
Book value = total assets (ignoring intangibles) minus liabilities.
Intangibles include brand power, patents
Liabilities include loans
Symbols for different types of funds
ETF = 3 or 4 letters
Mutual funds = 5 letters
Hedge fund = long name like Berkshire Hathaway
Market cap
share price * number of shares outstanding
Intrinsic value
PV = FV/DR
Example: if dividend is 1 dollar, and there are 1 million shares outstanding, and the DR is 5%, it’s 1 million / .05. Intrinsic value would be 20,000,000.
CAPM
Equation:
r-sub-i(t) = B-sub-i * r-sub-m(t) + alpha-sub-i(t)
- Tells us that a significant portion of a stock’s return is due to the market
- Alpha is random and expected value is 0
- Beta is relationship of stock to market. Make money by picking stocks with high B in up markets and low B in down markets.
- CAPM and EMH combined say you can’t beat market
- APT = multiple B for different sectors
RMSE equation
square_root_of((sigma(yTest - yPredict)^2) / N)
Roughly how big are training and testing chunks in general
60% training, 40% testing
Lin Reg vs KNN vs DT (using correlation), what is fastest to slowest training time?
KNN, LR, DT
Lin Reg vs KNN vs DT (using correlation), what is fastest to slowest query time?
LR, DT, KNN
Lin Reg vs KNN vs DT (using correlation), what is best for space needed to save model?
LR
Lin Reg vs KNN vs DT (using correlation), what is fastest for adding new data?
KNN
What are the benefits of ensemble learners?
lower error
less overfitting
Boosting and bagging
Same learner is used, just with different parts of its data
What is more likely to overfit as m increases? Boosting (ADA) or bagging?
ADA Boost
Technically, boosting and bagging are _____
wrappers for existing methods
Fundamental analysis
estimate a company’s value. buy stock when below company’s value. looks at earnings, dividends, cash flow, book value, etc.
Technical analysis
Don’t care about value of company. Look only for patterns or trends in a stock’s data. Looks only at price and volume.
Momentum equation
momentum[t] = (price[t] / price[t-n]) - 1
Where:
n = number of days, like 5 or 10
result is usually between -.5 and .5 (50% loss, 50% gain)
Simple moving average
SMA[t] = (price[t] / price[t -n : t].mean()) -1
Bollinger Bands
BB[t] = (price[t] - sma[t]) / (2 * std[t])
This is to see where price is relative to Bands
Normalization
Puts different metrics on same scale from -1 to 1, with mean of 0.
To norm something:
normed = (values - mean) / values.std()
what does bid mean?
buy
what does ask mean?
sell
market order book must always give stock to seller/buyer that is _______
the best price for them (highest if seller, lowest if buyer)
what does it mean if there are more asks than bids in the order book?
more selling pressure = stock price likely to go down
