Chapter 10 Regression, Multivariate, and Nonlinear Models Flashcards
Multiple Regression and What they Reveal About Alternative Asset Returns
Regression assumes:
- Error terms are normally distributed
- Error terms are not correlated with each other
- Error terms have the same variance throughout the data sample (homoskedastic).
Multiple Regression Model
Find values for alpahi, or Beta1 to BetaN to best fit data to a relationship (eg Famma French Equation)
Fama-French
Three factor model relies on a RISK factor, SIZE factor, and a VALUE factor.
Uses past returns on the market and a particular stock to find the alpha and Beta that best fit those ex post returns.
Multicollinearity
Exists when two or more independent variables are correlated.
Regression to allocate the explanatory power.
Stepwise Regression
Algorithim for choosing independent variables for a regression model by successively adding or removing variables based on t-test.
Outfitted Models
Regression models that include large numbers of independent variables that may fit the sample data but do not predict well going forward.
Nonlinear Exposure
Exposure refers to the change in the value of an asset when some other value changes.
Dummy Variables
Independent variables that are coded as 1 or 0. Provide a way to handle nonlinear exposure and, when combined with linear variables, replicate nonlinear exposures.
Conditional Correlation
Separate regression in that subsets of data are created and the correlations are calculated.
Two variables
Positive conditional correlation
Correlation from the subset of data observed market conditions.
Negative conditional correlation
Data observed in down market conditions
Rolling Window Analysis
Correlation on a subset of data, then drop the earliest data point and add the next available data point.
Style Analysis
Based on grouping funds by their investment strategies or styles.
Look-back option
Payoff that is based on the value of the underlying asset over a reference
