ERP Lecture 3

5 flashcards · Other · by Daniel F.

  1. Assessing your model’s goodness of fit

    Coefficient of determination R2 Proportion of the variation in 𝑦 that is explained by the linear combination of the 𝑥 variables Between 0 (no prediction) and 1 (perfect prediction 100% ) Adj. R2 R2 increases as we keep adding variables Modified measure of R2 that takes into account the number of independent variables and the sample size punishment for the inclusion of additional explanatory variables When comparing between models F-test test for the overall significance of a model Tests wheth…

  2. Assumtions of OLS

    Linearity (Y linear in X params) Zero mean errors (no X-ε correlation) No perfect multicollinearity (X not perfectly correlated) Constant error variance (homoscedasticity) No autocorrelation (errors independent) Normality (errors ~ normal, optional for inference) Random sampling (iid data) No extreme outliers Violations of these assumptions can lead to: Biased or inconsistent estimates of the regression coefficients. Incorrect standard errors, which can lead to invalid hypothesis tests. Ineffic…

  3. Outliers

    Data point of extreme values that does not follow the general trend of the data violates OLS assumptions: changes fit and slope

  4. How to detect outliers

    SD rule: outliers are outside 2-3 sd around mean quick and dirty Assumes normal distribution (often invalid for financial data) Sensitive to extreme values (outliers affect standard deviation calculation) Uses arbitrary threshold (3 std dev is convention, not theoretically justified) Doesn't account for expected skewness in financial data May identify too many or too few outliers depending on distribution shape IQE rule: outliers are outside Q1-1.5xIQR and Q3+1.5xIQR less affected by skewed dis…

  5. How to solve outliers

    does not change the general relationship, but affects conclusions: if dropped, should be acknowledged changes the general relationship, and affects conclusions: run with and without, acknowledge relationship between X and Y is clearly created by the outlier: drop the outlier, acknowledge If data error: Correct if possible Remove if correction impossible If valid observation: If theoretically important: Keep and possibly analyze separately If not central to theory: Transform data (log transforma…