my-markdown-library 0.1.0

data/F24LS_md/Lecture 11 - Econometrics.md

@@ -0,0 +1,345 @@
+---
+title: "Lecture 11 - Econometrics"
+type: slides
+week: 11
+source_path: "/Users/ericvandusen/Documents/Data88E-ForTraining/F24LS/Lecture 11 - Econometrics.pptx"
+---
+
+## Slide 1: Lecture 11: Econometrics
+
+- Lecture 11: Econometrics
+- Data 88E: Economic Models
+
+## Slide 2: Announcements
+
+- Project 4 will be released - Econometrics
+- No Lab
+
+## Slide 3: Academic Dishonesty and Attendance
+
+- Using ChatGPT to paste in code is dishonesty
+
+## Slide 4: Follow-ups
+
+- Last time I ranted about Facebook 2016 and 2020 - its in court today!!!
+- Did Facebook lie to shareholders about knowledge of risks?
+- Brent Kavanaugh’s best friend Joel Kaplan is Facebook Policy head?
+
+## Slide 5: Fed Meeting tomorrow!
+
+- What is forecast for tomorrow?
+- How might the election affect the Fed decisions
+
+## Slide 6: (untitled)
+
+- https://www.spglobal.com/marketintelligence/en/news-insights/latest-news-headlines/fed-s-autonomy-could-be-at-risk-if-trump-wins-in-november-85804211
+
+## Slide 7: Project 2025 has a chapter on the Fed - Chapter 24!
+
+- Do away with Dual Mandate
+- https://static.project2025.org/2025\_MandateForLeadership\_CHAPTER-24.pdf
+
+## Slide 8: Today’s class
+
+- Intro to econometrics
+- Intro to regression
+- Simple linear regression
+- Multiple linear regression
+- Dummy variables
+- Reading econometrics tables
+- Project demo
+- Upper division econometrics classes
+
+## Slide 9: Memes of the day
+
+## Slide 10: Intro to econometrics
+
+- An important part of economics is understanding the relationship between variables, especially cause-and-effect relationships
+  - How does price affect the quantity demanded?
+  - How does income tax affect consumption?
+  - How does schooling affect income?
+- Ideally, we would do randomized controlled trials to see if one variable (independent variable) causes another (dependent variable): gold standard for establishing causality
+- But we can’t usually conduct experiments in economics, the best we have is observational data
+  - What do you think are some challenges with conducting experiments in economics?
+- This is why we use econometrics: we use statistical techniques (like regression) to model relationships between economic variables
+
+## Slide 11: Intro to regression
+
+- Really important tool in econometrics
+- Used for modelling relationships between variables
+  - E.g. relationship between price and quantity demanded
+- 3 main purposes of regression:
+  - Describing associations: What kind of association do x and y have? Positive/negative? Strong/weak?
+  - Prediction: E.g. by how much will quantity demanded decrease if price increases by $1?
+  - Causal inference: Does x cause y or is it just a correlation?
+
+## Slide 12: Simple linear regression
+
+- You hypothesize that there’s some association between two variables – say price and quantity demanded
+- Start by creating a scatter plot to visualize the association
+  - This gives you an intuition for the association
+  - Gives you a sense for whether the association is positive/negative, strong/weak/none
+  - There is some randomness: e.g. in the example, y doesn’t always increase as x increases (otherwise the points would all lie on a straight line)
+
+## Slide 13: Simple linear regression
+
+- You can quantify this association by calculating the correlation coefficient or r (DATA 8 review!)
+- Review of correlation coefficient:
+  - Measure of linear association (won’t work for nonlinear association), i.e. slope is constant with respect to x
+  - Value between -1 and 1
+  - The sign tells you the direction of the association: positive or negative
+  - Take absolute value of r to determine strength of association
+  - 0 means no correlation
+  - Correlation is not causation!
+
+## Slide 14: Simple linear regression
+
+- How to calculate the correlation coefficient:
+  - Step 1: Convert data to standard units
+    - Express each value in terms of the number of standard deviations it is from the mean (“distance” from the mean)
+  - Step 2: Take the mean of product of x and y
+
+## Slide 15: Simple linear regression
+
+- We can draw a line of best fit through the data: the line that most accurately models the relationship between x and y
+- Recall that this line has the form y = slope \* x + intercept
+- We can use the correlation coefficient to find the slope and intercept
+  - Slope:
+  - Intercept:
+  - Equation of the line of best fit:                                 → Regression equation
+- A note notation: the hats indicate that the values are estimates (rather than actual values)
+- (fact: because average of x and y are always on the regression line)
+
+## Slide 16: Simple linear regression
+
+- Can plot the regression line (the line of best fit) using the equation we found
+- This is the same as the line of best fit you get by using a function you already looked at: np.polyfit
+- Not all the points lie on the regression line because of
+  - Random variation
+  - The model is not a perfect fit (not perfectly accurate)
+- Note: assuming all assumptions of doing linear regression are satisfied
+
+## Slide 17: Avocado Demand in Week 2?   np.polyfit(X,Y,1)
+
+## Slide 18: polyfit
+
+## Slide 19: Simple linear regression
+
+- Use the regression equation to generate predictions for y
+- Once we’ve generated predictions, we would be interested in knowing how accurate they are
+- We can calculate root mean squared error (RMSE) to determine accuracy:
+  - Calculate residuals → Square them → Take the average → Take the square root
+
+## Slide 20: Simple linear regression
+
+- RMSE: on average, how far are your predictions from the actual values?
+  - Think about whether your value for RMSE is a large number
+  - Most useful for comparing across models
+- Regression minimizes the RMSE of your data: of all the lines you can draw through your points, the regression line has the lowest RMSE
+  - This is why it’s called least squares (OLS) regression
+  - We can use the minimize() function to see this
+- Numpy minimizes RMSE when it’s calculating the slope and intercept (e.g. when you use np.polyfit)
+
+## Slide 21: Simple linear regression
+
+- So far: DATA 8 version of regression
+- We use the statsmodels library to do regression in Python
+  - First, import statsmodels.api as sm
+  - Code for regression:
+  - To get the coefficients (intercept, slope): result.params
+
+## Slide 22: (untitled)
+
+## Slide 23: Simple linear regression
+
+- Always include intercept term in your model (that is, don’t forget sm.add\_constant)
+  - Regression line passes through point of averages (property of the regression line)
+  - Best prediction for the average of x is the average of y
+  - Including the intercept term makes sure that your regression line passes through the point of averages
+  - Also, can’t generally assume that your model doesn’t have a y-intercept
+
+## Slide 24: Simple linear regression
+
+- Important things to focus on in the regression output:
+  - Slope (𝛃) : By how much does y increase/decrease when you increase x by 1 unit?
+  - Intercept (𝛂) : What is the value of y when x = 0?
+    - Not always meaningful (e.g. what are earnings when height is 0 inches?)
+  - Confidence interval: Is there a statistically significant association between x and y? (H0: 𝛃 = 0, H1: 𝛃 ≠ 0)
+
+## Slide 25: Simple linear regression
+
+- Confidence interval on coefficient estimates:
+  - Related to hypothesis testing:
+    - Null hypothesis: No association between x and y  (𝛃=0)
+    - Alternative hypothesis: There is an association between x and y (𝛃≠0)
+  - Regression output gives you 95% CI → test hypotheses at the 5% significance level
+    - If CI contains 0: evidence for the null
+    - If CI doesn’t contain 0: evidence against the null
+  - Can simulate using bootstrapping
+    - Center of the CI: regression slope
+
+## Slide 26: Regression Output ( Statsmodels)
+
+## Slide 27: Regression Output ( SM vs R vs Stata)
+
+- Python (statsmodels)
+- R - LM
+- Stata
+
+## Slide 28: Regression Output ( Statsmodels)
+
+- This is the amount of variation explained by the model
+- Number of Observations - reality check on size of data  N matters!
+- coef = 𝛃 = magnitude of estimated coefficient
+- std err = variability of estimate of coefficient
+- t=  a t-test testing whether 𝛃 = 0
+- P>[t]=  probability of that t-test
+
+## Slide 29: Jump to Notebook 1
+
+## Slide 30: Interactive Demo
+
+## Slide 31: (untitled)
+
+## Slide 32: (untitled)
+
+## Slide 33: Multiple linear regression
+
+- These variables that affect the dependent variable and are correlated with the independent variable that are not included in the model are called omitted variables
+  - E.g. your socioeconomic background affects your earnings and is positively correlated with how much education you get
+- They cause omitted variable bias: cause your estimate of the regression slope to be different from the actual slope
+  - What would a positive vs. negative value for omitted variable bias mean?
+- Can use multiple linear regression to account for omitted variables (really important purpose of regression!)
+
+## Slide 34: Multiple linear regression
+
+- These variables that affect the dependent variable and are correlated with the independent variable that are not included in the model are called omitted variables
+  - E.g. your socioeconomic background affects your earnings and is positively correlated with how much education you get
+- They cause omitted variable bias: cause your estimate of the regression slope to be different from the actual slope
+- Can use multiple linear regression to account for omitted variables
+
+## Slide 35: Multiple linear regression
+
+- Using statsmodels: just select multiple columns when defining the x-variable
+  - Multiple regression slopes: each independent variable has a slope
+- Compare slope on independent variable of interest from simple and multiple linear regression to see if it’s overstated or understated
+- Example of a MLR model:
+
+## Slide 36: Multiple linear regression
+
+- Mathematical intuition:
+- Each slope is the partial effect of the corresponding independent variable on y: effect holding all other independent variables constant
+
+## Slide 37: Multiple linear regression
+
+- Graphically with 2 independent variables: 3D plot with a regression plane (not line)
+- Y
+- X1 - continuous variable
+- X2 - dummy variable
+- Can only be 0,1
+
+## Slide 38: Dummy variables
+
+- Dummy variables are variables that take on a value of either 0 or 1 to indicate the presence or absence in a category
+  - E.g. smoker or non-smoker, went to college or didn’t go to college
+- Also known as indicator variables
+- Mutually exclusive: you can only be in one of the categories (you either went to college or you didn’t – not both)
+- Collectively exhaustive: you must be in one of the categories (only 2 possible scenarios)
+
+## Slide 39: Dummy variables
+
+- Example of regression model with dummy variable:
+  - col is a dummy variable indicating whether or not the person went to college (0 = didn’t go to college)
+  - Coefficient on col → difference of mean earnings when col = 1 and col = 0
+
+## Slide 40: Dummy variables
+
+- Beware of the dummy variable trap
+- Say you include all possible variables for a dummy variable
+  - E.g. col (for whether or not you went to college) and notcol (for whether or not you didn’t go to college)
+  - Example:
+
+## Slide 41: Dummy variables
+
+- There are infinite solutions for the coefficients (slopes and intercepts)
+- This is because the independent variable in the model have a perfect correlation – perfect multicollinearity
+- Happens any time one independent variable is a linear combination of another
+  - E.g. if you have age and schooling in your model and age = schooling + 3
+- Use pd.get\_dummies() to convert categorical variables to dummy variables
+
+## Slide 42: Jump to Lec NB2
+
+## Slide 43: Reading econometrics tables
+
+- In papers, the results of econometric analysis are summarized in regression tables
+- Example of regression table from project 4:
+
+## Slide 44: Reading econometrics tables
+
+- In papers, the results of econometric analysis are summarized in regression tables
+- Example of regression table from project 4:
+- Filtering Variable
+- X1
+- X2
+- X3,X4,X5…
+- Y
+- n
+
+## Slide 45: Reading econometrics tables
+
+- Always look at how the dependent variable is being measured: in the example, the authors are taking log earnings
+  - Recall what you learned about semi-log demand curves: interpreted as % change in y per unit change in x
+- Look at what is being controlled for, e.g.
+  - They are estimating separate models for men and women so they are controlling for gender
+  - In the second column for each group, they are controlling for test scores
+
+## Slide 46: Project demo
+
+- In this project, you will use regression to analyze the relationship between a person’s height and earnings
+- Based on a study conducted by Anne Case and Christina Paxson (very interesting study!)
+- Divided into 3 parts: simple linear regression, multiple linear regression and reading econometrics tables
+
+## Slide 47: Upper division econometrics classes
+
+- ECON 140: Economic Statistics and Econometrics
+  - Edwards - in R in Fall
+- ECON 141: Econometric Analysis (with Linear Algebra)
+- ECON 142: Applied Econometrics and Public Policy
+- ECON 143: Econometrics: Advanced Methods and Applications ( New)
+- ECON 144: Financial Econometrics
+- STAT 153: Time Series
+- ECON 148:  Data Science for Economists
+- Learn more here: http://guide.berkeley.edu/courses/econ/
+
+## Slide 48: Aside - Econometrics vs ML
+
+- The idea of Econometrics is that the model has an underlying structure
+- Economic Theory gives us a reason to structure the model
+- We seek to explain the effect of X on Y
+- We also need to hold multiple variables constant
+- We will adapt a lot of techniques to get an unbiased estimator
+- OLS is the starting point - variations depart from there
+- Machine Learning - we don’t need to have an underlying model
+- Whatever does the best job in prediction! Or modeling, or classification...
+- ML has many models that can inform econometrics!
+- Random Forest, Network Graph theory, nonparametric approaches
+- Neural Networks
+
+## Slide 49: Terminology - an aside econometrics vs ML
+
+- Dependent Variable ( Y)  ~  predictor and predicted values
+- Independent Variable ( X)   ~ regressors, explanatory variables     ~ coefficients, betas
+- X  in the ML world might be called “model features”
+- Y in the ML world might be called “target”
+- 0-1 variables when  they are in the Y variable - call for a different class of models- especially Logit model - in ML this would be a classifier model
+- 0-1 dummy variables - in the X variable would be  called “one hot encoding” of categorical variables
+- ML - split up your datasetTraining data - used for training the modelTesting data - used for measuring the accuracy of your model
+
+## Slide 50: Matrix notation for Linear Regression
+
+- https://online.stat.psu.edu/stat462/node/132/
+- Formula for Data Model
+- Formula for OLS
+
+## Slide 51: SKlearn
+