Assignment 3

  1. 1

    Change of variables: Income-Consumption model. The model Consumption=β0+β1 GDP+u\text{Consumption} = \beta_0 + \beta_1\,\text{GDP} + u is estimated first for observations from Israel and then for a US dataset:

    Estimates Israel United States
    intercept β0\beta_0 −21.894-21.894 515.784515.784
    slope β1\beta_1 0.8150.815 0.6450.645

    Consumption and GDP in Israel are measured in millions of NIS, in the US in billions of dollars. Assume the exchange rate is 3.4 NIS to $1, and translate the slope and intercept from Israel into identical units to the US.

  2. 2

    Change of variables: Wage model. The model ln⁡(wage$)=β0+β1⋅education+u\ln(\text{wage}_{\$}) = \beta_0 + \beta_1 \cdot \text{education} + u is estimated for women who work, aged 25 to 65, where wage is in dollars and education in years. The estimates are β^0=2.497\hat\beta_0 = 2.497 and β^1=0.0788\hat\beta_1 = 0.0788.

    Translate the results from dollars to NIS: what is the estimate for the intercept and what is the estimate for the slope? (Hint: use ln⁡(cY)=ln⁡(c)+ln⁡(Y)\ln(cY) = \ln(c) + \ln(Y) and assume 3.4 NIS to 1 USD.)

  3. 3a

    Regression through the origin: the Income-consumption model revisited. A colleague estimates Consumption=β0+β1 GDP+u\text{Consumption} = \beta_0 + \beta_1\,\text{GDP} + u and sends you the R output (in the assignment PDF).

    She claims it is not reasonable for the intercept to be negative, since negative consumption is impossible, so she estimates a model without an intercept (a regression through the origin), Consumption=β1 GDP+u\text{Consumption} = \beta_1\,\text{GDP} + u, and sends you that output too.

    Is the intercept of the first model significant?

  4. 3b

    Given the estimated results, what is the correct model to use in this context?

  5. 4a

    Part B: the housing model. We model the housing market with model (1):

    pricei=β0+β1 bdrmsi+β2 lotsizei+ui\text{price}_i = \beta_0 + \beta_1\,\text{bdrms}_i + \beta_2\,\text{lotsize}_i + u_i

    using HPRICE1_v2026 (N=68N = 68 house sales). price is the sale price in USD, bdrms the number of bedrooms, lotsize the lot area in square feet, and lprice, llotsize the logs of price and lot size. Sample means: price ≈$291,074\approx \$291{,}074, bdrms ≈3.68\approx 3.68, lotsize ≈7,528\approx 7{,}528 sq ft.

    The classical assumptions. Write the four classical assumptions A1 to A4 in terms of this model.

  6. 4b

    What are some other variables that might affect housing price, aside from size and number of bedrooms?

  7. 4c

    For each assumption A1 to A4, explain in a sentence or two whether you think the assumption is reasonable in the context of housing prices. (Hint: any variables not included in the model but that affect price will be part of the random error uu.)

  8. 5a

    Model estimation I and II. Estimate model (1) using the lm command in R and write the results as a formula.

  9. 5b

    Your colleague points out that when the lot size and the number of bedrooms are zero, the price should also be zero. Test this hypothesis using the estimation results in 5a.

  10. 5c

    Estimate model (2) below (hint: in lm use the formula price ~ 0 + bdrms + lotsize):

    pricei=β1 bdrmsi+β2 lotsizei+ui\text{price}_i = \beta_1\,\text{bdrms}_i + \beta_2\,\text{lotsize}_i + u_i
  11. 6a

    Linearity of the model. Create a scatterplot between price and lot size (price on the y-axis).

  12. 6b

    Create a scatterplot between log price and log lot size (log price on the y-axis).

  13. 6c

    Which of the two relationships (6a and 6b) seems more linear?

  14. 7a

    For the remainder of the problem we consider model (3):

    log⁡(price)i=β0+β1 bdrmsi+β2 log⁡(lotsize)i+ui\log(\text{price})_i = \beta_0 + \beta_1\,\text{bdrms}_i + \beta_2\,\log(\text{lotsize})_i + u_i

    Model estimation III. Estimate model (3) above and report the results as a formula.

  15. 7b

    By how much does the log price of a house change if the log lot size increases by 4 (all else equal)?

  16. 7c

    By how much does the log price of an apartment change if the log lot size increases by 4 and the number of bedrooms increases by 2? Compare this to the result in 7b.

  17. 7d

    A house has 4 bedrooms and a lot size of 6,000. What is the predicted price of this house?

  18. 8a

    Confidence intervals. Calculate the 95% confidence interval for β2\beta_2 and explain in a sentence or two the economic interpretation for this range.

  19. 8b

    Calculate the 90% and 99% confidence intervals for β2\beta_2 and, in a couple of sentences, explain which one is wider and why. Make sure to relate your answer to the definition of a confidence interval.

  20. 9a

    Hypothesis testing. Test the hypothesis that increasing the log lot size by 1 (holding the number of bedrooms fixed) will not change the house price (at 5% significance).

  21. 9b

    What is the lowest significance level for which you would reject the null hypothesis in 9a?

  22. 9c

    Test the hypothesis (at 5% significance) that increasing the number of bedrooms by one increases the log price by 1.

  23. 10a

    Change of variables. Suppose you want to translate the lot size from square feet to square metres. What will be the value of β2\beta_2 in the new model where lot size is measured in square metres? (Hint: 1 ft2=0.093 m21\ \text{ft}^2 = 0.093\ \text{m}^2.)

  24. 10b

    Confirm your answer in R: create a new variable llotsize.m which measures the log lot size in metres squared. Estimate model (3) using this new variable. Compare the results to the estimation in 7a.

Solutions are Alfie's submitted work, not an official key.

Related: Lec_08-Goodness of Fit & Multicollinearity