Scribd Logo
Upload

Welcome to Scribd!

  • 19 views

    Handout 9 IV2

    Uploaded by

    CHUA WEI JIN
    This document discusses using instrumental variables to estimate the demand for cigarettes. It uses state-level cigarette sales and price data, instrumenting price with sales tax rates. Controlling for income improves the model. Using two instruments, sales and cigarette taxes, yields more precise estimates. However, weak instruments can produce misleading results, so first-stage instrument significance must be checked.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Handout 9 IV2

    Uploaded by

    CHUA WEI JIN
    19 views18 pages
    This document discusses using instrumental variables to estimate the demand for cigarettes. It uses state-level cigarette sales and price data, instrumenting price with sales tax rates. Controlling for income improves the model. Using two instruments, sales and cigarette taxes, yields more precise estimates. However, weak instruments can produce misleading results, so first-stage instrument significance must be checked.

    Original Description:

    Econometrics - SMU

    Original Title

    handout_9_IV2

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses using instrumental variables to estimate the demand for cigarettes. It uses state-level cigarette sales and price data, instrumenting price with sales tax rates. Controlling for income improves the model. Using two instruments, sales and cigarette taxes, yields more precise estimates. However, weak instruments can produce misleading results, so first-stage instrument significance must be checked.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    19 views18 pages

    Handout 9 IV2

    Uploaded by

    CHUA WEI JIN
    This document discusses using instrumental variables to estimate the demand for cigarettes. It uses state-level cigarette sales and price data, instrumenting price with sales tax rates. Controlling for income improves the model. Using two instruments, sales and cigarette taxes, yields more precise estimates. However, weak instruments can produce misleading results, so first-stage instrument significance must be checked.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 18

    Handout 9: Instrumental variable II

    Yichong Zhang1

    1 School of Economics

    Singapore Management University

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 1 / 18


    Empirical Example: Demand for Cigarettes

    Policy debate over the health risks from smoking often revolves
    around government intervention.
    One policy option is to tax cigarettes heavily, so that smoking will be
    discouraged.
    But how big a tax hike is needed to make a dent in cigarette
    consumption?
    The answer depends on the elasticity of demand for cigarettes
    How much (by what %) will consumption decrease if price increases by
    1%?

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 2 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)

    Here, we will use 2SLS to estimate the elasticity of demand for


    cigarettes using state level data from 1995.
    Cigarette consumption (Qi ) is the number of packs of cigarettes sold
    per capita in each state.
    The price (Pi ) is the average real price per pack including all taxes.

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 3 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)

    Our model of demand is then

    ln Qi = β 0 + β 1 ln Pi + ui

    Since this is a model of demand (which is determined simultaneously


    with supply), OLS will be inconsistent (ln P will be correlated with u).
    To use 2SLS, we pick an instrument, SalesTax, which is the portion
    of the tax on cigarettes arising from the general sales tax (measured
    in dollars per pack).

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 4 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)

    Is it relevant (Cov (Zi , Xi ) 6= 0)?


    It seems good: taxes clearly increase the sales price.
    Is it exogenous (Cov (Zi , ui ) = 0)?
    Maybe.
    Is SalesTax correlated with Q only through P?
    That is, does it affect the demand for cigarettes only through price?
    Perhaps, but taxes can be affected by the same local conditions that
    influence demand (attitudes towards smoking, tobacco lobbyists, etc.).
    However, this is probably more of a problem for cigarette specific taxes
    than sales taxes.
    Also, wealthy states might not need such high taxes (an omitted
    variables problem).

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 5 / 18


    1 First Stage Regression (i.e. Xi = π0 + π1 Zi + vi ):

    ln
    d Pi = 4.62 + 0.031 SalesTaxi
    (0.03) (0.005)

    2 Second Stage Regression (i.e. Yi = β 0 + β 1 Xbi + ui ):

    Qi = 9.72 − 1.08ln
    lnd d Pi
    Econ 107 Handout 9 (SMU) Introduction to Econometrics 6 / 18
    Empirical Example: Demand for Cigarettes (Cont’d)

    Can we test relevance?


     2
    F = t2 = .031
    .005 = (6.35)2 = 38.44 > 10, which looks quite good.
    Can we test exogeneity?
    Not in this case (we are exactly identified).
    Also, you should not use the SE’s from the second stage regression,
    because they don’t account for the first stage.

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 7 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)

    We can estimate the two stages together (and get the correct HR
    standard errors) by using the TSLS command in EViews.
    Notice that you need to tell EViews which variable(s) are endogenous
    and what you will be using for the instrument(s).

    Qi = 9.72 − 1.08 ln Pi
    lnd
    (1.53) (0.32)

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 8 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)

    Since cigarettes are addictive, this seems like a very high elasticity (an
    increase in the price of 1% reduces consumption by 1.08%), but
    exogeneity might be a problem...
    We should also worry about omitted variables.
    We probably need to include more variables (some W ’s).

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 9 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)
    An obvious candidate for an omitted variable that’s possibly important
    is income. We’ll use the log of real per capita state income (ln Inci ).
    Cigarette demand probably depends on income,
    and it may also affect the tax structure (and level) used by the state.
    Here we will include it as a control: so it will appear in both stages!
    With one instrument (SalesTax), we are still exactly identified and
    the 2SLS estimates are:

    Qi = 9.43 − 1.14 ln Pi + 0.21 ln Inci


    lnd
    (1.26) (0.37) (0.31)

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 10 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)

    Relevance? You need to look at the first stage regression

    ln
    d Pi = 3.59 + 0.389 ln Inci + 0.027 SalesTaxi
    (0.173) (0.065) (0.004)

    .027 2
    Our first stage F = t 2 =
    
    .004 = 44.7 > 10, which looks good.
    Exogeneity?
    We still can’t test it (we’re still exactly identified).
    However, another candidate instrument is available: cigarette specific
    taxes (CigTax).
    Let’s see what happens when we use both.

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 11 / 18


    Empirical Example: Demand for Cigarettes (Cont’d)

    With two instruments (SalesTax, CigTax ), we are now over-identified


    and the 2SLS estimates are:

    Qi = 9.89 − 1.28 ln Pi + 0.28 ln Inci


    lnd
    (0.96) (0.24) (0.25)

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 12 / 18


    Relevance? Again, you need to look at the first stage regression

    ln
    d Pi = 4.10 + 0.108 ln Inci + 0.011 SalesTaxi + 0.009 CigTaxi
    (0.088) (0.040) (0.002) (0.001)

    We need to test the joint hypothesis that the coefficients on SalesTax


    & CigTax are both equal to zero.

    Our first stage F -stat ≈ 210 > 10, which looks very good.

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 13 / 18


    The Scary Regression (Weak Instruments)

    As we all now know, labor economists are always interested in


    estimating the effect of education on earnings.
    The problem is that education is correlated with an omitted variable,
    namely ability.
    Can we use IV to solve this problem?
    Well, what would our instrument be?

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 14 / 18


    The Scary Regression (Weak Instruments)

    Here’s a clever idea:


    Laws require students to stay in school until their 16th birthday.
    If you turn 16 in January, while you are in 10th grade you can drop out
    without completing 10th grade.
    If you turn 16 in July, you will have to complete 10th grade.
    So your date of birth is an exogenous instrument (it should have no
    direct effect on earnings, other than through more time in school).

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 15 / 18


    The Scary Regression (Weak Instruments)

    Angrist and Krueger (1991) used data from the US Census (with
    n = 329, 000) and instrumented for education using birth quarter
    (interacted with many other variables) in the wage regression

    ln Wagei = β 0 + β 1 educi + OtherStuff + ui

    They found a return to education of about 8%.


    This is actually higher than the estimate from OLS, which is surprising
    since we expect OLS to over-estimate the effect of education (why?)
    The t-stat on βb1 was also quite large (> 5).
    But is there a problem?

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 16 / 18


    The Scary Regression (Weak Instruments)

    Even though n = 329, 000 (a huge sample!), the F -stat from the first
    stage was less than 2, implying that their instruments were very weak.
    In fact, if you replace birth quarter with a randomly generated birth
    quarter (definitely irrelevant), you get nearly the same results.
    This was scary to labor economists since the 2SLS SEs computed
    using the real data suggested that the effect of education was
    precisely estimated, but so did the SEs from the fake data!

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 17 / 18


    Weak Instruments

    The Angrist and Krueger paper generated a long - and still evolving -
    literature on weak instruments
    The main lesson is that you should always check the significance of
    the instruments in the first stage
    You should also avoid the temptation to add lots of marginally
    significant instruments.

    Econ 107 Handout 9 (SMU) Introduction to Econometrics 18 / 18

    You might also like