Homework 10
Group 2: James Ottaway, James Pardy, Ed Mullen, Matthew Kourlinins
13 October 2015
Group2 Week10 Ed Mullen (25%), James Ottaway (25%), James Pardy (25%), Matthew Kourlinis (25%)
1. Draw a causal diagram including an outcome variable, a treatment variable, an observed confounding variable (control variable), an unobserved confounding variable, an instrumental variable & a bad control variable.
2. If I want to estimate the (linear) causal relationship between the treatment variable & an outcome variable under the assumption there are no unobserved confounders, what linear regression model should I estimate?
\(Outcome = β0 + β1(treatment variable) + Xβ2 + ε\) , where what we are trying to determine is the effect the treatment variable has on outcome (β1), β0 is some initial or current state of the outcome, X represents the control variables in the regression & B2 represents their effect on the outcome variable & ε simply represents everything else (it is an error term for the regression & includes all the unobserved variables).
3. Suppose I am concerned that there are unobserved confounders that affect both the treatment & the outcome, & as such I can’t assume the relationship between them is causal. How should I go about identifying the causal effect?
By introducing an instrumental variable that affects the treatment variable & through the treatment variable the outcome, but not the outcome directly.
Then we must ensure that the expected value of the treatment variable given some level of the instrumental variable replaces the treatment variable. This causes the link between the unobserved confounders (which directly affect treatment & outcome) & the treatment variable being analysed to be severed (zero). While the treatment may be have unobserved confounders, the instrument will not.
As of this point the unobserved confounders are not influencing the expected treatment variable & hence \(Cov(Δtreatmenti,t+1 , εi,t+1) = 0.\) We can therefore establish causation between the treatment variable & the outcome by varying the instrument variable (natural experiment) we have selected.
4. Describe the 2 properties that an instrumental variable must have. What is the exclusion restriction? What is the relationship between an instrumental variable & a natural experiment?
An instrumental variable must be: observable & quasi-random. Observable just means that we can observe the instrumental variable and its impacts on outcome. Quasi-random is when the treatment variable has been randomised from the point of view of the recipients & the relationship between the unobserved confounders & this treatment variable has been nullified. It avoids being biased in its distribution. This permits us to draw causation between the treatment variable & the outcome.
However, an exclusion restriction must be met: treatment & control groups must also be comparable among all dimensions other than the treatment variable. That way the effects shown are caused from the variation in the treatment variable. The exclusion restriction is that the instrumental variable should not impact on the outcome when the treatment is held constant.
A natural experiment is an event that has occurred in the real world available for an economist’s use to ascertain the effects of certain variables or the presence of a particular phenomenon. A real world event suitable for observation, an instrumental variable such as tax cuts or government pay-outs can be observed and so can changes in outcomes, this allows for big data from which a regression model can be developed. They’re a useful substitute for macroeconomists as they cannot carry out field or laboratory experiments akin to micro-economists, they are dealing with much larger concepts with the economy at large. Carrying out these experiments would be extremely difficult to achieve in practice due to their cost & widespread consequences. However, the data & information of natural experiments often needs to be treated so that it is appropriate for such use. After this treatment has taken place it can be used as an instrumental variable to achieve the desired results we have outlined above.
5. Matthew Kourlinis, reunification of Germany, The response of household saving to the large shock of German Reunification by Nicola Fuchs-Schündeln (2008).
Describe a natural experiments a researcher has used to examine the PIH.
The natural experiment I have chosen relates to the coming together of East & West Germany through reunification, East Germany left their communist ways behind in this endeavour. It is a large positive permanent income shock where East Germans responded to the event according to the permanent income hypothesis. ‘The response of household saving to the large shock of German Reunification’ by Nicola Fuchs-Schündeln (2008) is the primary research topic of my discussion.
The permanent Income hypothesis states that people are forward-looking and rational as they consume in response to their level of current assets & permanent life income over their life-cycle. It also states that the response to an event takes place at its announcement not its implementation, crippling liquidity constraints may limit this in practice. Therefore, only unexpected shocks to income affect consumption, those that are anticipated have already been accounted for by the recipients.
What data is used?
The data used is from the German Socio-Economic Panel (GSOEP), German Central Bank & ‘two additional German data sets, namely the Income and Expenditure Survey (EVS), and the Microcensus’ around the relevant periods prior, during & after the German reunification in ‘October 3 1990’.
What’s the natural experiment?
How do we know it’s a natural experiment? A natural experiment has two characteristics: observable and quasi-ransom. Let’s analyse if these two characteristics are present.
It is a historical event that is observable with sufficient data available to make conclusions & draw assumptions, note the data needed to be treated for suitability purposes. It was something that we saw occur in the real world, it was not consciously designed in a laboratory experiment.
East & West Germans were also considered to be quite similar in terms of their characteristics as people (Alesina and Fuchs-Schündeln (2007)), they were just living under different conditions (communism & capitalism respectively). With both groups effectively consisting of very similar persons we can see that the population of Germany has been randomised through the border separating the East & the West (Alesina and Fuchs-Schündeln (2007)). Both groups are comparable among all dimensions other than the treatment variable. The control group (West Germany) & the treatment group (East Germany) are compared against each other as we vary the treatment variable & all effects are caused by the change in the treatment variable. The event was also unexpected in nature.
We can therefore conclude that the conditions for a natural experiment has been met, it is observable & quasi-random. The treatment, what they’re interested in the effects of, was the change from communism to capitalism.
What are the exclusion restrictions & do you believe them?
When analysing precautionary saving she has chosen to exclude the presence of liquidity constraints, as I have mentioned liquidity constraints may force some people to wait until the implementation of a policy before they are capable of changing their behaviour. It is rational but still contradicts some of the findings of the permanent income hypothesis. By excluding such a factor that is present with positive shocks such as reunification of Germany she is somewhat biasing her results in favour of the permanent income hypothesis but her overall conclusions in the section are still valid in my opinion (this bias doesn’t render them useless), liquidity constraints would make the model much more complex and difficult to create. It’s reasonable for simplification but not perfect.
When assessing age differences she uses the birth year of the head of the household but excludes those that have retired, a reasonable assumption as we are primarily concerned with the working population here. What they do with their income as it receives an unexpected shock.
‘Since I can observe the empirical savings rate only from 1992 on, I calibrate and simulate the model from that year on’, this appears to be a reasonable approach in her statistical analysis.
What are the findings?
They found that this large permanent income shock caused a consumption re-optimisation procedure to initiate for the East Germans (treatment group) as their expected permanent incomes increased. This is in accordance with the concept of the permanent income hypothesis, the East Germans were acting rationally by determining the effect of emerging from a centrally planned economy on their permanent lifetime income. They were adjusting consumption and other factors, smoothing them, as they were looking forward. This is in direct contrast to the Keynesian consumption theory which argues the opposite, people consume from their current income, not their lagged income.
As their current assets grew their precautionary saving decreased, over time they were catching up to the West Germans. This is not contrary to the permanent income hypothesis as it was occurring within the re-optimisation process. Assets generate income, as they become more well-informed of the true effects of capitalism on their assets & hence income-producing ability they were calibrating their consumption to their new optimal value. Note the fact that the communist state was no longer going to finance their retirement also created a retirement motive that increased savings among East Germans especially over the older demographic, this was analysed by the paper.
Fuchs-Schündeln (2008) also found that the change to capitalism had a positive effect on East Germany & assisted in closing the distant gap between the living standards and wealth of the East & West of Germany through the higher savings rates of the East Germans. Inequality was decreasing and living standards increasing (if you measuring living standards by GDP per capita which is commonplace).
5. James Ottaway: Agarwal and Qian 2014 “Consumption and Debt Response to Unanticipated Income Shocks: Evidence from a Natural Experiment in Singapore.” American Economic Review, 104(12): 4205-30.
What data is used?
Panel data of 180,000 consumers’ cards in Singapore. Cards are behind 30% of aggregate personal consumption in Singapore. The natural experiment was a cash-payout by the Singaporean government of US$1.17 billion, from $78-702 per person. It was announced unexpectedly in February 2011 and paid expectedly in April. Foreigners did not qualify for the payouts and account for 40% of the population.
What are the exclusion restrictions & do you believe them?
The exclusion restriction is that foreigners do not differ from Singaporeans in a manner which impacts on their consumption. This was done by restricting the foreigners to those similar to Singaporean backgrounds, dropping foreigners and testing for heterogeneity of the payout amount, restricting to those unaffected by other potential treatments (payouts). It appears their tests are rigorous, furthermore while Savage claimed there was an issue with age, they claim that “[foreigners] are well represented across age”.
What are the findings?
After the announcement, for each $1 received, 80c was spent for ten months. Pre-announcement there was no difference in consumption between locals and foreigners. Between the announcement and payout (2 months) for each $1 received, 15c was spent. Consumers spent using credit during the two months, then debit post-payout.
Q5 Ed Mullen. Paper Examined: Serrato & Wingender, ‘Estimating Local Fiscal Multipliers’, Stanford Institute for Economic Policy Research, 2014.
DATA USED
-The instrumental variable is the census ‘error of closure’ at the county level for all counties not in Alaska and Hawaii (over 3,000). The error of closure is the difference between the population estimate based on the previous census and the national growth rate, and the population found by the following census.
-Federal funding to localities (county based, as above). The source is the Consolidated Federal Funds Report (annual publication by census bureau), which provides estimates of funding at a county level. Lag is at least 2 years (due to census results not being available for approx. 2 years)
-Data on county personal income, salaries and wages and employment from the Bureau of Economic Analysis’ Regional Economic Information System (REIS)
-Data on employment and earnings from the Quarterly Census of Employment and Wages (QCEW), from the Bureau of Labour Statistics.
The last 2 are used to estimate spending growth.
WHAT IS THE NATURAL EXPERIMENT?
The assumption is that there is an exogenous shock when federal funding flowing to a locality is altered due to a changed estimate of the population after a census (changed government spending from a so called ‘Census Shock’). This can be used to identify the fiscal multiplier by observing the change in growth following the funding changes. This assumes that any changes in the growth trajectory can be attributed to the changes in federal funds flowing into the county (see exclusion restriction below).
EXCLUSION RESTRICTION
-That the ‘census shock’ leading to alterations in federal funding allocated to a county only affects local economic growth through changes in the amount of federal funds received. In short, that alterations to the growth rate in a county can be attributed to the changes in federal funds flowing in.
IS THE RESTRICTION VALID?
The restriction is appealing so long as the existing growth trajectory of the locality is fairly stable. If the trajectory is erratic, or affected by other significant shocks, then the restriction may not be realistic due to confounding variables.
FINDINGS
They estimate the fiscal multiplier to be, on average, 1.57, and posit that fiscal policy has a measurable impact on spending. This leads to the conclusion that fiscal policy can be effective at increasing spending and output.
James Pardy: How Much Did the 2009 Australian Fiscal Stimulus Boost Demand? Evidence from Household-Reported Spending Effects
Andrew Leigh
Leigh uses survey data from 1201 individuals collated by the Social Research Centre in Melbourne for ANU. The natural experiment was the fiscal response to the global financial crisis in 2008-09. Reduced taxes and transfers were varied but the most common in 2008 were; Pensioners $1400, Child payments $1000 per child. Followed by a ‘working Australians tax bonus’ in early 09 which was a one off $900 payment for all taxable incomes under $80,000 dollars the previous tax year. Collectively the two packages totalled just over $21 billion.
The exclusion restriction is that the MPC was not influenced by current income. Although the payments were means-tested which ensured those who had less disposable income were more likely to spend. The use of the survey data was meant to be a comparison to quarterly sales data to give a better understanding of the MPC with or without the payments. In some ways I agree with their exclusion restriction but in other ways I don’t. The one off payments would be spent if the individual didn’t have pressing financial constraints and also ignored warnings of a slowing economy.
Leigh found an MPC of 0.41-42, although argued that the real MPC and therefore fiscal multiplier would be even larger in the LR seeing as only 68% said they received the payment. The government estimate / aim was for 80%, Leigh argued that not all had received payments or had forgotten about payments due to lack of prompt in survey question. A noteworthy finding was that Labor voters were 50% likely to spend the payment while Liberals and Nationals were only 29 and 24% respectively. The higher MPC rate in Aus compared to the US, about 1 ½ to 2 times, was said to be because of the bullish outlook of the Aus public. Another reason was the increased liquidity of Australian households with most US households stuck under crippling debt. There was no general trend of spending across wage levels, under $40,00 had a higher propensity to spend that $40-100 which sat 10 points lower than the other wage levels. Leigh compares the impact of fiscal policy on household consumption. Estimating a $9 billion rise after the $8.8 billion in -08 and $4 billion after the $12 billion in payments in 2009. Disregarding the first statistic due to highly seasonal behaviour leading up to Christmas the second round of payments saw an MPC of $4/$12b=0.3. This assumes that second multiplier effects are minimal which is consistent with economic academia, unless you work for Clive Palmer.