12/11/2009 Final Exam Review # How to study For any research-related paper, study name of authors, the question they try to answer, the strategy they employ to find answer, and what they find For any overview paper (Sen about famines, Kreuger's rant on world bank/imf), just cover intro/first section/conclusion Know concepts/papers covered in lectures that weren't required readings diff in diff randomized trial instrumental variables # Techniques Want to measure impact of treatment vs. no treatment (y_i(1) - y_i(0)). Ideally, for each person we'd make two universes, where they do and don't take pills, and see outcome. But we can't give treatment to each person (D_i = 1 or D_i = 0) - In reality, we make two random groups, treat each, and compare averages. ( E[y_i(1) | D_i = 1] - E[y_i(0) | D_i = 0] ) = E[y_i(1) - y_i(0) | D_i = 1] + (E[y_i(0) | D_i = 1] - E[y_i(0) | D_i = 0]) ^-------counterfactual-------^ ^----------------observed---------------------------------^ - counterfactual wasn't actually measured, only the observed stuff. = treatment effect + selection bias - treatment effect is what we wanted to measure. So the goal now is to eliminate selection bias ( 3 ways to remove selection bias - Randomized evaluation - if I randomly assign people to groups, then selection bias should equal out between treatment and control. -> stratification allows you to explicitly control for variables such as age, gender, etc. So that helps eliminate selection bias and avoid putting all old people into one group. - Difference in Differences - Compare across multiple dimensions, and use different regions as counterfactuals for each other. e.g. measure treatment (potatoes) in different regions of the world. So compare not only treatment and control across time, but subtract from each other. So measure suitability of potatoes in france and ireland. Then over time when potatoes were introduced, subtract difference in outcome before and after in each country, and then subtract out france. So if ireland is suitable, and france is not, then ireland after - ireland before is affected by potatoes and industrial revolution. To subtract industrial revolution, we subtract france before and after potatoes, which we assume would have been the same as ireland other than potatoes. Estimates effect of treatment by assuming that nontreatment based changes over time are captured by the control group. Papers used: Bleakley. Duflo (indonesia school building paper). - Instrumental variable - Look at Duflo indonesia paper for this as well. Want to study y (wages) = \beta * x (schooling) + \epsilon But there may be other factors that improve schooling and wages, such as ability. That might get you more schooling and wages, independent of how much schooling you got. So call y = \beta * x + \epsilon (stage II) Then find some input z such that x = \pi * z + \eta (stage I). Make sure z affects y only by x, and through no other way. [ y = f(x(z)) != f(x,z) ] [ so dy/dz = f'(x) * x'(z) ] Reduced form: y = kz + \nu, where k = dy/dz = \beta * \pi -> solve for \beta, which is what you actually care about acemuglu: y = gdp/capita. x = life expectancy. z = mortality rate. -> constructed z for 15 diseases to be - if you have treatment for disease: frontier death rate---best one for that disease - if you have notreatment for disease: death rate in 1940 for your country -> sure that might not be reasonable to assume, but remember you're dividing k by pi at the end, so the craziness cancels out. Papers: acemuglu johnson - disease/mortality acemuglo johnson robinson - settler mortality to compare instutitions vs. geography duflo - indonesia schooling Way to criticize: see if you can think of any reason that y would be affected by z other than through x # Introduction PPP - adjust purchasing power in countries to see what $1 can buy in that country. Millenium development goals (MDG)---cut poverty in half. reduce child/mortality rates. cut aids. sustainability. global partnership. Dollar & Krey: growth is good for the poor because their share of the pie stays the same with growth. Equiproportional growth between entire country and bottom quintile. Issue: plot log Y vs log S + log Y. That reduces variance and shows high correlation perhaps meaninglessly. Sala-i-Martin: Use survey data to correct for gaps in data, but finds same results. Banerjee + Duflo: "Economic Lives of the Poor"---People spend 40-80% of money on food. As money increases, spend more on alcohol, etc. Look at slides on the topic to get a flavor. Nutrition-based Poverty Traps - y = income. h = health. t = today. - y_t = f(h_t). h_{t+1} = g(y_t) - y_{t+1} = f(g(y_t)), so y = f(g(y)). - If y_{t+1} vs. y_t is above y=x 45 degree line, then equilibrium will tend toward first time y=x. - If y_{t+1} vs. y_t is below y=x 45 degree line, then everyone will get weaker, and tend toward 0. - If it keeps criss-crossing, then all sections below line will get to next lowest crossover, and all sections above will go to next highest crossover. - potatoes will shift the whole curve up. for all wealth values, you can get more health, and thus income. Potatoes: do potatoes increase population growth? Yes, using differences in differences in introduction in old world vs. new world (1700s and after) across countries w/ different potato growth sustainability. Sen famine paper: Initially, food availaibility decline (burma taken by japanese so they burnt earth), cyclone, and crop failure due to fungus. But despite all of this, in 1943 after decline, there was still more than 1942. So what happened? Prices went up significantly, but rural wages didn't rise. This led to panic in poor, and hoarding in people who sell rice. So that leads to man-made famine. Bezley + Burchase: govts respond faster to media coverage of famines Stromburg: olympics cancel coverage of famine. So slow news days lead to more coverage of disaster, which leads to more aid. Increased income->more food, but not better food. You buy sweeter/fatter foods. Giffen goods: increased price = increased demand. income effect outweighs substitution effect. Externality: Things good for you have good effect on people other than you. Education has lots of externalities, think about these. Despite externalities of education, people still do child labor. - Udry paper: people don't discount the future value of child's education. That makes them get money now from kids. Not necessarily present biased unless they will regret this later. - Basio has two models: - (reading) households try to reach subsistence nutrition. Need kids to work to make up for lower wages. Firms get fined for child labor, so firms work that into lower child wage. But that means children have to work harder. - (slides) wages drop as more kids enter market. so now need to send more kids into market. so two equillibria: either everyone has to send kids to work, reducing wages, or no one sends them, and makes more money for it. Easterly: studies change in education and shows it doesn't lead to more growth later on. Two important papers: - duflo in inpres (indonesia). what are the returns to education? indonesia constructs schools. uses diff in diff in first stage. different areas get more schooling. finds return to education is 7.5% increase in wage for each year of schooling. - banerjee, cole, duflo, linden, ... (balsaki paper). randomized trials to see effect of computerized learning and remedial education. remedial education helped remedial students, but not good students. focus on program evaluation in general. Udaipur paper (banerjee + duflo): people make mistakes, and are sick. Acemulu + Johnson: impact of disease eradication on growth. frontier (best) mortality rate for any country that had cure, and your own 1940 mortality rate if your country didn't have the cure. predicted mortality only affects growth through life expectancy. find that gdp doesn't change, and gdp/capita goes down. Bleakley: studies malaria in americas, which affects morbidity. How does malaria eradication affect wages? Uses diff in diff. Time period: before and after DDT, with two factors: age and high/low malaria areas. Found eradication->higher wages in the young. ___ insecticide treated bednets: free bednets better than cost sharing because of carryover effect and increased demand (look at slides). Finds no significance difference in usage when free, but high drop in purchases when it costs, so free is better. This doesn't mean jeff sachs was right. External validity is questionable due to information campaign going on at the same time and sample size being low. Duflo kramer robinson---fertilizer: different versions of delivering fertilizer. look at slides discussing each of these. Farmers are time-inconsistent/present biased/hyperbolic discounting in how they react to getting option for fertilizer. Why does this imply time inconsistency? Bezley: summary of different kinds of finance options. Think about ROSCAs. Banerjee + Newman: credit-based poverty trap. credit constraint + large fixed cost investment to get into market + poverty so no money on hand. Duflo + . What happens when microfinance comes to town and offers credit. Use randomized evaluation as identification strategy to introduce microfinance in half of regions. Other institutions are also involved though, and there may be some dynamic game that the other banks will enter that affects how treatment and control neighborhoods react. Still, found that borrowing went up in treatment areas. Some effects on profit, and increase in purchase of durable goods (they are time-inconsistent---can't save easily to improve business). No education, health, and women empowerment outcomes. Daily grind: survey on people's spending. Burges + Pande paper: ??? Microinsurance: first time we see moral hazard (coverage makes you lazier) and adverse selection (people who want your insurance are the worst-off ones). Insurance schemes handle this by making cheaper plans w/ less coverage, so that people will self-select into correct group (risky/less risky). Time inconsistency/hyperbolic discounting. Ashraf, karlin, yin: randomize three groups: control, educational visit, and educational visit + savings product. sophisticated present-biased people want savings product, since they know they would otherwise spend money. Muya: survey paper on land titling. Acemoglu + johnson + robinson: do institutions or geography cause poverty. Intrumental variable: settler mortality. Settler mortality affects economic growth in the future through the institutions set up there, and not any other (geographic) factor. Settler mortality has high effect on gdp by way of set-up institutions, but not geography. WB + IMF: perhaps outdated, but still have a purpose Trade: comparative advantage [ between two countries ] + returns to scale firms within countries should invest in one industry, even if they compete with another country on a similar product. minimum efficient scale (smallest scale you need in order to compete). thus france germany italy all trade cars, kind of killing idea of comparative advantage. comparative advantage losers: producers in country that's worse at producing. think about winners/losers. Capital flows: FDI (having more than 10%) vs. portfolio investment. So FDI is either having significant vote, or as having a larger firm in multiple countries trading with itself. Benefits of FDI---better use of resources, competition, lower transaction costs, transfer of knowledge. Pros + cons. Aid: Big paper---easterly. Transformationalists vs. marginalists. Transformationalist thinks of multiple equilibria, and think of shoving you into good equilibrium. Marginalists---look for small changes to push you in right direction, but no big jump into new equillibria.