Outline of the course. Lecture 2 Geography

Transcription

1 Outline of the course Lecture 2 Geography Matti Sarvimäki History of Economic Growth and Crisis 27 February The Malthusian Era 2 Fundamental causes of growth 1 Geography and luck 1 Models of spatial distribution of production: unique vs. multiple equilibria? 2 Empirical example: Tennessee Valley Authority 3 Papers for Essays: Portage, Bombs 2 Culture 3 Institutions 3 Innovation and crises 4 Unleashing talent Matti Sarvimäki Economic History Geography 1 / 40 Geography as destiny? Population density 1500 and GDP per capita 1995 Acemoglu, Johnson, Robinson (2002, QJE): Reversal of Fortune 10 - CAA NZL The simplest version time-invariant geographic variables (climate, topography, disease environment...) affect work effort and productivity, transportation costs etc. More nuanced versions exist Diamond (1997): Eurasia became powerful due to suitable environmental factors that were amplified by positive feedback loops. [see the National Geography TV version(!) in YouTube] Matti Sarvimäki Economic History Geography 2 / 40 LO C) CHL sme O> ARG URY VEN ZAF COr A CsBRA?ci o - - a) INAM CU DZA TUN PRY AM SWZ PHL - ION SUR CPV MAR GUY "-BOL EGY. LSO sdn a 4fO KEN BEN 0 7 UGA NPL 0) ZMEM R3.-3 MLI ER MWI MOZ TZA SLE ETH Log Population Density in 1500 Among the former European colonies, there is a negative relationship between log population density in 1500 and income per capita today. Similar pattern exists for urbanization rates in 1500 (another proxy for income). They are robust to controlling for many things and for excluding the "neo-europes" from the sample. MEX

2 Introduction Unique vs. multiple equilibria Tennessee Valley Authority Papers for the essays Unique vs. multiple equilibria This lecture will be about economic geography 50% of jobs in 0.03% of the land area 98% of non-water cells empty largest cluster: 25% of jobs, 0.4% of land tiny clusters even in the countryside a good starting point to think about multiple equilibria instead of countries, we focus on locations within countries useful because culture and institutions vary less Multiple equilibria implies path dependence if it is a prominent feature of the world, understanding history is much more important than in a unique equilibrium world it will also have very different implications for policy Econ geography particularly interesting because fundamentals provide a powerful reason for unique equilibrium Next slide: an example of the pattern to be explained (spatial distribution of jobs in Finland in 2010 using m grid) Matti Sarvimäki Introduction Economic History Unique vs. multiple equilibria Geography Tennessee Valley Authority 4 / 40 Papers for the essays Multiple equilibria in spatial structure: an example Krugman (1991): Geography and Trade Introduction Distribution of manufacturing employment Manufacturing: can be produced in East, West or both production only in East/West transportation costs production in both fixed setup cost monopolistic competition (each firm produces own variety) East only 50/50 split West only A numerical example Geography Cost for a typical firm East Both West There are three equilibria. Suppose that all other manufacturing firms are already in the East (first row). If one firm now locates to East, it will pay the fixed costs 4 once and transportation cost of 3. If it locates in both, it will pay the fixed costs twice and no transportation costs. If it locates to west, it will pay 4+7=11. So, everyone locates in the East because everyone else locates in the East. The case for everyone locating in the West is symmetrical (last row). But if manufacturing is initially splitted 50/50, the cost minimizing startegy for everyone is to have two plants. Homework: what happens when tranportation costs change? Are the equilbria stable? splitted manufacturing: local demands 5 and 5 concentrated manufacuring: local demands 7 and 3 Economic History Papers for the essays Krugman (1991): Geography and Trade Agriculture divided 50/50 between East and West fixed-cost of opening a plant: 4 transportation cost per unit: 1 total demand for a variety: 10 60% of labor force farmers Tennessee Valley Authority Multiple equilibria in spatial structure: an example Imagine a country with two locations: East and West Matti Sarvimäki Unique vs. multiple equilibria 6 / 40 Matti Sarvimäki Economic History Geography 7 / 40

3 Transformation costs can take many forms Marshall (1890): The Principles of Economics Transformation costs can take many forms Marshall (1890) as summarizes by Ed Glaeser When an industry has thus chosen a locality for itself, it is likely to stay there long: so great are the advantages which people following the same skilled trade get from near neighbourhood to one another. The mysteries of the trade become no mysteries; but are as it were in the air [...] if one man starts a new idea, it is taken up by others and combined with suggestions of their own; and thus it becomes the source of further new ideas. And presently subsidiary trades grow up in the neighbourhood, supplying it with implements and materials, organizing its traffic [...] Employers are apt to resort to any place where they are likely to find a good choice of workers with the special skill which they require; while men seeking employment naturally go to places where there are many employers who need such skill as theirs Proximity decreases the cost of moving ideas goods people Matti Sarvimäki Economic History Geography 8 / 40 Matti Sarvimäki Economic History Geography 9 / 40 Implications of multiple equilibria Tennessee Valley Authority (TVA) Currently rich and poor areas were not destined to be so historical accidents matter Geography may work through favorable initial conditions e.g. most great cities located by a river One-off shocks may have long-term effects wars, natural disasters, Big Push development policies... Use TVA to test a version of the Big Push argument hypothesis: sufficiently large public investments may push the economy to a new equilibrium traditional BP models focus on consumption externalities KM s version works through agglomeration economies Roadmap description of TVA a reduced-form evaluation of local impacts structural model to assess national effects Matti Sarvimäki Economic History Geography 10 / 40 Matti Sarvimäki Economic History Geography 11 / 40

4 Tennessee Valley Authority (TVA) TVA service area TVA is a federally owned corporation created in 1933 Objective: modernize the Tennessee Valley s economy Area: 163 counties inc. entire Tennessee, and parts of Kentucky, Alabama, and Mississippi Large investments in public infrastructure projects including series of hydroelectric dams 650-mile navigation canal extensive road network construction of new schools and flood control systems Matti Sarvimäki Economic History Geography 12 / 40 Federal transfers: timeline Proposed Authorities M il l i o ns o f D ol l a rs $4,000 $3,500 $3,000 $2,500 $2,000 $1,500 $1,000 $500 $0 -$ Total Per Capita $400 $350 $300 $250 $200 $150 $0 -$50 D ol l a rs p er R es i $100 d e n t $50 73% of the federal transfers occured in In the early 1950s, per household transfer roughly 10% of average household income In 1959, TVA power generation system made self-financing federal subsidies declined sharply TVA was ment to be merely the first such programs Roosevelt: If we are successful here we can march on Senate bill 1937 proposed the creation of seven new authorities failed (even though were considered likely to pass) Congress in bills proposing programs comparable to the TVA all failed (even though were, again, considered likely to pass) Proposed authorities useful for identifying the impact of TVA credible counterfactuals because modeled on the TVA Matti Sarvimäki Economic History Geography 14 / 40 Matti Sarvimäki Economic History Geography 15 / 40

5 Data Pre-TVA averages (1930) KM create a county-level panel for aggregate from various microdata published tables from the Population Census, the Manufacturing Census, and the Agricultural Census topographic variables collected by Fishback, Haines, and Kantor (2007) Quality issues for early years substantial measurement error likely, particularly for wages TVA Non-TVA Non-TVA South Non-TVA proposed authorities 1930 characteristics Log population Log employment Log # of houses Log average manufacturing wage Manufacturing employment share Agricultural employment share % White % Urbanized % Illiterate % of Whites foreign born Log average farm value Log median housing value Log median contract rent % Own radio Max elevation (meters) 1, , , , Elevation range (max min) 1, , , % Counties in South Matti Sarvimäki Economic History Geography 16 / 40 TVA counties were significantly more agricultural, rural, and had lower manufacturing wages, housing values, agricultural land values, literacy rates, immigrants, radios (likely due to low income and the lack of electricity). Proposed authorities are more similar to the TVA counties than to the average U.S. county, but are clearly not equal. Pre-TVA trends ( ) TVA Non-TVA Non-TVA South Non-TVA proposed authorities Changes Log population Log employment Log # of houses Log average manufacturing wage Manufacturing employment share Agricultural employment share % White % Urbanized % Illiterate % of Whites foreign born Log average farm value # of Observations 163 2, # of States Evaluating the long-term effects of TVA The challenge TVA different already before the program Solution: compare changes in TVA counties to changes in rest of the country rest of the South proposed authorities... and control for pre-program differences This backwardness in levels coincides with some trend differences consistent with simple models of regional convergence (e.g., Barro and Sala-i-Martin 1991). In particular, the TVA region exhibited greater growth in manufacturing share than the rest of the country, accompanied by a faster rate of retrenchment in agriculture. Importantly, the trends in manufacturing share are similar in proposed authorities. Matti Sarvimäki Economic History Geography 19 / 40

6 Identification Placebo test: Effect on the pre-tva outcomes Identifying assumption: if TVA would not have existed, average outcomes in the TVA-counties would have changed similarly as average outcomes in the non-tva counties (that had similar observable characteristics in 1920 and 1930) Is this plausible? proposed authorities seem quite compelling (to me) impact of TVA on other policies: defined as part of the treatment placebo tests: does TVA affect pre-tva outcomes Outcome Panel C: TVA region versus proposed authorities (1) (2) (3) (4) (5) (6) Point estimate Clustered Point estimate Clustered Spatial (unadjusted) std. err. (controls) std. err. HAC N Population (0.019) (0.016) 926 Total employment (0.017) (0.015) 926 Housing units (0.016) (0.013) 926 Average manufacturing wage (0.015) (0.017) 734 Manufacturing share (0.006) (0.006) 926 Agricultural share (0.006) (0.006) 926 Average agricultural land value 0.080*** (0.026) (0.018) 908 Estimates interpreted as average differences in 10-year growth rates experienced by TVA counties relative to non-tva counties between Only the change in agricultural land values statistically different (before adjusting for observables). Results using only the U.S. South and adjusted estimates for the entire U.S. are similar, see Table II in the paper. Matti Sarvimäki Economic History Geography 20 / 40 Matti Sarvimäki Economic History Geography 21 / 40 Impact of TVA on growth rates in Impact during and after the federal subsidies (1) (2) (3) (4) (5) (6) Entire U.S. South Proposed authorities Outcome Panel C: TVA region versus proposed authorities (1) (2) (3) (4) (5) (6) Point estimate Clustered Point estimate Clustered Spatial (unadjusted) std. err. (controls) std. err. HAC N Population (0.018) (0.017) 991 Average manufacturing wage 0.018*** (0.007) (0.006) 618 Agricultural employment 0.101*** (0.029) 0.071*** (0.027) 991 Manufacturing employment 0.066*** (0.024) 0.053** (0.024) 991 Value of farm production (0.026) (0.035) 989 Median family income ( only) 0.060*** (0.012) 0.025** (0.011) 991 Average agricultural land value 0.060*** (0.019) (0.016) 991 Median housing value 0.033** (0.016) (0.016) 991 Estimates interpreted as average differences in 10-year growth rates experienced by TVA counties relative to non-tva counties between The strongest effect of the program was on jobs in agriculture and manufacturing. There is little evidence that local prices, particularly manufacturing wages and housing prices, changed significantly. But median family income seems to have improved, driven presumably by the replacement of agricultural jobs with better paying manufacturing jobs. Outcome Population Average manufacturing wage * Agricultural employment 0.106*** 0.134*** 0.106*** 0.130*** 0.119*** 0.166*** Manufacturing employment 0.114*** 0.033** 0.116*** 0.035* 0.097** 0.032** Value of farm production 0.076* ** ** Median family income N/A N/A N/A 0.019* Average agricultural land value Median housing value Estimates interpreted as average differences in 10-year growth rates experienced by TVA counties relative to non-tva counties. During the period of generous federal funding, growth rate of employment in both agriculture and manufacturing is about 10 percentage points larger in the TVA region. These are remarkably large employment effects, probably explained by an increase in labor demand due to the rapid electrification of the region and the addition of new transportation infrastructure. In the second period employment growth in agriculture falls behind, reversing the gains of the previous period. This is consistent with the end of federal investment, and the lack of important agglomeration economies in agriculture. By contrast, even after the end of federal outlays, manufacturing employment keeps growing significantly faster in TVA counties (although less fast than in the early period). Matti Sarvimäki Economic History Geography 22 / 40

7 Reduced form impacts on the TVA area: summary The effect of TVA on the national economy TVA dramatically accelerated the pace of industrialization shift of employment from agriculture to manufacturing Limited long-run effect on local wage rates, housing values suggest a large supply of workers capable of moving into manufacturing (from outside TVA area, agriculture, home) and an elastic supply of housing and land Manufacturing grows even after cutting down the subsidies agglomeration effects in manufacturing (but not in agriculture) does this mean that manufacturing exhibits multiple equilibria? Did TVA affect also the national economy? economists typically critical to place-based policies workhorse models: such policies only move production around should focus on people rather than places Empirical challenge no plausible control group for the entire U.S. exist Solution: structural model to rationalize the reduced-form effects derive conditions for TVA to increase aggregate output estimate parameters and perform cost-benefit analysis Matti Sarvimäki Economic History Geography 24 / 40 Matti Sarvimäki Economic History Geography 25 / 40 The model: a roadmap An aside: exam requirements for this part Spatial equilibrium wages such that workers indifferent across locations Production TFP depends on the density of manufacturing workers The impact of TVA aggregate production may increase, but only in the presence of specific nonlinearities in agglomeration economies Estimation: see the appendix MSc students only need to get the intuition the concept of spatial equilibrium, how nonlinearities in agglomeration forces create multiple equilibria (the graphs), basic empirical conclusions you do not need to memorize the exact structure of the model, to master the subleties of the IV strategy etc. PhD students expected to understand everything (including the econometrics discussed in the appendix) We will not work through the algebra (you will not be asked to derive any of the expressions, but might be asked to interpret them) Matti Sarvimäki Economic History Geography 26 / 40 Matti Sarvimäki Economic History Geography 27 / 40

8 Assumptions of the model: spatial equilibrium Assumptions of the model: production and wages Counties small open economies (price-takers) that differ in local amenities fundamentals (unobs. locational productivity advantages) endogenous agglomeration externalities Capital and labor perfectly mobile + workers have homogeneous preferences utility equalized across counties lnw it + M it = ū t where w it is wages at location i in year t and M it local amenities low wages can be compensated with high amenties workers migrate until wages are such that this holds in equilibrium, everyone are indifferent across locations (otherwise they would move) Production function for manufacturing assumed to be Y it = A it Kit α F β i L 1 α β it where A it is local TFP, K it is local capital stock, F i is a fixed factor ( fundamentals ) and L it number of manufacturing workers Yields standard inverse labor demand curve, where wages decrease with local labor supply and price of capital increase with productivity and locational fundamentals Matti Sarvimäki Economic History Geography 28 / 40 Matti Sarvimäki Economic History Geography 29 / 40 Assumptions of the model: local productivity Implications of the model Local productivity assumed to be Lit 1 lna it = g + δ t D i + η i + γ t + it That is, local TFP is assumed to depend on past density of manufacturing employment: L it 1 (technological externalities/thick labor markets; is square milage) additional investment from TVA: D i (0/1 for TVA participation, note that δ t varies over time) time-invariant suitability of the county for manufacturing: η i (e.g. proximity to a body of water) calendar year: γ t (captures overall changes in technology etc.) idiosyncratic shocks: it (unobserved changes in local infrastructure, changes in the regulatory environment... assumed to evolve as it = it 1 + ξ it ) Matti Sarvimäki Economic History Geography 30 / 40 The steady-state impact of marginal increase in productivity due to TVA on output in county i dy i dδ = 1 1 α Y i D i + 1 α β + σ i dl i L i dδ Direct impact through new infrastructure: 1 1 α % larger than 1 and increases with capital share α, because capital adjustment augments a productivity change Indirect effect mechanically by increasing manufacturing labor: dl i /dδ and through agglomeration economies: σ i = g Li Li Impact on national output: sum over all counties Matti Sarvimäki Economic History Geography 31 / 40

9 Implications of the model Implications of the model Total effect unambiguously positive this is simpy because total infrastructure was increased Indirect effect due to labor reallocation ambiguous moving a worker from i to j raises aggregate output iff Y i L i (1 α β + σ i ) < Y j L j (1 α β + σ j ) i.e. depends on the average labor productivity, Y i L i, and agglomeration elasticity, σ i,ineachcounty Case 1: amenities equal across the two communities thus wages and productivity must also be equal Reallocation from i to j raises output if σ i <σ j if σ i = σ j benefits in j equal loses in i (no aggregate effect) Case 2: amenities differ, σ i = σ j aggregate output can be raised by moving workers to lower amenity areas where wages (and thus productivity) are higher however, this comes at a utility cost that perfectly offsets the value of increases in aggregate output Thus the shape of g ( ) determines whether TVA was welfare improving for the U.S. as a whole Matti Sarvimäki Economic History Geography 32 / 40 Matti Sarvimäki Economic History Geography 33 / 40 The model: dynamics under log-linear g ( ) The model: dynamics under log-linear g ( ) SR: short-run inverse demand curve. LR: long-run inverse demand curve incorporates the agglomeration effects. LR flatter than the SR because the agglomeration economies dampen the effects of the fixed factor on labor productivity. Initial equilibrium: the intersection of the LR curve with the horizontal labor supply curve determines the (unique) steady-state level of manufacturing employment. Introduction of TVA makes firms more productive and thus shifts both the SR and LR curves up by an amount δ t. Employment increases as manufacturing employment converges toward its new steady state. Once the direct productivity effects of TVA lapse, the LR curve will slowly revert back to its original position as the initial infrastructure investment depreciates and the employment gains are gradually reversed.

10 The model: dynamics under nonlinear g ( ) The model: dynamics under nonlinear g ( ) Here g ( ) exhibits strong threshold effects so that productivity increases rapidly once the sector reaches some critical level of density, but begins to decrease afterward due to the presence of the fixed factor. Due to nonlinearities TVA can have a positive effect on aggregate productivity. In fact, if workers come from developed regions on the downward-sloping portion of the LR curve, productivity in those areas also rise. Furthermore, TVA can have long-lasting effects if it pushes the county pass the tipping point. In this, the county eventually converges to the developed equilibrium (after TVA investments have fully depreciated). Estimates for agglomeration elasticities Estimates for the direct effect of TVA (1) (2) (3) (4) (5) (6) OLS OLS OLS 2SLS 2SLS 2SLS Change in log manufacturing density spline components: Low (0.037) (0.037) (0.037) (0.102) (0.108) (0.107) [177.17] [159.14] [157.20] Medium (0.045) (0.044) (0.045) (0.124) (0.123) (0.124) [106.74] [109.55] [110.13] High (0.051) (0.050) (0.050) (0.150) (0.150) (0.151) [206.66] [204.69] [200.36] Estimates for agglomeration elasticities for counties with low, medium, and high levels of density. Estimates suggest that a 10% increase in density yields a 4 4.7% increase in labor demand. They are robust to controlling for baseline density and regional trends. Notably, we cannot reject a constant elasticity relationship in any of the IV specifications. Matti Sarvimäki Economic History Geography 35 / 40 (1) (2) (3) Spline in levels (0.070) (0.041) (0.039) Spline in logs (0.081) (0.033) (0.035) Controls for 1920 and 1930 characteristics yes yes yes N 1,587 1,498 1,533 Consistent with the reduced-form estimates, the TVA is estimated to have substantially boosted productivity over the period Given (reasonable, but nevertheless uncertain) assumptions about α and β, thetvawouldhaveraisedlocal productivity by approximately 8.7% over the period. This was followed by insignificant negative direct effects in later periods, which is in keeping with the earlier evidence that TVA transfers scaled down over this horizon and that local infrastructure began to depreciate. Matti Sarvimäki Economic History Geography 36 / 40

11 Discussion Cost-benefit calculation Predictions from the structural model closely in line with the reduced-form estimates suggests that the required (strong) assumptions are reasonable The key results agglomeration economies have a constant elasticity wrt manufacturing density policies that just reallocate manufacturing unlikely to increase aggregate welfare implies a unique steady-state (recall the dynamics graphs) Costs: $17.3 billion 1940 present value of the year-by-year stream of federal appropriations to the TVA Benefits: $ billion relies on the structure (functional form) of the model constant elasticity no benefits from agglomeration but from the national investment that raise productivity Matti Sarvimäki Economic History Geography 37 / 40 Matti Sarvimäki Economic History Geography 38 / 40 Conclusions Papers for the essays Results suggest powerful agglomeration economies in manufacturing (but not in agriculture) BUT: this alone does not imply multiple equilibria structural estimates actually point towards a unique equilibrium you should note, however, that the external validity of this conclusion is not clear KM: We caution, however, that [...] our results are specific to the manufacturing sector and a period of U.S. history Davis and Weinstein (2002): Bones, Bombs, and Break Points: The Geography of Economic Activity. AER 92(5): This paper examines the distribution of regional population in Japan from the Stone Age to the modern era and exploits variation due to the Allied bombing of Japanese cities in WWII to test for the presence of multiple equlibria in the Japanese regional structure. Bleakley and Lin (2012): Portage and Path Dependence. QJE 127(2): This paper documents the continuing importance historical portage sites for the current economic geography of the U.S. (even though their original advantages have long since become obsolete). Matti Sarvimäki Economic History Geography 39 / 40 Matti Sarvimäki Economic History Geography 40 / 40

12 Selection into TVA Appendix Prioritized counties which rural, required additional electric power experienced severe flooding and/or had misguided land use experienced heavy deficits lacked public facilities (libraries, health services, schools...) were willing to receive technical and advisory assistance had planning agencies and enabling legislation agreed to experiment with new fertilizers were within reasonable transmission distance of power plants Thus TVA counties less developed by design Matti Sarvimäki Economic History Geography 42 / 40 Econometrics: reduced form 1 Using the non-tva counties, estimate y it y it 1 = α + X β +( it it 1 ) where y it y it 1 is the change in the relevant dependent variable between year t 1 and t for county i and X i is a vector of pre program characteristics 2 Use ˆβ to predict the counterfactual mean for the TVA counties 3 Estimated effect: average over the difference between the real outcome and the predicted counterfactual Slightly more complicated than standard regression framework advantage: identifies the average treatment effect on treated in the presence of treatment effect heterogeneity and has dual interpretation as a propensity score reweighting estimator (Kline 2011) In practice, standard regression models yield similar results Econometrics: structural model Starting point: the production function for manufacturing Y it = A it Kit α F β i L 1 α β it where A it is local TFP, K it is local capital stock, F i is a fixed factor ( fundamentals ) and L it number of manufacturing workers. implies a inverse labor demand curve lnw it = C β 1 α lnl it + β 1 α lnf i α 1 α lnr t α lna it where r t is the (nationwide) price of capital, C = ln(1 α β)+ α 1 α lnα Matti Sarvimäki Economic History Geography 44 / 40

13 Estimating the structural parameters Let s rearrange the inverse labor demand curve and take first differences to get the estimation equation L it = a + b w it + c t D i +Σ 3 k=1 d Lit k + controls where agglomeration economies are specified as a linear spline. Note that county FEs ( fundamentals ) are removed by differencing over time Correspondence with the model parameters lnðl it Þ lnðl it 1 Þ ¼ 1 ðln w it ln w it 1 Þþ t t 1 þ X3 k g L it 1 L it 2 k g k k¼1 þ X 0 i ~ þ ~ t ~ t 1 þ ~ it : Matti Sarvimäki Economic History Geography 45 / 40 D i Objects of interest. The coefficients t t 1, which give the change in the direct effects of TVA between decades;. The spline coefficients k, which determine the indirect effects of the program since they give the labor demand effects of increasing manufacturing density within the relevant density range. We refer to 1 as the agglomeration effect at low density, 2 the effect at medium density, and 3 the effect at high density. Matti Sarvimäki Economic History Geography 46 / 40 The identification challenge The causal chain productivity shock manu. workers productivity Need to separately identify initial employment response to a shock feedback effects of that response (agglomeration)... and to test whether feedback effects are stronger in underdeveloped counties Problem: shocks may be persistent one may mistake the persistence of the shocks for the feedback effects in econometrics parlance: need to separate state dependence from serial correlation in unobservables (serially correlated ν it ) bias: regression will attribute all of the serial correlation in employment changes to state dependence (agglomeration) when some of it is actually the result of additional shocks. Matti Sarvimäki Economic History Geography 47 / 40 IV strategy To address state dependece, KM instrument manufacturing changes with lagged manufacturing changes, i.e. Z (k) Lit 2 Lit 3 it g k g k is used as an instrument for g k Lit The identifying assumption is E [v it Z it ] k g k Lit 1 i.e. that unobserved productivity shocks are independent over ahorizonof20years Matti Sarvimäki Economic History Geography 48 / 40

14 Caveats of the IV strategy Counties could have long-lasting heterogeneous trends in productivity growth e.g. conditional convergence in manufacturing activity for reasons having little to do with agglomeration How to mitigate this issue? control for initial conditions (1920 and 1930 observables) control for 1940 manufacturing density control for fixed regional trends use earlier periods for constructing the instruments KM: Although these robustness exercises do not guarantee that all trend heterogeneity has been removed, we believe they suggest our results are not spurious. Moreover, we note that many of our conclusions rest on the relative magnitude of the three θ k /β parameters. Matti Sarvimäki Economic History Geography 49 / 40 Another challenge: wages lnðl it Þ lnðl it 1 Þ ¼ 1 ðln w it ln w it 1 Þþ t t 1 þ X3 k g L it 1 L it 2 k g k k¼1 þ X 0 i ~ þ ~ t ~ t 1 þ ~ it : Two potential problems in wage changes they may be correlated with ν it division bias KM deal with these issues by setting the coefficient for wages to -1.5 (taken from Hamermesh s 1993 review) later assess the sensitivity of estimates to different values of this elasticity Matti Sarvimäki Economic History Geography 50 / 40 D i