Banque nationale de Belgique SA · EURONEXT:BNB
Banque nationale de Belgique SA — Technical Presentation (March 2012)
48 slides from Banque nationale de Belgique SA's technical presentation, March 2012
Slide 1: Staggered Adjustment and Trade Dynamics
Staggered Adjustment and Trade Dynamics Costas Arkolakis Jonathan Eaton Samuel Kortum Yale, Penn State, Chicago March 2012
Slide 2: Motivation
Motivation Are new empirical trade models helpful for open economy macro? A fundamental incompatibility: Open economy macro models need a small trade elasticity to explain short-run responses to changes in relative cost Trade models need a large trade elasticity to explain long-run responses to chan
Slide 3: Motivation II
Motivation II Bilateral trade share: Xnit Tit (wit dnit ) θ π tni = Xnt ∑k T t (w t d t ) θ k k nk Compare i’s and j’s shipments to market n: Xnit Tit wit dnit ln = ln θ ln θ ln Xnjt Tjt wjt t dnj What do changes in relative wages explain?
Slide 4: Motivation II
Motivation II Bilateral trade share: Xnit Tit (wit dnit ) θ π tni = Xnt ∑k T t (w t d t ) θ k k nk Compare i’s and j’s shipments to market n: Xnit Tit wit dnit ln = ln θ ln θ ln Xnjt Tjt wjt t dnj Specify a regression in changes X̂nit ŵit ln = θ ln + εtni t X̂nn ŵnt
Slide 5: Motivation III: Pooled Regressions (Annual Differences)
Motivation III: Pooled Regressions (Annual Di¤erences): Explanatory power of the wage changes bt X bit w ln ni = θ ln + εtni b t Xnn bnt w b θ = 0.4 (s.e. 0.04) R 2 = 0.21 Explanatory power of encompassing source e¤ect (Sit = Tit (wit ) θ ) bt X bt S ln ni = ln i + etni bnn X t bnt S R 2 = 0.47
Slide 6: Our Goal
Our Goal Develop a model that works well at di¤erent frequencies Open economy model exploiting information in bilateral trade Sacri…ce optimizing behavior to make it easier to take to the data
Slide 7: Line of Attack: Staggered Adjustment by Consumers
Line of Attack: Staggered Adjustment by Consumers Basic idea: consumers are slow to switch sources of supply Consistent with evidence on “Customer Loyalty” (Reichheld & Teal ’96) Like …rms changing prices in Calvo model Analysis inspired by Ruhl ’05, related to Drozd & Nosal ’12 Common feature: slow
Slide 8: Main Idea in a Nutshell
Main Idea in a Nutshell A continuum of goods Consumer can search for cheapest supplier only for a fraction of goods each period Sluggish adjustment at the extensive margin (consistent with Baier et al ’11) Standard CES Dixit-Stiglitz demand at the intensive margin Firms perfectly competitive Take pr
Slide 9: Main Idea: Prices over time
Main Idea: Prices over time POSER pag MP meus P1 P3cer P*cEr ely P?GER Se
Slide 10: Main Idea: Prices over time & Consumer Choice
Main Idea: Prices over time & Consumer Choice
Slide 11: The Model
The Model
Slide 12: Theoretical Framework
Theoretical Framework Our model encompasses empirical Ricardian and Armington models Adds just one parameter to EK ’02, λ : fraction of consumers adjusting λ = 0 delivers Armington λ = 1 delivers EK In between is where trade dynamics get interesting
Slide 13: Basics
Basics Perfect competition, continuum of goods j 2 [0, 1], integer I countries (denoted n when importing and i when exporting) Wage wit , aggregate productivity Ati , iceberg trade costs dnit Unit cost of i’s labor for supplying n: t cni t wit dni /Ati A technology for making good j in country i wit
Slide 14: Technologies
Technologies New technologies rain down, independently across countries, as in EK’01 For any good, the number of techniques with e¢ ciency better than z is distributed Poisson (Tit z θ ) Each country more than one available techniques to produce good j
Slide 15: Old EK: Full Adjustment
Old EK: Full Adjustment Only the best technologies are used in each country, e¢ ciency zia,t (j ). Probability that no technique better than z has arrived (Poisson evaluated at 0) yields Fréchet: t Pr[zia,t (j ) θ z ] = e Ti z
Slide 16: Old EK: Results
Old EK: Results Price distribution for lowest cost in n: e Φn p t θ Pr[pna,t (j ) p ] = Gna,t (p ) = 1 How countries contribute to n: Φtn = ∑k Tkt cnk t θ Distribution same for all i conditional on n buying from i Share of goods for which i is lowest cost in n: Tit (cnit ) θ π̄ a,t ni = Φtn
Slide 17: New AEK: Staggered Adjustment
New AEK: Staggered Adjustment Consumer: With probability λ looks everywhere for the lowest-cost supplier, otherwise buys good from same supplier as last year Producers: perfectly competitive If Tit = Ti (constant technologies) consumer buys country’s lowest cost supplier If Tit increases consumer ma
Slide 18: New AEK: Staggered Adjustment
New AEK: Staggered Adjustment Fraction of goods n buys from i evolves as: π̄ tni = λπ̄ a,t ni + (1 λ) π̄ tni 1 Price distribution for goods purchased from country i evolves as: ! a,t t 1 λ π̄ ( 1 λ ) π̄ p Gnit (p ) = ni G a,t (p ) + ni Gnit 1 π̄ tni n π̄ tni cnit /cnit 1 (not the same for all i)
Slide 19: New AEK: Staggered Adjustment
New AEK: Staggered Adjustment Fraction of goods n buys from i evolves as: π̄ tni = λπ̄ a,t ni + (1 λ) π̄ tni 1 Price distribution for goods purchased from country i evolves as: ! a,t t 1 λ π̄ ( 1 λ ) π̄ p Gnit (p ) = ni G a,t (p ) + ni Gnit 1 π̄ tni n π̄ tni cnit /cnit 1 Note: We haven’t talked yet
Slide 20: Consumer Preferences
Consumer Preferences Standard Dixit-Stiglitz preferences as in EK (σ 1 ) p xnt (p ) = Xnt (1) Pnt with price index (σ 1 ) = ∑ π̄ tni Pnit (σ 1 ) Pnt i where Z ∞ 1/(σ 1 ) Pnit = p (σ 1 ) dGnit (p ) 0
Slide 21: Trade Shares
Trade Shares Multiplying each side of (1) by π̄ tni then integrating over Gnit (p ) gives the observed trade share: (σ 1 ) Pnit π tni = π̄ tni Pnt For λ low, π̄ tni responds slowly to a change in costs, but its long-run elasticity is always θ For λ low, the price index Pnit moves nearly in proportio
Slide 22: Three Key Parameters
Three Key Parameters Elasticity of substitution σ Elasticity of the extensive margin θ Speed of adjustment λ Their roles: In Armington σ 1 is the trade elasticity In EK θ is the trade elasticity For us, σ 1 matters for the short run and θ for the long run, with λ governing the arrival of the long ru
Slide 23: Closing the Model
Closing the Model Static equilibrium allowing for exogenous trade de…cits Dnt Labor supply Lti is also exogenous Labor market equilbrium requires: I wit Lti = ∑ π tni wnt Ltn + Dnt n =1 Want to include a non-traded sector and intermediates, but keep it simple for now.
Slide 24: Model Implications
Model Implications Trade elasticity ∂π tni wi = [λθ + (1 λ)(σ 1)] (1 π tni ). ∂wi π tni Insensitivity of trade shares to wages (if λ and σ are low) makes wages more sensitive to a shock, such as a change in D. Real wage (letting Znt = Tnt (Atn )θ combine the two productivity terms): wnt 1/θ 1/(σ 1 )
Slide 25: Numerical Illustrations
Numerical Illustrations
Slide 26: Parameters, Initial Conditions, and Shocks
Parameters, Initial Conditions, and Shocks Benchmark parameters Θ = fθ, σ, λg θ = 4 : long-run elasticity estimates (Simonovska Waugh ’10, Caliendo Parro ’10) σ = 0.5: short-run elasticity estimates (Reinert ’92, Bloningen Wilson ’99) λ = 0.05, λ = 0.20, or λ = 1: marketing case studies (' 0.15) Ful
Slide 27: Data
Data Bilateral trade in manufactures among G6 countries and ROW Annual 1991-2006 Manufacturing gross production for Yi
Slide 28: Experiment I: Balancing Trade
Experiment I: Balancing Trade Start from initial condition in 2006, then permanently adjust to balanced trade Japan starts in surplus; United Kingdom and the United States in de…cit Show outcomes for i = J, UK , USA: 1. Wage wit (relative to world manufacturing production) 2. Real wage wit /Pit 3. T
Slide 29: Experiment I: Trade Balance after 2006
Experiment I: Trade Balance after 2006 0.2 0.15 0.1 Net Exports to GDP 0.05 0 -0.05 -0.1 -0.15 FRA G ER ITA JAP UK USA
Slide 30: Figure 2: Balanced Trade with High Lambda (DEK)
Figure 2: Balanced Trade with High Lambda (DEK) Wages (Relative to World GDP) Real Wages Japan Japan 1.8 UK 1.4 UK 1.6 USA 1.3 USA 1.4 1.2 Ratio 1.2 1.1 1 1 0.8 0.9 0.6 0.8 0 5 10 0 5 10 Terms of Trade Share on Domestic Spending Japan Japan 1.8 UK 1.4 UK USA USA 1.6 1.3 1.4 1.2 Ratio 1.2 1.1 1 1 0.8
Slide 31: Figure 3: Balanced Trade with Medium Lambda
Figure 3: Balanced Trade with Medium Lambda Wages (Relative to World GDP) Real Wages Japan Japan 1.8 UK 1.4 UK 1.6 USA 1.3 USA 1.4 1.2 Ratio 1.2 1.1 1 1 0.8 0.9 0.6 0.8 0 5 10 0 5 10 Terms of Trade Share on Domestic Spending Japan Japan 1.8 UK 1.4 UK USA USA 1.6 1.3 1.4 1.2 Ratio 1.2 1.1 1 1 0.8 0.9
Slide 32: Figure 4: Balanced Trade with Low Lambda
Figure 4: Balanced Trade with Low Lambda Wages (Relative to World GDP) Real Wages Japan Japan 1.8 UK 1.4 UK 1.6 USA 1.3 USA 1.4 1.2 Ratio 1.2 1.1 1 1 0.8 0.9 0.6 0.8 0 5 10 0 5 10 Terms of Trade Share on Domestic Spending Japan Japan 1.8 UK 1.4 UK USA USA 1.6 1.3 1.4 1.2 Ratio 1.2 1.1 1 1 0.8 0.9 0.
Slide 33: Experiment II: Trade Liberalization
Experiment II: Trade Liberalization Start from initial condition in 2006, then d̂ni2007 = 1/(1.1) for all n 6= i Set λ = 0.05 Outcomes for i = J, UK , USA: 1. Wage wit (r-elative to world manufacturing production) 2. Real wage wit /Pit 3. Terms of trade (export price over import price) 4. Domestic s
Slide 34: Figure 5: Trade Liberalization Outcomes
Figure 5: Trade Liberalization Outcomes Wages (Relative to World GDP) Real Wages Japan Japan 1.8 UK 1.4 UK Ratio 1.6 USA 1.3 USA 1.4 1.2 1.2 1.1 1 1 0.8 0.9 0.6 0.8 0 5 10 0 5 10 Terms of Trade Share on Domestic Spending Japan Japan 1.8 UK UK USA 1.1 USA 1.6 Ratio 1.4 1 1.2 0.9 1 0.8 0.8 0.6 0 5 10
Slide 35: Trade Liberalization: Welfare Loss due to Slow Adjustment
Trade Liberalization: Welfare Loss due to Slow Adjustment 1.01 J apan UK U SA 1.005 Ratio 1 0.995 0.99 0 5 10 Y e a rs fro m In i ti a l S te a d y S ta te
Slide 36: Experiment III: Estimate Trade Elasticities
Experiment III: Estimate Trade Elasticities Simulate using actual de…cits 1992-2006 Productivity shocks log normal: mean and std. deviation of growth 0.01 Generate 100 separate data sets by repeatedly simulating the model Run the following short-run regression on both the simulated and the actual da
Slide 37: Short-Run Regression Results
Short-Run Regression Results λ = 0.05 λ = 0.2 λ=1 Data mean 0.384 0.017 1.17 0.41 std. deviation 0.003 0.017 0.239 0.04 model elasticity 0.275 0.400 4.00
Slide 38: Long-Run Regression
Long-Run Regression Simulate permanent removal of a 10% tari¤ on ROW imports, in 1992 Captured by d̂ni1992 = 1/(1.1) when n = ROW We also include the same shocks to de…cits and productivity as in the previous exercise Construct the Head and Ries ’01 index π tni π tin Hnit = π tnn π tii Run the long-
Slide 39: Towards Estimation
Towards Estimation
Slide 40: Equations for Estimation
Equations for Estimation Let τ ti = ln Ẑit and εtni = θ ln d̂nit be random shocks: exp (τ ti + εtni ) (ŵit ) θ b̄ a,t π ni = a,t 1 ∑k π̄ ni exp (τ tk + εtnk ) (ŵkt ) θ Let gT = T̂it : ! σθ 1 b̄ a,t π νtni = λπ̄ a,t ni + (1 λ) ni νtni 1 gT Given full-adjustment initial conditions at t = 0 and obse
Slide 41: Moment Conditions
Moment Conditions To identify the parameters Θ = fθ, σ, λ, gT g impose orthogonality conditions between the shocks and observed de…cits Dnt Gives a vector of moments f (τ, ε) that should be close to zero The identi…cation power of de…cits comes from the labor market equilibrium conditions, which imp
Slide 42: A Procedure
A Procedure Want to exploit recent work on structural estimation using Mathematical Programming with Equilibrium Constraints (MPEC): Su Judd ’10 Our model …ts nicely into the setting explored by Dube Fox Su ’09: minΘ,τ,ε f (τ, ε)0 Ωf (τ, ε) subject to π (τ, ε; Θ) = Π Note I: since π is a constraint,
Slide 43: Appendix
Appendix
Slide 44: Motivation II
Motivation II Bilateral trade share: Xnit t Tit (wit dnit ) θ = π ni = Xnt ∑k T t (w t d t ) θ k k nk Compare i’s and j’s shipments to market n: Xnit Tit wit dnit ln = ln θ ln θ ln Xnjt Tjt wjt t dnj What do changes in relative wages explain? Set: United Kingdom (i), Japan (j), each G6 country (n’s)
Slide 45: Motivation III
Motivation III UK/Japan wage (falls then rises). Manufacturing shipments (UK vs. Japan) to G6 destinations: Destination France Germany Italy 2 1 Japan United Kingdom United States 2 .5 1 1990 1995 2000 2005 1990 1995 2000 2005 1990 1995 2000 2005 .5 Time ireltrade irelwage
Slide 46: Three Key Equations for Simulation
Three Key Equations for Simulation θ Letting Ẑit = T̂it Âti capture productivity change (where x̂ t = x t /x t 1 ): θ Ẑit ŵit d̂nit b̄ a,t π ni = a,t 1 θ ∑k π̄ ni t Ẑkt ŵkt d̂nk P nt (σ 1 ) Letting νtni = π tni P na,t so that trade shares are π tni = νtni / ∑ νtnk : ! σθ 1 π̄ a,t 1 b̄ a,t π b̄
Slide 47: Experiment I: Trade Balance after 2006
Experiment I: Trade Balance after 2006 0.2 0.15 0.1 Net Exports to GDP 0.05 0 -0.05 -0.1 -0.15 FRA G ER ITA JAP UK USA
Slide 48: Trade Liberalization: Welfare Loss due to Slow Adjustment
Trade Liberalization: Welfare Loss due to Slow Adjustment 1.01 J apan UK U SA 1.005 Ratio 1 0.995 0.99 0 5 10 Y e a rs fro m In i ti a l S te a d y S ta te
Banque Nationale De Belgique — Technical Presentation (March 2012) on 6ix