General Equilibrium Dampened

Transcription

1 General Equilibrium Dampened (i) from Micro to Macro (ii) Forward Guidance George-Marios Angeletos Chen Lian Edinburgh December 31, 2016

2 Motivation GE effects key to macroeconomics (and elsewhere) upend partial-equilibrium (PE) intuitions limit usefulness of micro-based evidence a la Mian-Sufi drive interpretations of phenomena + policy implications But: GE effects hinge on common knowledge (CK) of structure and state of economy immense coordination in beliefs and behavior all hardwired in solution concept + info assumptions

3 Motivation GE effects key to macroeconomics (and elsewhere) upend partial-equilibrium (PE) intuitions limit usefulness of micro-based evidence a la Mian-Sufi drive interpretations of phenomena + policy implications But: GE effects hinge on common knowledge (CK) of structure and state of economy immense coordination in beliefs and behavior all hardwired in solution concept + info assumptions

4 This talk, Part I: Dampening GE Formalize notion GE Adjustment Takes Times Framework: abstract but flexible supply and demand Main result: Equivalence between relax solution concept Tattonment ( off equilibrium ) relax info / CK imperfect coordination ( on equilibrium ) Broader lessons/implications: lack of CK = relaxation of RE solution concept = dampen GE resuscitate PE intuitions in GE settings enhance value of empirical work a la Mian-Sufi

5 This talk, Part I: Dampening GE Formalize notion GE Adjustment Takes Times Framework: abstract but flexible supply and demand Main result: Equivalence between relax solution concept Tattonment ( off equilibrium ) relax info / CK imperfect coordination ( on equilibrium ) Broader lessons/implications: lack of CK = relaxation of RE solution concept = dampen GE resuscitate PE intuitions in GE settings enhance value of empirical work a la Mian-Sufi

6 This talk, Part II: Application to Forward Guidance Context: A New-Keynesian economy during a liquidity trap Question: Ability to stimulate economy my promising low interest rates after ZLB has ceased to bind Puzzling prediction: Ability is large and increases with horizon at which forward guidance operates Our contribution: puzzle driven solely by GE effects lack of CK anchors expectations of income and inflation attenuates relevant GE effects reduces power of forward guidance additional results: paradox of flexibility, discounted Euler/NKPC...

7 This talk, Part II: Application to Forward Guidance Context: A New-Keynesian economy during a liquidity trap Question: Ability to stimulate economy my promising low interest rates after ZLB has ceased to bind Puzzling prediction: Ability is large and increases with horizon at which forward guidance operates Our contribution: puzzle driven solely by GE effects lack of CK anchors expectations of income and inflation attenuates relevant GE effects reduces power of forward guidance additional results: paradox of flexibility, discounted Euler/NKPC...

8 Related Literature Part I: Higher-order uncertainty in macroeconomics Morris and Shin (1998, 2000, 2002), Woodford (2003), Angeletos and Pavan (2007), Angeletos and La O (2009), Nimark (2011), etc Angeletos and Lian (2016): chapter in Handbook of Macroeconomics Part II: Forward guidance Wiederholt (2016), Farhi and Werning (2016) McKay, Nakamura and Steinsson (2016a,b), Gabaix (2016) Garcia-Schmidt and Woodford (2015)

9 First paper: Angeletos and Lian (2016a) Dampening General Equilibrium: Macro is Micro in the Short Run

10 Framework Minimal framework for studying PE vs GE, and micro vs macro many locations, competitive firms and households idiosyncratic and aggregate shocks two (relative) prices three goods = numeraire + another tradable + one non-tradable per location What s next? micro-foundations, demand and supply review standard predictions two variants: (i) Tatonnment and (ii) Incomplete Info/Lack of CK

11 Framework Minimal framework for studying PE vs GE, and micro vs macro many locations, competitive firms and households idiosyncratic and aggregate shocks two (relative) prices three goods = numeraire + another tradable + one non-tradable per location What s next? micro-foundations, demand and supply review standard predictions two variants: (i) Tatonnment and (ii) Incomplete Info/Lack of CK

12 Supply Representative competitive firm at each location i [0, 1]. Technology (production possibilities set): F (q i,q i,q Z i ;a i ) 0 q i,q i,qz i = production of non-tradables, tradables, numeraire a i = technology shock = supply shock

13 Supply Firm s problem { max p i q i + p qi + qi Z } s.t. F (q i,q i,q Z i ;a i ) 0 pi,p = price of non-tradables, tradables Supply of local non-tradable q i = S (p i,p,a i ) Supply of tradable and numeraire: qi = S (...) qi Z = S Z (...)

14 Demand Representative competitive household at each location i [0, 1]. Preferences: u i = U(c i,c i,c Z i ;ξ i ) ξi = preference shock = demand shock Budget constraint: p i c i + p c i + c Z i = y i yi = income = p i q i + p q i + q Z i

15 Demand Demand for local non-tradable: c i = D (p i,p,ξ i,y i ) Demand for tradable and numeraire ci = D (...) ci Z = D Z (...)

16 Partial Equilibrium Partial equilibrium market clearing for non-tradable, but arbitrary p i.e., momentarily allow market for tradable to be off equilibrium Let θ i = (a i,ξ i ). Quantity and price of non-tradable c i = q i = Q (p,θ i ) and p i = P (p,θ i ) Net (excess) demand for tradable n i ci qi = N (p,θ i )

17 General Equilibrium Market clearing for tradable: p solves N (p,θ i )di = 0 GE imposes p = P ( θ ) Assumptions p N di < 0 (stable equilibrium) N θ di 0, or equivalently θ P 0 (non-zero GE effects)

18 Macro Effect of an Aggregate Shock How does the economy respond to a shock that moves θ i for all i? demand shock: housing wealth, consumer deleveraging... supply shock: productivity, payroll taxes... To simplify, work with log-linearized conditions all variables in log-deviations from steady state To be concrete, focus on expenditure on non-tradable local expenditure on non-tradable: x i q i + p i = X (p,θ i ) corresponding aggregate: x x i di = X (p, θ)

19 Macro Effect of an Aggregate Shock How does the economy respond to a shock that moves θ i for all i? demand shock: housing wealth, consumer deleveraging... supply shock: productivity, payroll taxes... To simplify, work with log-linearized conditions all variables in log-deviations from steady state To be concrete, focus on expenditure on non-tradable local expenditure on non-tradable: x i q i + p i = X (p,θ i ) corresponding aggregate: x x i di = X (p, θ)

20 Micro vs Macro Micro elasticity, or PE effect at local level aggregate counterpart ε micro ε micro i dx i dθ i εi micro di = d x d θ p constant p constant = X θ Macro effect ε Macro d x d θ p adjusts in GE = X θ }{{} PE + X P p ε micro θ }{{} GE

21 Micro vs Macro Micro elasticity, or PE effect at local level aggregate counterpart ε micro ε micro i dx i dθ i εi micro di = d x d θ p constant p constant = X θ Macro effect ε Macro d x d θ p adjusts in GE = X θ }{{} PE + X P p ε micro θ }{{} GE

22 PE effect p D(,p * 0,ξ 0 ) D(,p * 0,ξ 1 ) PE O S(,p * 0 ) PE effects q

23 GE amplifies PE p D(,p * 0,ξ 0 ) D(,p * 0,ξ 1 ) D(,p * 1,ξ 1 ) GE PE S(,p * 1 ) S(,p * 0 ) O PE effects GE effects q

24 GE attenuates PE p D(,p * 0,ξ 0 ) D(,p * 1,ξ 1 ) D(,p * 0,ξ 1 ) PE S(,p * 0 ) S(,p * 1 ) O GE GE effects PE effects q

25 Connection to Empirical Work Recent empirical macro: exploits cross-sectional variation provides estimate of ε micro Mian-Sufi, Nakamura-Steinsson, etc Tension between what is estimated (ε micro ) what is of interest (ε Macro ) Key problem: GE effect partialed out as time fixed effect in regressions Our contribution: lessens the problem (at least in the short run)

26 Preview Standard paradigm: adjustment in p is instantaneous perfect coordination What we are after: slow adjustment in p How? relax solution concept: Tâtonnement relax info assumption: remove common knowledge

27 Tâtonnement Let t index round of iteration in Tatonnement process soon to reinterpret t as time i,t, local market for non-tradables clears with given perception ˆp t gives PE outcomes with p = ˆp t to reinterpret t as time Walrasian auctioneer adjusts ˆp slowly from old GE level to new one ˆp solves the following ODE d ˆp t dt = b t [N (ˆp t, θ new )] with initial condition ˆp 0 = P ( θ old ) and for some exogenous {b t } with b t b > 0 t.

28 Tâtonnement Let t index round of iteration in Tatonnement process soon to reinterpret t as time i,t, local market for non-tradables clears with given perception ˆp t gives PE outcomes with p = ˆp t to reinterpret t as time Walrasian auctioneer adjusts ˆp slowly from old GE level to new one ˆp solves the following ODE d ˆp t dt = b t [N (ˆp t, θ new )] with initial condition ˆp 0 = P ( θ old ) and for some exogenous {b t } with b t b > 0 t.

29 Tâtonnement: micro vs macro Macro effect at t: ε T ât (t) = ε micro + w(t) (ε Macro ε micro) } {{ } GE at t where w(t) is increasing in t, with w(0) = 0 and w( ) = 1 PE same as in benchmark because of local market clearing GE is dampened by factor w because of erroneous perceptions of p

30 Tâtonnement: micro vs macro p D(,p * 0,ξ 0 ) D(,p * 0, ξ 1 ) D(,p * 1,ξ 1 ) GE PE S(,p * 1 ) S(,p * 0 ) O q

31 Tâtonnement: micro vs macro Corollary. ε T ât is monotone and continuous in t, with ε T ât (0) = ε micro and ε T ât ( ) = ε Macro That is, we can span the gap between the micro and the macro by varying the round t in Tâtonnement

32 Incomplete Information Goal: translate from off equilibrium to on equilibrium Same payoff environment Non-tradable decisions in the morning under incomplete information perfect knowledge of local conditions (θ i,q i,p i ) lack common knowledge (CK) of global conditions ( θ,p ) private signal about the latter: si = θ + v i Tradable decisions in the afternoon global conditions ( θ,p ) become common knowledge

33 Equilibrium Rational-Expectations Equil with inco info (similar to PBE) Morning: local markets for non-tradable clear, giving q i = Q (E i [p ],θ i ) p i = P (E i [p ],θ i ) where E i [p ] is the rational expectation of p conditional on s i Afternoon: p clears global market for tradable, giving p = P q ( q, θ) where q is the realized agg quantity of non-tradable

34 Characterization Lemma. Equilibrium outcomes satisfy Isomorphic to a beauty contest q i = αe i [ q] + ηe i [ θ] + ζ θ i GE effect akin to strategic interaction in games α < 1 : degree of strategic complementarity/substitutability Corollary. Rational expectation of p = hierarchy of beliefs about θ Ē [p ] = Ω h=1 α h 1 Ē h [ θ ] GE effects = HOB

35 Characterization Lemma. Equilibrium outcomes satisfy Isomorphic to a beauty contest q i = αe i [ q] + ηe i [ θ] + ζ θ i GE effect akin to strategic interaction in games α < 1 : degree of strategic complementarity/substitutability Corollary. Rational expectation of p = hierarchy of beliefs about θ Ē [p ] = Ω h=1 α h 1 Ē h [ θ ] GE effects = HOB

36 Lack of CK = Anchored Expectations Beliefs first-order beliefs: Ē 1 θ E i θdi = λ θ where λ κ (0, 1) captures deviation of common knowledge κ+σθ 2 higher-order beliefs (HOB): ] Ē h θ [Ē Ē h 1 [ θ] = λ h θ HOB vary less than lower-order beliefs expectations of p are anchored GE is stronger HOB more important the stronger the GE effect, the stronger its own attenuation

37 Micro vs Macro PE as in benchmark due to perfect knowledge of local conditions GE dampened due to lack of common knowledge of global conditions Macro effect revisited: ε Inc (λ) d x d θ = εmicro + g(λ) (ε Macro ε micro) }{{} GE effect parameterized by λ g(λ) is monotone in λ, with g(0) = 0 and g(1) = 1 g(λ) decreases with α tends to be closer to zero when GE effect is larger

38 Equivalence Result Proposition. For any {b t } and any t, there exists a λ such that 1 rational expectations Ē [p ] in inco-info economy same as ad hoc perceptions ˆp t in Tâtonnement economy 2 outcomes in inco-info economy same as in Tâtonnement economy 3 equal GE attenuation ε T ât (t) = ε Inc (λ) The converse is also true.

39 Complementary Results and Take-home Lesson Similar equivalence results for adaptive expectations reflective equilibrium (Garcia-Schmidt & Woodford, 2015) limited-depth reasoning Take-home lesson: lack of CK = relaxation of solution concept = GE dampened

40 Extension: GE Takes Time A dynamic extension Essentially repeated version of static economy u i = F (q i,t,q i,t,q Z i,t;a i ) 0 t=0 e ρt U(c i,t,c i,t,z i,t ;ξ i ) Slow learning about bar θ (or, equivalently, about global response)

41 Learning and GE Adjustment To avoid perfect aggregation of information: idiosyncratic iceberg costs for tradable good noisy private learning through realized prices Lemma. There exists an increasing sequence {λ t } such that, t, Ē h t [ θ ] = λ h t θ Ē t [p ] = Ω h=1 Proposition. Macro elasticity at t α h 1 Ē h t [ θ ] = f (λ t ) θ ε t = ε Inc (λ t ) = ε micro + w t (ε Macro ε micro) } {{ } GE t where w t = g(λ t ) is increasing in t, with w 0 < 1 = w Similar to static model, except that now λ t increases with time

42 Slow GE Adjustment Formalization of notion that GE adjustment takes time in short run, macro effect is close to micro/pe effect but as time passes, it converges to what predicted by standard model Speed of convergence? not surprisingly, it depends on quality of learning more interestingly, it depends on magnitude of GE effect Prop. For any given {λ t }, the sequence {w t } converges to 1 at a rate that is decreasing in α. In this sense, stronger GE effect slower GE adjustment!

43 Slow GE Adjustment Formalization of notion that GE adjustment takes time in short run, macro effect is close to micro/pe effect but as time passes, it converges to what predicted by standard model Speed of convergence? not surprisingly, it depends on quality of learning more interestingly, it depends on magnitude of GE effect Prop. For any given {λ t }, the sequence {w t } converges to 1 at a rate that is decreasing in α. In this sense, stronger GE effect slower GE adjustment!

44 Second paper: Angeletos and Lian (2016b) Forward Guidance without Common Knowledge

45 Forward Guidance Puzzle Context: an economy during a liquidity trap zero-lower bound (ZLB) binds for t T 1 Forward Guidance = promise at t to keep interest rates low at t T The Puzzle: standard NK model predicts that the stimulating effect is quantitatively large, increases with T, and explodes as T

46 Our contribution Prelim result: beauty-contest representation of NK model disentangle PE and GE effects recast GE effects as HOB Main result: remove CK anchor expectations of income and inflation attenuate relevant GE effects reduce power of forward guidance Complementary results: paradox of flexibility, discounted Euler/NKPC

47 Differences from Earlier Abstract Framework Concrete context, precise micro-foundations, policy focus Truly dynamic environment forward-looking expectations dynamic beauty contest Specific novel insights GE effects tied to expectations of future income and future inflation longer horizons map to beliefs of higher order dampening increases with T lessen forward-guidance puzzle dampening increases with price flexibility lessen paradox of flexibility

48 Textbook NK Model demand block: Euler condition (aka IS curve) y t = E t [y t+1 (R t π t+1 )] supply block: NK Philips Curve π t = βe t [π t+1 ] + κy t Monetary Policy: ZLB and forward guidance Rt = 0 t T 1 RT free forward guidance moves E 0 [R T ] yt = π t = 0 t T + 1 ex post optimal The puzzle: y 0 E 0 [R T ] increases with T and explodes as T

49 Textbook NK Model demand block: Euler condition (aka IS curve) y t = E t [y t+1 (R t π t+1 )] supply block: NK Philips Curve π t = βe t [π t+1 ] + κy t Monetary Policy: ZLB and forward guidance Rt = 0 t T 1 RT free forward guidance moves E 0 [R T ] yt = π t = 0 t T + 1 ex post optimal The puzzle: y 0 E 0 [R T ] increases with T and explodes as T

50 NK as Multi-Layer Beauty Contest Remove CK (of policy and/or of responses of others) Euler condition dynamic BC among consumers feedback from future spending to future income to current spending NKPC dynamic BC among firms feedback from future inflation to future MCs to current inflation Equilibrium: higher-layer BC between consumers and firms feedback from future inflation to current spending

51 NK as Multi-Layer Beauty Contest Euler condition dynamic BC among consumers y t = } c it di = c it = f (expected PV of income) = f (E it [y t+k ]) + y t = β k 1 { Ē t [R t+k 1 ] Ē t [π t+k ] } { } + + (1 β) β k 1 Ē t [y t+k ] k=1 k=1 NKPC dynamic BC among firms p t = θp t 1 + (1 θ) p it di p it = f (expected PV of nominal MC) = f (E it[p t+k ]) π t = κy t + κ + k=1 (βθ) k Ē f t [y t+k ] + 1 θ θ } = + (βθ) k Ēt f [π t+k ] + κµ t k=1

52 NK as Multi-Layer Beauty Contest Euler condition dynamic BC among consumers y t = } c it di = c it = f (expected PV of income) = f (E it [y t+k ]) + y t = β k 1 { Ē t [R t+k 1 ] Ē t [π t+k ] } { } + + (1 β) β k 1 Ē t [y t+k ] k=1 k=1 NKPC dynamic BC among firms p t = θp t 1 + (1 θ) p it di p it = f (expected PV of nominal MC) = f (E it[p t+k ]) π t = κy t + κ + k=1 (βθ) k Ē f t [y t+k ] + 1 θ θ } = + (βθ) k Ēt f [π t+k ] + κµ t k=1

53 GE Attenuation three GE mechanisms = three types of strategic complementary within demand block: income multiplier within supply block: pricing complementarity between two blocks: inflationary/deflationary spiral key insight: lack of CK attenuates all three at once!

54 Forward Guidance Dampened Proposition. With non-vanishing lack of CK, y 0 variant Ē 0 R T y 0 standard 0 E 0 R T as T Proposition. When lack of CK is sufficiently large, y 0 Ē 0 R 0 as T, T variant whereas y 0 E 0 R T standard.

55 Forward Guidance Dampened Proposition. With non-vanishing lack of CK, y 0 variant Ē 0 R T y 0 standard 0 E 0 R T as T Proposition. When lack of CK is sufficiently large, y 0 Ē 0 R 0 as T, T variant whereas y 0 E 0 R T standard.

56 Paradox of Flexibility, Discounting, and More Dampening is stronger when prices are more flexible low κ med κ high κ Lack of CK manifests as discounting of future expectations in Euler and NKPC of isomporphic representative-agent model Insights relevant also for shocks at ZLB, deflationary spirals, paradox of flexibility, eq. selection, neo-fisherian predictions...

57 Conclusion Worth revisiting solution concept and GE effects in macro even if we maintain individual rationality and PE effects Lack of CK = relaxation of solution concept = GE dampened formalization of GE takes time in short run, Macro is (close) to Micro Topical application: Forward Guidance Other applications... aggregate demand and Keynesian multipliers Ricardian equivalence, fiscal stimuli