Quantifying Confidence George-Marios Angeletos Fabrice Collard Harris Dellas Bank of Portugal, June 11, 2015 Angeletos, Collard, Dellas Quantifying Confidence 1 / 29 Standard approach coordination failure = multiple equilibria aggregate demand = gaps from flex prices, NKPC Angeletos, Collard, Dellas Quantifying Confidence 2 / 29 An alternative coordination failure aggregate demand Angeletos, Collard, Dellas = strategic uncertainty / beliefs Quantifying Confidence 3 / 29 An alternative coordination failure aggregate demand = strategic uncertainty / beliefs This Paper 1. tractable formalization 2. quantitative evaluation Angeletos, Collard, Dellas Quantifying Confidence 3 / 29 Contribution 1 explore observable implications of I imperfect coordination I relaxed solution concept 2 accommodate fluctuations in “confidence” 3 decouple AD from sticky prices 4 I bypass empirical failures of old and NK Philips curves I great recessions 6= great deflations explain multiple salient features of the data Angeletos, Collard, Dellas Quantifying Confidence 4 / 29 Roadmap Baseline Model and Methodological Contribution Quantitative Evaluation Extension to Medium-Scale DSGE & Estimation Complementary Empirical Work Angeletos, Collard, Dellas Quantifying Confidence 5 / 29 Baseline Model belief-enrichment of textbook RBC model geography I islands: differentiated intermediate goods, local L and K markets I mainland: final good (→ consumption and investment) I heterogeneous beliefs across islands sources of volatility I permanent shock to technology: At I transitory shock to HOB, or “confidence”: ξt Angeletos, Collard, Dellas Quantifying Confidence 6 / 29 Modeling Beliefs Stage 1 Stage 2 observe xit = log At + εit observe (At , Yt , prices) form beliefs about (Yt , Yt+1 , ...) update beliefs make production choices consume and invest t Angeletos, Collard, Dellas Quantifying Confidence t +1 7 / 29 Modeling Beliefs Stage 1 Stage 2 observe xit = log At + εit observe (At , Yt , prices) form beliefs about (Yt , Yt+1 , ...) update beliefs make production choices consume and invest t t +1 heterogeneous priors: εit ∼ N (0, σ) εjt ∼ N (ξt , σ) ξt → aggregate variation in HOB → “confidence” or “AD” Angeletos, Collard, Dellas Quantifying Confidence 7 / 29 ξt as a proxy for strategic uncertainty standard: Yt = Ēt [Yt ] = YtRBC ≡ χ At strategic uncertainty: Yt 6= Ēt [Yt ] = YtRBC + “belief wedge” Angeletos, Collard, Dellas Quantifying Confidence 8 / 29 ξt as a proxy for strategic uncertainty standard: Yt = Ēt [Yt ] = YtRBC ≡ χ At strategic uncertainty: Yt 6= Ēt [Yt ] = YtRBC + “belief wedge” Angeletos and La’O (Ecma 2013) I impose common prior (no biases) I abstract from capital, add market segmentation ⇒ observational equivalence Angeletos, Collard, Dellas Quantifying Confidence 8 / 29 ξt as a belief enrichment DSGE models vs “beauty contests”: behavior depends on beliefs of many endogenous outcomes (prices, wages, sales...) in many dates ξt = disciplined, parsimonious, and tractable belief enrichment research task: understand observable implications & quantify Angeletos, Collard, Dellas Quantifying Confidence 9 / 29 Methodological Contribution tractability, tractability, tractability.... take the limit as σ → 0 ⇒ I no learning, no Kalman filter I no cross-sectional heterogeneity, no Krusell-Smith I ξt is sufficient statistic for gap between higher- and first-order beliefs ⇒ small state spaces! solution almost as in representative-agent models Angeletos, Collard, Dellas Quantifying Confidence 10 / 29 Recursive equilibrium recursive equilibrium = PBE among fictitious local planners key objects: G, P, V1 , V2 I G = aggregate policy rule for capital: Kt+1 = G(At , ξt , Kt ) I P = local beliefs about prices (demand): I V1 , V2 = value functions of local planner in stages 1, 2 p̂it = P(xit , ξt , Kt ) heterogeneous priors → tractable fixed point → solution “almost” as in representative-agent models Angeletos, Collard, Dellas Quantifying Confidence 11 / 29 Recursive equilibrium stage-1 problem: V1 (k; x, ξ, K ) = max V2 (m̂; x, ξ, K ) − n s.t. 1 1+ν 1+ν n m̂ = p̂ ŷ + (1 − δ)k ŷ = xk θ n1−θ p̂ = P(x, ξ, K ) stage-2 problem: R V2 (m; A, ξ, K ) = max{c,k 0 } U(c) + β V1 (k 0 ; A0 , ξ 0 , K 0 )df (A0 , ξ s.t. c + k0 = m K 0 = G(A, ξ, K ) n(k, x, ξ, K ) & g(m, A, ξ, K )= policy rules for (n, k) y(x, A, ξ, K ) = output implied by policy rules Angeletos, Collard, Dellas Quantifying Confidence 12 / 29 Recursive equilibrium belief consistency: P(x, ξ, K ) = y(x + ξ, x, ξ, K ) y(x, x, ξ, K ) aggregation: X G(A, ξ, K ) = g y(A, A, ξ, K ) + (1 − δ)K ; A, ξ, K bottom line: tractable fixed-point problem Angeletos, Collard, Dellas Quantifying Confidence 13 / 29 Log-linear solution original model: (Ct , It , Nt ; Kt+1 ) = Γk · Kt + Γa · At + Γξ · ξt belief-augmented model: (Ct , It , Nt ; Kt+1 ) = Γk · Kt + Γa · At + Γξ · ξt generalization to arbitrary linear DSGE models (see Appendix) → simulate/calibrate/estimate as in standard DSGE models Angeletos, Collard, Dellas Quantifying Confidence 14 / 29 Calibration fix all familiar params to conventional values Parameter Role Value β δ ν α ψ Discount Rate Depreciation Rate Inverse Elasticity of Labor Supply Capital Share in Production Inverse Elasticity of Utilization 0.990 0.015 0.500 0.300 0.300 fix persistence of belief shock to ρ = .75 choose σa and σξ so as to match of BC volatilities of Y , H, I , C Angeletos, Collard, Dellas Quantifying Confidence 15 / 29 Observable implications: IRFs to confidence shock Output % deviation 2 Productivity 0.5 1.5 Consumption 0.2 0.15 1 0 1.5 1 2 0.5 0.05 0 0 -0.5 0 10 Quarters 20 Hours Worked 2 4 0.1 0.5 Investment 6 0 0 0 10 Quarters 20 0 10 Quarters 20 0 10 Quarters 20 0 10 Quarters 20 co-movement patterns very different from I I I investment- or consumption-specific shocks news or noise shocks any shock that works through TFP (e.g., uncertainty shocks) similar to monetary shock in NK, but w/o inflation Angeletos, Collard, Dellas Quantifying Confidence 16 / 29 Pony Race: Confidence Shocks vs NK Demand Shocks NK with TFP plus... Data Our RBC stddev(y ) stddev(h) stddev(c) stddev(i) 1.42 1.56 0.76 5.43 corr(c, y ) corr(i, y ) corr(h, y ) corr(c, h) corr(i, h) corr(c, i) corr(y , y /h) corr(h, y /h) corr(y , sr ) corr(h, sr ) I shock C shock News Monetary 1.42 1.52 0.76 5.66 1.24 1.18 0.86 7.03 1.15 0.97 0.95 7.04 1.29 1.02 0.84 7.24 1.37 1.44 0.77 6.20 0.85 0.94 0.88 0.84 0.82 0.74 0.77 0.92 0.85 0.34 0.99 0.47 0.42 0.82 0.80 -0.19 1.00 -0.17 0.37 0.75 0.77 -0.29 1.00 -0.33 0.43 0.84 0.86 -0.07 1.00 -0.13 0.73 0.90 0.84 0.24 0.99 0.35 0.08 -0.41 0.82 0.47 0.15 -0.37 0.85 0.47 0.37 -0.24 0.92 0.52 0.54 -0.10 0.92 0.49 0.61 0.13 0.94 0.65 0.20 -0.36 0.94 0.61 Angeletos, Collard, Dellas Quantifying Confidence 17 / 29 Take-home lesson (so far) a simple formalization of non-monetary demand shocks superior performance within “textbook” models key to quantitative success: I waves of optimism/pessimism about “demand” in the short run I disconnect from TFP and labor productivity Angeletos, Collard, Dellas Quantifying Confidence 18 / 29 Extensions medium-scale DSGE → robustness and structural estimation multiple shocks → multiple competing mechanisms I I I I I I permanent and transitory TFP shock permanent and transitory investment-specific shock news about future productivity discount-factor shock fiscal shock monetary shock also: IAC and HP → endogenous persistence, plus help NK two versions: flexible vs sticky prices Angeletos, Collard, Dellas Quantifying Confidence 19 / 29 Observable Implications Output 1 Consumption 0.5 0.8 0.4 0.6 0.3 0.4 0.2 0.2 0.1 0 10 Quarters 20 1 1 0.5 0 0 -1 0 10 Quarters 20 Hours Worked 1.5 2 0 0 Investment 3 -0.5 0 10 Quarters 20 Inflation Rate 0.08 Nom. Interest Rate 2 0.06 0 0.04 -2 0.02 -4 0 -6 -0.02 0 10 Quarters Flexible Prices (RBC) 20 -8 0 10 Quarters 20 0 10 Quarters 20 Sticky Prices (NK) similar effects in RBC vs NK, or in textbook vs medium-scale models important: that’s NOT the case for other shocks/mechanisms Angeletos, Collard, Dellas Quantifying Confidence 20 / 29 Estimated Contribution despite multiple competing forces, estimation attributes more than half of the observed business cycles to “confidence” contribution to volatility (6-32 quarters) Flexible Prices Sticky Prices Y C I h π R 50.98 47.73 43.72 40.89 54.63 44.24 76.04 65.66 0.00 11.95 99.15 32.64 contribution to covariances (6-32 quarters) Flexible Sticky (Y , h) (Y , I ) (Y , C ) (h, I ) (h, C ) (I , C ) 75.80 68.53 60.06 53.23 56.34 58.40 75.67 62.64 96.53 106.30 84.75 107.41 Angeletos, Collard, Dellas Quantifying Confidence 21 / 29 What Is Aggregate Demand? posterior odds of competing models: RBC <<< NK << RBC with confidence interpretation: a potent theory of AD (even) w/o nominal rigidity Angeletos, Collard, Dellas Quantifying Confidence 22 / 29 Complementary Empirical Work (Angeletos-Collard-Dellas) Angeletos, Collard, Dellas Quantifying Confidence 23 / 29 Complementary Empirical Work (Angeletos-Collard-Dellas) bypass any particular model or any structural VARs “anatomy” of comovement I construct factors designed to capture volatility of certain variables I inspect comovement patterns across variables and/or frequencies Angeletos, Collard, Dellas Quantifying Confidence 23 / 29 Complementary Empirical Work (Angeletos-Collard-Dellas) bypass any particular model or any structural VARs “anatomy” of comovement I construct factors designed to capture volatility of certain variables I inspect comovement patterns across variables and/or frequencies Angeletos, Collard, Dellas Quantifying Confidence 23 / 29 Complementary Empirical Work (Angeletos-Collard-Dellas) bypass any particular model or any structural VARs “anatomy” of comovement I construct factors designed to capture volatility of certain variables I inspect comovement patterns across variables and/or frequencies show that this anatomy points towards shocks/mechanisms that are I highly transitory I disconnected from both productivity and inflation I unlike usual suspects Angeletos, Collard, Dellas Quantifying Confidence 23 / 29 Identifying a “Business Cycle Factor” 1 VAR/VECM on {Y , H, I , C , PI /PC , π, R, G , ...} with 2 unit-root components 2 “business cycle factor” = combination of VAR innovations that maximizes band-pass volatility of Y and/or (H, I ) at 6-32 quarters Angeletos, Collard, Dellas Quantifying Confidence 24 / 29 Factor: Variance Contribution Y Baseline, with permanent 6–32 quarters 49.62 32–80 quarters 21.49 80–∞ quarters 0.00 Angeletos, Collard, Dellas I h C components excluded 55.70 49.22 24.34 28.52 28.19 8.89 0.00 3.37 0.00 Quantifying Confidence Y /h Pi π R 15.03 6.44 0.00 5.92 4.24 0.00 17.74 14.63 2.08 31.33 31.38 5.15 25 / 29 Factor: Variance Contribution Y Y /h Pi π R components excluded 55.70 49.22 24.34 28.52 28.19 8.89 0.00 3.37 0.00 15.03 6.44 0.00 5.92 4.24 0.00 17.74 14.63 2.08 31.33 31.38 5.15 Variant, with permanent components included 6–32 quarters 47.97 55.87 58.97 21.45 32–80 quarters 17.27 25.01 26.55 9.46 80–∞ quarters 6.67 6.67 7.26 6.66 23.23 12.89 6.67 4.96 6.22 6.62 15.87 15.86 6.68 44.39 43.44 9.52 Baseline, with permanent 6–32 quarters 49.62 32–80 quarters 21.49 80–∞ quarters 0.00 Angeletos, Collard, Dellas I h C Quantifying Confidence 25 / 29 Factor: Comovement Patterns Output Consumption Investment 0.5 2 0.5 0 0 0 -0.5 -2 -0.5 5 10 15 Quarters Hours Worked 20 5 10 15 Quarters Productivity 20 1 0 -0.5 0 0 -0.5 -0.5 -1 5 1 10 15 Quarters Gov. Spending 10 15 Quarters Rel. Price of Inv. 20 20 0.5 0.5 0.5 5 5 10 15 Quarters Inflation Rate 20 0.2 0.2 0.5 0.1 0.1 0 0 0 -0.5 -0.1 -0.1 -0.2 5 10 15 Quarters Nom. Int. Rate 20 5 10 15 Quarters 20 -0.2 -1 5 10 15 Quarters baseline Angeletos, Collard, Dellas 20 5 10 15 Quarters 20 variant Quantifying Confidence 26 / 29 Model Counterpart? has to be transitory has to trigger strong comovement in (Y , H, I , C ), without strong comovement in either (Y /H, TFP, Pi/PC ) or π unlike any of the “usual suspects” in standard models I not technology I not news/noise I not financial or uncertainty shock that work through TFP I not I- or C-specific shocks what can this be? “confidence”, or something else Angeletos, Collard, Dellas Quantifying Confidence 27 / 29 Confidence shock (model) vs factor (data) 4 Output 15 Investment 6 Hours Worked Consumption 2 3 10 4 5 2 1.5 2 1 1 0.5 0 0 0 -5 -2 -10 -4 0 -0.5 -1 -1 -2 -3 -1.5 -2 -4 -15 1970 1980 1990 2000 -6 1970 1980 1990 2000 1970 1980 1990 2000 1970 1980 1990 2000 this explains why structural estimation favors confidence shock evidence in favor of our theory and/or against standard theories Angeletos, Collard, Dellas Quantifying Confidence 28 / 29 Conclusion Methodological contribution: I embed tractable higher-order beliefs in a large class of macro models I accommodate a certain relaxation of solution concept Applied contribution: I reveal observable implications of HOB I accommodate waves of optimism and pessimism about SR I accommodate “aggregate demand” without sticky prices I explain multiple salient features of the data Angeletos, Collard, Dellas Quantifying Confidence 29 / 29