Technological Revolutions and Debt Hangovers: Is There a Link? Dan Cao Jean-Paul L Huillier January 9th, 2014 Cao and L Huillier 0/41
Introduction Observation: Before Great Recession: IT (late 1990s) Before Japanese crisis 1990s: electronics (1980s) Before Great Depression: combustion/electricity (1910-1925?) Write a business cycle model Tech revolutions = consumption slumps Structural estimation Cao and L Huillier 1/41
Goal and Features of the Model Key role: anticipations about the future News (Beaudry & Portier AER 2006) But: noisy information Goal: Measure beliefs about the long-run Then, use model to characterize implied dynamics of 1. Consumption 2. Household debt Cao and L Huillier 2/41
Finding: A Sequence of Events 1. Start of tech rev increase in growth rates of pty 2. Wave of optimism, increase in spending 3. Pty slowdown, accumulation of debt 4. Deleveraging and consumption slump Altogether: 20 to 25 years to be completed Shifts in the economy are rarely forecast and often not fully recognized until they have been underway for some time. Larry Summers Financial Times, March 25, 2012 Cao and L Huillier 3/41
Procedure: Structural Estimation Focus: 1. Consumption 2. Medium-run dynamics Estimated medium-run dynamics allow to generate: model predicted beliefs about the long run use these for out-of-sample checks Also present reduced form evidence Cao and L Huillier 4/41
Main Results 1. Delayed adjustment of beliefs Half life is 5.25 years 2. Large accumulation of debt, slow deleveraging Large Reason: Optimism when productivity is slowing down Slow Reason: Income is low when deleveraging 3. Pty-to-consumption ratio has common medium freq. shape This is consistent with our model Cao and L Huillier 5/41
Related Literature News and business cycle fluctuations Beaudry & Portier 2006; Schmitt-Grohe & Uribe 2012; Lorenzoni 2009 Permanent and transitory technology shocks Aguiar & Gopinath 2007; Boz, Daude & Durdu 2011 Household debt cycles Mian & Sufi 2012; Primiceri, Justiniano, Tambalotti; Midrigan and Philippon Cao and L Huillier 6/41
Model: Productivity (or TFP) Productivity sum of two components: a t = x t + z t permanent component x t = ρ x t 1 + ε t transitory component z t = ρz t 1 + η t. Cao and L Huillier 7/41
Information News and Noise information structure: Agents observe: 1. current productivity a t 2. noisy signal regarding the permanent component x t s t = x t + ν t Plan: 1. Study Kalman filter of these agents 2. Put into model a la Aguiar & Gopinath (2007) 3. Do structural estimation for the 3 episodes Cao and L Huillier 8/41
Key: Slow Adjustment of Beliefs Borrow idea from Hobijn & Jovanovic (1999): Technological revolutions come in waves 1. Start of tech rev Increase in growth of permanent productivity (from the old, deterministic, trend) 2. End of tech rev Decrease in growth of permanent productivity (from the new trend) Consumers use Kalman filter to update beliefs Try to track path of permanent component Slow to adjust beliefs after slowdown Remain optimistic for a while Cao and L Huillier 9/41
Beliefs About the Long Run Notation: E t [X τ ] E [X τ I t ] Definition Beliefs about the long run (BLR) are the conditional expectation about the long-run level of productivity: lim E t[a t+τ ] τ Consumers use Kalman filter to update beliefs Form beliefs about path of permanent component Derive BLR Cao and L Huillier 10/41
Formula for BLR BLR are: lim τ E t [a t+τ ] = lim τ E t [x t+τ + z t+τ ] Permanent component: x t = ρ x t 1 + ε t lim τ E t[x t+τ ] = E t [ x t + τ=1 ρ τ x t ] Transitory component: z t = ρz t 1 + η t lim E t [z t+τ ] = 0 τ = E t[x t ] ρe t [x t 1 ] 1 ρ = lim τ E t [a t+τ ] = lim τ E t [x t+τ ] = E t[x t ] ρe t [x t 1 ] 1 ρ Cao and L Huillier 11/41
An Example: Off-trend Permanent Tech Shocks Cao and L Huillier 12/41
An Example: Off-trend Permanent Tech Shocks, cont. Cao and L Huillier 13/41
Delayed Adjustment of Beliefs To fix ideas, a definition Definition After ε t = 1: under perfect info, BLR jumps long-run level 1/(1 ρ) under noisy info, it takes time for the BLR to adjust delay is the time it takes BLR to reach half of 1/(1 ρ). Quantify this delay later Will also show reduced form evidence Cao and L Huillier 14/41
Model Open economy DSGE Use previous information structure Permanent income consumers Form beliefs about the future path of xt (BLR) These beliefs affect consumption and net exports Cao and L Huillier 15/41
Consumers and Production Representative consumer maximizes [ ] E β t ln(c t ) subject to t=0 C t +B t 1 = W t N+Q t B t B t is external debt Linear production and competitive goods market Y t = e a t N Cao and L Huillier 16/41
Resource Constraint and Interest Rate Resource constraint C t + NX t = Y t Interest rate 1 } = R t = R + ψ {e B t Yt b 1 Q t b is steady-state level of B/Y ratio Cao and L Huillier 17/41
Effect of Permanent Shock: Parametrization Fixed parameters Parameter Description Value β Discount Factor 0.99 ψ Sensitivity Interest Rate 0.0010 ψ as previous literature Calibrate b from flow of funds (C/Y =.9979) ML estimate of ρ, σ ε, σ η, σ ν using data for Great Recession Cao and L Huillier 18/41
BLR and Consumption Proposition In the loglinear solution of the model, when β 1 and ψ/(1 β) 0, consumption is only a function of BLR. Specifically, ĉ t = 1 C/Y x t+ t Relevant for parametrization adopted in the literature: ψ is small ( 0.0010), β close to one Conceptually useful Also: Result hold in more general model Cao and L Huillier 19/41
IRFs to a Permanent Tech Shock ε t 0.8 0.03 1.2 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.02 0.01 0 0.01 0.02 1 0.8 0.6 0.4 0.2 0 0.2 0.4 0 0.03 0 10 20 30 0 10 20 30 Productivity Net Exports 0 10 20 30 Debt to output Ratio Delay in the accumulation of debt Cao and L Huillier 20/41
ML Estimates: US 1990 2013 Observables: labor productivity and net exports Estimates: Param. Description Value s.e. ρ Persistence tech. process 0.97 0.01 σ a Std. dev. productivity 0.64 0.04 σ ε Std. dev. permanent shocks (implied) 0.02 σ η Std. dev. transitory shocks (implied) 0.63 σ ν Std. dev. noise shocks 7.39 2.04 Delay: half-life of BLR is 5.25 years Cao and L Huillier 21/41
Robustness Estimation ρ σ a σ ν ψ Delay 1. Baseline 1990 2013 0.9720 (0.0057) 2. 1985 2013 0.9737 (0.0046) 3. 1980 2013 0.9777 (0.0034) 4. Full 1948 2013 0.9893 (0.0010) 5. Estimating ψ 0.9659 (0.0099) 6. TFP 0.9739 (0.0056) 7. Cons. (limit model) 0.9736 (0.0073) 8. 1920 1935 (limit model) 0.9493 (0.0145) (Standard errors are reported in parenthesis) 0.6433 (0.0446) 0.6183 (0.0390) 0.7020 (0.0415) 1.0999 (0.0544) 0.5604 (0.0382) 0.6918 (0.0483) 0.5254 (0.0303) 1.7849 (0.1398) 7.3922 (2.0451) 6.2959 (1.7023) 9.0652 (2.1890) 10.6144 (3.2643) 10.0022 (2.4024) 9.3421 (2.4848) 2.0509 (0.6153) 8.9030 (3.1423) 0.0010 5.25 0.0010 5.50 0.0010 6.50 0.0010 9.75 0.0001 (0.0002) 4.75 0.0010 6.00 0 3.75 0 2.50 Cao and L Huillier 22/41
Variance Decomposition of BLR at Different Horizons 1 0.9 0.8 Variance Share 0.7 0.6 0.5 0.4 0.3 0.2 0.1 Permanent Transitory Noise 0 0 5 10 15 20 25 30 Years Cao and L Huillier 23/41
Smoothed Permanent Shocks ε t 0.05 0.04 0.03 0.02 0.01 0 0.01 0.02 0.03 0.04 0.05 1995 2000 2005 2010 Consistent with Fernald (2012) Cao and L Huillier 24/41
Smoothed Transitory and Noise Shocks ((η t and ν t ) 2 Transitory 1 0 1 2 20 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 Noise 10 0 10 20 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 Cao and L Huillier 25/41
Effect of Permanent Shocks on BLR 15 10 5 0 5 10 1995 2000 2005 2010 Lag in optimism, peak between 00 and 05 Cao and L Huillier 26/41
Out-of-Sample Check: Comparison with Survey Evidence From Hoffman, Krause, and Laubach: Cao and L Huillier 27/41
Out-of-Sample Check: Comparison with Survey Evidence Comparing: Cao and L Huillier 28/41
Estimates for Japan and Great Depression Same procedure as for Great Recession Japan Great Dep. Param. Description Value s.e. Value s.e. ρ Persistence 0.94 0.02 0.86 0.05 σ a Std. dev. productivity 1.00 0.06 1.66 0.15 σ ε Std. dev. permanent 0.06 0.24 σ η Std. dev. transitory 0.97 1.53 σ ν Std. dev. noise 14.49 3.50 20.05 8.06 Similar estimates: large ρ, large σ ν Delays: 2.5 years (Japan) and 1 year (GD) Cao and L Huillier 29/41
Smoothed Permanent Shocks ε t 0.15 Japan 1 Great Depression 0.1 0.05 0.5 0 0 0.05 0.1 0.5 1980 1990 2000 1 1920 1925 1930 1935 Cao and L Huillier 30/41
Effect of Permanent Shocks on BLR 30 Japan 10 Great Depression 20 10 5 0 5 0 10 10 20 15 20 25 30 1980 1990 2000 30 1920 1925 1930 1935 Cao and L Huillier 31/41
Reduced-Form Evidence Look at the ratio of productivity-to-consumption Productivity: measure of income Consumption: measure of BLR Features common medium-run dynamics Cao and L Huillier 32/41
Productivity-to-consumption Ratio, US (1990 2013) Wavy-form : 4 stages Cao and L Huillier 33/41
Productivity-to-consumption Ratio, US (1990 2013) Wavy-form : (in red, HP-trend, λ = 800) 4 stages Cao and L Huillier 33/41
No-news Benchmark σ ν : Uninformative signal Pty is random walk = BLR = a t So if ψ small = c t = a t = Ratio is flat Does not fit the data Cao and L Huillier 34/41
Perfect Foresight Benchmark c t = x and flat Pty goes up and then down Then: Ratio goes up and then down as well Does not fit the data Cao and L Huillier 35/41
Japan (1975 2005) Wavy-form : Cao and L Huillier 36/41
Japan (1975 2005) Wavy-form : (in red, HP-trend, λ = 800) Cao and L Huillier 36/41
Great Depression (US 1920 1935) Wavy-form : Cao and L Huillier 37/41
Great Depression (US 1920 1935) Wavy-form : (in red, HP-trend, λ = 800) Cao and L Huillier 37/41
Dynamics of Debt Depends on three elements: 1. Persistence of permanent technology process: ρ (income effect + persistence of beliefs) 2. Standard deviation of noise: σ ν (speed of learning) 3. Timing of the shocks (degree of optimism before slowdown) Cao and L Huillier 38/41
Simulation of the Dynamics of Debt 10 9 8 7 6 5 4 3 2 1 0.6 0.4 0.2 0 0.2 0.4 0.6 25 20 15 10 5 0 0 1990 2000 0.8 2010 1990 2000 2010 Productivity Net Exports 5 1990 2000 2010 Debt to output Ratio In 2010: low productivity and high debt Cao and L Huillier 39/41
Simulation: Path of Debt After 2010 25 20 15 10 5 0 5 1990 1995 2000 2005 2010 2015 2020 Debt to output Ratio Deleveraging process is slow Cao and L Huillier 40/41
Conclusion Main point: Medium-run look at the data of cons and pty: useful to understand build-up of debt and deleveraging Key: delay in adjustment of BLR Highlight common features for 3 episodes Sequence of events Wavy-form Cao and L Huillier 41/41