Oil and macroeconomic (in)stability

Oil and macroeconomic (in)stability Hilde C. Bjørnland Vegard H. Larsen Centre for Applied Macro- and Petroleum Economics (CAMP) BI Norwegian Business School CFE-ERCIM December 07, 2014 Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 1 / 20

Our Question: Large literature debating the role of good luck versus good policy as cause of general decline in macroeconomic volatility ( The Great Moderation ) Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 2 / 20

Our Question: Large literature debating the role of good luck versus good policy as cause of general decline in macroeconomic volatility ( The Great Moderation ) We question if reduced oil price volatility can be responsible for some of the decline in macroeconomic volatility Yes, according to some: Nakov and Pescatori (2009, EJ), estimate a DSGE with a split sample (84Q1) Use counterfactual simulations to find that smaller oil sector shocks alone explains 17% and 11% of the decline in volatility of inflation and GDP growth respectively. Blanchard and Gaĺı (2008), estimate a S-VAR with a split sample (84Q1) Find that the effects of oil price shocks have changed over time, with steadily smaller effects on prices and wages, as well as on output and employment. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 2 / 20

Stylized facts (US macro variables and the real price of oil) Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 3 / 20

Decline in US volatility since 1984 Series 1970Q1 1983Q4 1984Q1 2014Q1 Inflation 0.56 0.25 (-55 %) GDP growth 1.18 0.61 (-48 %) Interest rate 0.87 0.70 (-17 %) Real oil price 18.79 13.72 (-27 %) Note: Standard deviations are reported. Percentage volatility reduction in parentheses. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 4 / 20

Decline in US volatility since 1984 Series 1970Q1 1983Q4 1984Q1 2014Q1 Inflation 0.56 0.25 (-55 %) GDP growth 1.18 0.61 (-48 %) Interest rate 0.87 0.70 (-17 %) Real oil price 18.79 13.72 (-27 %) Real oil price (1974Q1 is taken out) 7.80 13.72 (76 %) Note: Standard deviations are reported. Percentage volatility reduction in parentheses. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 5 / 20

What we do We revisit the role of oil prices in reducing US macroeconomic volatility. Address question by estimating a New-Keynesian model that allows for macroeconomic effects of oil price changes to the US (and vice versa). Use a Markov switching set up that allows for changes in both the volatility of shocks and in the monetary policy responses throughout the sample. Good luck (stochastic volatility change) Changes in the volatility of oil price shocks Good policy (did the FED changed its policy responses?) Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 6 / 20

Why Markov Switching? Previous studies split sample (exogenous break). Constant parameter model in each sub-sample. But constant parameter models abstract from considerations of structural changes; recession, policy changes, central bank interventions etc. If something happens in the past, it can happen again. Is the great moderation a thing of the past? Forward looking models - information is taken into account when agents form their expectations. Need a model that assigns probabilities to being in the various states. Markov-switching model. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 7 / 20

Results and contribution Novel attempt to explicitly model and analyze the implications of a change in volatility and the change in policy in explaining the Great Moderation where we especially look at oil price volatility. Allow for interaction between oil and the macroeconomy Find: There has been no decline in oil price volatility coinciding with the Great Moderation. BUT Several short periods of heightened oil price volatility throughout sample - prior to many NBER recessions. Changing volatility of US macro shocks gives the most substantial model performance boost. But policy also matters! Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 8 / 20

A New-Keynesian Model (ala Blanchard and Gali (2008)) IS-equation: y t = E t y t+1 ( r t E t [π t+1 ] ) + Λs t + z d,t (1) Phillips-equation: π t = βe t [π t+1 ] + κy t + Γs t + z s,t (2) Taylor rule: r t = ρ r r t 1 + (1 ρ r )(φ π π t + φ y y t ) + σ r ɛ r,t (3) where ɛ r,t i.i.d. N(0, 1) Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 9 / 20

Model cont. Oil price process: s t p o,t p t = ρ o s t 1 + ζy t + σ o ɛ o,t where ɛ o,t i.i.d. N(0, 1) (4) Demand shock: Supply shock: z d,t = ρ d z d,t 1 + σ d ɛ d,t where ɛ d,t i.i.d. N(0, 1) (5) z s,t = ρ d z s,t 1 + σ s ɛ s,t where ɛ s,t i.i.d. N(0, 1) (6) Data Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 10 / 20

The general Markov switching framework E t { A + st+1 x t+1 (s t+1, s t ) + A 0 s t x t (s t, s t 1 ) + A s t x t 1 (s t 1, s t 2 ) + B st ε t } = 0 x t is a n 1 vector including all the endogenous (predetermined and non-predetermined) variables ε t N (0, I ), is the vector of structural shocks s t {1, 2,..., h} (s t, s t 1 ) denotes the state today s t and the state in the previous period s t 1 Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 11 / 20

The general framework II We have a transition matrix with entries p st,s t+1 denoting the prob of going from state s t in the current period to state s t+1 next period. s t, s t+1 {1, 2,..., h} This allows us to define the expectation E t A + s t+1 x t+1 (s t+1, s t ) h p st,s t+1 E t A s + t+1 x t+1 (s t+1, s t ) s t+1 =1 All models are estimated with Bayesian techniques. All calculations for solving and estimating the models are done by the RISE toolbox (an object-oriented Matlab toolbox). Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 12 / 20

Definition of the regimes Constant regime (M 1 ) Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 13 / 20

Definition of the regimes Constant regime (M 1 ) Macroeconomic volatility regime (M 2 ) Two regimes for general macroeconomic shock volatility where σ d, σ s switch between a high and low regime. Oil price volatility regime (M 3 ) Two regimes for the shock volatility of the oil price where σ o switch between a high and a low regime Monetary policy response regime (M 4 ) Two regimes for the central banks monetary response where φ π, φ y, ρ r switch between a high and low monetary policy response regime. We make a normalization: The high policy response regime is the periods where the FED responds the highest to inflation. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 13 / 20

Model performance Model Description Switching parameters Log-MDD Rank M 1 No parameters are allowed to 2223 #8 change. M 2 The volatility of the demand and σ d, σ s 2262 #4 supply shock can change. M 3 The variance of the oil price shock σ o 2253 #6 can change. M 4 The parameters in the Taylor rule φ π, φ y, ρ r 2224 #7 can change. M 8 A combination of M 2, M 3 and M 4. σ d, σ s, σ o, φ π, φ y, ρ r 2301 #1 Letting the variance of the shocks to the US macroeconomic variables switch, is giving the largest improvement in model score Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 14 / 20

Model performance Model Description Switching parameters Log-MDD Rank M 1 No parameters are allowed to 2223 #8 change. M 2 The volatility of the demand and σ d, σ s 2262 #4 supply shock can change. M 3 The variance of the oil price shock σ o 2253 #6 can change. M 4 The parameters in the Taylor rule φ π, φ y, ρ r 2224 #7 can change. M 8 A combination of M 2, M 3 and M 4. σ d, σ s, σ o, φ π, φ y, ρ r 2301 #1 Letting the variance of the shocks to the US macroeconomic variables switch, is giving the largest improvement in model score Allowing for a switch in oil price volatility is also improving the model Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 14 / 20

Model performance Model Description Switching parameters Log-MDD Rank M 1 No parameters are allowed to 2223 #8 change. M 2 The volatility of the demand and σ d, σ s 2262 #4 supply shock can change. M 3 The variance of the oil price shock σ o 2253 #6 can change. M 4 The parameters in the Taylor rule φ π, φ y, ρ r 2224 #7 can change. M 8 A combination of M 2, M 3 and M 4. σ d, σ s, σ o, φ π, φ y, ρ r 2301 #1 Letting the variance of the shocks to the US macroeconomic variables switch, is giving the largest improvement in model score Allowing for a switch in oil price volatility is also improving the model Allowing for switches in the Taylor rule is the least important improvement, but still an improvement Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 14 / 20

Model performance Model Description Switching parameters Log-MDD Rank M 1 No parameters are allowed to 2223 #8 change. M 2 The volatility of the demand and σ d, σ s 2262 #4 supply shock can change. M 3 The variance of the oil price shock σ o 2253 #6 can change. M 4 The parameters in the Taylor rule φ π, φ y, ρ r 2224 #7 can change. M 8 A combination of M 2, M 3 and M 4. σ d, σ s, σ o, φ π, φ y, ρ r 2301 #1 Letting the variance of the shocks to the US macroeconomic variables switch, is giving the largest improvement in model score Allowing for a switch in oil price volatility is also improving the model Allowing for switches in the Taylor rule is the least important improvement, but still an improvement Allowing all parameter sets to switch gives the best fit Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 14 / 20

High volatility regime (Model M 8 ) The probability of being in the high macroeconomic volatility regime: The shaded areas represent NBER recessions. Parameters Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 15 / 20

High oil price volatility regime (Model M 8 ) The probability of being in the high oil volatility regime: The shaded areas represent NBER recessions. Parameters Episodes with high oil price volatility Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 16 / 20

High monetary policy response regime (Model M 8 ) The probability of being in the high policy response regime: The shaded areas represent NBER recessions. Parameters FED governors Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 17 / 20

The effects from an oil price shocks: Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 18 / 20

Extensions/Robustness Various measures of the output gap/unemployment End sample before the financial crisis (great moderation versus great recessions) More backward looking behavior Different policy rules. Only φ π changes. What about responding directly to the oil price? Model for oil prices: Global demand (versus US) drives oil prices; see for instance Aastveit, Bjørnland and Thorsrud (2014) Reduced oil in consumption and production, response in economy has changed. Λ and Γ has changed. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 19 / 20

Conclusion There has been no decline in oil price volatility coinciding with the Great Moderation. Instead: Several short periods of heightened oil price volatility throughout sample - prior to many NBER recessions. Allowing for changing volatility of US macro variables gives the most substantial performance boost compared to other specifications. Yet, best performing model is where policy, oil and macro volatility are allowed to change independently. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 20 / 20

Episodes with high oil price volatility: 1 1973 1974: OPEC embargo against countries supporting Israel during the Syrian and Egypt led attack on Israel. 2 1978 1979: Iranian revolution. 3 1986: Saudi Arabia increased oil production. 4 1990 1991: First Gulf war. 5 1997 2000: Asian financial crisis and Resumed Growth (Hamilton (2013)). 6 2001 2004: U.S. recession, Venezuelan unrest, and the second Gulf war. 7 2006 2010: The financial crisis and Growing demand and stagnant supply (Hamilton (2013)). Back Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 21 / 20

Prior and posterior distributions No-switching parameters. Prior distribution Posterior distribution Parameter Distribution 5% 95% Mean 5% 95% β Gamma 0.20 4.00 0.70 0.64 0.82 κ Beta 0.05 0.30 0.19 0.01 0.33 ζ Gamma 0.05 1.00 0.25 0.23 0.28 ρ d Beta 0.05 0.95 0.95 0.92 0.97 ρ s Beta 0.05 0.15 0.92 0.68 0.99 ρ o Beta 0.05 0.15 0.96 0.94 0.98 100Λ Gamma 0.05 1.00 0.20 0.08 0.32 100Γ Gamma 0.05 1.00 0.17 0.04 0.33 Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 22 / 20

Prior and posterior distributions The distribution of the parameters related to the macroeconomic volatility regimes. Prior distribution Posterior distribution Parameter Distribution 5% 95% Mean 5% 95% 100σ d (high) Inv. Gamma 0.05 1.00 0.31 0.22 0.44 100σ d (low) Inv. Gamma 0.05 1.00 0.09 0.07 0.14 100σ s (high) Inv. Gamma 0.05 1.00 0.25 0.10 0.40 100σ s (low) Inv. Gamma 0.05 1.00 0.09 0.04 0.15 p12 m Beta 0.05 0.15 0.09 0.05 0.13 p21 m Beta 0.05 0.15 0.08 0.05 0.11 Back Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 23 / 20

Prior and posterior distributions The distribution of the parameters related to the oil price volatility regimes. Prior distribution Posterior distribution Parameter Distribution 5% 95% Mean 5% 95% σ o (high) Inv. Gamma 0.05 1.00 0.23 0.20 0.27 σ o (low) Inv. Gamma 0.05 1.00 0.07 0.06 0.08 p12 o Beta 0.05 0.15 0.13 0.07 0.20 p21 o Beta 0.05 0.15 0.08 0.04 0.12 Back Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 24 / 20

Prior and posterior distributions The distribution of the parameters related to the policy response regimes. Prior distribution Posterior distribution Parameter Distribution 5% 95% Mean 5% 95% φ π (high) Gamma 0.50 5.00 3.63 3.22 4.54 φ π (low) Gamma 0.50 4.00 1.46 1.01 1.67 φ y (high) Gamma 0.05 2.00 0.11 0.02 0.24 φ y (low) Gamma 0.05 2.00 0.26 0.06 0.67 ρ r (high) Beta 0.05 0.95 0.78 0.56 0.85 ρ r (low) Beta 0.05 0.95 0.73 0.66 0.84 p p 12 Beta 0.05 0.95 0.06 0.03 0.10 p p 21 Beta 0.05 0.95 0.13 0.06 0.20 Back Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 25 / 20

Data We use quarterly data from 1970Q1 2014Q1 for the following variables: y t The U.S. output gap is calculated using the Hodrick-Prescott filter. π t The US inflation, calculated as the quarterly difference in logarithms of the GDP deflator. r t The U.S. Federal Funds Rate. s t The real oil price is the log of the WTI index divided by the GDP deflator. Back Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 26 / 20

FED governors and the Taylor rule coefficient on inflation Back Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 27 / 20

Blanchard, O. J. and J. Gali (2008). The macroeconomic effects of oil shocks: Why are the 2000s so different from the 1970s? NBER Working Papers 13368, National Bureau of Economic Research. Hamilton, J. D. (2013). Historical oil shocks. In R. E. Parker and R. Whaples (Eds.), Routledge Handbook of Major Events in Economic History, pp. 239 265. New York: Routledge Taylor and Francis Group. Bjørnland and Larsen Oil and macro (in)stability Pisa December 2014 20 / 20