Pushing on a string: US monetary policy is less powerful in recessions Silvana Tenreyro and Gregory Thwaites LSE and Bank of England September 13
Disclaimer This does not represent the views of the Bank of England
Summary US monetary policy shocks have much more powerful effects on output and inflation during economic expansions than during recessions.
Summary US monetary policy shocks have much more powerful effects on output and inflation during economic expansions than during recessions. A shock that raises the policy rate by 1pp has an impulse response many times larger in a boom than in a recession.
Summary US monetary policy shocks have much more powerful effects on output and inflation during economic expansions than during recessions. A shock that raises the policy rate by 1pp has an impulse response many times larger in a boom than in a recession. The effect is most pronounced in household durables expenditure, but also present in business investment and nondurable consumption.
Summary US monetary policy shocks have much more powerful effects on output and inflation during economic expansions than during recessions. A shock that raises the policy rate by 1pp has an impulse response many times larger in a boom than in a recession. The effect is most pronounced in household durables expenditure, but also present in business investment and nondurable consumption. What is not going on
Summary US monetary policy shocks have much more powerful effects on output and inflation during economic expansions than during recessions. A shock that raises the policy rate by 1pp has an impulse response many times larger in a boom than in a recession. The effect is most pronounced in household durables expenditure, but also present in business investment and nondurable consumption. What is not going on No systematic difference across regimes in the responses of taxes, government spending, credit volumes or credit spreads to monetary shocks
Summary US monetary policy shocks have much more powerful effects on output and inflation during economic expansions than during recessions. A shock that raises the policy rate by 1pp has an impulse response many times larger in a boom than in a recession. The effect is most pronounced in household durables expenditure, but also present in business investment and nondurable consumption. What is not going on No systematic difference across regimes in the responses of taxes, government spending, credit volumes or credit spreads to monetary shocks No difference in proportion of positive or negative shocks across regimes
Summary US monetary policy shocks have much more powerful effects on output and inflation during economic expansions than during recessions. A shock that raises the policy rate by 1pp has an impulse response many times larger in a boom than in a recession. The effect is most pronounced in household durables expenditure, but also present in business investment and nondurable consumption. What is not going on No systematic difference across regimes in the responses of taxes, government spending, credit volumes or credit spreads to monetary shocks No difference in proportion of positive or negative shocks across regimes No effect from shocks being larger or smaller in different regimes
Plan for today Literature Method Data Results Discussion
Literature Mixed results on state-dependence of monetary policy multipliers Identification problems (common to many studies on effects of policy) General form is (y t y t ) = α (L) (y t 1 y t 1 ) + βx t + ε t where x is shock and y is outcome Older studies typically allow only a proper subset of {α (L), β, y} to depend on the state of the cycle We allow generalise β to β (L) and allow it to depend flexibly on the cycle (Auerbach and Gorodnichenko) This appears to make a big difference
Method (1) Local projection method (Jorda, AER 5) Regress leads of the response variable y on the shocks x Run H separate regressions for h {1,.., H} Coefficient β h on the shock when the lead is h periods is the IRF at horizon h y t+h = β h x t + γ z t + u t, h {1,.., H}
Method () Smooth transition regression (Granger and Terasvirta, 1993) Multiply the RHS variables by the state of the economy F (z) and (1 F (z)) where z is an indicator of the state y t = F (z t ) β b x t + (1 F (z t )) β r x t + u t ( ) F (z t ) = exp θ (z t c) σ z ( ) 1 + exp θ (z t c) σ z β b is the coefficient in an expansion, β r is the coefficient in a recession.
Method (3) Smooth transition, local projection model (Auerbach and Gorodnichenko, Ramey et al). Run H regressions of the following form ( ) ) y t+h = F (z t ) β h b x t + γ b z t + (1 F (z t )) (β h r x t + γ r z t + u t β h b h {1,.., H} is the IRF at horizon h if you were in an expansion when the shock hit
Data US national accounts data 1969-8 Exclude the crisis and the ZLB Extended series of Romer monetary policy shocks (Coibion et al.) State variable centred 7-quarter moving average of GDP growth (Auerbach and Gorodnichenko)
Results (1) - GDP.15.1.5 GDP level impulse response linear Expansion Recession 1.5 1.5 T stat: boom = recession Log points -.5 -.5 -.1-1 -1.5 -.15 - -. 5 1 15 -.5 5 1 15
Results () - Household durable consumption and housing investment.4. Durables level impulse response linear Expansion Recession 1.5 T stat: boom = recession 1.5 Log points -. -.4 -.5-1 -.6-1.5 -.8 5 1 15-5 1 15
Results (3) - Nondurable household consumption Log points Nondurables level impulse response.1 linear.8 Expansion Recession.6.4. -. -.4 -.6 -.8.5 1.5 1.5 -.5-1 -1.5 - T stat: boom = recession -.1 5 1 15 -.5 5 1 15
Results (4) - GDP deflator inflation (quarterly annualised) 4 x 1-3 Inflation impulse response linear Expansion Recession 1 T stat: boom = recession - -4 Log points -6-8 -1-1 - -1-3 -14-16 5 1 15-4 5 1 15
What is not going on Positive v negative shocks: y t+h = τt + α b h + β+ h max [, ε t] + β h min [, ε t] + γ b x t + u t Positive shocks (tightenings) have bigger effects, but they are equally common in booms and recessions So cannot explain our finding
What is not going on Positive v negative shocks: y t+h = τt + α b h + β+ h max [, ε t] + β h min [, ε t] + γ b x t + u t Positive shocks (tightenings) have bigger effects, but they are equally common in booms and recessions So cannot explain our finding Large v small shocks y t+h = τt + α b h + β s h ε t + β l h ε3 t + γ b x t + u t If anything, bigger shocks are proportionally more powerful, but noally signficant Fatter tails during recessions So cannot explain our finding
What is not going on Positive v negative shocks: y t+h = τt + α b h + β+ h max [, ε t] + β h min [, ε t] + γ b x t + u t Positive shocks (tightenings) have bigger effects, but they are equally common in booms and recessions So cannot explain our finding Large v small shocks y t+h = τt + α b h + β s h ε t + β l h ε3 t + γ b x t + u t If anything, bigger shocks are proportionally more powerful, but noally signficant Fatter tails during recessions So cannot explain our finding No significant difference in responses of fiscal policy or credit spreads across the cycle
Issues to explore Nonlinear identification scheme Normalisation of shocks across regimes
Outline of a model - intuition The difference in the effect of policy is most pronounced in durables purchases
Outline of a model - intuition The difference in the effect of policy is most pronounced in durables purchases Durables purchases are lumpy, characterised by long periods of inaction
Outline of a model - intuition The difference in the effect of policy is most pronounced in durables purchases Durables purchases are lumpy, characterised by long periods of inaction More likely to be in inaction zone during a recession (Berger and Vavra)
Outline of a model - intuition The difference in the effect of policy is most pronounced in durables purchases Durables purchases are lumpy, characterised by long periods of inaction More likely to be in inaction zone during a recession (Berger and Vavra) In an expansion: You were thinking of buying a new car The central bank cuts (raises) interest rates You buy a bigger (smaller) one
Outline of a model - intuition The difference in the effect of policy is most pronounced in durables purchases Durables purchases are lumpy, characterised by long periods of inaction More likely to be in inaction zone during a recession (Berger and Vavra) In an expansion: You were thinking of buying a new car The central bank cuts (raises) interest rates You buy a bigger (smaller) one In a recession: You have just lost your job You don t need your car and don t want to buy a new one A small cut in interest rates doesn t change this
Outline of a model - intuition The difference in the effect of policy is most pronounced in durables purchases Durables purchases are lumpy, characterised by long periods of inaction More likely to be in inaction zone during a recession (Berger and Vavra) In an expansion: You were thinking of buying a new car The central bank cuts (raises) interest rates You buy a bigger (smaller) one In a recession: You have just lost your job You don t need your car and don t want to buy a new one A small cut in interest rates doesn t change this
Extra charts
Shocks and state 1.5 Monetary policy shocks Probability of an expansion 1.5.5 1 1965 197 1975 198 1985 199 1995 5
Distribution of shocks 4 3 1 PDF expansion recession average 1.5.5 1 1.5 1.8.6.4. CDF expansion recession average 1.5.5 1 1.5
Baseline parameters - expenditure volumes. Consumption of durables and housing investment volume.5.. Log points.. 1..5 1. 1.4 1 Consumption 5 x of nondurables and services volume 1 3.5 x 1 3 1 5 Log points. 1. 5 1..5 1 4 1 Fixed business investment volume.. 5 1 Log points...4.. 1 1 1 1 Three models Linear model Expansion Recession 4 1 Expansion = Recession
Baseline parameters - expenditure prices. Consumption 5 x of durables and housing investment inflation 1 3.5 5 x 1 3 Log points. 5 5.4 1. 1 1.5 1 1 1 Consumption 5 x of nondurables and services inflation 1 3.5 5 x 1 3 4 1 Log points 5 5. 1.5 1 1.1.5 1 1 1 Fixed business investment inflation.1.1 4 1 Log points.1.1.5..1. 1 1 1 1 Three models Linear model Expansion Recession 4 1 Expansion = Recession
Baseline parameters - other response variables Log points Log points Log points Log points.. 1.1.1 1.5.5 1.5.1 Government consumption volume.5.1.1.5.1 1 1 1 Tax GDP ratio.... 1.. 1 GZ spread 1. 1.. 1. 1 1 1 Private nonfinancial debt GDP ratio..1..5..1. 1 1 1 1 Three models Linear model Expansion Recession 1 5 5 1 1 1 Expansion = Recession
Headline variables - positive v negative shocks 5 x 1 3.1 GDP volume.1.1 Log points 5.1 1 1 5 x 1 3.1..1 1 1 1 5 x 1 3 GDP inflation.1.1 4 1 Log points 5 5 Percentage points 1 1 1 1 1 Three models 1 1 1 1 1 Linear model.1.1 1 1 Federal funds rate 1 1 1 Positive shocks 1 Negative shocks 1 4 1 Positive = Negative
Headline variables - coefficient on cube of shocks GDP volume 5 1 15 GDP inflation 5 1 15 Romer shocks quarterly average 5 5 5 1 15