When Inequality Matters for Macro and Macro Matters for Inequality - - PowerPoint PPT Presentation

‘When Inequality Matters for Macro and Macro Matters for Inequality’

Discussion by Christopher Carroll1 (with help from Edmund Crawley1)

1Johns Hopkins University

ccarroll@jhu.edu edmundcrawley@gmail.com

NBER Macro Annual Meeting, April 7, 2017

What’s Wrong With Macroeconomics?

◮ Since 2007: Lots of criticism of benchmark macro models

What’s Wrong With Macroeconomics?

◮ Since 2007: Lots of criticism of benchmark macro models ◮ Larry Summers (2011)’s (in)famous quote:

What’s Wrong With Macroeconomics?

◮ Since 2007: Lots of criticism of benchmark macro models ◮ Larry Summers (2011)’s (in)famous quote:

◮ “Almost nothing from the academic macroeconomics literature

• ver the prior 30 years was useful in understanding what to do

in the crisis”

What’s Wrong With Macroeconomics?

◮ Since 2007: Lots of criticism of benchmark macro models ◮ Larry Summers (2011)’s (in)famous quote:

◮ “Almost nothing from the academic macroeconomics literature

• ver the prior 30 years was useful in understanding what to do

in the crisis”

◮ Many others have said similar things (if usually more politely)

Can This Paper Fix It?

After thinking it over, a number of policymakers in recent years have said that an important part of solving the problem will be to take heterogeneity much more seriously:

◮ Fed Chair Janet Yellen (2016)

Can This Paper Fix It?

After thinking it over, a number of policymakers in recent years have said that an important part of solving the problem will be to take heterogeneity much more seriously:

◮ Fed Chair Janet Yellen (2016) ◮ IMF Chief Economist Olivier Blanchard (2016)

Can This Paper Fix It?

After thinking it over, a number of policymakers in recent years have said that an important part of solving the problem will be to take heterogeneity much more seriously:

◮ Fed Chair Janet Yellen (2016) ◮ IMF Chief Economist Olivier Blanchard (2016) ◮ ECB Governing Board Member Benoit Coeure (2013)

Can This Paper Fix It?

After thinking it over, a number of policymakers in recent years have said that an important part of solving the problem will be to take heterogeneity much more seriously:

◮ Fed Chair Janet Yellen (2016) ◮ IMF Chief Economist Olivier Blanchard (2016) ◮ ECB Governing Board Member Benoit Coeure (2013) ◮ Bank of England Chief Economist Andy Haldane (2016)

This Paper Is a Major Step in the Right Direction

• 1. (Almost) Nobody objects in principle to adding heterogeneity

This Paper Is a Major Step in the Right Direction

• 1. (Almost) Nobody objects in principle to adding heterogeneity
• 2. Obstacle has been math/computational difficulty in practice

This Paper Is a Major Step in the Right Direction

• 1. (Almost) Nobody objects in principle to adding heterogeneity
• 2. Obstacle has been math/computational difficulty in practice
• 3. Methods here promise to enormously reduce that barrier

This Paper Is a Major Step in the Right Direction

• 1. (Almost) Nobody objects in principle to adding heterogeneity
• 2. Obstacle has been math/computational difficulty in practice
• 3. Methods here promise to enormously reduce that barrier

◮ But: Methods work only in continuous time

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ...

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices ◮ Ultimately they discretize everything anyway

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices ◮ Ultimately they discretize everything anyway ◮ Errors are more easily ...

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices ◮ Ultimately they discretize everything anyway ◮ Errors are more easily ... ◮ Avoided

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices ◮ Ultimately they discretize everything anyway ◮ Errors are more easily ... ◮ Avoided ◮ Detected

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices ◮ Ultimately they discretize everything anyway ◮ Errors are more easily ... ◮ Avoided ◮ Detected ◮ Understood

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices ◮ Ultimately they discretize everything anyway ◮ Errors are more easily ... ◮ Avoided ◮ Detected ◮ Understood ◮ Fixed

We Left the Continuum For Good Reasons

Should We Return? Depends whether it is:

◮ Magic ... ◮ Black Magic

History makes me wary:

• 1. Finance: Many disasters
• 2. Computational Modeling:

◮ Computers are Digital Devices ◮ Ultimately they discretize everything anyway ◮ Errors are more easily ... ◮ Avoided ◮ Detected ◮ Understood ◮ Fixed ◮ ... if you don’t rely on ‘magic’

What Is a ‘Shock’ In Continuous Time?

Two issues:

• 1. Nature

What Is a ‘Shock’ In Continuous Time?

Two issues:

• 1. Nature
• 2. Timing

Definition: ‘Friedman’ Income Process

Friedman (1957) got it right (almost):1 yt = pt + z1,t pt+1 = pt + z2,t Et[z•,t+n] = In logs, still an amazingly good description of annual HH y dynamics (pace time aggregation issues):

◮ DeBacker, Heim, Panousi, Ramnath, and Vidangos (2013)

1He thought in levels; today we interpret y and p as logs.

Definition: ‘Friedman’ Income Process

Friedman (1957) got it right (almost):1 yt = pt + z1,t pt+1 = pt + z2,t Et[z•,t+n] = In logs, still an amazingly good description of annual HH y dynamics (pace time aggregation issues):

◮ DeBacker, Heim, Panousi, Ramnath, and Vidangos (2013)

◮ Millions of IRS records 1He thought in levels; today we interpret y and p as logs.

Definition: ‘Friedman’ Income Process

Friedman (1957) got it right (almost):1 yt = pt + z1,t pt+1 = pt + z2,t Et[z•,t+n] = In logs, still an amazingly good description of annual HH y dynamics (pace time aggregation issues):

◮ DeBacker, Heim, Panousi, Ramnath, and Vidangos (2013)

◮ Millions of IRS records ◮ Improvement from more complexity is minimal 1He thought in levels; today we interpret y and p as logs.

• 1. Nature of Their Shocks

zt = z1,t + z2,t

• 1. Nature of Their Shocks

zt = z1,t + z2,t

• 1. Nature of Their Shocks

zt = z1,t + z2,t where

◮ z1,t supposed to capture ‘transitory’

• 1. Nature of Their Shocks

zt = z1,t + z2,t where

◮ z1,t supposed to capture ‘transitory’ ◮ z2,t supposed to capture ‘persistent’

• 1. Nature of Their Shocks

zt = z1,t + z2,t where

◮ z1,t supposed to capture ‘transitory’ ◮ z2,t supposed to capture ‘persistent’

◮ Let the data say whether ‘permanent’

• 1. Nature of Their Shocks

zt = z1,t + z2,t where

◮ z1,t supposed to capture ‘transitory’ ◮ z2,t supposed to capture ‘persistent’

◮ Let the data say whether ‘permanent’

• 1. Nature of Their Shocks

zt = z1,t + z2,t where

◮ z1,t supposed to capture ‘transitory’ ◮ z2,t supposed to capture ‘persistent’

◮ Let the data say whether ‘permanent’

So far, looks good

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down)

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1)

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

Calibration of ρ1 and ρ2?

◮ Optimize to best match

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

Calibration of ρ1 and ρ2?

◮ Optimize to best match

◮ Guvenen, Karahan, Ozkan, and Song (2015); ‘GKOS’

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

Calibration of ρ1 and ρ2?

◮ Optimize to best match

◮ Guvenen, Karahan, Ozkan, and Song (2015); ‘GKOS’

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

Calibration of ρ1 and ρ2?

◮ Optimize to best match

◮ Guvenen, Karahan, Ozkan, and Song (2015); ‘GKOS’

Results:

◮ ‘Transitory’: Half life is a quarter

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

Calibration of ρ1 and ρ2?

◮ Optimize to best match

◮ Guvenen, Karahan, Ozkan, and Song (2015); ‘GKOS’

Results:

◮ ‘Transitory’: Half life is a quarter ◮ ‘Persistent’: ρ2 ≈ 0.99 (quarterly)

• 1. Both Shocks Work The Same Way

‘Jump-drift’ process:

◮ Income flow rate might jump (up or down) ◮ Decays toward zero like an AR(1) ◮ Call the two ‘decay’ parameters ρ1 and ρ2

Calibration of ρ1 and ρ2?

◮ Optimize to best match

◮ Guvenen, Karahan, Ozkan, and Song (2015); ‘GKOS’

Results:

◮ ‘Transitory’: Half life is a quarter ◮ ‘Persistent’: ρ2 ≈ 0.99 (quarterly)

◮ It’s (nearly) a permanent shock

• 1. Assessment?

‘Persistent’ shock:

◮ Yay!

• 1. Assessment?

‘Persistent’ shock:

◮ Yay! ◮ Common interpretation of GKOS (including by GKOS):

• 1. Assessment?

‘Persistent’ shock:

◮ Yay! ◮ Common interpretation of GKOS (including by GKOS):

◮ ‘Friedman’ permanent shocks wrong way to think about it

• 1. Assessment?

‘Persistent’ shock:

◮ Yay! ◮ Common interpretation of GKOS (including by GKOS):

◮ ‘Friedman’ permanent shocks wrong way to think about it ◮ Looks pretty good to me!

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time ◮ AKMWW: $600 shock (for someone with weekly paycheck) is:

Week Amount 1 $35.07 2 $33.09 ... ... 13 $17.39 ... ... ∞ Sum: $600

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time ◮ AKMWW: $600 shock (for someone with weekly paycheck) is:

Week Amount 1 $35.07 2 $33.09 ... ... 13 $17.39 ... ... ∞ Sum: $600

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time ◮ AKMWW: $600 shock (for someone with weekly paycheck) is:

Week Amount 1 $35.07 2 $33.09 ... ... 13 $17.39 ... ... ∞ Sum: $600 Is this equivalent to $600 lump sum?

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time ◮ AKMWW: $600 shock (for someone with weekly paycheck) is:

Week Amount 1 $35.07 2 $33.09 ... ... 13 $17.39 ... ... ∞ Sum: $600 Is this equivalent to $600 lump sum?

◮ For nondurables spending – see next slides

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time ◮ AKMWW: $600 shock (for someone with weekly paycheck) is:

Week Amount 1 $35.07 2 $33.09 ... ... 13 $17.39 ... ... ∞ Sum: $600 Is this equivalent to $600 lump sum?

◮ For nondurables spending – see next slides ◮ For durables, surely not:

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time ◮ AKMWW: $600 shock (for someone with weekly paycheck) is:

Week Amount 1 $35.07 2 $33.09 ... ... 13 $17.39 ... ... ∞ Sum: $600 Is this equivalent to $600 lump sum?

◮ For nondurables spending – see next slides ◮ For durables, surely not:

◮ Several papers find lump sums are used as car down payments

• 1. Transitory Shock

Possibly more problematic. Take 2008 stimulus checks:

◮ Arrived as a lump sum of $600 (or $1200) at instant of time ◮ AKMWW: $600 shock (for someone with weekly paycheck) is:

Week Amount 1 $35.07 2 $33.09 ... ... 13 $17.39 ... ... ∞ Sum: $600 Is this equivalent to $600 lump sum?

◮ For nondurables spending – see next slides ◮ For durables, surely not:

◮ Several papers find lump sums are used as car down payments ◮ $35.07 (or $70.14) would not suffice!

• 1. ‘Deaton’ Permanent Income

In a certainty equivalent model: Dt = Et ∞

• n=0

R−nyt+n

• (1)

Ct = (r/R)Dt (2)

◮ AKMWW ‘transitory’ shock makes Dt ↑ by $600

• 1. ‘Deaton’ Permanent Income

In a certainty equivalent model: Dt = Et ∞

• n=0

R−nyt+n

• (1)

Ct = (r/R)Dt (2)

◮ AKMWW ‘transitory’ shock makes Dt ↑ by $600 ◮ Exactly equivalent to lump sum of $600 if

• 1. ‘Deaton’ Permanent Income

In a certainty equivalent model: Dt = Et ∞

• n=0

R−nyt+n

• (1)

Ct = (r/R)Dt (2)

◮ AKMWW ‘transitory’ shock makes Dt ↑ by $600 ◮ Exactly equivalent to lump sum of $600 if

◮ Perfect foresight

• 1. ‘Deaton’ Permanent Income

In a certainty equivalent model: Dt = Et ∞

• n=0

R−nyt+n

• (1)

Ct = (r/R)Dt (2)

◮ AKMWW ‘transitory’ shock makes Dt ↑ by $600 ◮ Exactly equivalent to lump sum of $600 if

◮ Perfect foresight ◮ No liquidity constraints

• 1. ‘Deaton’ Permanent Income

In a certainty equivalent model: Dt = Et ∞

• n=0

R−nyt+n

• (1)

Ct = (r/R)Dt (2)

◮ AKMWW ‘transitory’ shock makes Dt ↑ by $600 ◮ Exactly equivalent to lump sum of $600 if

◮ Perfect foresight ◮ No liquidity constraints ◮ Perfect capital markets

• 1. ‘Deaton’ Permanent Income

In a certainty equivalent model: Dt = Et ∞

• n=0

R−nyt+n

• (1)

Ct = (r/R)Dt (2)

◮ AKMWW ‘transitory’ shock makes Dt ↑ by $600 ◮ Exactly equivalent to lump sum of $600 if

◮ Perfect foresight ◮ No liquidity constraints ◮ Perfect capital markets

• 1. ‘Deaton’ Permanent Income

In a certainty equivalent model: Dt = Et ∞

• n=0

R−nyt+n

• (1)

Ct = (r/R)Dt (2)

◮ AKMWW ‘transitory’ shock makes Dt ↑ by $600 ◮ Exactly equivalent to lump sum of $600 if

◮ Perfect foresight ◮ No liquidity constraints ◮ Perfect capital markets

That is, if there is no reason to do any of the incredibly hard and impressive work they do to deal with liquidity constraints, uncertainty, time-varying interest rates, etc etc etc

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down ◮ A week? Indistinguishable from ‘Friedman’ shock

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down ◮ A week? Indistinguishable from ‘Friedman’ shock ◮ A quarter? Starts to be problematic

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down ◮ A week? Indistinguishable from ‘Friedman’ shock ◮ A quarter? Starts to be problematic

◮ Say, for analyzing 2008 stimulus

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down ◮ A week? Indistinguishable from ‘Friedman’ shock ◮ A quarter? Starts to be problematic

◮ Say, for analyzing 2008 stimulus ◮ Summers would not be impressed

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down ◮ A week? Indistinguishable from ‘Friedman’ shock ◮ A quarter? Starts to be problematic

◮ Say, for analyzing 2008 stimulus ◮ Summers would not be impressed

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down ◮ A week? Indistinguishable from ‘Friedman’ shock ◮ A quarter? Starts to be problematic

◮ Say, for analyzing 2008 stimulus ◮ Summers would not be impressed

Am pretty confident that it can be made to work OK ...

◮ ... with recalibration

• 1. The Good News

As ρ2 ↓ 0

◮ In theory:

◮ AKMWW shock approaches a ‘Friedman’ transitory shock

◮ In practice, numerics must break down somewhere ◮ Interesting to know half-life mark where it breaks down ◮ A week? Indistinguishable from ‘Friedman’ shock ◮ A quarter? Starts to be problematic

◮ Say, for analyzing 2008 stimulus ◮ Summers would not be impressed

Am pretty confident that it can be made to work OK ...

◮ ... with recalibration ◮ ... and for some Q’s (like stimulus) maybe need 3 not two z’s

• 2. Timing of Shocks

Frequency of arrival of shocks:

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years Yikes!

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years Yikes!

◮ Low, Meghir, and Pistaferri (2010): z2 shocks are mostly:

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years Yikes!

◮ Low, Meghir, and Pistaferri (2010): z2 shocks are mostly:

◮ Promotions

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years Yikes!

◮ Low, Meghir, and Pistaferri (2010): z2 shocks are mostly:

◮ Promotions ◮ Job Changes

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years Yikes!

◮ Low, Meghir, and Pistaferri (2010): z2 shocks are mostly:

◮ Promotions ◮ Job Changes

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years Yikes!

◮ Low, Meghir, and Pistaferri (2010): z2 shocks are mostly:

◮ Promotions ◮ Job Changes

which happen maybe once every 5 years, not 38

• 2. Timing of Shocks

Frequency of arrival of shocks: Transitory Arrives once every 3 years Permanent Arrives once every 38 years Yikes!

◮ Low, Meghir, and Pistaferri (2010): z2 shocks are mostly:

◮ Promotions ◮ Job Changes

which happen maybe once every 5 years, not 38

◮ Vast lit: Trans shocks large at annual frequency

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data ◮ Am sure it would make huge miss

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data ◮ Am sure it would make huge miss

◮ Good news: With recalibration, it could be serious

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data ◮ Am sure it would make huge miss

◮ Good news: With recalibration, it could be serious ◮ Many ‘millions of data points’ papers out there besides GKOS

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data ◮ Am sure it would make huge miss

◮ Good news: With recalibration, it could be serious ◮ Many ‘millions of data points’ papers out there besides GKOS

◮ Most of them estimate Friedmanesque process

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data ◮ Am sure it would make huge miss

◮ Good news: With recalibration, it could be serious ◮ Many ‘millions of data points’ papers out there besides GKOS

◮ Most of them estimate Friedmanesque process ◮ GKOS specialized in measuring leptokurtosis, recessions, tails

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data ◮ Am sure it would make huge miss

◮ Good news: With recalibration, it could be serious ◮ Many ‘millions of data points’ papers out there besides GKOS

◮ Most of them estimate Friedmanesque process ◮ GKOS specialized in measuring leptokurtosis, recessions, tails ◮ They aren’t aiming at any of these targets anyway

Micro: Not Quite There Yet ...

Is their description of HH income ‘good enough’?

◮ In short: Not yet.

◮ Sim process at HH level, compare results to data ◮ Am sure it would make huge miss

◮ Good news: With recalibration, it could be serious ◮ Many ‘millions of data points’ papers out there besides GKOS

◮ Most of them estimate Friedmanesque process ◮ GKOS specialized in measuring leptokurtosis, recessions, tails ◮ They aren’t aiming at any of these targets anyway ◮ ⇒ Match rest of literature, not GKOS

Macro: Aggregate C and Y Dynamics

Claim to solve two (related) puzzles:

• 1. Campbell and Deaton (1989): ‘Excess smoothness’:

Macro: Aggregate C and Y Dynamics

Claim to solve two (related) puzzles:

• 1. Campbell and Deaton (1989): ‘Excess smoothness’:

◮ var(∆ log C) < var(∆ log Y )

Macro: Aggregate C and Y Dynamics

Claim to solve two (related) puzzles:

• 1. Campbell and Deaton (1989): ‘Excess smoothness’:

◮ var(∆ log C) < var(∆ log Y )

• 2. Campbell and Mankiw (1989): ‘Excess sensitivity’:

Macro: Aggregate C and Y Dynamics

Claim to solve two (related) puzzles:

• 1. Campbell and Deaton (1989): ‘Excess smoothness’:

◮ var(∆ log C) < var(∆ log Y )

• 2. Campbell and Mankiw (1989): ‘Excess sensitivity’:

◮ Flavin (1981)

Macro: Aggregate C and Y Dynamics

Claim to solve two (related) puzzles:

• 1. Campbell and Deaton (1989): ‘Excess smoothness’:

◮ var(∆ log C) < var(∆ log Y )

• 2. Campbell and Mankiw (1989): ‘Excess sensitivity’:

◮ Flavin (1981) ◮ ∆ log C inappropriately ‘sensitive’ to predictable ∆ log Y

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = ∆ log Yt+1

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = ∆ log Yt+1 ◮ Variances of ∆ log Ct+1 and ∆ log Yt+1 are equal

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = ∆ log Yt+1 ◮ Variances of ∆ log Ct+1 and ∆ log Yt+1 are equal

• 3. Growth is serially correlated: ∆Zt+1,2 = γ∆Zt,2 + ǫt+1

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = ∆ log Yt+1 ◮ Variances of ∆ log Ct+1 and ∆ log Yt+1 are equal

• 3. Growth is serially correlated: ∆Zt+1,2 = γ∆Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = (1/(1 − γ))∆ log Yt+1

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = ∆ log Yt+1 ◮ Variances of ∆ log Ct+1 and ∆ log Yt+1 are equal

• 3. Growth is serially correlated: ∆Zt+1,2 = γ∆Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = (1/(1 − γ))∆ log Yt+1 ◮ var(∆ log Ct+1) = (1/(1 − γ))2var(∆ log Yt+1)

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = ∆ log Yt+1 ◮ Variances of ∆ log Ct+1 and ∆ log Yt+1 are equal

• 3. Growth is serially correlated: ∆Zt+1,2 = γ∆Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = (1/(1 − γ))∆ log Yt+1 ◮ var(∆ log Ct+1) = (1/(1 − γ))2var(∆ log Yt+1)

• 1. Excess Smoothness: Campbell and Deaton (1989)

Consequences for ‘Deaton’ permanent income Dt of shock to Yt depend critically on time series process for Y Consider three options:

• 1. No permanent shocks to log Y : (Zt,2 = 0 ∀ t)

◮ Then ∆ log Ct+1 = (r/R)∆ log Yt+1 ◮ C vastly smoother than Y

• 2. log Y is random walk: Zt+1,2 = Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = ∆ log Yt+1 ◮ Variances of ∆ log Ct+1 and ∆ log Yt+1 are equal

• 3. Growth is serially correlated: ∆Zt+1,2 = γ∆Zt,2 + ǫt+1

◮ Then ∆ log Ct+1 = (1/(1 − γ))∆ log Yt+1 ◮ var(∆ log Ct+1) = (1/(1 − γ))2var(∆ log Yt+1)

They assume the last

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large ◮ 90% interval: 0.88 to 1.01

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large ◮ 90% interval: 0.88 to 1.01

◮ Much of vast DSGE literature assumes an AR(1) in levels:

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large ◮ 90% interval: 0.88 to 1.01

◮ Much of vast DSGE literature assumes an AR(1) in levels:

◮ log Yt+1 = 0.95 log Yt

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large ◮ 90% interval: 0.88 to 1.01

◮ Much of vast DSGE literature assumes an AR(1) in levels:

◮ log Yt+1 = 0.95 log Yt

◮ Corr(∆ log Yt, ∆ log Yt−1) = 0.37 need not imply unexpected

shocks have long lasting effects

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large ◮ 90% interval: 0.88 to 1.01

◮ Much of vast DSGE literature assumes an AR(1) in levels:

◮ log Yt+1 = 0.95 log Yt

◮ Corr(∆ log Yt, ∆ log Yt−1) = 0.37 need not imply unexpected

shocks have long lasting effects

◮ e.g.:

(1 − aL)(1 − ρL) log Yt = ut

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large ◮ 90% interval: 0.88 to 1.01

◮ Much of vast DSGE literature assumes an AR(1) in levels:

◮ log Yt+1 = 0.95 log Yt

◮ Corr(∆ log Yt, ∆ log Yt−1) = 0.37 need not imply unexpected

shocks have long lasting effects

◮ e.g.:

(1 − aL)(1 − ρL) log Yt = ut

◮ Their model is a = 0.37, ρ = 1

• 1. Problem: Who Knows the Right Income Process?

◮ Huge literature in 1980s and 1990s on this exact subject

◮ Stock (1991): Confidence bands for the largest root are large ◮ 90% interval: 0.88 to 1.01

◮ Much of vast DSGE literature assumes an AR(1) in levels:

◮ log Yt+1 = 0.95 log Yt

◮ Corr(∆ log Yt, ∆ log Yt−1) = 0.37 need not imply unexpected

shocks have long lasting effects

◮ e.g.:

(1 − aL)(1 − ρL) log Yt = ut

◮ Their model is a = 0.37, ρ = 1 ◮ But a = 0.44, ρ = 0.9 also has 0.37 autocorrelation

• 1. Relative Volatility is Highly Sensitive to ρ

One asset, small open economy model (CSTW)

0.90 0.92 0.94 0.96 0.98 1.00

ρ

0.4 0.5 0.6 0.7 0.8

σ(∆logCt) σ(∆logYt)

logYt = ρlogYt − 1 + εt

◮ Basically, any number between 0.4 and maybe 2 is consistent

with some statistically unrejectable description

• 2. Do They Explain Campbell and Mankiw (1989)?

In principle, it might:

◮ Shock arrives at t that has important info about

Et[∆ log Yt+1]

• 2. Do They Explain Campbell and Mankiw (1989)?

In principle, it might:

◮ Shock arrives at t that has important info about

Et[∆ log Yt+1]

◮ But lots of AMKWW consumers are ‘nearly constrained’

• 2. Do They Explain Campbell and Mankiw (1989)?

In principle, it might:

◮ Shock arrives at t that has important info about

Et[∆ log Yt+1]

◮ But lots of AMKWW consumers are ‘nearly constrained’ ◮ So, don’t respond until they get cash in hand

• 2. Do They Explain Campbell and Mankiw (1989)?

In principle, it might:

◮ Shock arrives at t that has important info about

Et[∆ log Yt+1]

◮ But lots of AMKWW consumers are ‘nearly constrained’ ◮ So, don’t respond until they get cash in hand

◮ N.B.: This is exactly the circumstance in which z1 shock is

NOT equivalent to Friedman shock

• 2. The Test

Campbell and Mankiw (1989) say C can be modeled by ∆ log Ct+1 = (1 − λ)σ(rt+1 − ϑ) + λ∆Et[∆ log Yt+1]

• 2. The Test

Campbell and Mankiw (1989) say C can be modeled by ∆ log Ct+1 = (1 − λ)σ(rt+1 − ϑ) + λ∆Et[∆ log Yt+1] λ is income of ‘rule-of-thumb’ (c = y) consumers. Right test? ∆ log Ct+1 = γ0 + γ1Et−1[rt+1] + γ2∆Et−1[∆ log Yt+1]

• 2. The Test

Campbell and Mankiw (1989) say C can be modeled by ∆ log Ct+1 = (1 − λ)σ(rt+1 − ϑ) + λ∆Et[∆ log Yt+1] λ is income of ‘rule-of-thumb’ (c = y) consumers. Right test? ∆ log Ct+1 = γ0 + γ1Et−1[rt+1] + γ2∆Et−1[∆ log Yt+1] Hall (1978): γ2 = 0 in standard representative agent model

• 2. We Can’t Figure Out Their Results

From their paper:

• 2. We Can’t Figure Out Their Results

From their paper: Hall: Coefficient in IV reg for RA model is 0 AKMWW: They find 0.247 Why?

◮ rt+1 omitted from RHS of regression? They say no.

• 2. We Can’t Figure Out Their Results

From their paper: Hall: Coefficient in IV reg for RA model is 0 AKMWW: They find 0.247 Why?

◮ rt+1 omitted from RHS of regression? They say no. ◮ Time aggregation? (Their notation says t − 1 ...)

• 2. We Can’t Figure Out Their Results

From their paper: Hall: Coefficient in IV reg for RA model is 0 AKMWW: They find 0.247 Why?

◮ rt+1 omitted from RHS of regression? They say no. ◮ Time aggregation? (Their notation says t − 1 ...) ◮ Something else?

• 2. We Can’t Figure Out Their Results

From their paper: Hall: Coefficient in IV reg for RA model is 0 AKMWW: They find 0.247 Why?

◮ rt+1 omitted from RHS of regression? They say no. ◮ Time aggregation? (Their notation says t − 1 ...) ◮ Something else? ◮ If we can’t understand RA model, no hope for 2-asset model

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

◮ ∆ log Ct+1 = σ(rt+1 − ϑ)

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

◮ ∆ log Ct+1 = σ(rt+1 − ϑ) ◮ Says that when r ↑, Ct+1 ↓ instantly

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

◮ ∆ log Ct+1 = σ(rt+1 − ϑ) ◮ Says that when r ↑, Ct+1 ↓ instantly ◮ Data: Half life of response of C to MP shocks looks like a year

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

◮ ∆ log Ct+1 = σ(rt+1 − ϑ) ◮ Says that when r ↑, Ct+1 ↓ instantly ◮ Data: Half life of response of C to MP shocks looks like a year ◮ Their (embarrassed) solution: Assume C is just stuck for a year

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

◮ ∆ log Ct+1 = σ(rt+1 − ϑ) ◮ Says that when r ↑, Ct+1 ↓ instantly ◮ Data: Half life of response of C to MP shocks looks like a year ◮ Their (embarrassed) solution: Assume C is just stuck for a year ◮ (Embarrassed) motivation: You plan your budget a year in

advance and can’t change it

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

◮ ∆ log Ct+1 = σ(rt+1 − ϑ) ◮ Says that when r ↑, Ct+1 ↓ instantly ◮ Data: Half life of response of C to MP shocks looks like a year ◮ Their (embarrassed) solution: Assume C is just stuck for a year ◮ (Embarrassed) motivation: You plan your budget a year in

advance and can’t change it

The Real Excess Smoothness/Sensitivity Problem

◮ First ‘structural’ model of monetary policy I know of is

Rotemberg and Woodford (1997)

◮ One big problem they faced:

◮ ∆ log Ct+1 = σ(rt+1 − ϑ) ◮ Says that when r ↑, Ct+1 ↓ instantly ◮ Data: Half life of response of C to MP shocks looks like a year ◮ Their (embarrassed) solution: Assume C is just stuck for a year ◮ (Embarrassed) motivation: You plan your budget a year in

advance and can’t change it

Vast subsequent literature captures same fact using ‘habits’

Problems With Habits

◮ No micro evidence for habits

Problems With Habits

◮ No micro evidence for habits

◮ Deaton (1992)

Problems With Habits

◮ No micro evidence for habits

◮ Deaton (1992) ◮ Dynan (2000)

Problems With Habits

◮ No micro evidence for habits

◮ Deaton (1992) ◮ Dynan (2000) ◮ Subsequent literature: No micro evidence for habits

Problems With Habits

◮ No micro evidence for habits

◮ Deaton (1992) ◮ Dynan (2000) ◮ Subsequent literature: No micro evidence for habits

◮ Habits deepen problem RA model has with κ

Problems With Habits

◮ No micro evidence for habits

◮ Deaton (1992) ◮ Dynan (2000) ◮ Subsequent literature: No micro evidence for habits

◮ Habits deepen problem RA model has with κ

◮ Standard habits parameterization might imply ¯

κ = 0.01.

Problems With Habits

◮ No micro evidence for habits

◮ Deaton (1992) ◮ Dynan (2000) ◮ Subsequent literature: No micro evidence for habits

◮ Habits deepen problem RA model has with κ

◮ Standard habits parameterization might imply ¯

κ = 0.01.

◮ Gap between ‘rule-of-thumb’ and ‘rational’ widens to chasm

Problems With Habits

◮ No micro evidence for habits

◮ Deaton (1992) ◮ Dynan (2000) ◮ Subsequent literature: No micro evidence for habits

◮ Habits deepen problem RA model has with κ

◮ Standard habits parameterization might imply ¯

κ = 0.01.

◮ Gap between ‘rule-of-thumb’ and ‘rational’ widens to chasm

◮ Hard to maintain straight face and say this is ‘microfounded’

Macro Consumption Puzzles Solved?

Summary:

• 1. Campbell and Deaton (1989): It’s not really a puzzle

Macro Consumption Puzzles Solved?

Summary:

• 1. Campbell and Deaton (1989): It’s not really a puzzle
• 2. Campbell and Mankiw (1989): We’re deeply confused

Macro Consumption Puzzles Solved?

Summary:

• 1. Campbell and Deaton (1989): It’s not really a puzzle
• 2. Campbell and Mankiw (1989): We’re deeply confused
• 3. Real ‘excess smoothness’ problem: They don’t examine it

Macro Consumption Puzzles Solved?

Summary:

• 1. Campbell and Deaton (1989): It’s not really a puzzle
• 2. Campbell and Mankiw (1989): We’re deeply confused
• 3. Real ‘excess smoothness’ problem: They don’t examine it

◮ Though they could; maybe should do this instead

What Should They Do Instead?

◮ Their huge contribution is methodological

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

My preference: Model where

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

My preference: Model where

• 1. Both micro and macro have pure permanent shocks

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

My preference: Model where

• 1. Both micro and macro have pure permanent shocks
• 2. Without the liquid and illiquid assets

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

My preference: Model where

• 1. Both micro and macro have pure permanent shocks
• 2. Without the liquid and illiquid assets
• 3. In partial and general equilibrium

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

My preference: Model where

• 1. Both micro and macro have pure permanent shocks
• 2. Without the liquid and illiquid assets
• 3. In partial and general equilibrium
• 4. They examine how close they can get to Friedman z1

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

My preference: Model where

• 1. Both micro and macro have pure permanent shocks
• 2. Without the liquid and illiquid assets
• 3. In partial and general equilibrium
• 4. They examine how close they can get to Friedman z1

What Should They Do Instead?

◮ Their huge contribution is methodological ◮ They need to convince skeptics it’s right ◮ Wrong way to do that:

◮ Solve model so innovative and complicated that authors

themselves are confused about how findings relate to literature

◮ Right approach:

◮ Solve cleanest, clearest, simplest models possible ◮ Where everyone can see their answers are right ◮ They do this with the 20-year-old KS model ◮ Should do it with some others

My preference: Model where

• 1. Both micro and macro have pure permanent shocks
• 2. Without the liquid and illiquid assets
• 3. In partial and general equilibrium
• 4. They examine how close they can get to Friedman z1

Then write 3 or 4 other papers containing the (remaining) substantive (vs. methodological) contributions of the present paper

A Final Concern: It Will Be a Black Box

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

◮ Distribution of MPC may be close to ‘sufficient statistic’

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

◮ Distribution of MPC may be close to ‘sufficient statistic’ ◮ Results robust to alternative treatments of aggregate stuff

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

◮ Distribution of MPC may be close to ‘sufficient statistic’ ◮ Results robust to alternative treatments of aggregate stuff

◮ Microeconomic data abundant, becoming more so (‘big data’)

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

◮ Distribution of MPC may be close to ‘sufficient statistic’ ◮ Results robust to alternative treatments of aggregate stuff

◮ Microeconomic data abundant, becoming more so (‘big data’)

◮ Both Houseold Level and Regional

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

◮ Distribution of MPC may be close to ‘sufficient statistic’ ◮ Results robust to alternative treatments of aggregate stuff

◮ Microeconomic data abundant, becoming more so (‘big data’)

◮ Both Houseold Level and Regional

◮ Microeconomic theory is often more intuitive, transparent

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

◮ Distribution of MPC may be close to ‘sufficient statistic’ ◮ Results robust to alternative treatments of aggregate stuff

◮ Microeconomic data abundant, becoming more so (‘big data’)

◮ Both Houseold Level and Regional

◮ Microeconomic theory is often more intuitive, transparent

◮ It’s about behavior of individuals

A Final Concern: It Will Be a Black Box

The new DYNARE?

◮ Unlimited Theory, Little Understanding, and Almost No Data

Differences:

◮ For key macro questions (fiscal, monetary policy):

◮ Distribution of MPC may be close to ‘sufficient statistic’ ◮ Results robust to alternative treatments of aggregate stuff

◮ Microeconomic data abundant, becoming more so (‘big data’)

◮ Both Houseold Level and Regional

◮ Microeconomic theory is often more intuitive, transparent

◮ It’s about behavior of individuals ◮ Most macroeconomists are individuals

Conclusion

This is a hugely important paper

◮ For its methodological contribution

Conclusion

This is a hugely important paper

◮ For its methodological contribution ◮ More work is needed before that will become obvious

References I

Blanchard, Olivier (2016): “Do DSGE Models Have a Future?,” Discussion paper, Petersen Institute for International Economics, Available at https://piie.com/system/files/documents/pb16-11.pdf. Campbell, John, and Angus Deaton (1989): “Why is Consumption So Smooth?,” The Review of Economic Studies, 56(3), 357–373, http://www.jstor.org/stable/2297552. Campbell, John Y., and N. Gregory Mankiw (1989): “Consumption, Income, and Interest Rates: Reinterpreting the Time-Series Evidence,” in NBER Macroeconomics Annual, 1989, ed. by Olivier J. Blanchard, and Stanley Fischer, pp. 185–216. MIT Press, Cambridge, MA, http://www.nber.org/papers/w2924.pdf. Coeure, Benoit (2013): “The relevance of household-level data for monetary policy and financial stability analysis,” . Deaton, Angus S. (1992): Understanding Consumption. Oxford University Press, New York. DeBacker, Jason, Bradley Heim, Vasia Panousi, Shanthi Ramnath, and Ivan Vidangos (2013): “Rising Inequality: Transitory or Persistent? New Evidence from a Panel of U.S. Tax Returns,” Brookings Papers on Economic Activity, Spring, 67–122. Dynan, Karen E. (2000): “Habit Formation in Consumer Preferences: Evidence from Panel Data,” American Economic Review, 90(3), http://www.jstor.org/stable/117335. Flavin, Marjorie B. (1981): “The Adjustment of Consumption to Changing Expectations About Future Income,” Journal of Political Economy, 89, 974–1009, http://www.jstor.org/stable/1830816. Friedman, Milton A. (1957): A Theory of the Consumption Function. Princeton University Press. Guvenen, Fatih, Fatih Karahan, Serdar Ozkan, and Jae Song (2015): “What do data on millions of US workers reveal about life-cycle earnings risk?,” Discussion paper, National Bureau of Economic Research. Haldane, Andy (2016): “The Dappled World,” Discussion paper, Bank of England, Available at http://www.bankofengland.co.uk/publications/Pages/speeches/2016/937.aspx. Hall, Robert E. (1978): “Stochastic Implications of the Life-Cycle/Permanent Income Hypothesis: Theory and Evidence,” Journal of Political Economy, 96, 971–87, Available at http://www.stanford.edu/~rehall/Stochastic-JPE-Dec-1978.pdf. Low, Hamish, Costas Meghir, and Luigi Pistaferri (2010): “Wage risk and employment risk over the life cycle,” The American economic review, 100(4), 1432–1467.

References II

Rotemberg, Julio J., and Michael Woodford (1997): “An Optimization-Based Econometric Model for the Evaluation of Monetary Policy,” in NBER Macroeconomics Annual, 1997, ed. by Benjamin S. Bernanke, and Julio J. Rotemberg, vol. 12, pp. 297–346. MIT Press, Cambridge, MA. Summers, Lawrence H. (2011): “Larry Summers and Martin Wolf on New Economic Thinking,” Financial Times video interview, http://tinyurl.com/dl201108a. Yellen, Janet (2016): “Macroeconomic Research After the Crisis,” Available at https://www.federalreserve.gov/newsevents/speech/yellen20161014a.htm.