the effects of financial crises october 12, 2016 · 10/12/2016  · and financial chronicle. • a...

LECTURE 8

The Effects of Financial Crises

October 12, 2016

Economics 210C/236A Christina Romer Spring 2016 David Romer

I. OVERVIEW

Central Issue

• What are the macroeconomic effects of financial crises?

• Papers for today look at aggregate, time-series evidence.

• Next week look at more micro, cross-section evidence.

What Is a “Financial Crisis?”

• Many candidates: Could involve sovereign debt, the exchange rate, intermediation, asset prices, ….

• Today’s papers all focus on developments involving financial intermediation—something causes a rise in the cost of credit intermediation.

• And if the goal is to focus on “crises,” need some way of distinguishing crises from more run-of-the-mill disruptions to intermediation.

Papers

• Reinhart-Rogoff: Aftermaths of crises in a large sample of countries.

• Jalil: Detailed study of the United States, 1825–1929.

• Krishnamurthy-Muir: A more statistical approach to identifying crises and their effects.

II. REINHART AND ROGOFF, “THE AFTERMATH OF

FINANCIAL CRISES,” CHAPTER 14 OF THIS TIME IS DIFFERENT: EIGHT CENTURIES OF FINANCIAL FOLLY

Two Key Steps

• Identifying crises.

• Estimating their effects.

Reinhart and Rogoff’s Definition

“We mark a banking crisis by two types of events: (1) [systemic, severe] bank runs that lead to the closure, merging, or takeover by the public sector of one or more financial institutions and (2) [financial distress, milder] if there are no runs, the closure, merging, takeover, or large-scale government assistance of an important financial institution (or group of institutions) that marks the start of a string of similar outcomes for other financial institutions.”

From: Reinhart and Rogoff, This Time Is Different, p. 11.

Reinhart and Rogoff’s Application of Their Definition

• Secondary sources.

• Limited discussion of why they classified things as they did.

Japan

From: Reinhart and Rogoff, This Time Is Different, p. 371.

Issues with Their Identification of Crises

• Accuracy?

• Could the procedures for identifying crises introduce bias?

• Is a binary classification appropriate?

Key Components of Good Narrative Analysis

• Evaluate reliability of narrative sources.

• Know what you are looking for in the sources.

• Read as carefully and objectively as possible.

• Document classification extensively.

Digression on Romer and Romer’s New Measure of Financial Distress

• Use a single, real-time narrative source.

• Define financial distress as a rise in the cost of credit intermediation.

• Try to come up with a scaled series on financial distress.

• Series is quite different from Reinhart and Rogoff and the IMF.

Comparison of Chronologies for Key Pre-2008 Episodes

0

2

4

6

8

10

12

14

1988

:2

1989

:2

1990

:2

1991

:2

1992

:2

1993

:2

1994

:2

1995

:2

1996

:2

1997

:2

1998

:2

New

Dis

tress

Mea

sure

a. Finland

0

2

4

6

8

10

12

14

1983

:1

1984

:1

1985

:1

1986

:1

1987

:1

1988

:1

1989

:1

1990

:1

1991

:1

1992

:1

1993

:1

New

Dis

tress

Mea

sure

d. United States

0

2

4

6

8

10

12

14

1990

:1

1991

:1

1992

:1

1993

:1

1994

:1

1995

:1

1996

:1

1997

:1

1998

:1

1999

:1

2000

:1

2001

:1

2002

:1

2003

:1

2004

:1

2005

:1

2006

:1

New

Dis

tress

Mea

sure

b. Japan

0

2

4

6

8

10

12

14

1986

:1

1987

:1

1988

:1

1989

:1

1990

:1

1991

:1

1992

:1

1993

:1

1994

:1

1995

:1

1996

:1

New

Dis

tress

Mea

sure

c. Norway

Reinhart and Rogoff’s Empirical Technique

• Average peak-to-trough change in GDP “around” crises.

• Very simple; no standard errors or tests of statistical significance.

Sample of Crises Considered

• 21 major banking crises.

• 6 recent; 13 other postwar (5 in advanced countries, 8 in developing); 2 others (Norway 1899, U.S. 1929).

Reinhart and Rogoff’s Evidence on The Aftermath of Financial Crises

Percent Decrease in Real GDP Per Capita Duration in Years

From: Reinhart and Rogoff, This Time Is Different.

Issues with Their Estimation of the Impact of Financial Crises

• Might reverse causation be important?

• Simple statistics may lead them astray (example: Finland).

• What is the logic behind the sample of countries included?

Real GDP in Finland, 1985–1996

11.4

11.4

11.5

11.5

11.6

11.6

11.7

11.7

1985-I 1987-I 1989-I 1991-I 1993-I 1995-I

Loga

rithm

s

From: Reinhart and Rogoff, This Time Is Different.

Romer and Romer’s Regression Technique

• Jordà regressions of outcome variable at various horizons on financial distress at time t.

• Timing assumption: Financial distress in t can affect GDP in t, but not vice versa.

• Panel data: 24 countries, 1967–2015.

• Use weighted least squares to deal with heteroskedasticity.

Figure 6 Impulse Response Function, Outcome to Distress

a. Real GDP, Full Sample, WLS

-10

-8

-6

-4

-2

0

2

4

0 1 2 3 4 5 6 7 8 9 10

Res

pons

e of

Rea

l GD

P (P

erce

nt)

Half-Years After the Impulse

From: Romer and Romer, “New Evidence on the Aftermath of Financial Crises in Advanced Countries”

III. JALIL, “A NEW HISTORY OF BANKING PANICS IN THE

UNITED STATES, 1825-1929: CONSTRUCTION AND IMPLICATIONS”

Overview

• Like Reinhart and Rogoff, interested in the macroeconomic effects of financial crises.

• But focuses on one country over a defined period: United States, 1825–1929.

• Again, two key steps:

• Identifying crises.

• Estimating their effects.

Previous Panic Series for the U.S.

• Bordo-Wheelock • Thorp • Reinhart-Rogoff (2 versions) • Friedman-Schwartz • Gorton • Sprague • Wicker • Kemmerer • DeLong-Summers

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

Jalil’s Definition of a Panic

• A financial panic occurs when fear prompts a widespread run by private agents … to convert deposits into currency (a banking panic).” (p. 300)

• “A banking panic occurs when there is an increase in the demand for currency relative to deposits that sparks bank runs and bank suspensions.” (p. 300)

• “A banking panic occurs when there is a loss of depositor confidence that sparks runs on financial institutions and bank suspensions.” (p. 302)

Implementing the Definition • Use articles in Niles Weekly Register, the Merchants’

Magazine and Commercial Review, and The Commercial and Financial Chronicle.

• A banking panic requires accounts of a cluster of bank suspensions and runs.

• A cluster means 3 or more, and excludes ones mentioned in articles that do not reference other suspensions or runs or general panic.

• A panic ends if there are no references to panics or suspensions for a full calendar month.

• A panic is major if it is mentioned on the front page of the newspaper and if its geographic scope is greater than a single state and its immediately bordering states.

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929,” Appendix

Documentation from the Online Appendix

From: Jalil, “A New History of Banking Panics in the United States, 1825-1929”

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

Seasonality of Panics

Issues in Jalil’s Identification of Crises

• Very different from other series—is this a problem?

• Should NYC panics be counted as local?

• 3 of his 7 major panics are in the 1830s—does that raise questions about his procedures?

• Is there corroborating evidence?

• Is his narrative work of high quality?

From: Jalil, “A New History of Banking Panics in the United States, 1825-1929,” Online Appendix

Interest Rates during Major Panics

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

Peak-to-Trough Change in IP around Crises

-0.20

-0.15

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

1820

1825

1830

1835

1840

1845

1850

1855

1860

1865

1870

1875

1880

1885

1890

1895

1900

1905

1910

1915

Standard Deviation 1820-1889 0.060 1890-1915 0.089

Percentage Change in Industrial Production

Jalil’s VAR Specification

𝐹𝑡 = 𝑎 + �𝛼𝑖𝐹𝑡−𝑖

3

𝑖=1

+ �𝛽𝑖∆𝑌𝑡−𝑖 + 𝑢𝑡

3

𝑖=1

∆𝑌𝑡 = 𝑐 + �𝛾𝑖𝐹𝑡−𝑖

3

𝑖=1

+ �𝛿𝑖∆𝑌𝑡−𝑖 + 𝑣𝑡

3

𝑖=1

• Where F is the crisis dummy and ∆Y is the change in log output, and u and v are uncorrelated with one another and over time.

• Notice timing assumption: Neither variable is allowed to affect the other contemporaneously.

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

How Does Jalil Attempt to Deal with Endogeneity?

• Narrative evidence on the cause of the crises.

• Restrict sample to major crises that were not caused by a decline in output.

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

Does he need Dimension 2, given he uses a VAR?

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

Hint on Tables: They should be self-explanatory. Many readers just flip through the tables.

From: Jalil, “A New History of Banking Panics in the United States, 1825-1929,” Online Appendix

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

Looking for Trend and Level Effects

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

From: Jalil, “A New History of Banking Panics in the United States, 1825–1929”

Evaluation

• Very careful and an impressive attempt to get more information.

• Takes identification seriously.

• Does the study have implications for modern financial disruptions?

IV. KRISHNAMURTHY AND MUIR, “HOW CREDIT CYCLES

ACROSS A FINANCIAL CRISIS”

Goals

• Bringing information about credit spreads and credit growth into the analysis of crises:

• Predictive power of behavior of credit spreads and credit growth.

• How the behavior of credit spreads and credit growth interacts with financial crises as identified by traditional chronologies.

Constructing Spreads

• 1869–1929:

• Monthly data on individual bonds.

• For a given month for a given country: Average of the top 90% of yields minus average of the bottom 10%.

• Use the average for the last 3 months of the year.

• After 1929: More conventional series (for example, Moody’s BAA–AAA spread for the U.S.).

From: Krishnamurthy and Muir, “How Credit Cycles across a Financial Crisis”

Business Cycle Peaks with and without Crises

From: Jordà, Schularick, and Taylor, “When Credit Bites Back”

Predictive Power of Spreads – Full Sample

From: Krishnamurthy and Muir, “How Credit Cycles across a Financial Crisis”

Defining Crises Based on Spreads

From: Krishnamurthy and Muir, “How Credit Cycles across a Financial Crisis”

Discussion

From: Krishnamurthy and Muir, “How Credit Cycles across a Financial Crisis”

From: Krishnamurthy and Muir, “How Credit Cycles across a Financial Crisis”

Discussion

V. A LITTLE ABOUT GLS, HETEROSKEDASTICITY-CONSISTENT STANDARD ERRORS, CLUSTERING, AND

ALL THAT

The Big Picture

• We spend much of the course worrying about the possibility that coefficient estimates (or other estimates of economic relationships) may be biased.

• But standard errors can also be biased – sometimes greatly.

• Just as there is no mechanical way to solve the problem of potential bias in point estimates, there is no mechanical way to solve the problem of potential bias in standard errors.

Basics

• Consider 𝑌 = 𝑋𝛽 + 𝜀. The standard errors of the OLS estimates of 𝛽 are the square roots of the diagonal elements of (𝑋′𝑋)−1𝑋′𝛺𝑋 𝑋′𝑋 −1, where Ω is the variance-covariance matrix of ε.

• Thus, standard errors can be computed using (𝑋′𝑋)−1𝑋𝑋𝛺�𝑋(𝑋′𝑋)−1, where 𝛺� is an estimate of Ω.

• The basic idea of corrected standard errors is to use information from the estimated residuals to construct 𝛺.�

The “Original Sin” of Corrected Standard Errors

• Since 𝛺 ≡ E 𝜀𝜀′ , it is tempting to estimate Ω as 𝛺� = 𝜀̂𝜀̂′, where 𝜀̂ is the vector of regression residuals.

• With this choice of a 𝛺,� our estimated variance-covariance matrix for �̂� − 𝛽 is (𝑋′𝑋)−1𝑋𝑋𝜀̂𝜀̂′𝑋(𝑋′𝑋)−1.

• We can write this as 𝑋′𝑋 −1(𝑋𝑋𝜀̂)(𝑋𝑋𝜀̂)𝑋(𝑋′𝑋)−1.

• Since 𝑋𝑋𝜀̂ = 0, this gives us standard errors of zero.

• Oops!

For a Little More on These Issues

• See the handout.