Transcript Was it Predictable? - Athabasca University

The First Global Crisis of the Twenty-first Century:
Was it Predictable?
Saktinil Roy
Athabasca University
Introduction
The US Subprime Crisis
by now a global crisis
 “unprecedented” since the Great Depression

However
some authors suggest: historical bank-centered
crises in several other parts of the world could
foretell the story
 My research reviews and confirms this
proposition

Introduction (contd...)
In particular, Reinhart & Rogoff (2008a)

find “striking” similarities between post world-war banking
crises in advanced economies and the US subprime crisis

consider a few crisis indicators and suggest certain changes
in their behaviour as “precursors” of crisis
Reinhart & Rogoff (2008b)

arrive at the same conclusion with an extended database that
goes back to 1800 and includes a core sample of sixty-six
countries
Introduction (contd...)
But the questions really are

Could we actually “predict” the subprime crisis
with historical experiences?

How well in advance?

Which country experiences would have given the
best prediction?

Which indicators are most relevant?
Introduction (contd...)
Some other observations
 Shiller (2008):
 The recent housing boom in the United States was lot more
pronounced in the low-price tier than in the high-price tier.
 After the bubble burst the sharpest fall in house prices was
observed in the low-price market
 Krugman (2007):
 Increase in risky mortgage debts in the United States partly as an
effect of protracted increase in income inequality over the last
three decades
 Iacoviello (2008) suggests a similar causality:
 Examines the US time series of income distribution and
household debt from 1963 through 2003
 Finds that long run increase in household debt can be explained
only by increasing income gaps
Introduction (contd..)
Figure 1. 1950--2007: US Labor Productivity and Real Wage Rate
100
90
80
70
60
50
40
30
50
55
60
65
70
75
80
Labor productivity
85
90
95
Real wage
00
05
Introduction (contd..)
Hence, a couple of other questions

The historical crisis episodes better
understood in relation to rise in income
inequality?

This could help predict the US experience
with greater confidence?
Reinhart & Rogoff, 2008a
Postwar Crisis episodes:

The “Big Five” Crises:


Spain (1977), Norway (1987), Finland (1991), Sweden
(1991), Japan (1992)
Other Postwar Bank-centered Financial Crises:
 Australia (1989), Canada (1983), Denmark (1987),
France (1994), Germany (1977), Greece (1991),
Iceland (1985), Italy (1990), New Zealand (1987),
Britain (1974, 1984,1991, 1995), and the United States
(1984)
Reinhart and Rogoff, 2008a (contd..)
Indicators:





Real Estate Real Price
Real Equity Price
% of Current Account in GDP
Growth of Real GDP per Capita
Public Debt
Reinhart & Rogoff, 2008a (contd..)
Figure 1: Real Housing Prices and Banking Crises
135
130
125
US, 2003=100
120
Index
115
110
Average for banking crises in
advanced economies
105
100
Average for the "Big 5"
Index t-4=100
95
t-4
t-3
t-2
t-1
T
t+1
t+2
t+3
Reinhart & Rogoff, 2008a (contd..)
Figure 2: Real Equity Prices and Banking Crises
135
130
125
US, 2003=100
120
115
110
Index
105
100
Average for banking crises in
advanced economies
95
90
85
Average for the "Big 5"
Index t-4=100
80
t-4
t-3
t-2
t-1
T
t+1
t+2
t+3
Reinhart & Rogoff, 2008a (contd..)
Figure 3: Current Account Balance/GDP on the
Eve of Banking Crises
t-4
t-3
t-2
t-1
t
0
-1
percent of GDP
-2
-3
-4
-5
U.S.
-6
-7
Average for banking crises in
advanced economies
Reinhart and Rogoff, 2008a (contd..)
Figure 4: Real GDP Growth per Capita and
Banking Crises (PPP basis)
5
4
Percent
3
US
Average for banking crises in
advanced economies
2
1
Average for the "Big 5" Crises
0
-1
-2
1
Reinhart & Rogoff, 2008a (contd..)
Figure 5: Public Debt and Banking Crises
250
230
210
Index
190
Average for banking crises in
advanced economies
170
150
Average for the "Big 5" Crises
130
US, 1997=100
Index t-10=100
110
90
t-4
t-3
t-2
t-1
T
t+1
t+2
t+3
Reinhart & Rogoff, 2008a (contd..)
Limitations:



If a certain change in the behavior of an indicator must be accepted
as a “precursor of crisis” is an issue that must be settled only in
practice – specifically with an actual out-of-sample forecasting
exercise
If we assume that the four year period foreshadows a crisis
episode, then for “prediction” purposes it is also important to
know if there were similarities even before this period – this will
then allow us to distinguish between a period that is “tranquil” and
a period that is “pre-crisis”
A comparison of any crisis with the “average” of historical crises
could be elusive
 There could be differences across countries which might be
cloaked under the “average” construct
 For prediction purposes such differences need to be controlled
Predicting the US Subprime Crisis
The Problem:

Specifically, could we predict the crisis
starting from 2003?

Same as asking: going back to the year of
2002 or earlier could we predict that the
United States was not going to experience
any financial crisis in 2006 or earlier (except
the savings and loan crisis in 1984)?
Predicting the US Subprime Crisis
(contd..)
I consider two additional indicators

Growth difference between average
productivity and real wage rate


In the absence of adequate data on Gini coefficient
and other measures of income inequality this is taken
as the measure of growth of income gap
Inflation

To account for growth of cost of living
Predicting the US Subprime Crisis
(contd..)
Data, Methodology & Criteria:

For any historical crisis include only eight
observations prior to the actual occurrence

Labelling


“Pre-crisis” period: four years just prior to the crisis
episode
“Tranquil” period: four years even prior to the “precrisis” period
Predicting the US Subprime crisis
(contd..)

A panel probit model

The binary dependent variable
= 1 when “pre-crisis”;
= 0 when “tranquil”

To conclude in favour of a crisis incidence within
a period of four years
The predicted probability is compared to a pre-specified
threshold value
 If the probability exceeds the threshold then a crisis
incidence is predicted
 I consider both 50% and 75% as the threshold

Predicting the US Subprime Crisis
(contd..)
Results:

Four specifications
 Specification 1A (all postwar crisis episodes)
 Except Public Debt all indicators considered by Reinhart &
Rogoff (2008a) are significant
 Specification 1B (all postwar crisis episodes)
 The two additional indicators are introduced – only “inequality
growth” is significant but with a wrong sign
 Specification 2 (only the BIG FIVE crises)
 Only Real Estate Price, Public Debt & Inflation are significant
 Specification 3 [Finland (1991), Italy (1990), Japan (1992), New
Zealand (1987), UK (1995) & US (1984)]
 Real Estate Price, Public Debt, Current Account and Inequality
Growth are significant with the correct signs
Predicting the US Subprime Crisis
(contd..)
Within Sample Performance Results
Spec 1A
Spec1B
Spec 2
Spec 3
______________________________________________________________________________________________
Threshold Probability = 50%
Percent of pre-crisis years
correctly called
86.8
85.3
80.0
79.2
True alarms as percent
of total alarms
77.6
79.4
94.1
82.6
69.1
64.7
80.0
70.8
73.9
74.6
94.1
94.4
Threshold Probability = 75%
Percent of pre-crisis years
correctly called
True alarms as percent
of total alarms
Predicting the US Subprime Crisis
(contd..)
Out-of-Sample Performance Results
Spec 1A
Spec1B
Spec 2
Spec 3
______________________________________________________________________________________________
Threshold Probability = 50%
Percent of pre-crisis years
correctly called
True alarms as percent
of total alarms
100.0
75.0
100.0
100.0
80.0
100.0
80.0
100.0
50.0
50.0
75.0
100.0
100.0
100.0
100.0
100.0
Threshold Probability = 75%
Percent of pre-crisis years
correctly called
True alarms as percent
of total alarms
Predicting the US Subprime Crisis
(contd..)
Interpretations:

Significance of Real Estate Price:
 Consistent with Shiller (2008) – “irrational exuberance” -economists, policy makers & market experts believed in everrising house prices
 Similar thing was observed during the housing & stock market
booms in Japan before the financial crisis in 1992 and also in
Sweden and Finland before the financial crises in 1991.
Predicting the US Subprime Crisis
(contd..)

Significance of Current Account
 Consistent with Reinhart & Rogoff (2008a, 2008b):
persistent current account deficit & capital flow
bonanza are precursors of crisis

Significance of Public Debt
 Again, consistent with Reinhart and Rogoff (2008a)

Significance of Inequality Growth
 Like with other historical experiences, rise in earnings
inequality contributed significantly – consonant with
Krugman (2007)
Conclusions
 The US Subprime fiasco could be predicted consistently starting
in 2003
 Earlier than 2005 when Robert Shiller (one of the few
economists) predicted the crisis
 Despite Alan Greenspan (2007) talking about “froths” in the local markets,
thus rejecting the possibility of a speculative bubble
 The crisis could be predicted much earlier than February 2007 -- a date
that the Global Financial Stability Report (ch. 3, April 2009) by the IMF states
as the earliest when increasing systemic pressures could be signaled
 One got to look at both “similarities” and “dissimilarities” across historical
crisis experiences
 Apart from the traditional indicators, inequality growth contributed
significantly