Adam Gulan, Markus Haavio, Juha Kilponen. Kiss Me Deadly: From Finnish Great Depression to Great Recession
1. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Kiss Me Deadly:
From Finnish Great Depression to Great Recession
Adam Gulan Markus Haavio Juha Kilponen
Bank of Finland1
April 10 2015
1The views expressed are those of the authors and do not necessarily reflect
the views of the Bank of Finland
3. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Finnish Great Depression: Two stories
Real:
Collapse of Finnish-Soviet trade in 1991:
and ToT reversal
reinforced by labor market frictions
“From Russia with Love”, Gorodnichenko et al., AER 2012
(see also Tarkka, 1994)
Financial:
Finnish Great Depression was preceded by financial
liberalization, asset price boom, credit boom
asset price collapse
severe banking crisis and credit crunch
e.g. Vihri¨al¨a, 1997, Honkapohja and Koskela, 1999
4. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Finnish Great Depression: Two stories
Real:
Collapse of Finnish-Soviet trade in 1991
and ToT reversal
reinforced by labor market frictions
“From Russia with Love”, Gorodnichenko et al., AER 2012
(see also Tarkka, 1994)
Financial:
Finnish Great Depression was preceded by financial
liberalization, asset price boom, credit boom
asset price collapse
severe banking crisis and credit crunch
e.g. Vihri¨al¨a, 1997, Honkapohja and Koskela, 1999
6. ECB-RESTRICTED
Finnish GDP and exports to Russia,
change from a year earlier, million €2000
3.4.2014 Adam Gulan 2
-5000
-4000
-3000
-2000
-1000
0
1000
2000
3000
1985 1990 1995 2000 2005 2010
Exports to Russia GDP
13. 1985 1990 1995 2000 2005 2010 2015
-100
0
100%
New bank loans, change from a year earlier
1985 1990 1995 2000 2005 2010 2015
-50
0
50
%
Real house prices, change from a year earlier
1985 1990 1995 2000 2005 2010 2015
-100
0
100
%
Real stock prices, change from a year earlier
15. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Model specification
Estimate a partially identified SVAR(1) model of 9 variables
External block:
- World trade (real total global imports)
- Finnish ToT (price of exports over price of imports)
- financial market stress indicator CISS (Hollo et al. 2012)
New Keynesian block:
- real GDP
- inflation (GDP deflator)
- interest rate spread (lending rate - 3m MM rate)
Financial block:
- asset prices (first PCA of stock and house prices)
- new loan volumes (to nonfinancial private sector)
- loan losses (total real losses of banks)
16. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Model specification
Estimate a partially identified SVAR(1) model of 9 variables
External block:
- World trade (real total global imports)
- Finnish ToT (price of exports over price of imports)
- financial market stress indicator CISS (Hollo et al. 2012)
New Keynesian block:
- real GDP
- inflation (GDP deflator)
- interest rate spread (lending rate - 3m MM rate)
Financial block:
- asset prices (first PCA of stock and house prices)
- new loan volumes (to nonfinancial private sector)
- loan losses (total real losses of banks)
17. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Model specification
Estimate a partially identified SVAR(1) model of 9 variables
External block:
- World trade (real total global imports)
- Finnish ToT (price of exports over price of imports)
- financial market stress indicator CISS (Hollo et al. 2012)
New Keynesian block:
- real GDP
- inflation (GDP deflator)
- interest rate spread (lending rate - 3m MM rate)
Financial block:
- asset prices (first PCA of stock and house prices)
- new loan volumes (to nonfinancial private sector)
- loan losses (total real losses of banks)
18. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Model specification
Estimate a partially identified SVAR(1) model of 9 variables
External block:
- World trade (real total global imports)
- Finnish ToT (price of exports over price of imports)
- financial market stress indicator CISS (Hollo et al. 2012)
New Keynesian block:
- real GDP
- inflation (GDP deflator)
- interest rate spread (lending rate - 3m MM rate)
Financial block:
- asset prices (first PCA of stock and house prices)
- new loan volumes (to nonfinancial private sector)
- loan losses (total real losses of banks)
22. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Financial shocks
Asset price shock:
exuberance and bubbles (Bernanke and Gertler, 1999)
net worth shock (Bernanke and Gertler, 1989
risk shock (Christiano et al. 2014)
news about future TFP?
(Christiano et al., 2010 vs Gilchrist and Leahy, 2002)
Loan supply shock:
changes in lending standards and regulatory environment
monitoring costs (De Fiore et al. 2011, Fuentes-Albero, 2014)
Which financial shocks matter?
Bassett et al., 2010
Helbling et al., 2011
23. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Financial shocks
Asset price shock:
exuberance and bubbles (Bernanke and Gertler, 1999)
net worth shock (Bernanke and Gertler, 1989
risk shock (Christiano et al. 2014)
news about future TFP?
(Christiano et al., 2010 vs Gilchrist and Leahy, 2002)
Loan supply shock:
changes in lending standards and regulatory environment
monitoring costs (De Fiore et al. 2011, Fuentes-Albero, 2014)
Which financial shocks matter?
Bassett et al., 2010
Helbling et al., 2011
24. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Financial shocks
Asset price shock:
exuberance and bubbles (Bernanke and Gertler, 1999)
net worth shock (Bernanke and Gertler, 1989
risk shock (Christiano et al. 2014)
news about future TFP?
(Christiano et al., 2010 vs Gilchrist and Leahy, 2002)
Loan supply shock:
changes in lending standards and regulatory environment
monitoring costs (De Fiore et al. 2011, Fuentes-Albero, 2014)
Which financial shocks matter?
Bassett et al., 2010
Helbling et al., 2011
25. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Financial shocks
Asset price shock:
exuberance and bubbles (Bernanke and Gertler, 1999)
net worth shock (Bernanke and Gertler, 1989
risk shock (Christiano et al. 2014)
news about future TFP?
(Christiano et al., 2010 vs Gilchrist and Leahy, 2002)
Loan supply shock:
changes in lending standards and regulatory environment
monitoring costs (De Fiore et al. 2011, Fuentes-Albero, 2014)
Which financial shocks matter?
Bassett et al., 2010
Helbling et al., 2011
26. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Financial shocks
Asset price shock:
exuberance and bubbles (Bernanke and Gertler, 1999)
net worth shock (Bernanke and Gertler, 1989
risk shock (Christiano et al. 2014)
news about future TFP?
(Christiano et al., 2010 vs Gilchrist and Leahy, 2002)
Loan supply shock:
changes in lending standards and regulatory environment
monitoring costs (De Fiore et al. 2011, Fuentes-Albero, 2014)
Which financial shocks matter?
Bassett et al., 2010
Helbling et al., 2011
27. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Financial shocks
Asset price shock:
exuberance and bubbles (Bernanke and Gertler, 1999)
net worth shock (Bernanke and Gertler, 1989
risk shock (Christiano et al. 2014)
news about future TFP?
(Christiano et al., 2010 vs Gilchrist and Leahy, 2002)
Loan supply shock:
changes in lending standards and regulatory environment
monitoring costs (De Fiore et al. 2011, Fuentes-Albero, 2014)
Which financial shocks matter?
Bassett et al., 2010
Helbling et al., 2011
28. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Financial shocks
Asset price shock:
exuberance and bubbles (Bernanke and Gertler, 1999)
net worth shock (Bernanke and Gertler, 1989
risk shock (Christiano et al. 2014)
news about future TFP?
(Christiano et al., 2010 vs Gilchrist and Leahy, 2002)
Loan supply shock:
changes in lending standards and regulatory environment
monitoring costs (De Fiore et al. 2011, Fuentes-Albero, 2014)
Which financial shocks matter?
Bassett et al., 2010
Helbling et al., 2011
29. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Partial identification
There are fewer identified domestic shocks (4) than domestic
variables (6)
Hence the model is partially identified: there are two
unidentified shocks
Some shocks cannot be identified, given the set of variables,
and the time range (e.g. monetary policy shocks)
More generally: the limitations of economic modelling
30. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - sign restrictions
Reduced-from VAR(1) model
yt = Ayt−1 + ut, ut ∼ N(0, Σ)
Structural shocks: linked to residuals through some
identification matrix W
ut = W εt, Σ = WW
Start with Cholesky decomposition on Σ
Σ = BB
where B is a lower triangular matrix
Draw some orthonormal matrix Q (such that QQ = I)
Σ = BB = BQQ B
so that W = BQ and ut = BQεt.
31. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - sign restrictions
Reduced-from VAR(1) model
yt = Ayt−1 + ut, ut ∼ N(0, Σ)
Structural shocks: linked to residuals through some
identification matrix W
ut = W εt, Σ = WW
Start with Cholesky decomposition on Σ
Σ = BB
where B is a lower triangular matrix
Draw some orthonormal matrix Q (such that QQ = I)
Σ = BB = BQQ B
so that W = BQ and ut = BQεt.
32. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - sign restrictions
Reduced-from VAR(1) model
yt = Ayt−1 + ut, ut ∼ N(0, Σ)
Structural shocks: linked to residuals through some
identification matrix W
ut = W εt, Σ = WW
Start with Cholesky decomposition on Σ
Σ = BB
where B is a lower triangular matrix
Draw some orthonormal matrix Q (such that QQ = I)
Σ = BB = BQQ B
so that W = BQ and ut = BQεt.
33. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - sign restrictions
Reduced-from VAR(1) model
yt = Ayt−1 + ut, ut ∼ N(0, Σ)
Structural shocks: linked to residuals through some
identification matrix W
ut = W εt, Σ = WW
Start with Cholesky decomposition on Σ
Σ = BB
where B is a lower triangular matrix
Draw some orthonormal matrix Q (such that QQ = I)
Σ = BB = BQQ B
so that W = BQ and ut = BQεt.
34. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks
et = B 1
ut
where BB0 = Σ and B is obtained by a the Cholesky
decomposition
3 Draw an orthonormal rotation matrix Q, and produce an
alternative set of structural shocks
εt = Q0
et
35. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks
et = B 1
ut
where BB0 = Σ and B is obtained by the Cholesky
decomposition
3 Draw an orthonormal rotation matrix Q, and produce an
alternative set of structural shocks
εt = Q0
et
36. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks
et = B 1
ut
where BB0 = Σ and B is obtained by a the Cholesky
decomposition
3 Draw an orthonormal rotation matrix Q, and produce an
alternative set of structural shocks
εt = Q0
et
37. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks et
3 lternative set of structural shocks εt
4 Let
W = BQ
The impulse responses to the structural shocks εt are given by
Φ0 = W (on impact, or period 0)
Φ1 = AW (period 1)
...
Φj = Aj
W (period j)
5 If the impulse responses satisfy the sign restricitons, keep the
rotation matrix Q and the stuctural shocks εt . Otherwise
discard Q and the stuctural shocks εt .
38. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks et
3 lternative set of structural shocks εt
4 Let
W = BQ
The impulse responses to the structural shocks εt are given by
Φ0 = W (on impact, or period 0)
Φ1 = AW (period 1)
...
Φj = Aj
W (period j)
5 If the impulse responses satisfy the sign restricitons, keep the
rotation matrix Q and the stuctural shocks εt . Otherwise
discard Q and the stuctural shocks εt .
39. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks et
3 lternative set of structural shocks εt
4 Let
W = BQ
The impulse responses to the structural shocks εt are given by
Φ0 = W (on impact, or period 0)
Φ1 = AW (period 1)
...
Φj = Aj
W (period j)
5 If the impulse responses satisfy the sign restricitons, keep the
rotation matrix Q and the stuctural shocks εt . Otherwise
discard Q and the stuctural shocks εt .
40. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks et
3 lternative set of structural shocks εt
4 Let
W = BQ
The impulse responses to the structural shocks εt are given by
Φ0 = W (on impact, or period 0)
Φ1 = AW (period 1)
...
Φj = Aj
W (period j)
5 If the impulse responses satisfy the sign restricitons, keep the
rotation matrix Q and the stuctural shocks εt . Otherwise
discard Q and the stuctural shocks εt .
41. The procedure step by step
1 Reduced form residuals ut
2 Cholesky-based structural shocks et
3 lternative set of structural shocks εt
4 Let
W = BQ
The impulse responses to the structural shocks εt are given by
Φ0 = W (on impact, or period 0)
Φ1 = AW (period 1)
...
Φj = Aj
W (period j)
5 If the impulse responses satisfy the sign restricitons, keep the
rotation matrix Q and the stuctural shocks εt . Otherwise
discard Q and the stuctural shocks εt .
42. The procedure step by step
We repeat the procedure N times
In our case, N = 8 1010 (or 80 billion)
... and we get a certain number K of structural models (or Q
matrices) that satisfy all the sign restrictions (and the
Fry-Pagan …lter)
In our case, K = 2700
43. Model selection
Which of the K structural models do we choose, when we try
to interprete Finnish business cycles, and economic crises?
We base model selection on the historical shock decomposition
Θt = ∑
j
Φj Et j
where Et j is matrix with εt j on the diagonal (and zeros
elsewhere)
The cumulative contribution of current and past structural
shocks to model variables
44. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - choosing the model
Each draw of Q gives rise to a different model and IRFs.
Keep only those draws of Q which satisfy sign restrictions.
Choosing the final model from all admissible candiates.
Modification of Fry and Pagan, JEL, 2011
Median historical decomposition:
x∗
= argmin
N
∑
n=1
J
∑
j=1
T
∑
t=1+p
(θx
n,j,t − ¯θn,j,t )2
θx
n,j,t is normalized cummulative effect of shock j on variable n
up to period t, obtained via vector MA representation,
¯θn,j,t is the median over all model candidates.
45. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - choosing the model
Each draw of Q gives rise to a different model and IRFs.
Keep only those draws of Q which satisfy sign restrictions.
Choosing the final model from all admissible candiates.
Modification of Fry and Pagan, JEL, 2011
Median historical decomposition:
x∗
= argmin
N
∑
n=1
J
∑
j=1
T
∑
t=1+p
(θx
n,j,t − ¯θn,j,t )2
θx
n,j,t is normalized cummulative effect of shock j on variable n
up to period t, obtained via vector MA representation,
¯θn,j,t is the median over all model candidates.
46. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - choosing the model
Each draw of Q gives rise to a different model and IRFs.
Keep only those draws of Q which satisfy sign restrictions.
Choosing the final model from all admissible candiates.
Modification of Fry and Pagan, JEL, 2011
Median historical decomposition:
x∗
= argmin
N
∑
n=1
J
∑
j=1
T
∑
t=1+p
(θx
n,j,t − ¯θn,j,t )2
θx
n,j,t is normalized cummulative effect of shock j on variable n
up to period t, obtained via vector MA representation,
¯θn,j,t is the median over all model candidates.
47. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - choosing the model
Each draw of Q gives rise to a different model and IRFs.
Keep only those draws of Q which satisfy sign restrictions.
Choosing the final model from all admissible candiates.
Modification of Fry and Pagan, JEL, 2011
Median historical decomposition:
x∗
= argmin
N
∑
n=1
J
∑
j=1
T
∑
t=1+p
(θx
n,j,t − ¯θn,j,t )2
θx
n,j,t is normalized cummulative effect of shock j on variable n
up to period t, obtained via vector MA representation,
¯θn,j,t is the median over all model candidates.
48. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Methodology - multiple shocks problem
To avoid contaminating identified shocks with unidentified ones,
disregard all draws of Q for which unidentified shocks give rise to
same impulse response patterns as the identified ones.
Then, all unidentifed shocks remain orthogonal to identified ones.
+ − ?
+ + ?
+ + ?
W 1
≡
+ − +
+ + +
+ + +
and W 2
≡
+ − +
+ + −
+ + +
57. Introduction Methodology Results Concluding remarks Impulse Responses Robustness
Concluding remarks
“From Russia with Love” can explain at most half of Finnish
Great Depression.
Financial crisis started the depression and prolonged it
Great recession was very different. Imported recession.
No financial crisis in Finland.
Initial financial conditions much more robust.
Large overall role of financial shocks, esp. related to loan
supply and banking