Transcript Slides

Financial cycles
Mathias Drehmann
DNB Workshop on “Estimating and Interpreting Financial Cycles”
Amsterdam, 2 September 2016
The views presented are those of the author and do not necessarily represent those of the Bank for
International Settlements
Restricted
What is the financial cycle?
 The broad concept of the financial cycle encapsulates joint
fluctuations in a wide set of financial variables
 Financial cycles are characterised by financial booms and busts
that can lead to serious financial and macroeconomic strains
Restricted
2
There are many ways to measure the financial cycle
 Initial literature focused on early warning indicators (EWIs) for
crises or identified credit booms statistically
 Does not allow for characterizing the full financial cycle
 But EWIs can be used to construct financial cycle measures
 Use variants of the classical business cycle dating method
 Claessens et al (2011), Drehmann et al (2012),…
 Use statistical filters
 Aikman et al (2015), Drehmann et al (2012), De Bonis &
Silvestrini (2014), Runstler & Vlekke (2015), Galati et al (2016)..
 Use long-run relationships (Juselius and Drehmann (2015))
Restricted
3
Drehmann, Borio, Tsatsaronis (2012)
 Frequency based filters (Christiano and Fitzgerald, 2005, Comin
and Gertler, 2006)
 Turning-point method (Burns and Mitchell, 1946, Harding and
Pagan, 2002)
Short term cycle
(business cycle)
Medium term
cycle
Frequency
based filters
Turning-points
5-32 quarters
Local maxima\minima: 5q window
Minimum cycle length: 5 quarters
8-30 years
Local maxima\minima: 9q window
Minimum cycle length: 5 years
Restricted
4
Explore potential series
 Business cycle:
 GDP
 Financial cycle:
 Credit to the private non-financial sector
 Credit-to-GDP ratio
 Property prices
 Equity prices
 Asset prices
 Because of lack of data do not include other financial
variables (e.g. spreads, profits, write-offs, leverage…)
 7 countries (AU, DE, JP, NO, SE, UK, US) with 11 banking crises
 Quarterly data from 1960 to 2011
Restricted
5
Medium-term versus short–term: Filters
 Volatility of medium-term component greater than short-term
one for all variables
 Medium-term component becomes larger after 1985
Table 1
Relative volatility of short- and medium-term cycles: individual series1
(Frequency-based analysis)
AU
DE
GB
JP
NO
SE
US
Credit
4.52
1.80
3.73
4.34
6.28
6.78
3.87
Credit/GDP
7.36
2.83
5.28
3.39
4.99
5.98
4.92
House prices
1.75
2.19
2.42
3.05
2.21
4.91
3.91
Equity prices
1.72
1.40
1.77
2.14
1.30
1.42
1.41
AAP2
1.95
3.94
2.56
3.36
1.60
1.48
1.75
GDP
3.25
1.73
1.93
3.06
2.55
1.84
1.51
1
The figures refer to the ratio of the standard deviation of the medium-term cyclical component to that of the
short-term component over the entire sample period. A number greater (smaller) than 1.0 means that the
medium-term cyclical component is more (less) volatile than the short-term component. Cells shown in bold
denote cases where the ratio of medium- to short-term component volatility for the corresponding series is higher
that the corresponding ratio for GDP. Acronyms used here and in subsequent tables and graphs:
AU = Australia; DE= Germany; GB= United Kingdom; JP = Japan; NO = Norway; SE = Sweden; US = United
States. 2 Aggregate asset price index.
Restricted
6
What happens around crisis?
 All domestic crises coincide with medium term cyclical peaks
 Independent of the method
 Peaks in medium term cycles often coincide with crises
 For credit and property prices 40-50% of peaks coincide
with crises (65-70% after 1985)
 For short term cycles the relationship much weaker (1830%)
→ Medium-term frequencies are key!
Restricted
7
Aggregation
 Frequency based filters
 Average of individual series
 Turning-point method (Harding and Pagan (2005))
 Peak in the common cycle if
- there is a cluster when all individual series peak
- individual series are closest to their peak within the
cluster
 Impose same constraints as on dating method for individual
series
 Cluster width
 3 years
 3 to 6 yeas (weak)
Restricted
8
Which series should underpin the financial cycle?
 The financial cycle is derived from credit, the credit-to-GDP
ratio and property prices
 Equity prices (and thus aggregate asset prices) are not included
 Greater short-term volatility
 Medium-term cyclical peaks occur often without crisis
 Medium-term cycle not well aligned with credit series or
property prices
- Low concordance (turning-point method)
- Low correlation (frequency based filters)
Restricted
9
Peaks in the financial cycle are close to crises or
periods of stress
Country
Date
Close to crises
GB
2009q1
SE
2009q1
US
2007q3
JP
1992q2
GB
1991q1
AU
1990q3
US
1990q3
SE
1990q2
NO
1989q3
GB
1973q4
Average
Time to
closest
Time to closes
crises peak using filters Country
Date
Not close to crises
-6
5
NO
2009q2
-2
4
AU
2009q1
0
0
DE
1998q4
2
-3
SE
1980q4
-3
-2
US
1979q3
-3
-2
DE
1973q4
-2
-5
JP
1973q4
5
3
5
-2
0
0
-0.4
-0.2
Average
Time to
closest
Time to closes
crises peak using filters
-74
-77
35
43
42
135
76
3
1
9
-2
-1
-1
-2
25.7
1.0
Restricted
10
Different financial cycle dating methods generally coincide
Spain
United Kingdom
United States
Source: BIS, 2016, 86th Annual Report
Restricted
11
The EWI method
 Built on the EWI literature
 Large credit-to-GDP gaps, i.e. deviations of the credit-toGDP ratio from a long-run trend, are useful early warning
indicators. (eg Drehmann et al (2011), Detken et al (2014),
Schularick and Taylor (2012)…)
 Real residential property prices start to fall ahead of
financial crises
Restricted
12
The signalling quality of different EWIs
AUCs for different forecast horizons
Graph 2
Credit-to-GDP gap
Debt-service-ratio (DSR)
Non-core liabilities
Credit growth
Property price growth
GDP growth
The horizontal-axis denotes the forecast horizons in quarters before crises. The vertical-axis denotes AUC. The horizontal line at 0.5
highlights the value of an uninformative indicator. A solid blue line indicates that the specific variable for the given horizon is statistically
different from an uninformative indicator, while a dashed blue line indicates the opposite. A hollow blue circle shows that the signal is
stable in the sense that it does not reverse direction within the forecast horizon until the crisis. Red diamonds highlight that the specific
variable is statistically the best indicator for this particular horizon. Other indicators that are not statistically different from best-performing
indicator are marked by solid blue circles.
Source: Drehmann and Juselius (2013).
Restricted
13
The EWI method (II)
 Define peak (trough) in financial cycle:
 Peak: positive credit-to-GDP gap and real property price
growth turns persistently negative
 Trough: negative credit-to-GDP gap and real property price
growth turns persistently positive
Restricted
14
Different financial cycle dating methods generally coincide
Spain
United Kingdom
United States
Source: BIS, 2016, 86th Annual Report
Restricted
15
Decomposing the financial cycle
 Juselius and Drehmann (2015) (JD) decompose the financial
cycle into two long run relationships:
 Leverage (credit-to-assets ratio): (𝑐𝑡 − 𝑦𝑡 ) − (𝑝𝐴,𝑡 − 𝑝𝑡 )
 Debt service ratio (DSR): 𝑐𝑡 − 𝑦𝑡 + 𝛽𝑖𝐿,𝑡
Restricted
16
Estimation strategy
 JD estimate VAR in error correction form
Δ𝑐𝑟 𝑟
Δ𝑦 𝑟
Δ𝑝𝐴𝑟
Δ𝑖𝑟
with
2
= 𝜇 + 𝛼 𝑙𝑒𝑣
𝑑𝑠𝑟
𝑡
+
𝑡−1
Π𝑖
𝑖=1
Δ𝑐𝑟 𝑟
Δ𝑦 𝑟
Δ𝑝𝐴𝑟
Δ𝑖𝑟
+ Γ𝑠𝑡 + 𝜖𝑡
𝑡−𝑖
𝑙𝑒𝑣 = 𝑐𝑟𝑡𝑟 − 𝑦𝑡𝑟 − 𝑝𝐴𝑟 − 𝜇𝑙
𝑑𝑠𝑟 = 𝑐𝑟𝑡𝑟 − 𝑦𝑡𝑟 + 𝛽𝑖𝑟 − 𝜇𝑑
𝑐𝑟𝑡𝑟 real credit to the private non-financial sector, 𝑦𝑡𝑟 real GDP, 𝑝𝐴𝑟 real aggregate asset price index, 𝑖𝑟 nominal lending rate
on the stock of debt
 Quarterly US data from 1985 until 2013.
Restricted
17
DSR and leverage in the US
Restricted
18
Effects on growth
 Effects can work in same direction or off-set each other
 Big shocks or non-linearities not needed for sharp downturns
 DSR gap effect on expenditure explosive
 Long-lasting and large real effects
 Interaction between gaps lead to endogenous cycles
Restricted
19
(Pseudo) real-time prediction of the Great Recession
-1
0
-1
1
0
2
1
2
Expenditure growth
3
Credit growth
2005q3
2007q3
-2
Actual
Simulation 05
Long-run average
2009q3
2011q3
2013q3
2005q3
2007q3
2009q3
2011q3
2013q3
Restricted
20
The gaps and the financial cycle?
DSR gap
negative
positive
negative
Boom
(credit ↑, GDP↑)
Late boom
(credit ↑, GDP ~)
positive
Recovery
(credit ↓, GDP ~)
Bust
(credit ↓, GDP ↓)
Leverage
gap
Restricted
21
1
1985q1
1990q1
1995q1
2000q1
2005q1
2010q1
2015q1
-40
0
0
-20
-10
-20
0
0
10
20
20
1
40
GB
30
US
1985q1
1990q1
1995q1
2000q1
2005q1
2010q1
2015q1
boom
late_boom
boom
late_boom
bust
recovery
bust
recovery
lev
dsr
lev
dsr
Restricted
22
Different financial cycle dating methods generally coincide
Spain
United Kingdom
United States
Source: BIS, 2016, 86th Annual Report
Restricted
23
Conclusion
 Key empirical findings
 Financial cycles are longer than business cycles
 Peaks of the financial cycle coincide with serious strains in
financial systems\crisis.
 There are different methods to identify financial cycles
 From a high-level perspective, they lead to similar results
 From a day-to-day policy perspective, differences are large
Restricted
24