Real Time Early Warning Indicators Alessi Detken for Costly Asset - - PowerPoint PPT Presentation

▶

Dec 01, 2022 272 likes •593 views

Real Time Early Warning for Costly Boom/Bust Cycles Real Time Early Warning Indicators Alessi Detken for Costly Asset Price Boom/Bust Cycles: Introduction Methodology A Role for Global Liquidity Boom identification Data

SLIDE 1

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

‘Real Time’ Early Warning Indicators for Costly Asset Price Boom/Bust Cycles: A Role for Global Liquidity

Lucia Alessi Carsten Detken

European Central Bank Milan, 10 May 2010

SLIDE 2

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Motivation

The need for Early Warning Models

The set up of an Early Warning System is one of the key tasks of the European Systemic Risk Board With the aim of identifying threats to financial stability in a timely manner, and allow for the adoption of targeted macro-prudential regulatory measures Implications of financial stability for price stability in the medium to long run EWMs are necessary input for “leaning against the wind” New generation of EWMs

SLIDE 3

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Introduction

Literature and contributions

Signalling approach of Early Warning Indicator Models (e.g. Kaminsky/Reinhart 1999, AER) Application to predict costly aggregate asset price boom/bust cycles

Real vs financial variables Global vs domestic financial variables Money vs credit Prediction whether mid-2000s asset price boom wave would be costly

Pseudo Real Time Ranking according to policy maker loss function, not noise to signal ratio (similar to Bussière/Fratzscher 2008, JPM), and focus on usefulness

SLIDE 4

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

Recursive detrending

Boom quarter if at least 3 consecutive quarters where QAAPRi > recursive HP trendi + 1.75 ∗ recursive stdevi

SLIDE 5

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

Recursive detrending

Boom quarter if at least 3 consecutive quarters where QAAPRi > recursive HP trendi + 1.75 ∗ recursive stdevi

50 100 150 200 250 300 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Asset Prices

SLIDE 6

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

Recursive detrending

Boom quarter if at least 3 consecutive quarters where QAAPRi > recursive HP trendi + 1.75 ∗ recursive stdevi

50 100 150 200 250 300 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Asset Prices Ex-Post Trend

SLIDE 7

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

Recursive detrending

Boom quarter if at least 3 consecutive quarters where QAAPRi > recursive HP trendi + 1.75 ∗ recursive stdevi

50 100 150 200 250 300 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Asset Prices Ex-Post Trend

Rec. Trend

SLIDE 8

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

Recursive detrending

Boom quarter if at least 3 consecutive quarters where QAAPRi > recursive HP trendi + 1.75 ∗ recursive stdevi

50 100 150 200 250 300 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Asset Prices Ex-Post Trend

Rec. Trend

SLIDE 9

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

Recursive detrending

Boom quarter if at least 3 consecutive quarters where QAAPRi > recursive HP trendi + 1.75 ∗ recursive stdevi

50 100 150 200 250 300 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Asset Prices Ex-Post Trend

Rec. Trend

SLIDE 10

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

Recursive detrending

Boom quarter if at least 3 consecutive quarters where QAAPRi > recursive HP trendi + 1.75 ∗ recursive stdevi bridged periods if less than 3 quarters in between booms artificially ended if QAAPR drops by more than 35% 60 booms identified results robust to classification

SLIDE 11

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Country-specific boom identification

High cost vs Low cost booms

High Cost Booms if real GDP growth 1 pp p.a. lower than potential growth on average over 3 post boom years 45 classifiable booms:

29 are HC 16 are LC (control group)

costly banking crises (FI ’91-’94, IT ’90-’95, SE ’91-’94) follow HC booms

SLIDE 12

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Boom/bust cycles

3 waves: mid-late 80s, 90s, 00s

Number of Countries with Aggregate Asset Price Booms

2 4 6 8 10 12 14 1 9 7 Q 1 1 9 7 1 Q 1 1 9 7 2 Q 1 1 9 7 3 Q 1 1 9 7 4 Q 1 1 9 7 5 Q 1 1 9 7 6 Q 1 1 9 7 7 Q 1 1 9 7 8 Q 1 1 9 7 9 Q 1 1 9 8 Q 1 1 9 8 1 Q 1 1 9 8 2 Q 1 1 9 8 3 Q 1 1 9 8 4 Q 1 1 9 8 5 Q 1 1 9 8 6 Q 1 1 9 8 7 Q 1 1 9 8 8 Q 1 1 9 8 9 Q 1 1 9 9 Q 1 1 9 9 1 Q 1 1 9 9 2 Q 1 1 9 9 3 Q 1 1 9 9 4 Q 1 1 9 9 5 Q 1 1 9 9 6 Q 1 1 9 9 7 Q 1 1 9 9 8 Q 1 1 9 9 9 Q 1 2 Q 1 2 1 Q 1 2 2 Q 1 2 3 Q 1 2 4 Q 1 2 5 Q 1 2 6 Q 1 2 7 Q 1

High Cost Booms Low Cost or Unclassified Booms

SLIDE 13

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Data

18 variables for 18 countries

Real: GDP , private consumption, total investment, housing investment, consumer prices Financial: equity, private housing, aggregate asset prices (including also commercial housing), long rates, short rates, term spread, M1, M3, private credit, domestic credit, real effective exchange rates, global M1, global M3, global short rates, global private credit, global domestic credit Sample is 1970:Q1 to 2007:Q4

SLIDE 14

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Indicators

89 indicators

(Up to) 6 transformations per variable: ratios to GDP annual growth rates 6 quarters cumulated growth rates recursively HP detrended (constant and variable history) ratios to GDP recursively HP detrended (constant and variable history) cumulated shocks from recursive VAR in growth rates (for money and credit variables)

SLIDE 15

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Global private credit gap and optimal threshold

Evaluation period 1979-2002

0.5 1 1.5 1979Q1 1983Q1 1987Q1 1991Q1 1995Q1 1999Q1 2003Q1 2007Q1

Housing/ Savings and Loans dot.com Credit

SLIDE 16

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Signalling approach

Policy Maker’s Loss Function Costly No Costly Boom Boom Signal A B No signal C D

L = θ C A + C

booms not called

+(1 − θ) B B + D

false alarms

θ= policy maker’s relative aversion of missing a call versus receiving a false alarm

Usefulness = min[θ; 1 − θ] − L

SLIDE 17

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

The trade-off between missed crises and false alarms

Optimal thresholds for the Global M1 Gap

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Frequency

Type I errors (missing crises) Type II errors (false alarms)

0.90 0.85 0.40 0.10 0.95

SLIDE 18

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Optimal threshold

Given θ, we optimize threshold quantile (by means of grid search in [0.05; 0.95], step 0.05) for each indicator with respect to loss function (for same θ) Three levels of aggregation

18 individual countries (AU, BE, CA, CH, DE, DK, ES, FI, FR, GB, IE, IT, JP , NL, NO, NZ, SE, US) average over all countries GDP weighted average over 8 EA countries

SLIDE 19

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Joint indicators

A signal is issued when both indicators breach their respective thresholds Global PC and Global M1 plus 16 best individual indicators in 2 dimensional grid search ⇒ reduce false alarms

SLIDE 20

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Evaluation period

Evaluation period is 1979:Q1 to 2002:Q1 Horizon is 6 quarters Boom quarters are excluded from evaluation as of the 4th boom quarter Last boom wave is excluded from evaluation Same analysis on the control group of Low Cost Booms ⇒ QAAPR is the best

SLIDE 21

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

(Pseudo) Real Time

Data availability at each point in time is (roughly) taken into account by means of using appropriately lagged variables (usually -1 quarter, often more for housing and aggregate asset prices, up to

4). We do not use a true real time dataset (with original vintages of

data) The distribution of the indicators and thus the actual threshold for issuing a signal changes at each point in time. What is not real time is the optimal quantile, which is derived using all booms for which data availability allowed a classification as high

r low cost

SLIDE 22

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

(Pseudo) Real Time

Data availability at each point in time is (roughly) taken into account by means of using appropriately lagged variables (usually -1 quarter, often more for housing and aggregate asset prices, up to

4). We do not use a true real time dataset (with original vintages of

data) The distribution of the indicators and thus the actual threshold for issuing a signal changes at each point in time. What is not real time is the optimal quantile, which is derived using all booms for which data availability allowed a classification as high

r low cost

SLIDE 23

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Best 3 indicators

Average over all countries

θ = 0.2 θ = 0.3 usefulness good false usefulness good false calls alarms calls alarms GlobM1-detr 0.03 0.38 0.06 GlobM1-detr 0.07 0.38 0.06 Shock-GlobalM1 0.01 0.09 0.01 GlobPC-detr 0.06 0.55 0.15 GlobM1-HP 0.01 0.11 0.02 QEPR-detr 0.04 0.47 0.14 θ = 0.4 θ = 0.5 usefulness good false usefulness good false calls alarms calls alarms GlobPC-HP 0.14 0.82 0.32 GlobPC-HP 0.25 0.82 0.32 GlobPC-detr 0.13 0.55 0.15 INV-cum 0.22 0.85 0.42 GlobM1-detr 0.12 0.48 0.12 QAAPR-yoy 0.21 0.9 0.47 θ = 0.6 θ = 0.8 usefulness good false usefulness good false calls alarms calls alarms GlobPC-HP 0.17 0.88 0.41 GDPR-HP 0.07 0.99 0.63 GDPR-HP 0.16 0.94 0.51 QAAPR-cum 0.06 0.99 0.64 QAAPR-cum 0.15 0.95 0.54 GlobSR-HP 0.06 0.98 0.63

SLIDE 24

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Best 3 indicators

Weighted-average over EA countries

θ = 0.2 θ = 0.3 usefulness good false usefulness good false calls alarms calls alarms GlobPC-detr 0.02 0.48 0.1 GlobPC-detr 0.09 0.63 0.14 LRN-detr 0.01 0.48 0.1 LRN-detr 0.07 0.48 0.1 HINV-yoy 0.01 0.13 0.02 QEPR-detr 0.06 0.56 0.15 θ = 0.4 θ = 0.5 usefulness good false usefulness good false calls alarms calls alarms GlobPC-detr 0.17 0.63 0.14 GlobPC-HP 0.26 0.85 0.34 GlobPC-HP 0.14 0.85 0.34 GlobPC-detr 0.24 0.63 0.14 QEPR-detr 0.14 0.69 0.23 M1toGDP-detr 0.24 0.73 0.26 θ = 0.6 θ = 0.8 usefulness good false usefulness good false calls alarms calls alarms GlobPC-HP 0.18 0.85 0.34 QAAPR-cum 0.08 0.98 0.55 QAAPR-cum 0.17 0.98 0.56 GDPR-HP 0.08 1 0.61 QAAPR-HP 0.16 0.93 0.49 GlobSR-HP 0.08 1 0.63

SLIDE 25

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Relative usefulness of money and credit

Average results over all countries

5 10 15 20 25 0.2 0.3 0.4 0.5 0.6 0.7 0.8

θ

Usefulness (%) Global Credit/GDP Global M1/GDP

SLIDE 26

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Joint indicators

Weighted-average over EA countries

usefulness % booms good false aNtS cond. diff. alt called calls alarms prob. prob. GlobPC-detr 0.19 0.63 0.60 0.09 0.14 0.57 0.42 5.4 LRR-HP GlobPC-detr 0.18 0.63 0.58 0.09 0.15 0.59 0.44 5.4 QAAPR-HP GlobPC-detr 0.18 0.63 0.57 0.08 0.13 0.63 0.47 5.2 QRPR-HP GlobPC-detr 0.17 0.63 0.63 0.14 0.23 0.51 0.35 5.5 θ = 0.4

SLIDE 27

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Main results

Useful? Yes, if balanced preferences Real vs financial? Financial Global vs domestic financial? Global Money vs credit? Credit

SLIDE 28

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Mid-2000s boom

Prediction of the best indicators

GlobPC-HP GlobM1-detr Optimal Number of Optimal Number of threshold signals threshold signals 70 (all) 7 90 (all) 85 (EA) 3 95 (EA) If last wave included in the evaluation, GlobPC best indicator Signals should be only one of the inputs for decision makers

SLIDE 29

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Receiving early warning signals for costly asset price booms in ‘real time’ is possible (aNtS in the 0.20s feasible). If central bankers’ preferences are relatively balanced, monitoring credit and money provides value added (weighted sum of errors reduced 14-25 p.p. for best indicators). Aggregate asset price cycles are correlated across countries ⇒ (standard) global financial variables are useful for ‘leaning against the wind’ type of policies.

SLIDE 30

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions

Outlook

Produce a weighted composite indicator Improve real time concept Test (other) balance sheet items of (other) financial intermediaries ⇒ factor models Build indicators which take into account cross-country differences in the financial structure Complement statistical models with structural analyses of the deep causes of the predicted risks

SLIDE 31

‘Real Time’ Early Warning for Costly Boom/Bust Cycles Alessi Detken Introduction Methodology

Boom identification Data Evaluation

Results

Best indicators Mid-2000s boom

Conclusions