Transcript Prezentace aplikace PowerPoint

Early-Warning Indicators
International Macroeconomics
April 7, 2015
Outline
1) Bussière (2007)
2) Jan Babecký et al. (2011)
3) Causality x Correlation Discussion
Bussière (2007)
Which types of crisis are reflected by this class
of models?
Characteristics, examples?
Bussière (2007)
Currency crisis in emerging countries in
past decades
 Examples

◦ E.g. Russia 1998
 floating peg→ free float
 Inflation 84 % in 1998
 band of 5.3 to 7.1 (1997) → 21 RUR/USD (21.9.98)
Bussière (2007)
What were the two basic assumptions currency
crisis models before his own model?
Bussière (2007)
Assumptions of currency crisis models
before:
 Static models
◦ Lagged dependent variable not included

All explanatory variables with same lag
Bussière (2007)
Are they problematic?
Why do we include lagged dependent variable
(occurrence of crisis)?
In which situations does it have positive or
negative sign? In short and long run?
Bussière (2007)
Static feature
 State dependence problem
Short run effects of crisis
 Large outflow of capital → negative sign
(capital is out, no other outflow expected…)
 Indicator of vulnerability → other outflows
→ positive sign
Long run effects of crisis
 Regarded as stronger and more reliable by
investors (motivation for reforms) →  Regarded as vulnerable by investors → +
Bussière, M. (2007)
Same lag of explanatory variables
 Neglects long x short run effects of
explanatory variables
Problem of idiosyncrasy in case of lagged
variables?
How do they solve it?
Bussière (2007)
Methodology
Dataset
 27 countries
 7 years starting 1994, monthly
observations
Dependent variable
 CI (currency crisis index) as (0;1)
Bussière (2007)
Crisis index II
 Contemporaneous index
 Forward index
Models
 Static model
 Dynamic model
 Static model with fixed effects
 Dynamic model with fixed effects
Bussière (2007)



Explanatory variables
Debt ratio
Current account
Government budget balance
◦ Related to 1st generation crisis. HOW?




Pre-crisis real exchange rate overappreciation
“Lending boom” measure
Real growth rate
Contagion across emerging markets
Bussière, M. (2007)
Anything interesting looking at the descriptive
statistics of the explanatory variables?
Bussière (2007)
Average x pre-crisis year
 Exchange rate is over-appreciated with
respect to trend
 Current account and trade balances in
higher deficit
 The government deficit not necessarily
larger → 1st generation models?
 Excessive bank lending before the crisis
Pre-crisis year x 2 years before crisis
 Problems visible even two years before…
Bussière (2007)
What were the main results of the static
model (with and without FE)?
Is there any significant difference between the
model with and without FE?
Bussière (2007)
Static model without FE
 Increase in probability
◦
◦
◦
◦

Deviation of ER from trend
Faster “lending boom”,
High debt to reserves ratios
Contagion from other financially integrated countries
Decrease in probability
◦ Strong GDP growth
Static model with FE
 Short-term debt twice higher coefficient
 CA surplus significant
Bussière (2007)
Any difference between dynamic model and
static one?
State dependence significant?
Bussière (2007)
Differences between static and dynamic model
 Lending boom, financial contagion insignificant
 CA significant
State dependence
 5,6 month lags significant with FE
Lag analysis
 Short run impact
◦
E.g. debt-reserve ration
Long and short run impact

◦
Contagion
Short to medium impact

◦
Lending boom
Bussière (2007)
Is the “lag analysis” relevant for policy
decisions?
What do you think about the predictive power
of his model?
Bussière (2007)
Lag analysis
 “Short term” variable → immediate
action
 “Long run” variable → time for less
drastic policies
Predictive power (R2)
 Contemporaneous index:0.045 - 0.122
 Forward index: 0.167 - 0.288
Bussière (2007)
Any other comment, critical remark to the
paper?
Babecký et al. (2011)
How would you summarize the aim of the
authors? Their contribution?
Babecký et al. (2011)
Contribution
 Only developed countries (40 c., 40
years) and various types of crises
 Construction own early warning system
◦ Discrete model
◦ Continuous model
Any (dis)advantage of continuous model?
Babecký et al. (2011)
Authors use rich set of econometric and
statistical techniques compared to Bussière
(2007).
Do you find it contributive? And why?
Which weakness of Bussière (2007) are
overcome by Babecký et al. (2011)?
Babecký et al. (2011)
Used techniques
 VAR → abandon fixed horizon (2 years)
 BMA → refines the selection of leading
indicators
 Dynamic panel estimation techniques →
allow cross-country heterogeneity
 Cluster analysis
 Bussière’s approach very limited
compared to Babecký et al.
Babecký et al. (2011)
What were the main results?
Which factors as warning indicators are the
most important?
Babecký et al. (2011)
Results
 Most robust internal factors: housing
prices (all clusters)
 Other key global factors: world credit
growth, world output growth
 Core countries: external debt
 Non-core: capital market, money and
credit indicators
Papers and Impressions
Have you found the papers interesting?
Have you found any other weakness not
mentioned so far?
Causality x Correlation
What do you think about the EWI approach?
Can it help to warn us about the coming
crises?
Do they reveal causality or just sophisticated
correlations? Why is it important?
Causality x Correlation
Why not causality?
 EWI – huge amount of factors (nearly data-mining
approach)
 Various crises, different mechanisms
 Lucas critique – without structural models (based
and derived from theories) we measure only
correlations which are time variant
Suspicion of prediction efficiency
 Uncertainty (Knight, 1921) → world
unpredictable → focus on underlying causal
chains and explanatory theories
◦ Critical realism (Maki, Lawson,…), Post-keynesians
(Davidson),…
Other Comments
Questions, remarks, etc.???