Transcript The passthrough coefficient has fallen in developing countries.

Contractionary Currency Crashes
In Developing Countries
Mundell-Fleming Lecture, Nov. 5, 2004;
Fifth IMF Annual Research Conference.
Forthcoming, IMF Staff Papers, 2005.
IMF Institute, May 27, 2005.
Jeffrey A. Frankel
Harpel Professor
KSG, Harvard University
1
Why are devaluations so
costly?
Political costs
Economic costs
If the answer is contractionary effects
of devaluation,
what is the mechanism?
And what can a country do to minimize
them?
2
After a devaluation,
leaders lose their jobs…
Twice as often within 1 yr. (30%)
as in control group (14%) in 1960s.
– R.Cooper, 1971. (Criterion is 10% deval.)
 Updated to 1970-2003,
within 12 mo.s of devaluation (27%), vs, otherwise
(21%)
within 6 mo.s (19%), vs. as otherwise (12%).
Difference is statistically significant at 0.5% level
(Criterion is 25% deval, incl. 10% acceleration.)
3
Premier 2/3 more likely to lose his
job within 6 mos. of currency crash
1970-2003
Change in premier
No change in
premier
Total cases
Within 6 mos.
Of devaluation
Other
periods
36
812
(19.1%)
(11.6%)
153
(81.0%)
6192
(88.4%)
189
7004
4
Fin.min.s / governors are even
more likely to lose office in yr.
Updated to
Within 12 mos.
Of devaluation
1971-2003
Change in fin.min.
14
or CB governor
(58.3%)
No change in
fin.min. or CB gov.
Total cases
Other
periods
212
(35.8%)
10
(41.7%)
380
(64.1%)
24
592
5
Political price of devaluations
in developing countries
Change in Premiership and Finance Minister or CB Governor*
Comparison between Devaluation vs. Normal Times
70%
58.3%
60%
50%
35.8%
40%
29.3%
30%
20%
19.1%
20.5%
11.6%
10%
0%
Premier 6-month
(1971-2003)
Premier 12-month
(1971-2003)
Normal period
* IMF Governors
IMF Governor 12-month
(1995-1999)
Post-devaluation period
6
Citations (1972-2003)
Of Cooper (1971): 84
Of Mundell (1963): 319
Of Fleming (1962): 257
7
Possible reasons currency crashes
correlate with change of
leadership
Elections cause devaluations, rather than the
other way around (tho our timing is post-deval.)
Devaluation as proxy for unpopular IMF
austerity programs
The government is perceived as having broken a
promise
High economic costs associated with
devaluations.
8
Conditioning on IMF program does
not affect leader turnover
Dummy variable defined for initiation of IMF
program within 3 months of currency crash
does not raise frequency of leader job loss,
relative to other devaluations.
Thus conditioning on the IMF dummy variable
has no discernible effect on frequency of leader
job loss within 6 mo.s:
21.1% with an IMF program, vs.
21.9% without.
Both similar to overall rate of job turnover
following devaluations (22.8%) in full sample -still almost double the 11.6% rate in normal times.
9
Executives twice as likely to lose their jobs
if the government had said “no devaluation.”
Leader was changed
within
12 mos
unchanged Frequency
within 12 mos of changes
Of 6 promises
4
2
2/3
Of 15 with no
prior promises
5
10
1/3
9
12
(whether by
premier, fin
min, or CB)
not to devalue
Total = 21
case studies
10
Executives > twice as likely lose their jobs if
the government had said “no devaluation.”
Leader was Changed
within 6 mo.s
Unchanged Frequency
within 6 mo.s of changes
Of 6 promises
3
3
.5
Of 15 with no
prior promises
3
12
.2
6
15
(whether by
premier, fin
min, or CB)
not to devalue
Total = 21
case studies
11
Narrowing down source of
political costs of devaluation
As noted, IMF programs make no difference.
Even in those cases where no assurances had
been given over preceding month, rate of job
loss (20%) still > no devaluation cases (11.6%)
(or 33% at 12-month horizon > 20.5%) .
Thus, although “broken promise” effect is there,
political costs must also reflect economic pain.
12
Why did output fall sharply in
many recent crises?
Excessive expenditure-reduction (Sachs & Stiglitz)?
Excessive fiscal contraction
Excessive monetary contraction (high i => default)
Versus what alternative?
More expenditure-switching instead?
Devaluations were very large as it was.
Magical relaxation of external finance constraint?
Who pays?
Doesn’t change the graph
But if devaluations are contractionary,
internal balance line slopes the other way
New intersection is hard to find
May lose output regardless the policy mix.
13
Textbook model: right combination of
i & E should attain new adverse external
finance constraint, without recession
14
1990s version: External balance line
slopes down, like internal balance line
15
Possible contractionary
effects of devaluation
Some require rapid passthrough, from
exchange rate to prices of either:
imported inputs
consumer imports, & so W
All TGs, & so the distribution of income
CPI, & so M/CPI
Other effects do not need passthrough
Balance sheet effect
16
The passthrough coefficient has
fallen in developing countries.
Slow/incomplete passthrough of exchange
rate changes has long been a property of US
market & other large rich countries.
After large devaluations in Asia and other
emerging market countries from 1994
(Mexico) to 2001 (Argentina), most
observers feared correspondingly large rises
in local currency prices.
It never happened.
17
Passthrough is greatest for prices of
imports at dock, but less for retail and CPI
Source: Frankel, Parsley & Wei (2004) – effect within one year
passthrough coefficient
Exchange Rate Passthrough
to Domestic Prices
0.6
0.4
0.2
at the dock
imported good
prices
local
competitor
prices
consumer price
index
18
Passthrough for less developed
countries > for rich, historically.
Source: Frankel, Parsley & Wei (2004) – effect within one year
Passthrough and Income
passthrough coefficient
(Average 1990-2001)
(Country Grouping Based on World Bank Classification)
0.7
0.6
12 countries
36 countries
0.5
0.4
0.3
0.2
28 countries
0.1
0
Low Income
Middle Income
High Income
19
Passthrough to prices of 8
narrow imported products
Coefficient initially higher for developing
countries (.8) than rich (.3), in 1990
Downward trend in coefficient during decade:
Significant, regardless of income level
Twice as fast for developing countries as for rich
Partly explained, for rich, by rising real wages
Speed of adjustment (ECM)
Initially higher for developing countries than rich
Significant downward trend for developing c.s (only)
20
 exchange rate
 exporter's price
( s)*trend
( s) *
log [RGDP per cap: (importer)/
exporter)]
( s)* tariff levels
( s)* log distance
( s)* log[RGDP: (importer)/
(exporter)]
( s)* log real wage ($)
( s)* l.t. inflation
( s)* l.t. exchange rate variability
( s)* US Importer dummy
ECM terms omitted
Equation 1
Rich
 Dev.
0.133*** 0.365***
(0.031)
(0.045)
0.108 ***
-0.052
(0.025)
(0.042)
Equation2
Rich
 Dev.
0.310 *** 0.496 ***
(0.075)
(0.101)
0.108 ***
-0.023
(0.025)
(0.042)
Equation 8
Rich  Dev.
2.489 **
-1.607
(1.144)
(1.595)
0.082***
-0.069
(0.028)
(0.062)
-0.025 ***
(0.009)
-0.014
(0.012)
-0.022
(0.034)
-0.046 **
(0.019)
-0.050
(0.048)
-0.431
(0.265)
-0.029
(0.048)
0.016
(0.030)
0.368
(0.314)
0.049
(0.091)
0.029
(0.044)
-0.019 *
(0.010)
2.135
(1.871)
-2.262
(4.932)
-0.442**
(0.190)
0.017
(0.013)
-1.781
(1.884)
0.890
(5.112)
-0.026 **
(0.013)
21
Passthrough coefficient fell substantially in the 1990s.
For developing countries, out of total sample of 76
Passthrough
coefficient
start of
sample
(1990)
Overall
trend
sample
For import
prices of 8
narrowly
defined goods
.81
.50
-.05/
yr.
For CPI
.68
.33
-.05/
yr.
Estimates from Frankel, Parsley & Wei, NBER WP no. 11199 (2005), “Slow
Passthrough Around the World: A Recent Import to Developing Countries?”
22
Balance sheet effect is most important
of the contractionary devaluation
effects
Analytical literature on balance sheet effect
includes:
Kiyotaki & Moore (1997), Krugman (1999), Aghion, Banerjee &
Bacchetta (2000), Cespedes, Chang & Velasco (2000, 2003),
Chang & Velasco (1999), Caballero and Krishnamurty (2002),
Christiano, Gust & Roldos (2002), Dornbusch (2001), Mendoza
(2002), Calvo, Izquierdo, & Talvi (2003), Cavallo (2004),
Eichengreen & Hausmann (1999) and many others.
Empirical evidence of its output cost includes:
Cavallo, Kisselev, Perri and Roubini (2002)
Guidotti, Sturzenneger and Villar (2003)
23
Foreign-denominated debt in crises
x real devaluation => Output loss
Source: M.Cavallo, K.Kisselev, F.Perri, & N.Roubini, 2002.
Some suggested influences
on the balance sheet effect
Eichengreen & Hausmann (1999) made the
structural inability to borrow in local currencies
famous under the name “original sin.”
Is it historical fate?
Some argue that the culprit is adjustable pegs.
Short run: During the procrastination phase,
composition of debt typically shifts to $.
Medium run: Trade openness affects
vulnerability to currency crashes.
25
Procrastination phase ≡
period after reserves peak,
but before the speculative attack
Typically lasts 6-13 mo.s – Frankel & Wei (WB, 2005)
E.g., Mexico, 1994
Four ways to “gamble for resurrection”
Officials announce won’t devalue
Run down reserves
Shift composition of debt to short-term
Shift composition to $-denominated
Each ploy may gain time, but raises cost if
devaluation comes.
26
Exhaustion of Mexico’s Reserves
Up to December 1994 Crisis
Data source: IMF International Financial Statistics.
35 0 0 0. 00
30 0 0 0. 00
25 0 0 0. 00
20 0 0 0. 00
IMF PROGRAM
15 0 0 0. 00
CRISIS
10 0 0 0. 00
5 0 0 0. 00
L eve l
1995M4
1995M3
1995M2
1995M1
1994M12
1994M11
1994M10
1994M9
1994M8
1994M7
1994M6
1994M5
1994M4
1994M3
1994M2
1994M1
1993M12
1993M11
1993M10
1993M9
1993M8
1993M7
1993M6
1993M5
1993M4
1993M3
1993M2
1993M1
1992M12
0. 00
3m Mo ving A vg
27
Mexico’s shift to $-linked debt, in 1992-94
Data source: Mexican Ministry of Finance and Public Credit.
Evolution of Mexican debt according to currency denomination
80%
70%
60%
50%
40%
30%
20%
10%
tesobonos/(tesobonos+cetes)
Nov-95
Sep-95
Jul-95
May-95
Mar-95
Jan-95
Nov-94
Sep-94
Jul-94
May-94
Mar-94
Jan-94
Nov-93
Sep-93
Jul-93
May-93
Mar-93
Jan-93
Nov-92
Sep-92
Jul-92
May-92
Mar-92
Jan-92
0%
tesobonos/total domestic debt
28
Mexico’s shift to short-term debt in 1992-94
Data source: Mexican Ministry of Finance and Public Credit.
Evolution of Mexican debt according to maturity
% long term/(long term+short term) - left axis
Nov-95
Sep-95
Jul-95
May-95
Mar-95
Jan-95
Nov-94
Sep-94
Jul-94
May-94
Mar-94
Jan-94
Nov-93
0
Sep-93
90%
Jul-93
100
May-93
92%
Mar-93
200
Jan-93
94%
Nov-92
300
Sep-92
96%
Jul-92
400
May-92
98%
Mar-92
500
Jan-92
100%
average maturity in days - right axis
29
Lesson
After inflows cease, adjust to new
external balance early
while it is still possible to maintain internal
balance;
Rather than procrastinating by running
down reserves & shifting to s.t. $ debt.
Because after the crash, any combination of
high i & devaluation may be contractionary.
30
In long run, how does openness to
trade affect a country’s vulnerability?
 Some say openness raises vulnerability to sudden stops
Vulnerability to real shocks, financial shocks; loss of trade credit
Empirical support:
Milesi-Ferretti & Razin (1998, 2000)
 Others say it reduces vulnerability
Sachs (1985): explains Asian resilience, vs. Latin America
Trade is a hostage that reassures creditors
Eaton & Gersovitz (1981), Rose (2002)
If m is lower, cost of adjustment to given cut-off is higher
measured by necessary expenditure-reduction at given prices
or by necessary devaluation (with corresponding contractionary effects)
Argentina is favorite example of toxic interaction of low trade/GDP and
balance sheet effect
Guidotti, Sturzenegger & Villar, (2003), Calvo & Talvi (2004), Desai & Mitra
(2004), Cavallo (2004), and Secy. Paul O’Neill.
Empirical support:
Calvo, Izquierdo & Mejia (2003); Edwards (2004a,b)
31
Countries of currency crashes tend to
be less open to trade, especially those
with sudden stops as well
Average Opennes and Fitted Opennes by category of Crisis
0.14
0.65
0.64
0.12
0.63
0.62
0.61
0.08
0.6
0.06
Openness
Fitted Openness
0.1
Fitted Open
Open
0.59
0.58
0.04
0.57
0.02
0.56
0
0.55
No crash, no SS (1815 events)
Crash but no SS (317 events)
Crash and SS (18 events)
32
Countries that are less open to trade
are more prone to
sudden stops & currency crashes
Sudden Stops (SS1) and Currency Crises (Crash) by level of openness
350
292
300
Number of Episodes
250
200
> Mean open
< Mean open
150
127
100
52
50
34
0
SS1 (86 events)
Crashes (419 events)
33
Endogeneity of trade/GDP
Possible channels of endogeneity of openness
Via income: richer countries tend to liberalize
Part of a more general reform strategy driven by proglobalization philosophy or “Washington Consensus” forces.
Experience with crises -- the dependent variable -- may
itself cause liberalization, via an IMF program.
Feedbacks betw. trade & financial openness. Aizenman
(2003)
Solution: Use gravity as IV for trade:
“Does Openness to Trade Make Countries More Vulnerable
to Sudden Stops, Or Less? Using Gravity to Establish
Causality” – Cavallo & Frankel (2004)
34
Predicting sudden stops,
while treating endogeneity of trade
Source: Cavallo & Frankel (2004)
Trade
openness
Foreign
Debt/GDP
t-1
Liability
Dollarization
CA/GDP
observations
t-1
t-1
Ordinary probit
IV
(gravity)
-0.53
(0.259)**
-0.080
(0.217)
(0.316
(0.195)
-2.45
(0.813)**
0.196
(0.275)
0.591
(0.256)***
-4.068
(1.297)**
-7.386
(2.06)***
778
1062
35
Predicting currency crashes,
while treating endogeneity of trade
Source: Cavallo & Frankel (2004)
Ordinary probit
Trade openness
Foreign Debt/GDP t-1
Liability
Dollarizationt-1
observations
-0.37
(0.21)*
0.35
(0.18)**
0.04
(0.20)
IV
(gravity)
-0.97
(0.59)*
0.57
(0.23)**
-0.11
(0.18)
798
1196
36
S Stops & crashes rise also when it
is gravity-fitted openness that is low
Sudden Stops (SS1) and Currency Crashes (Crash) by level of fitted openness
350
313
300
250
200
> Mean fitted open
< Mean fitted open
150
106
100
67
50
19
0
SS1 (86 events)
Crashes (419 events)
Number of Episodes
37
Conclusion
 An increase in trade openness of 10 percentage points
decreases likelihood of a sudden stop (definition of
Calvo, et al.) by approximately 32%.
E.g., moving from Argentina’s trade share (.2) to Australia’s (.3)
CA/GDP is the other strongest predictor
 Increase in openness also decreases the likelihood of
currency crash, defined as 25% increase in exchange
market pressure≡ (exchange rate * reserves)
Foreign debt/GDP or Reserves/imports become the other strong
predictors
 Also some evidence that openness reduces the output cost
associated with crises
38
“Take-aways”
Leaders are twice as likely to lose office
after a devaluation.
Passthrough has declined sharply.
To minimize contractionary effects of
devaluation, countries should
avoid weakening of their balance sheets
during “procrastination phase” of debt cycle;
in the long term, open to trade.
39
The highway analogy
 Superhighways are useful, get you where you want to
go quicker. But accidents at high speeds are more likely
fatal. -- Merton
 Sudden stops:
“It’s not the speed that kills, it’s the sudden stops” – Dornbusch
  sharp disappearance of private capital inflows, reflected (esp.
at 1st) in reserve declines, & (soon) in disappearance of
previously large CA deficit - Calvo
 Superhighways: Modern financial markets get you
where you want to go fast, but accidents are bigger, and
so more care is required. -- Merton
 Is it the road or the driver? Even when multiple
countries have accidents in the same stretch of road,
their own policies are also important determinants; it’s
not just the fault of the system. – Summers
 Contagion is also a contributor to multi-car accidents.40
Highway analogy, continued
 Moral hazard -- G7/IMF bailouts reduce impact of given crisis
in the LR undermine the incentive for investors & borrowers to
be careful. Like air bags and ambulances.




But to claim that moral hazard means we should abolish the
IMF would be like claiming that drivers would be safer with a
spike in the center of the steering wheel column – Mussa.
Optimal sequence: Highway off-ramp should not dump highspeed traffic into center of a village before streets are paved,
intersections regulated, & pedestrians learn to walk on
sidewalks. So a country with a primitive domestic financial
system perhaps should not open to the full force of international
capital flows before domestic reforms & prudential regulation.
Slowing down: There may be a role for controls on capital
inflow (speed bumps & posted limits).
Reaction time: How driver reacts in short interval between
appearance of hazard and moment of impact (speculative
attack) influences outcome. Adjust, rather than procrastinate
by using up reserves & switching to short-term $ debt – JF
Routine defensive driving: Keeping high reserves and an
economy open to trade is like leaving ample following-distance.
41