Transcript Arrufat

Arnoldshain Seminar XI.
Migration, Development, and Demographic Change –
Problems, Consequences, Solutions
June 25 – 28, 2013, University of Antwerp, Belgium
ARGENTINE TERMS OF TRADE VOLATILITY
HANDLING STRUCTURAL BREAKS
AND EXPECTATION ERRORS
José Luis Arrufat
Alberto M. Díaz Cafferata
Santiago Gastelú
Instituto de Economía y Finanzas. Facultad de Ciencias Económicas
Universidad Nacional de Córdoba
2
I. Introduction
II. Literature review
III. Breaks in Argentine terms of trade and
GDP
IV. Approaches to measuring volatility
V. Empirical estimation of GDP and TOT
volatility
VI. Exploratory analysis of causality
VII. Concluding remarks
3
I
Introduction
Current prominence of volatility in
development economics:
impact on growth.
What is volatility?
How high?
How does it behave along time?
4
Argentina TOT 1810-2010. Index 1993=100.
Large & sudden changes. Extreme peaks and valleys
Four structural breaks 1882, 1913, 1945, 1975.
* Structural breaks
1909: 146
1948: 150
1922: 71
1987: 85
2000: 106
2010: 141
1839
1917
1950
*
*
*
5
Argentina TOT index, 1810-2012
High observed fluctuations.
Mean = 97.05; SD = 22.46;
CV = 0.23
High TOT volatility is a characteristic of
commodity-exporter developing countries.
TOT volatility developing countries,
3 times higher than industrial countries.
(Aizenman et al. 2011, Mendoza 1995)
Does it matter?
6
SOE “vulnerability” to external shocks and volatility.
Do TOT matter?
The answer, two temporal frameworks.
Macroeconomic perspective
o-f-all unexpected TOT shocks
→ “cause” CA shifts? Harberger-Laursen-Metzler effect.
Sign of transitory or permanent, shocks.
Models w/wo investment.
Long-term development
Effects of uncertainty: volatility
on rate & volatility of growth, distribution and poverty.
Shocks & macro
Literature on the HLM effect. “TOT matter”
Harberger, Arnold C. 1950 "Currency Depreciation, Income, and the
Balance of Trade." JPE . (58).
Laursen Sven & Metzler Lloyd A., 1950 “Flexible Exchange Rates and
the Theory of Employment.” Review of Ec & Statistics, (32) 3 .
Obstfeld Maurice, 1982 "Aggregate Spending and the Terms of Trade:
Is There a Laursen-Metzler Effect?“ Quarterly J Economics (97) 2.
1950. The “HLM effect”: positive relationship between TOT shocks
and the CA. Income rises and C rises less.
1981. Obstfeld, Sachs, Svenson & Razin: the CA improves only if the
TOT shock is transitory (otherwise there is not a smoothing role
for the CA)
1990. Mendoza & Otto: there is an HLM effect with both transitory
and permanent shocks)
Barone Sergio V. , Ricardo L. Descalzi, Alberto M. Díaz Cafferata (2009)
“Terms of Trade Shocks and Current Account Adjustment”. XXIV Jornadas
Anuales de Economía. BCU
18 LACs, in 1976-2007. Data: BM y FMI. Model FGLS
TOT matter for the CA:
Estimated coefficient for the permanent TOT shock
significantly different from zero, and positive sign.
9
Our focus: volatility & growth
• Perceived costs of high and irregular
fluctuations along time.
• Attention shifts from SR impact of shocks
towards effects of volatility on growth
Problem: shared intuition, but not an
agreed empirical measure
of TOT volatility
in quantitative estimations,
10
Methodological issues
To quantify magnitude, and effects
What is formally “volatility”?
How high?
It depends on how you measure it.
11
Volatility is not an inequivocal
concept
Several definitions depict
different temporal profiles!
How do different methods compare?
What criteria to choose to depict
stylized facts and estimate
association?
Compare below patterns with three
methods
12
FIGURE V.1. TOT VOLATILITY
Detrended cum breaks (BLUE) Detrened cum breaks +
decycled (RED, lower volatility!)
*
Sample:
1840 to
2012.
* 1839
* 1917
* 1951
30-year
rolling
sample SD.
13
FIGURE V.3 IS VOLATILITY REALLY
OVERESTIMATED? (Friedman-Cavallo)
14
II
Literature review.
Empirical estimation of
volatility and structural breaks
15
Magnitude and impacts of volatility
Broad range of topics
How high is volatility (empirical estimation),
Measure uncertainty (methods to portray)
Causes (specialization & markets)
Channels and effects on GDP growth and
distribution
Weaknesses of developing countries.
Policy recommendations
16
“Prominence of volatility”
Aizenman and Pinto 2005, p2. Volatility has a central
place in development economics.
What has catapulted volatility
into this prominence?
Negative impacts on trend growth,
Effects on saving & investment, and links between
technological progress and the capital stock
Understanding the nature of volatility, anticipating
and managing its consequences, is of considerable
interest to policymakers in developing countries.
17
Volatility matters for growth
Mendoza 1997 “TOT are typically a significant and
robust determinant of economic growth”. Model
savings under uncertainty.
Aizenman and Pinto (2005) large growth cost
especially for developing countries.
Wolf (2005) a growing body of research suggests
that higher volatility is causally associated with
lower growth.
Loayza and Raddatz (2007) 25% of the variation in
growth volatility.
Koren and Tenreyro (2007)
18
WOW HIGH IS TOT VOLATILITY?
How does it evolves?
Our focus, tackle
EMPIRICAL ESTIMATION OF VOLATILITY
Note it
INVOLVES METHODOLOGICAL ISSUES
We adopt an
EXPECTATIONS-BASED PERSPECTIVE
19
Metodological issues
in the empirical estimation of volatility
Critique to the purely statistical approach: distinguish
volatility from variability (Dehn, Wolf)
Filter perceived trend (problem: choice of detrending
method; Canova, Bee de Dagum)
Time varying volatility (use of rolling window; Ramey and
Ramey; Arrufat et al))
Large jumps vs smooth trends (finding breaks; Ocampo &
Parra, Bai-Perron)
Filter perceived regular cycles (Bolch & Huang; determine
cycles included)
Deal with temporal anachronism (agent´s dataset and
knowledge of DGP; Cavallo, Friedman)
20
(a)
Empirical estimation of volatility:
the statistical approach
Perry (2009) SD of cyclical component from the trend
Aizenman et al. (2011), “Adjustment patterns to commodity
terms of trade shocks: the role of exchange rate and
international reserves policies”, NBER WP 17692.
Larrain & Parro (2006), “Chile menos volátil”, Instituto de
Economía, U. Católica de Chile.
This method depict observed fluctuations. Does it
measure volatility?
21
(b)Expectations-based volatility
Decompose observed data on
predictable (regular part)
and unpredictable (uncertainty) components.
• Kim (2007)“Openness, external risk, & volatility: implications for the
compensation hypothesis”, Cambridge UP
• Wolf “Volatility: Definitions and Consequences”, In Aizenman & Pinto
Managing Volatility and Crises.
• Dehn (2000), "Commodity price uncertainty in developing countries”,
World Bank (Series 2426)
• Baxter (2000), “International trade and business cycles”, in Grossman and
Rogoff .
22
Abrupt changes in TOT.
Several authors note the presence of breaks
• Ocampo and Parra-Lancourt (2010b) barter TOT for
commodities vs manufactures improved declined since the
early 20th century with a stepwise deterioration in 1920 and
1979.
• Cuddington and Urzúa (1989) the real commodity price
index drops abruptly in 1921; there is no evidence of an
ongoing secular deterioration.
• Bleaney & Greenaway (1993)
•
23
Reasons to identify breaks.
Empirical research has found significant episodes
of large jumps in TOT.
Portraying stylized facts.
Improve analysis identifying changes in DGP and
structural differences in regimes between breaks.
Detrending method in the presence of breaks,
Are there breaks in Argentine
TOT and GDP?
24
III
Structural breaks
in Argentina.
TOT and GDP
First step in the estimations.
Breakpoints: Bai-Perron test.
Different regimes.
25
Reasons to test for breaks
Avoid erroneous characterizations of the nature of
the series. (e.g. mistakenly arriving at the
conclusion that a series is stationary in differences
when it is in fact trend stationary but with a
segmented trend).
Severe pitfalls may arise in the process to isolate
cycles. An important outlying observation may
lead the researcher to identify a bogus cycle the
period of which is excessively lengthy.
26
Bai – Perron test for m breaks
A segmented trend for the first subperiod :T0 to T1-1
If there is one break, the second subperiod runs between T1 and T2-1
…
If there are m breaks, the expression for the m+1 regime is:
All summations run from 0 to m
27
ESTIMATION OF TOT BREAKS
Notice that breaks
occur in
1839, 1917, and
1951.
Dummies that were
not significantly
different from zero
were dropped to
ensure the most
parsimonious model.
28
• CAMBIAR LA FILMINA POR OTRA
• CON ERRORES ESTÁNDAR ROBUSTOS
• DADA LA PRESENCIA DE ALTA AUTO
CORRELACIÓN
29
The Bayesian Information Criterion
• There is a trade-off between goodness of fit (the
residual sum of squares RSS) measured on the
right axis , which is monotonically nonincreasing with the number of breaks, and
parsimony.
• The Bayesian Information Criterion (BIC) takes
into account both goodness of fit and parsimony.
• The minimum BIC in the TOT is for three breaks.
There are four breaks in the GDP series.
30
Argentina 1810-2012. TOT Break-points
-80
7.5
-90
7
-100
6.5
-110
6
-120
5.5
-130
5
-140
4.5
-150
4
-160
3.5
-170
3
-180
RSS
BIC
Figure 3.1
BIC and Residual Sum of Squares - TOT
2.5
0
1
2
3
4
5
Number of Breaks
BIC
RSS
Four TOT regimes. Break years: 1839, 1917, 1951
31
TOT Breakpoints
5.2
2.9
2.7
2.5
2.3
2.1
1.9
1.7
1.5
1.3
1.1
0.9
0.7
0.5
0.3
0.1
-0.1
-0.3
5
4.8
4.6
4.2
4
3.8
3.6
3.4
3.2
Year
Estimated Values
TOT
Residuals
2010
2000
1990
1980
1970
1960
1950
1940
1930
1920
1910
1900
1890
1880
1870
1860
1850
1840
1830
1820
3
1810
TOT (ln)
4.4
Residuals
Figure 3.2
Argentina, TOT trends and breaks
32
ESTIMATION OF GDP BREAKS
Notice that breaks occur in
1882, 1913, 1945, and 1975.
Dummies not significantly
different from zero were
dropped to ensure a
parsimonious model.
33
Argentina 1810-2012. GDP Break-points
Figure 3.3
BIC and Residual Sum of Squares - GDP
50
14
-50
12
BIC
8
-250
6
-350
RSS
10
-150
4
-450
2
-550
0
0
1
2
3
4
5
Number of Breaks
BIC
RSS
Five GDP regimes. Break years: 1882, 1913, 1945, 1975
34
GDP Breakpoints
Figure 3.4
Argentina, GDP trends and breaks
14
0.95
12
0.75
11
0.65
10
0.55
9
0.45
8
0.35
0.25
7
0.15
6
0.05
Year
GDP
Estimated Values
Residuals
2020
2010
2000
1990
1980
1970
1960
1950
1940
1930
1920
1910
1900
1890
1880
1870
-0.25
1860
3
1850
-0.15
1840
4
1830
-0.05
1820
5
Residuals
0.85
1810
GDP (ln)
13
35
Dating of breakpoints for logTOT and logGDP, and
other sources of epochs
20-year
subperiods
1810- 1830- 1850- 1870- 1900- 1920- 1940- 1950- 1970- 19901829
logTOT
logGDP
CortésConde
growth*
Díaz C.
long-run
growth**
Epochs
1849
1839
1869
1899
1919
1939
1917
1882 1913
1875
1884
Argentina: accelerating LR
growth
1949
1969
1989
2012
1951
1945 1975
1980
Interwar
Globalization
The Baring Crisis was in 1890. Cfr. Cortés Conde, la economía argentina en el largo plazo.
Díaz Cafferata “Inercia estructural del crecimiento”: Academia Nac Cs Económicas, after
Max trend growth decline secularly with trade openness until the 1980s
36
Detrending cum breaks
• Both TOT and GDP exhibit breaks that shall be taken into
account in the decycling.
• The break-points point out a transformation or transition
zones.
• Years of breaks estimated make sense: portray three great
economic history epochs of Argentine: first one the open,
golden XIXth Century high growth, like other land
abundant countries, until the first World War (Baring
crisis 1890) with four decades of transition between 1875
and 2014. A second one is the interwar period of relatively
low trade openness. The third one the last half-century of
globalization.
37
IV
Measuring volatility with
alternative methods.
A discussion.
38
How much “volatility”?
Volatility analytical interpretation:
associated with uncertainty.
Proxy in standard empirical practice,
through two approaches.
39
Measuring volatility. Our taxonomy of
approaches to volatility.
Different definitions of volatility in the literature,
can be grouped in two main empirical approaches
(a) Statistical
SD of a time series
SD of detrended residuals
(b)Expectations-based
b.1. Detrending + Decycling
b.2. Forecasting errors
40
(a) Statistical approach
Original Series. Statistical approach.
Descriptive measures of dispersion.
SD of a time series
SD of detrended residuals.
Single value or rolling sample. Volatility measured by the
SD: may be a single global value of the period, or a rolling
window which provides a temporal profile.
Measures fluctuations of observed series
With or without filtering: alternatives. HP Filter /
Polynomial detrending.
41
(b)Expectations-based
approach
Identification ex-post
of uncertainty ex-ante of economic agents.
Expectation based,
detecting breaks and removing regularities:
Detrended residuals + decycling
b.1) Detrending + decycling
b.2) Forecasting errors (the best you can do)
42
First expectations adjustment: detrending
“Much care has to be dedicated to the detrending
procedure since a wrong specification can bias severely
the subsequent analysis” (Bee Dagum)
“Different detrending procedures are alternative windows
which look at the series from different perspectives”
(Canova)
• Bee Dagum et al. (2006), “A critical investigation on
detrending procedures for non-linear processes”, J. of
Macroeconomics (vol 28).
• Kauermann et al. (2011), "Filtering time series with
penalized splines", Studies in Nonlinear Dynamics and
Econometrics, (vol 15(2))
• Canova (1998), “Detrending and business cycle facts: A
user’s guide”, Journal of Monetary Economics (vol 41).
43
b.1) Detrending + decycling
Distinction between variability and volatility.
Implicit assumptions about decomposition of data:
knowledge and ignorance: agents perceive regular but
not irregular movements of economic time series.
Unexpected portion, the unpredictable component of
variability.
•
SD of Hodrick Prescott (HP) filtered residuals
•
SD of polynomial detrending residuals
44
Decycling: Fourier decomposition
Bolch and Huang
Periodic components of a time series
101

Z at    i cos 2 i t
i 0
T

101
i 0
Zat  Yat  Yˆat
Y1t  log TOTt 

  i sin 2 i t
Y2t  log  GDPt 
T

45
Choice of the best method
The best empirical method should be determined
by the modeling of economic agents´choices and
the channels of effects on activity and
distribution.
But there is not a canonical model to take as a
reference.
For empirical measuring TOT volatility:
Volatility is associated with uncertainty.
TOT fluctuations are exogenous in the small
open economy (SOE)
46
(b2) Expectations-based
approach
Previous methods suffer a
temporal inconsistency
Is tackled through:
b.2) Forecasting errors (the best you can do)
Out of sample estimation and errors
47
V
Empirical identification of GDP and TOT
volatility.
Temporal volatility profiles for Argentina:
stylized facts.
Cathegories to compare: amplitude, breaks,
asymmetry, thresholds …
48
Modeling and estimating uncertainty (3)
Original Series
HP Filter / Polynomial
Detrending
Detrended Residuals
Fourier
Decomposition
Detrended + Decycled Residuals
Standard Deviation
Volatility
49
Data and methods
Data: Argentina TOT and GDP logged from index
1993=100. 1810 – 2012 (Ferreres & INDEC)
Throughout the exercises all series are logged
Detrending
Cubic polynomial detrending
HP filter detrending (lambda = 100)
Detrending + decycling
Fourier decomposition
Time inconsistency and the best you can do.
Out of sample forecasting
50
Statistical measures for Argentina: single SD
and 30 previous years rolling window RW
• TOT, four comparative graphs
• Figure V.1. Statistical approach. A single SD of logged TOT
and GDP for the whole period.
• Figure V.2. Statistical approach and expectations approach:
detrended with breaks. SD of 30 previous years RW
represents observed data and perception of the data
generating process DGP.
• Figure V.3. Detrended with breaks + decycling
• Figure V.4. The best you can do
• GDP only detrending
• Figure V.3. Expectations approach
Dependent Variable: LOGGDP
Method: Least Squares
• Sample: 1810 2012 Included observations: 203
•
Variable
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
C
T
T2
T3
Coefficient
6.264786
0.011277
0.000354
-1.26E-06
Std. Error
t-Statistic
Prob.
0.045981
0.001947
2.22E-05
7.14E-08
136.2466
5.791189
15.99986
-17.70645
0.0000
0.0000
0.0000
0.0000
R-squared
0.994496
Mean dependent var
Adjusted R-squared
0.994413
S.D. dependent var
9.649786
2.150790
S.E. of regression 0.160767
Akaike info criterion
Sum squared resid
5.143348
Schwarz criterion
Log likelihood
85.01890
Hannan-Quinn criter.
F-statistic
11984.95
Durbin-Watson stat
Prob(F-statistic) 0.000000
-0.798216
-0.732931
-0.771804
0.122200
TOT VOLATILITIES – 6 APPROACHES
LOGTOT
v_logtot_cubic_nb
V_LOGTOT_
DET
V_LOGT
OT_DET_
DEC
SEP_1
SEP_2
Mean
4.6066
0.1621
0.1336
0.1064
0.1130
0.1531
Median
4.6052
0.1715
0.1295
0.1085
0.1006
0.1101
Maximum
5.0136
0.2260
0.1902
0.1486
0.2342
0.4442
Minimum
4.1750
0.0949
0.0909
0.0743
0.0466
0.0503
Std. Dev.
0.1884
0.0324
0.0250
0.0191
0.0384
0.0860
-0.0211
-0.3392
0.7113
0.1564
1.0196
1.2809
2.4773
1.9555
2.8026
2.0094
3.3873
3.7038
Skewness
Kurtosis
53
FIGURE V.1. TOT VOLATILITY
Detrended cum breaks (BLUE) Detrened cum breaks +
decycled (RED, lower volatility!)
Sample:
1840 to
2012.
30-year
rolling
sample SD.
54
FIGURE V.3 IS VOLATILITY REALLY
OVERESTIMATED? (Friedman-Cavallo)
55
Comments on TOT decomposition
The most important cycle:
• period: … years
• frequency: observed …times in 203 years
• acounts for ….% of the total sum of squares.
The first … most important cycles account for …% of the
total sum of squares
56
VI
Exploratory analysis of
causality
57
TOT volatility and economic activity
• How much or in what ways is the ESTIMATED impact of
TOT volatility influenced by the approach in measuring
volatility?
• Identify lags, other influences: control variables usually
are: real exchange rate, trade and financial openness,
labor markets, fiscal deficit, exports to external debt ratio,
etc.
• Unique episodes (Keynes) the default
58
Impact of TOT volatility
• What dimensions of activity are affected by TOT
volatility?
• Investment and growth. Consumption and
macroeconomic fluctuations …
• Previous results in the literature mixed sometimes small
or non-significant
The characteristics of volatility:
• amplitude of fluctuations, shocks permanent or transitory,
presence of breaks, symetry, thresholds, …
The structural environment
• Institutions, governance, …
• Government response
59
Testing Granger Causality
•
60
Testing Granger Causality
TOT volatility (definition 1) and contemporaneous GDP growth
TOT volatility (definition 2) and contemporaneous GDP growth
Lag
(a) TOT volatility causes growth
(b) GDP growth causes TOT
volatility
(c) TOT volatility causes growth
(d) StError causes growth
1
0.853
0.476
0.690
0.572
0.521
0.517
0.581
0.380
0.464
0.603
0.698
0.720
0.769
0.812
0.859
0.865
0.907
0.978
0.923
0.869
0.933
0.954
0.102
0.193
0.344
0.442
0.248
0.321
0.408
0.398
0.439
0.462
0.532
0.588
0.608
0.403
0.376
0.358
0.491
0.588
0.394
0.350
0.242
0.302
0.650
0.791
0.904
0.969
0.801
0.884
0.881
0.742
0.443
0.378
0.421
0.475
0.128
0.175
0.245
0.282
0.210
0.178
0.236
0.126
0.156
0.172
0.246
0.542
0.496
0.436
0.550
0.533
0.641
0.658
0.765
0.612
0.602
0.623
0.683
0.750
0.755
0.824
0.726
0.779
0.714
0.803
0.657
0.646
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
61
VII
Concluding remarks
62
About volatility in Argentine time series
• Features that matter: Check for breaks. Variability and
volatility. Rolling window.
• Need further work to define the theoretical interpretation
of different algorithms
• Need to check the long-run data.
• Current TOT volatility is not high for historical standard,
still relatively high for international standards
• Patterns for Argentina show coincidently rising volatility
in a growing scenario in the late XIXth Century, a “U”
until the 1950s and a reduced volatility in the last six
decades: Why? Will it last? A research topic.
63
Thank You
Arnoldshain Seminar XI.
Migration, Development, and Demographic Change –
Problems, Consequences, Solutions
June 25 – 28, 2013, University of Antwerp, Belgium
José Luis Arrufat, Alberto M. Díaz Cafferata, Santiago Gastelú
ARGENTINE TERMS OF TRADE VOLATILITY
HANDLING STRUCTURAL BREAKS
AND EXPECTATION ERRORS
Instituto de Economía y Finanzas. Facultad de Ciencias Económicas
Universidad Nacional de Córdoba
65
Question: does volatility influence
development?
Usual perception that TOT volatility matters for
growth, but evidence is mixed. Why?
Jorrat
Find effect of TOTV smaller than domestic shocks
Cerro & Meloni; Lagos & Llach
Bour et al aaep 2011 significant effects of TOT
66
Impact of volatility
Counter-intuitive small effects found.
The reason: may be there is not such effects, or the
association is not correctly formulated, or volatility is
not adequately measured.
Breakpoins and different regimes?
Thresholds and non-linearities? Asymmetries? Lags?
Other variables?
Which is the correct experiment?
Multiple determinants: the currency regime, real
exchange rate, degree of commercial and financial
openness, trade taxes, fiscal solvency, institutions,
exports to debt ratio, …
67
Comments on GDP decomposition (1)
The most important cycle:
Another important cycle:
Cycles should not be taken mechanically
Their economic relevance has not a clear interpretation
For analytical purposes we have kept ,,, cycles
• For TOT we extracted approximately …% of variability
• For GDP we extracted approximately …% of variability
The results obtained proved to be robust to different
choices of end points
68
References
Aizenman
Edwards
Riera-Crichton (2011)
Larrain, Parro ?? (2006)
Mendoza (1994)
Kim (2007)
Wolf (2004)
Dehn (2000)
Baxter (2000)