Documente Academic
Documente Profesional
Documente Cultură
Objectives
Forecast ination Better understand the transmission of demand shocks through the economy Better understand the eects of macroeconomic policy (interest rates and scal decits) on the economy Introduce a new methodology for estimating the output gap, compatible with existing ones, such as the HP lter
Bjrnland and Leitemo (2009) estimated the interdependence of US monetary policy and the stock market using a SVAR with both short-run and long-run restrictions. The identication of these models has been a ourishing subject of study.
Rubio-Ramrez, Waggoner, and Zha (2010), RWZ from now on, provide a framework for global identication of SVAR's with non-linear restrictions. Binning (2013) provides a framework for combining short-run and long-run restrictions with sign restrictions
Estimating the Output Gap: a SVAR Approach
Algorithms for estimation of SVARs with short and long-run restrictions may be innecient. The absence of ecient algorithms for the estimation of SVARs with both short-run and long-run restrictions makes particularly dicult the calculation of condence intervals (since the Monte Carlo procedure involves repeating these estimations several times).
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
Framework
Comparison of results from four reduced forms for the output gap
(Benchmark): HP Filter SVAR with short-run (triangular) restrictions SVAR with short-run (non-triangular) restrictions SVAR with short-run and long-run restrictions
Decomposition of demand shocks from the residuals of the GDP equation (in order to compare these results with those of HP lter) Estimation of small model for the Brazilian economy
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
VAR Results
The VAR(2) was chosen by AIC, FPE, LR, and HQ criteria. Its residuals correlogram shows almost no residual autocorrelation. Variable Steady-state (annual %) Brazilian GDP 2.9 External GDP 2.3 Hours worked 1.6 Capacity utilization 0.1 Gross xed capital investment 4.2
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Y t = 0 +
Y t j j + Au t
j =1
All j coecients are easily identied using OLS. A is identied by the residuals' variance-covariance matrix (COV), using the identities below.
E t [COV ] = E t [Au t u t A ] Inasmuch as E t [u t u t ] = I and A is xed, E t [COV ] = AA However, A has 25 variables and COV is symmetric.
Therefore, we have only 15 equations and we need at least 10 identication restrictions in order to have 25 equations and 25 variables.
Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
LR =
j =1
AT
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Identication Restrictions
SVAR (SR and LR): We will assume three dierent hypotheses (10 identication restrictions): Brazil is a small open economy.
Therefore, Brazilian series cannot aect the external component (under the short and long run). This accounts for 6 restrictions (3 on the short run and 3 on the long run) This accounts for 2 restrictions: one on the short run and another on the long run. This accounts for 2 restrictions: one on the short run and another on the long run.
Estimating the Output Gap: a SVAR Approach
An increase in GDP may be caused by an increase in the hours worked, but not the opposite. An increase in GDP may be caused by an increase in the NUCI, but not the opposite.
Vincius Botelho, Samuel Pessoa, Silvia Matos
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Identication Restrictions
Cholesky decomposition (hours worked is the 'most endogenous' process
Hours worked Capacity utilization Gross formation of xed capital Brazilian GDP US GDP
SVAR (SR)
US GDP is exogenous Hours worked, NUCI, and gross formation of xed capital are not aected contemporaneously by the GDP (but the opposite). Hours worked and NUCI are not aected by the gross formation of xed capital (but the opposite). Brazilian GDP is not directly aected by US GDP.
Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Identication Conditions
The number of restrictions rule out only the order condition violation Only the Cholesky estimates are always identied (since they rely solely on the sign of the variance-covariance matrix, which is always positive) The rank condition found in Hamilton (1994) classify model SVAR (SR) as identied The rank condition test in RWZ has concluded our SVAR (SR and LR) is almost globally identied (the set of parameters not identiable, if it exists, has zero measure)
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
D n is the duplication matrix: D n vech() = vec vec (B 0 ) = S B B + s B (B are the parameters of B 0 ) vec (D ) = S D D + s D (D are the parameters of D )
Vincius Botelho, Samuel Pessoa, Silvia Matos
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Partial identication condition: the matrix J must have full column () vech Alternatively: rank. J = vech T T
B D
1 J = 2D + n ( B 0 )S B
1 1 D+ n [(B 0 ) (B 0 )]S D
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
E(
T t t )
=I
another in which the non-linear restrictions become linear) If M j (f (A0 , A+ )) is of rank n for 1 j n the SVAR is globally identied at the parameter point (A0 , A+ )
M j (f (A0 , A+ )) = Q j f (A0 , A+ ) I j 0jx (nj )
Q j represents the SVAR's linear and non-linear restrictions: Q j f (A0 , A+ )e j = 0 (e j is the j-th column of the identity matrix I n and f () transforms the original parameter space to
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Decomposition of Shocks
Be u t a vector for the ve underlying shocks on our economy (external, productivity, investment, capacity utilization, and hours worked) and e t the residuals obtained from our VAR. Therefore u t = A1 e t provides us a data series for each of these ve shocks. Then it is possible to compute the expected value for our system of equations if only some of these ve shocks happened. We dene the part of GDP growth caused by the shocks in hours worked and capacity utilization as output gap.
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
4 2 0
2 4
3Q1997
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
3Q1997
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
4 2 0
2 4
3Q1997
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
All Methodologies
102
4 2 0
2 4
3Q1997
3Q2002
3Q2007
3Q2012
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Outlines
Although the HP Filter indicates the economy is under its growth potential, two dierent SVAR measures indicate the economy is overheated (the ones which provide the better models for free and services prices). Both indicate a deceleration in the last quarter, but are still positive. On 4Q2002 there was a monetary policy shock, starting a tightening cycle that would end by 1Q2003. Full SVAR Gap indicates a negative demand shock on 4Q2002 (while the HP Gap increased in this period). This is a pattern for the Full SVAR and for the SVAR with only short-term restrictions.
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Outlines
There was a monterary policy shock on 2Q2008 captured by the HP Filter only on 4Q2008. However, all the 3 SVAR methodologies showed a contraction (indicating the negative demand shock) on 3Q2008. Since 2012 the HP lter indicates a deationary output gap (while there is no deation). Our SVAR indicates the presence of a residual demand pressure in the economy since then (up to 3Q2013). The Cholesky and SVAR with short-term restrictions show a slightly negative gap for the last three and two quarters, respectively.
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
HP lter yields smoother results than SVAR Maybe HP results can be predicted by the SVAR
Correlation of SVAR(t) and HP(t): 6.1% Correlation of SVAR(t-1) and HP(t): 31.0% Correlation of SVAR(t-2) and HP(t): 26.8% OLS: HP t = 0.69HP t 1 has 46.8% of adj-R 2 OLS: HP t = 0.44SVAR t 1 + 0.67HP t 1 has 54.6% of adj-R 2
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Reduced form: it = 0 it 1 + 1 E t [ t +1 ] +
j ht j +
j =0
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
j h t j +
j =0
SVAR SVAR HP Cholesky SR SR and LR 58.6% 56.4% 65.7% 59.7% 4.42 4.47 4.23 4.39 4.78 4.84 4.60 4.76 4.55 4.60 4.36 4.53
Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
j ht j +
j =0
SVAR SVAR HP Cholesky SR SR and LR 40.6% 40.5% 39.3% 43.9% 3.24 3.25 3.27 3.19 3.44 3.44 3.46 3.38 3.32 3.32 3.34 3.26
Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
j ht j +
j =0
SVAR SVAR HP Cholesky SR SR and LR 74.2% 72.4% 78.3% 75.3% 3.18% 3.25 3.01 3.14 3.55% 3.62 3.38 3.51 3.31% 3.38 3.14 3.27
Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
5 102
0.1
It is important to emphasize that great part of this uncertainty was caused by parameter revisions after the shock of the recent global recession.
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Therefore, this simple model guesses right 66.7% of the prices increases or decreases in our sample. Using the core prices (ex food and regulated prices) or the prices of services we obtain very similar results (68.8% and 58.3%, respectively). From 2008Q1 to 2013Q3 this percentage goes to 73.9%.
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 2Q2010
102
1.2
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 3Q2010
102
1.6
1.4
1.2
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 4Q2010
102
1.5
0.5
HP Filter SVAR SR
4Q2010
2Q2011
4Q2011
2Q2012
4Q2012
2Q2013
3Q2013
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 1Q2011
102
2
1.5
4Q2010
2Q2011
4Q2011
2Q2012
4Q2012
2Q2013
3Q2013
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 2Q2011
102
2
1.5
2Q2011
4Q2011
2Q2012
4Q2012
2Q2013
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 3Q2011
102
1.5
0.5
HP Filter SVAR SR
0.5
4Q2011 2Q2012 4Q2012 2Q2013
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 4Q2011
1
102
0.5
0.5
HP Filter SVAR SR
4Q2011
2Q2012
4Q2012
2Q2013
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 1Q2012
102
0.5
HP Filter SVAR SR
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 2Q2012
102
0.4
0.6
0.8
1.2
HP Filter SVAR SR
1.4
2Q2012
3Q2012
4Q2012
1Q2013
2Q2013
3Q2013
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 3Q2012
102
0.5
HP Filter
1.5
SVAR SR
3Q2012
4Q2012
1Q2013
2Q2013
3Q2013
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Stability 4Q2012
102
0.5
0.5
1
HP Filter SVAR SR
1.5
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Conclusions
The demand shocks estimated by a SVAR can signicantly aggregate information to the HP Filter. The SVAR can provide estimates for the output gap much more stable than the HP Filter does
Some identication strategies explain ination robustly better. This result is similar to previous ndings in the literature: Orphanides and van Norden (1999) and Cayen and van Norden (2002) nd that HP Filter estimates for the output gap are much more sensitive to data revisions than those from other methologies. In some cases negative HP gaps turned into positive ones, specially under economic transitions (it is taking approximately 10 quarters to converge in the last years).
VAR Estimation SVAR Identication Estimation of Output Gap from Demand Shocks Results Phillips Curve Fit Uncertainty of Estimation
Conclusions
Dierent identication restrictions can explain better some disaggregate measures of ination than others.
The SVAR estimates seem to anticipate the HP Filter ones. The SVAR estimates seem to respond quickly to monetary policy.
Phillips Curve
+4 t = 0 t 1 + 1 t 2 +(1 0 1 )E t (t t +1 )+ k1 j 1 =0 k2
j 1 H t j 1 +
j 2 =0
j 2 q t j 2 + u P t
Adj-R 2
Estimates 1.07*** (0.00) -0.27* (0.05) 0.29*** (0.01 ) 0.03*** (0.01) 82.6%
IS Curve
f1 f2
H t = c + R r t +
w 1 =1
w 1 H t w 1 +
w 2 =0
w 2H t w 2 +
f3 w 3 =0
w 3 g t w 3 + u H t
Variable
rt H t gt g t 2 H t 2
Adj-R 2
Estimates -0.11*** (0.00) 0.50*** (0.00) 0.13** (0.03) 0.11** (0.04) 0.36*** (0.00) 35.7%
w 1 H t w 1 +
w 2 =0
w 2H t w 2 +
f3
f4
w 3 g t w 3 +
w 3 =0 w 4 =0
w 4 b t w 4 + u H t
Variable
rt H t g t 1 bt H t 2
Adj-R 2
Estimates -0.11*** (0.00) 0.43*** (0.00) -0.34*** (0.00) 0.47*** (0.00) 0.46*** (0.00) 38.1%
Expectations
+4 t +4 On the long run: t t +1 = E t (t +1 )
+4 t t t E E t (t t +1 ) E t 4 (t 3 ) = (t 3 E t 4 (t 3 )) + u t +1 t +1 t t E E t (t t ) E t 1 (t ) = (t 1 E t 1 (t 1 )) + u t
Variable Model 1 Model 2 Prediction 0.45*** 0.13*** (0.00) (0.00) 2 Adj-R 50.1% 57.1% Estimation by OLS. Data from 2004Q1 to 2013Q2. Only specication 1 was used in the simulations.
Vincius Botelho, Samuel Pessoa, Silvia Matos Estimating the Output Gap: a SVAR Approach
Taylor Rule
+ Y H t j ) + u it i t = R i t 1 + (1 R )(i + (t ) + Y H t j + u it i t = 1 i t 1 + 2 i t 2 + (t )
Variable
R 1 2 Y
Alternative Model 0.87*** (0.00) -0.43*** (0.01) 0.18* (0.05) 0.33** (0.01) 75.0%
Adj-R 2
Fiscal Equilibrium
For each real interest rate (r), public debt (b), and natural growth (g) scenario there is primary surplus that stabilizes public debt according to the equation below.
r =
Therefore, this equation and the previous estimates for the real interest rate of equilibrium provide, for each level of public debt, a unique primary surplus and interest rate equilibrium. A 40% level of debt implies a primary surplus of 3% and real interest rates of 9% (US 10Yr Bonds at 4% and real growth at 2.5% yearly).
0.15
s b
+g
0.1
4 102
Fiscal Equilibrium
As a consequence, there each point in the IS Equilibrium Curve implies a stable debt level.
1.5
0.5
0.2
10
t
20
30
Baseline Alternative
0.2
0.4
10
t
20
30
Baseline Alternative
Estimating the Output Gap: a SVAR Approach
10
t
20
30
Conclusion
Fiscal policy may be an important driver to ination International interest rates matter a lot
The interest-rate spread matters An important part of the recent adjustment in the real interest rates may be due to the international ease on monetary conditions