Extracting GDP signals from the monthly indicator of economic - - PowerPoint PPT Presentation

SLIDE 1

B A N C O C E N T R A B A N C O C E N T R A L D E C H I L D E C H I L E L E 26-29 September 2010

Extracting GDP signals from the monthly indicator of economic activity: Evidence from Chilean real-time data

Michael Pedersen Central Bank of Chile 6th colloquium on modern tools for business cycle analysis

SLIDE 2

2

Four times every year, at the time of the MP meeting the quarterly IMACEC is available but not the first GDP release

IMACEC GDP MPM

Source: Central Bank of Chile.

05-Mar-08 (Jan-08) 07-Apr-08 (Feb-08) 05-May-08 (Mar-08) 05-Jun-08 (Apr-08) 07-Jul-08 (May-08) 05-Aug-08 (Jun-08) 04-Sep-08 (Jul-08) 06-Oct-08 (Aug-08) 05-Nov-08 (Sep-08) 05-Dec-08 (Oct-08) 05-Jan-09 (Nov-08) 05-Feb-09 (Dec-08) 13-Mar-08 10-Apr-08 10-Jun-08 10-Jul-08 14-Aug-08 04-Sep-08 09-Oct-08 13-Nov-08 11-Dec-08 08-Jan-09 12-Feb-09 12-Mar-09 23-Mar-09 (Q4-08) 24-Nov-08 (Q3-08) 26-May-08 (Q1-08) 25-Aug-08 (Q2-08) 08-May-08

Q1.08 Q2.08 Q3.08 Q4.08 Q1.09

SLIDE 3

3

The presentation

• 1. Description of data and revisions
• 2. Signals in quarterly IMACEC vs. first GDP

release

• 3. Signals in monthly IMACEC data
• 4. Conclusions

SLIDE 4

4

The monthly indicator of economic activity

Source: Central Bank of Chile.

• IMACEC: Indicador Mensual de AC

ACtividad EC EConómica.

• Classification: Synthetic indicator with an accounting approach
• Published (aggregate form) by the Central Bank of Chile since

1987.

• Revised on a quarterly basis in order to fit published GDP data.
• 10
• 5

5 10 15 87 89 91 93 95 97 99 01 03 05 07 09

IM IMAC ACEC: Fi EC: First release rst release

(annual change, percentage)

SLIDE 5

5

IMACEC and GDP of the current year are revised every quarter and once a year major revisions take place

Final GDP data: Has been subject to at least two annual revisions. Period analyzed: 1987-2007.

Q1 Q2 Q3 Q4 Year 2009 Type of data Estimation 18 May 09 18 Aug. 09 18 Nov. 09 18 Mar. 10 18 Mar. 10 Prelimanary Revision Q1 18 Aug. 09 18 Nov. 09 18 Mar. 10 Revision Q2 18 Nov. 09 18 Mar. 10 Revision Q3 18 Mar. 10 Annual revision 1 18 Mar. 11 18 Mar. 11 18 Mar. 11 18 Mar. 11 18 Mar. 11 Provisory Annual revision 2 19 Mar. 12 19 Mar. 12 19 Mar. 12 19 Mar. 12 19 Mar. 12 Revised

Calend Calendar of GDP revisions, 200 ar of GDP revisions, 2009

Source: Boletín Mensual 82(973), March 2009, Central Bank of Chile.

SLIDE 6

6

Only GDP data of the current year are revised quarterly

A section of the re A section of the real-time GDP databas al-time GDP database

(constant prices, million pesos)

Note: The rows contain the published data of the quarter indicated in the first column. The data were published in the issue of the “Boletín Mensual” indicated in the first row. Marked numbers indicate that the data has been revised. The data published in 2006 are in constant 1986- prices; while those published since March 2007 are in 2003-prices. Source: Central Bank of Chile.

2006 2007 2008 May Aug. Nov. Mar. May Aug. Nov. Mar. I/06 11.464.700 11.488.754 11.478.428 14.767.446 14.767.446 14.767.446 14.767.446 14.668.483 II/06 11.668.947 11.657.484 15.128.813 15.128.813 15.128.813 15.128.813 15.189.667 III/06 11.092.285 14.462.037 14.462.037 14.462.037 14.462.037 14.537.983 IV/06 15.230.521 15.230.521 15.230.521 15.230.521 15.352.820 I/07 15.629.265 15.617.348 15.634.239 15.580.886 II/07 16.045.098 16.071.464 16.130.990 III/07 15.057.612 15.110.208 IV/07 15.971.384

SLIDE 7

7

IMACEC data are revised with the publication of the quarterly GDP

A section of the real-time A section of the real-time IMACEC IMACEC database database

(constant prices, index)

Note: The rows contain the published data of the month indicated in the first column. The data were published in the issue of “Avance Estadístico del Boletín Mensual” indicated in the first row. Marked numbers indicate that the data has been revised. Until February 2007 the base year is 1996 and from March 2007 it is 2003. Source: Central Bank of Chile.

2006 2007 2008 Mar. Apr . May Jun. Jul. Aug . Sep. Oct. Nov. Dec. Jan. Feb. Mar . Apr. May Jun. Jul. Aug. Sep. Oct. Nov . Dec. Jan. Feb. Mar. 1/06 147.7 147.7 147.2 147.2 147.2 147.1 147.1 147.1 147.0 147.0 147.0 147.0 114.5 114.5 114.5 114.5 114.5 114.5 114.5 114.5 114.5 114.5 114.5 114.5 112.2 2/06 136.9 136.8 136.8 136.8 137.3 137.3 137.3 137.1 137.1 137.1 137.1 107.0 107.0 107.0 107.0 107.0 107.0 107.0 107.0 107.0 107.0 107.0 107.0 106.5 3/06 156.5 156.5 156.5 157.0 157.0 157.0 156.9 156.9 156.9 156.9 124.9 124.9 124.9 124.9 124.9 124.9 124.9 124.9 124.9 124.9 124.9 124.9 125.4 4/06 147.9 147.9 147.8 147.8 147.8 147.8 147.8 147.8 147.8 118.0 118.0 118.0 118.0 118.0 118.0 118.0 118.0 118.0 118.0 118.0 118.0 119.9 5/06 153.0 153.2 153.2 153.2 153.1 153.1 153.1 153.1 121.7 121.7 121.7 121.7 121.7 121.7 121.7 121.7 121.7 121.7 121.7 121.7 122.2 6/06 147.3 147.3 147.3 146.9 146.9 146.9 146.9 115.2 115.2 115.2 115.2 115.2 115.2 115.2 115.2 115.2 115.2 115.2 115.2 114.1 7/06 142.4 142.4 141.6 141.6 141.6 141.6 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 114.6 8/06 142.6 142.3 142.3 142.3 142.3 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.5 113.8 9/06 142.2 142.2 142.2 142.2 112.2 112.2 112.2 112.2 112.2 112.2 112.2 112.2 112.2 112.2 112.2 112.2 112.6 10/06 150.2 150.2 150.2 118.3 118.3 118.3 118.3 118.3 118.3 118.3 118.3 118.3 118.3 118.3 118.3 118.5 11/06 148.3 148.3 117.0 117.0 117.0 117.0 117.0 117.0 117.0 117.0 117.0 117.0 117.0 117.0 117.7 12/06 155.4 122.0 122.0 122.0 122.0 122.0 122.0 122.0 122.0 122.0 122.0 122.0 122.0 124.0 1/07 120.6 120.6 120.7 120.7 120.7 120.7 120.7 120.7 120.8 120.8 120.8 120.8 118.7 2/07 113.1 112.7 112.7 112.7 112.6 112.6 112.6 112.8 112.8 112.8 112.8 112.6 3/07 133.2 133.2 133.2 133.1 133.1 133.1 133.1 133.1 133.1 133.1 134.1 4/07 125.8 125.8 126.2 126.2 126.2 126.3 126.3 126.3 126.3 128.6 5/07 127.6 127.9 127.9 127.9 128.1 128.1 128.1 128.1 128.6 6/07 122.3 122.3 122.3 122.6 122.6 122.6 122.6 121.2 7/07 118.1 118.1 118.9 118.9 118.9 118.9 119.9 8/07 118.6 118.7 118.7 118.7 118.7 118.8 9/07 115.6 115.6 115.6 115.6 115.7 10/07 123.4 123.4 123.4 123.5 11/07 122.4 122.4 122.9 12/07 126.6 128.2

SLIDE 8

8

Annual growth rates are calculated with same vintage data

, 4 , 4,

100* 1

t t t

x x x

τ τ τ −

⎛ ⎞ Δ = − ⎜ ⎟ ⎜ ⎟ ⎝ ⎠

GDP: IMACEC: xt,τ: GDP level quarter t published at time τ yt,τ: IMACEC level quarter t published at time τ yti,τ: IMACEC level month i of quarter t

3 , 1 4 , 3 12, 1

1 3 100* 1 1 3

i i

t i t t i

y y y

τ τ τ = − =

⎛ ⎞ ⎜ ⎟ Δ = − ⎜ ⎟ ⎜ ⎟ ⎝ ⎠

∑ ∑

SLIDE 9

9

The revisions between IMACEC and GDP publications do not seem to be systematic

• 2.5
• 2.0
• 1.5
• 1.0
• 0.5

0.0 0.5 1.0 1.5 87 89 91 93 95 97 99 01 03 05 07 Note: Solid straight lines indicate averages and the dotted lines are the averages plus / minus two times the standard deviation. Source: Author’s elaboration.

Revisions m visions made betwee de between n th the firs e first publica t publicatio ions of

• f IM

IMACEC ACEC and and GD GDP

(percentage points)

GDP re GDP revision visions: last mi s: last minus first p first publ blication ication

(percentage points)

• 3
• 2
• 1

1 2 3 4 5 87 89 91 93 95 97 99 01 03 05 07

SLIDE 10

10 10

The general model allows for up to four lags and an MA representation

4 4 4 4 4 1 1

, ,

f t t k t k t t j t j k j

x c z x β β ε ε ρ μ

− − = =

Δ = + Δ + Δ + =

∑ ∑

xf: Final GDP x1: First released GDP z: Explanatory variable: 1st exercise: First released GDP vs. quarterly IMACEC 2nd exercise: i = 1,2,3 months of IMACEC data for the quarter

SLIDE 11

11 11

De Depen penden ent va t variable le: Fina Final GD GDP

Note: Numbers in parentheses are Newey and West (1987) HAC consistent standard errors. The evaluation of the in-sample performance is made with data from 1988 to 2007, while the out-of-sample nowcasts are made from 1999. RMSE: Root mean square error calculated in-sample. RMSNE: Root mean square nowcast error calculated out-

• f-sample. Wald: p-value of the Wald test for reduction of model compared to the less parsimonious. Numbers in hard brackets are p-values for the comparison with the

model reported in the column to the left. D-M: p-value for the Diebold-Mariano test of equality with the model with lowest RMSE / RMSNE (in bold) against the alternative that this model is better. Best: Percentage of times where the model makes better projections than the model with minimum RMSE / RMSNE. Source: Author’s elaboration.

In-sample: Possible to make nowcasts better than first released GDP. Out-of- sample: not that evident

Absolut Absolute erro errors rs

(pe (percen centage age points)

• ints)

3.A 3.B 3.C 3.D 3.E 3.F 3.G 3.H Constant 0.49

(0.46)

Δ4x1 0.97

(0.05)

0.99

(0.06)

0.99

(0.06)

1.01

(0.05)

1.01

(0.05)

1.10

(0.03)

1.11

(0.03)

1 Δ4x1(-1)

• 0.003

(0.07)

Δ4x1(-2) 0.003

(0.05)

Δ4x1(-3) 0.21

(0.07)

0.13

(0.05)

0.13

(0.05)

0.11

(0.04)

0.11

(0.04)

Δ4x1(-4)

• 0.16

(0.09)

MA(1) 0.59

(0.11)

0.57

(0.12)

0.61

(0.12)

0.50

(0.11)

0.58

(0.08)

0.52

(0.09)

MA(2) 0.61

(0.13)

0.32

(0.16)

0.28

(0.15)

MA(3) 0.33

(0.11)

0.27

(0.13)

0.29

(0.11)

0.14

(0.13)

MA(4)

• 0.40

(0.11)

• 0.12

(0.11)

• Adj. R2

0.92 0.92 0.92 0.92 0.92 0.91 0.89 0.85 Wald 0.41 0.00 [0.26] 0.00 [0.05] 0.00 [0.29] 0.00 [0.00] 0.00 [0.00] 0.00 [0.00] In-sample nowcasting RMSE 0.92 0.96 0.96 0.98 1.00 1.04 1.18 1.34 D-M 0.30 0.28 0.20 0.17 0.12 0.01 0.01 Best 46% 51% 44% 43% 43% 44% 44% Out-of-sample nowcasting RMSNE 2.48 0.83 0.71 0.71 0.69 0.91 0.95 0.70 D-M 0.08 0.13 0.25 0.21 0.02 0.03 0.41 Best 36% 53% 50% 50% 39% 44% 53%

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1999 2000 2001 2002 2003 2004 2005 2006 2007 Simple (3.H) Best (3.E)

SLIDE 12

12 12

Source: Author’s elaboration.

De Depen penden ent va t variable le: Fina Final GD GDP

Out-of-sample: Quarterly IMACEC nowcasts can be improved when including historical data, although not significantly so

4.A 4.B 4.C 4.D 4.E 4.F 4.G Constant 0.29

(0.53)

Δ4y3 1.08

(0.08)

1.13

(0.07)

1.09

(0.05)

1.07

(0.05)

1.12

(0.04)

1.14

(0.04)

1 Δ4x1(-1) 0.07

(0.07)

Δ4x1(-2)

• 0.16

(0.08)

• 0.09

(0.07)

Δ4x1(-3) 0.06

(0.06)

Δ4x1(-4) 0.14

(0.05)

0.17

(0.04)

0.16

(0.02)

0.07

(0.03)

MA(1) 0.80

(0.15)

0.75

(0.09)

0.82

(0.17)

0.61

(0.08)

0.65

(0.09)

MA(2) 0.06

(0.19)

MA(3) 0.44

(0.13)

0.43

(0.12)

0.50

(0.14)

MA(4)

• 0.08

(0.15)

• Adj. R2

0.93 0.93 0.93 0.91 0.91 0.87 0.82 Wald 0.65 0.48 [0.21] 0.00 [0.00] 0.00 [0.04] 0.00 [0.00] 0.00 [0.00] In-sample nowcasting RMSE 0.88 0.91 0.89 1.04 1.06 1.26 1.50 D-M 0.19 0.36 0.02 0.01 0.00 0.00 Best 55% 50% 48% 44% 31% 41% Out-of-sample nowcasting RMSNE 2.74 1.10 1.83 0.71 0.66 0.98 0.69 D-M 0.06 0.01 0.02 0.16 0.01 0.34 Best 31% 25% 22% 36% 28% 50%

Absolut Absolute erro errors rs

(pe (percen centage age points)

• ints)

0.5 1 1.5 2 1999 2000 2001 2002 2003 2004 2005 2006 2007 Simple (4.G) Best (4.E)

SLIDE 13

13 13

Final GDP signals extracted from quarterly IMACEC are as good as those of the first released GDP

Diebold-Mariano tests for equality of models (out-of-sample): Simple models: (3.H) = (4.G): p-value = 0.70 (58% / 42%) Best models: (3.E) = (4.E): p-value = 0.70 (44% / 56%)

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1999 2000 2001 2002 2003 2004 2005 2006 2007 GDP-Model (simple) IMACEC-Model (simple) Source: Author’s elaboration.

Absolut Absolute erro errors rs

(pe (percen centage age points)

• ints)

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 1999 2000 2001 2002 2003 2004 2005 2006 2007 GDP-Model (best) IMACEC-Model (best)

SLIDE 14

14 14

Source: Author’s elaboration.

De Depen penden ent va t variable le: Fina Final GD GDP

Out-of-sample: The best benchmark model is a random walk

5.A 5.B 5.C 5.D 5.E 5.F 5.G Constant 4.75

(2.15)

5.66

(1.05)

5.81

(0.76)

5.81

(0.73)

5.81

(0.70)

1.39

(0.55)

Δ4x1(-1)

• 0.08

(0.20)

0.85

(0.09)

1 Δ4x1(-2)

• 0.15

(0.11)

Δ4x1(-3) 0.25

(0.18)

Δ4x1(-4) 0.16

(0.11)

0.03

(0.15)

MA(1) 1.06

(0.22)

0.96

(0.07)

0.95

(0.07)

0.94

(0.14)

0.86

(0.07)

MA(2) 1.09

(0.20)

0.78

(0.07)

0.76

(0.08)

0.24

(0.13)

MA(3) 0.99

(0.19)

0.79

(0.07)

0.78

(0.07)

MA(4) 0.11

(0.19)

• Adj. R2

0.71 0.71 0.71 0.57 0.54 0.58 0.53 Wald 0.07 0.00 [0.83] 0.00 [0.00] 0.00 [0.07] 0.00 0.00 [0.04] In-sample nowcasting RMSE 1.78 1.85 1.85 2.27 2.36 2.27 2.40 D-M 0.15 0.16 0.03 0.01 0.01 0.01 Best 51% 54% 48% 41% 40% 38% Out-of-sample nowcasting RMSNE 42.86 2.19 2.09 2.21 2.37 1.98 1.65 D-M 0.13 0.03 0.10 0.08 0.06 0.11 Best 36% 50% 58% 42% 36% 36%

Absolut Absolute erro errors rs

(pe (percen centage age points)

• ints)

1 2 3 4 5 6 7 1999 2000 2001 2002 2003 2004 2005 2006 2007 Simple (5.G) Alternative (5.C)

SLIDE 15

15 15

Source: Author’s elaboration.

De Depen penden ent va t variable le: Fina Final GD GDP

The nowcast performance does not improve significantly when including the first IMACEC of the quarter

6.A 6.B 6.C 6.D 6.E 6.F 6.G 6.H Constant 1.80

(0.75)

1.82

(0.62)

2.51

(0.52)

2.07

(0.43)

2.02

(0.42)

1.42

(0.37)

Δ4y1 0.66

(0.09)

0.64

(0.09)

0.67

(0.09)

0.77

(0.08)

0.78

(0.08)

0.90

(0.07)

1.10

(0.05)

1 Δ4x1(-1) 0.21

(0.10)

0.16

(0.09)

Δ4x1(-2)

• 0.12

(0.09)

Δ4x1(-3) 0.08

(0.11)

Δ4x1(-4)

• 0.02

(0.12)

MA(1) 0.57

(0.14)

0.62

(0.14)

0.66

(0.12)

0.58

(0.16)

0.57

(0.09)

MA(2) 0.40

(0.15)

0.38

(0.12)

0.45

(0.10)

0.07

(0.15)

MA(3) 0.41

(0.11)

0.45

(0.10)

0.48

(0.10)

MA(4)

• 0.16

(0.13)

• Adj. R2

0.82 0.83 0.82 0.79 0.79 0.73 0.68 0.65 Wald 0.19 0.18 [0.07] 0.00 [0.00] 0.00 [0.62] 0.00 [0.00] 0.00 [0.00] 0.00 [0.05] In-sample nowcasting RMSE 1.39 1.41 1.43 1.57 1.57 1.80 1.97 2.06 D-M 0.30 0.17 0.02 0.01 0.00 0.00 0.00 Best 46% 48% 35% 39% 34% 34% 33% Out-of-sample nowcasting RMSNE 3.36 1.52 1.52 1.83 1.81 2.16 1.90 1.59 D-M 0.09 0.49 0.02 0.02 0.01 0.07 0.32 Best 39% 53% 47% 47% 33% 50% 44%

Absolut Absolute erro errors rs

(pe (percen centage age points)

• ints)

D-M: (6.C)=(5.G): p-value= 0.36 (50% / 50%)

1 2 3 4 5 6 7 1999 2000 2001 2002 2003 2004 2005 2006 2007 t1-model (6.C) RW

SLIDE 16

16 16

Source: Author’s elaboration.

De Depen penden ent va t variable le: Fina Final GD GDP

Nowcast performance improves when two months of IMACEC data have been published

7.A 7.B 7.C 7.D 7.E 7.F 7.G Constant 0.99

(0.66)

1.12

(0.40)

0.96

(0.34)

0.99

(0.32)

Δ4y2 0.94

(0.07)

0.93

(0.07)

0.97

(0.06)

0.96

(0.05)

1.10

(0.04)

1.13

(0.04)

1 Δ4x1(-1)

• 0.005

(0.07)

Δ4x1(-2)

• 0.02

(0.07)

Δ4x1(-3) 0.09

(0.08)

Δ4x1(-4)

• 0.05

(0.09)

MA(1) 0.72

(0.14)

0.74

(0.14)

0.62

(0.11)

0.71

(0.07)

0.67

(0.08)

MA(2) 0.36

(0.22)

0.30

(0.19)

MA(3) 0.35

(0.14)

0.37

(0.08)

0.22

(0.11)

MA(4)

• 0.14

(0.11)

• Adj. R2

0.89 0.89 0.89 0.88 0.88 0.83 0.78 Wald 0.52 0.15 [0.12] 0.00 [0.06] 0.00 [0.00] 0.00 [0.00] 0.00 [0.00] In-sample nowcasting RMSE 1.10 1.11 1.13 1.17 1.22 1.45 1.64 D-M 0.24 0.23 0.01 0.01 0.01 0.01 Best 51% 49% 45% 44% 40% 40% Out-of-sample nowcasting RMSNE 1.71 2.05 2.01 1.25 1.08 1.39 1.08 D-M 0.01 0.09 0.10 0.05 0.09 0.39 Best 36% 42% 39% 33% 39% 56%

Absolut Absolute erro errors rs

(pe (percen centage age points)

• ints)

1 2 3 4 5 6 7 1999 2000 2001 2002 2003 2004 2005 2006 2007 RW t2-model (7.E)

Absolut Absolute erro errors rs

(pe (percen centage age points)

• ints)

0.5 1 1.5 2 2.5 3 3.5 4 4.5 1999 2000 2001 2002 2003 2004 2005 2006 2007 t1-model (6.C) t2-model (7.E)

D-M: (7.E)=(6.C): p-value= 0.00 (67% / 33%) D-M: (7.E)=(5.G): p-value= 0.08 (58% / 42%)

SLIDE 17

17 17

Out-of-sample: Nowcast improvement is visible when two months of IMACEC data are available

Diebold-Mariano tests (1) Diebold-Mariano tests (1)

(p-values)

Best point nowcasts Best point nowcasts (2) (2)

(percentage of total)

(1) p-values for the hypotheses of equal models against the alternative that the row model is better. ti (i = 0,1,2,3): i months of IMACEC data available for the quarter. t3+: First released GDP data available. The model used are 3.E, 4.E, 5.G, 6.C and 7.E. (2) Percentage that the row model makes better point nowcasts that the column model. ti (i = 0,1,2,3): i months of IMACEC data available for the quarter. t3+: First released GDP data available. Source: Author’s elaboration.

t0 t1 t2 t3 t1 0.36 t2 0.08 0.00 t3 0.02 0.00 0.00 t3+ 0.03 0.00 0.01 0.70 t0 t1 t2 t3 t1 50% t2 58% 67% t3 72% 78% 69% t3+ 67% 75% 75% 44%

SLIDE 18

18 18

Conclusions

• Systematic behavior seems to be present in

revisions between first and final released GDP.

• In-sample:

Preliminary GDP estimate can be improved using only historical data. Out-of-sample: Not that evident.

• Quarterly IMACEC signals are as good as those of

the first released GDP data.

• One month of IMACEC data does not improve the

nowcast performance. With two months of data the improvement is more visible.

SLIDE 19

B A N C O C E N T R A B A N C O C E N T R A L D E C H I L D E C H I L E L E 26-29 September 2010

Extracting GDP signals from the monthly indicator of economic activity: Evidence from Chilean real-time data

Michael Pedersen Central Bank of Chile 6th colloquium on modern tools for business cycle analysis