Discussion“The Changing Relationship Between Commodity Prices and Prices

of Other Assets with Global Market Integration”by Barbara Rossi

Domenico GiannoneUniversite libre de Bruxelles, ECARES and CEPR

IMF-TMB ConferencePolicy Responses to Commodity Price Movements

Istanbul, April 2012

1 / 32

Once upon a time...

American Economic Association meetings, 2009

2 / 32

Once upon a time...

American Economic Association meetings, 2009

2 / 32

Once upon a time...

Discussion of Can

exchange rates forecast

commodity prices? Hélène Rey

London Business School, CEPR,

NBER

3 / 32

Once upon a time...

Intriguing result

!  Exchange rates of small commodity exporters

have forecasting power for global commodity

index. True in and out of sample.

!  Reverse is not true, i.e. commodity prices do not

seem to forecast exchange rates very well (and

not out of sample).

4 / 32

Once upon a time...

Conclusions

!  Very interesting set of results

!  Calls for extensions

!  Stock market index has also some forecast ability

in and out of sample.

!  Seems a slightly noisier predictor than the

exchange rate.

5 / 32

The Data

95−Q1 00−Q1 05−Q1 10−Q14

4.5

5

5.5(log) Global Commodity Price

95−Q1 00−Q1 05−Q1 10−Q1100

150

200

250

300

350(log) NZ Equity Price

6 / 32

The Sample

95−Q1 00−Q1 05−Q1 10−Q14

4.5

5

5.5(log) Global Commodity Price

95−Q1 00−Q1 05−Q1 10−Q1100

150

200

250

300

350(log) NZ Equity Price

7 / 32

Forecasting Global Commodity Prices (CP)

Forecasting Global Commodity Prices (CP) using asset prices ofsmall commodity producers (e.g. New Zealand (NZ))- Exchange Rate (EXR)- Equity Prices (EP)

This paper (EP)

Et∆CPt+1 = α + β∆EPNZt + ρ∆CPt

Barbara’s previous work ( EXR):

Et∆CPt+1 = α + β∆EXRNZt + ρ∆CPt

The Naive Benchmark: Random Walk

Et∆CPt+1 = 0

8 / 32

Forecasting Global Commodity Prices (CP)

Forecasting Global Commodity Prices (CP) using asset prices ofsmall commodity producers (e.g. New Zealand (NZ))- Exchange Rate (EXR)- Equity Prices (EP)

This paper (EP)

Et∆CPt+1 = α + β∆EPNZt + ρ∆CPt

Barbara’s previous work ( EXR):

Et∆CPt+1 = α + β∆EXRNZt + ρ∆CPt

The Naive Benchmark: Random Walk

Et∆CPt+1 = 0

8 / 32

Forecasting Global Commodity Prices (CP)

Forecasting Global Commodity Prices (CP) using asset prices ofsmall commodity producers (e.g. New Zealand (NZ))- Exchange Rate (EXR)- Equity Prices (EP)

This paper (EP)

Et∆CPt+1 = α + β∆EPNZt + ρ∆CPt

Barbara’s previous work ( EXR):

Et∆CPt+1 = α + β∆EXRNZt + ρ∆CPt

The Naive Benchmark: Random Walk

Et∆CPt+1 = 0

8 / 32

Forecasting Global Commodity Prices (CP)

Forecasting Global Commodity Prices (CP) using asset prices ofsmall commodity producers (e.g. New Zealand (NZ))- Exchange Rate (EXR)- Equity Prices (EP)

This paper (EP)

Et∆CPt+1 = α + β∆EPNZt + ρ∆CPt

Barbara’s previous work ( EXR):

Et∆CPt+1 = α + β∆EXRNZt + ρ∆CPt

The Naive Benchmark: Random Walk

Et∆CPt+1 = 0

8 / 32

The Input

95−Q1 00−Q1 05−Q1 10−Q1

−10

−5

0

5

10

Commodity Price

Equity Price

9 / 32

The Output

04−Q1 06−Q1 08−Q1 10−Q1−5

0

5

10

15

Model Forecast

Commodity Price

Random Walk

10 / 32

The Output

04−Q1 06−Q1 08−Q1 10−Q1−5

0

5

10

15

Model Forecast

Commodity Price

Random Walk

11 / 32

The Evaluation: Quadratic Loss

04−Q1 06−Q1 08−Q1 10−Q1−5

0

5

10

15

Model Forecast

Commodity Price

Random Walk

04−Q1 06−Q1 08−Q1 10−Q10

20

40

60

80

100

120Squared Errors

Model

Random Walk

12 / 32

The Evaluation: Fluctuation Test

04−Q1 06−Q1 08−Q1 10−Q1−5

0

5

10

15

Model Forecast

Commodity Price

Random Walk

04−Q1 06−Q1 08−Q1 10−Q10

20

40

60

80

100

120Squared Errors

Model

Random Walk

Smooth Mod.

Smooth RW

13 / 32

The empirical finding

[...] the appearance of the out-of-sample predictive ability of theequity market predictor [...] dated around mid-2000.

Since the mid-2000s marked a large increase in investment incommodity markets, ...

... our empirical evidence suggests a decrease in marketsegmentation at approximately the same time, ...

... and therefore the possibility that shocks in equity markets mighthave started to spill-over onto commodity markets in that period.

14 / 32

The empirical finding

[...] the appearance of the out-of-sample predictive ability of theequity market predictor [...] dated around mid-2000.

Since the mid-2000s marked a large increase in investment incommodity markets, ...

... our empirical evidence suggests a decrease in marketsegmentation at approximately the same time, ...

... and therefore the possibility that shocks in equity markets mighthave started to spill-over onto commodity markets in that period.

14 / 32

The empirical finding

[...] the appearance of the out-of-sample predictive ability of theequity market predictor [...] dated around mid-2000.

Since the mid-2000s marked a large increase in investment incommodity markets, ...

... our empirical evidence suggests a decrease in marketsegmentation at approximately the same time, ...

... and therefore the possibility that shocks in equity markets mighthave started to spill-over onto commodity markets in that period.

14 / 32

The empirical finding

[...] the appearance of the out-of-sample predictive ability of theequity market predictor [...] dated around mid-2000.

Since the mid-2000s marked a large increase in investment incommodity markets, ...

... our empirical evidence suggests a decrease in marketsegmentation at approximately the same time, ...

... and therefore the possibility that shocks in equity markets mighthave started to spill-over onto commodity markets in that period.

14 / 32

Other Predictors

00−Q1 10−Q14

4.5

5

5.5(log) Global Commodity Price

00−Q1 10−Q1100

150

200

250

300

350(log) NZ Equity Price

00−Q1 10−Q1−1

−0.8

−0.6

−0.4

−0.2

0(log) Exch. Rate

00−Q1 10−Q14.3

4.4

4.5

4.6

4.7

4.8

4.9

5(log) Global IP

15 / 32

Other Predictors

04−Q1 06−Q1 08−Q1 10−Q10

0.5

1

1.5

2

2.5

3

Equity Price

16 / 32

Other Predictors

04−Q1 06−Q1 08−Q1 10−Q10

0.5

1

1.5

2

2.5

3

Equity Price

Exch. Rate

17 / 32

Other Predictors

04−Q1 06−Q1 08−Q1 10−Q10

0.5

1

1.5

2

2.5

3

Equity Price

Exch. Rate

Global IP

18 / 32

Other Predictors

04−Q1 06−Q1 08−Q1 10−Q10

0.5

1

1.5

2

2.5

3

Equity Price

Exch. Rate

Global IP

Pool

19 / 32

Summing up

The appearance of predictive ability is not specific to the theequity market predictor!!

Factors other that the decrease in market segmentation andspillovers from equity markets might have been at work!

20 / 32

Forecasting Commodity Prices at the Time of the GreatRecession

00−Q1 10−Q14

4.5

5

5.5(log) Global Commodity Price

00−Q1 10−Q1100

150

200

250

300

350(log) NZ Equity Price

00−Q1 10−Q1−1

−0.8

−0.6

−0.4

−0.2

0(log) Exch. Rate

00−Q1 10−Q14.3

4.4

4.5

4.6

4.7

4.8

4.9

5(log) Global IP

21 / 32

Forecasting Commodity Prices at the Time of the GreatRecession

00−Q1 10−Q14

4.5

5

5.5(log) Global Commodity Price

00−Q1 10−Q1100

150

200

250

300

350(log) NZ Equity Price

00−Q1 10−Q1−1

−0.8

−0.6

−0.4

−0.2

0(log) Exch. Rate

00−Q1 10−Q14.3

4.4

4.5

4.6

4.7

4.8

4.9

5(log) Global IP

22 / 32

Commodity and Equity Prices at the Time of the GreatRecession

95−Q1 00−Q1 05−Q1 10−Q1

−30

−25

−20

−15

−10

−5

0

5

10

15

Commodity Price

Equity Price

23 / 32

Forecasting Commodity Prices at the Time of the GreatRecession

04−Q1 06−Q1 08−Q1 10−Q10

0.5

1

1.5

2

2.5

3

Equity Price

Exch. Rate

Global IP

Pool

24 / 32

Forecasting Commodity Prices at the Time of the GreatRecession

04−Q1 06−Q1 08−Q1 10−Q10

0.5

1

1.5

2

2.5

3

Equity Price

Exch. Rate

Global IP

Pool

25 / 32

Forecasting Commodity Prices at the Time of the GreatRecession

04−Q1 06−Q1 08−Q1 10−Q10

0.5

1

1.5

2

2.5

3

Equity Price

Exch. Rate

Global IP

Pool

26 / 32

Can we really forecast commodity prices??

27 / 32

An Alternative Look at the Data:Modeling Structural Change

Model: Time-Varying Vector Autoregressions (TV-VAR)Cogley and Sargent, 2001, 2005, Primiceri, 2006

yt = A0,t + A1,tyt−1 + εt , εt ∼ N(0,Σt) (1)

Parameters evolve according to

Coefficients : θt = θt−1 + ωt , ωt ∼ N(0,Ω) (2)

Variances : log σt = log σt−1 + ξt , ξt ∼ N(0,Ξ) (3)

Autocovariance : φi ,t = φi ,t−1 + ψi ,t , ψi ,t ∼ N(0,Ψi ) (4)

ψi ,t , ξt , ωt , εt all mutually uncorrelated at all leads and lags.

Reliable Forecasting Tool: D’Agostino, Gambetti and Giannone, 2009.

⇒ Flexible but parsimonious model of structural changes

28 / 32

An Alternative Look at the Data:Modeling Structural Change

Model: Time-Varying Vector Autoregressions (TV-VAR)Cogley and Sargent, 2001, 2005, Primiceri, 2006

yt = A0,t + A1,tyt−1 + εt , εt ∼ N(0,Σt) (1)

Parameters evolve according to

Coefficients : θt = θt−1 + ωt , ωt ∼ N(0,Ω) (2)

Variances : log σt = log σt−1 + ξt , ξt ∼ N(0,Ξ) (3)

Autocovariance : φi ,t = φi ,t−1 + ψi ,t , ψi ,t ∼ N(0,Ψi ) (4)

ψi ,t , ξt , ωt , εt all mutually uncorrelated at all leads and lags.

Reliable Forecasting Tool: D’Agostino, Gambetti and Giannone, 2009.

⇒ Flexible but parsimonious model of structural changes

28 / 32

An Alternative Look at the Data:Modeling Structural Change

Setting the Prior

yt = A0,t + A1,tyt−1 + εt , εt ∼ N(0,Σt) (5)

Parameters evolve according to

θt = θt−1 + ωt , ωt ∼ N(0,Ω) (6)

Estimate the model by ML over the training sample: 1992-2001 (θ)

θ0 = θ V (θ0) = V (θ)

Ω = V (θt − θt−1) = V (θ)× λ2

Quite loose prior to favor time variation:λ2 = 1

10

Estimation: Gibbs Sampling on entire sample

29 / 32

An Alternative Look at the Data:Time Varying Coefficients

Q1−00 Q1−05 Q1−10 Q1−150

0.1

0.2

0.3

0.4

0.5Commodity to Commodity

Q1−00 Q1−05 Q1−10 Q1−15−0.1

−0.05

0

0.05

0.1

0.15Equity to Commodity

Q1−00 Q1−05 Q1−10 Q1−15−0.5

0

0.5

1Commodity to Equity

Q1−00 Q1−05 Q1−10 Q1−15−0.2

−0.1

0

0.1

0.2

0.3Equity to Equity

If anything, Commodity Prices are becoming Less Predictable30 / 32

An Alternative Look at the Data:Stochastic Volatility

Q1−02 Q1−04 Q1−06 Q1−08 Q1−10 Q1−120

50

100

150

200Variance of Commodity Price Innovations

Q1−02 Q1−04 Q1−06 Q1−08 Q1−10 Q1−120

100

200

300

400

500

600Variance of Equity Price Innovations

Strong comovement in volatility: might be exploited in riskassessment 31 / 32

Conclusions

Fascinating, relevant and innovative research agenda

Vey interesting and well executed paper.

I strongly advice you to read it!!

32 / 32