Relationship Between Commodities and Currency Pairs Derrick Hang - - PowerPoint PPT Presentation

Relationship Between Commodities and Currency Pairs

Derrick Hang Econ 201FS April 14, 2010

Agenda

 Wrapping up the Bayesian  Commodities and Currency Pairs

• Intuition

 Data  Volume and Volatility  HAR-RV  Jump Test - Co-Jump Test  Further research

Closure on Previous Analysis

 Past analysis attempted to find a useful predictors for

prices of currency pairs in the framework of a Bayesian- style dynamic linear model in order to improve portfolio allocations of a basket of currencies

 Problems:

• Sensitivity to initial values and difficulty in determining/justifying

these values

• Complicated and fragile model prone to error and required an

unexpectedly large amount of debug time

• Unclear economic intuition behind results, if any
• General familiarity with the model/Lack of correlating work

Continuation and Intuition

 Retain foreign exchange topic but use other

frameworks to assess relationships

 Realization: Majority of the currency pairs in my

possession are/can be considered “commodity currency”

 Hypothesis: Commodity currencies mirror various

changes in their respective commodity

 Empirically explore these relationship using high-

frequency data

Data

 5 minute price and volume data for 9 currency pairs,

Brent Crude Futures, Comex Gold Futures, SPY

 “Oil Currency Pairs”

• CAD/USD, NOK/USD

 “Gold Currency Pairs”

• AUD/USD, NZD/USD, CHF/USD, ZAR/USD

 Other pairs

• JPY/USD, EUR/USD, GBP/USD

 Data from 9:35AM-3:55PM weekdays from Jan – Jun

2009

• Exclude Jan 1st, Jan 19th, Feb 16th, Apr 10th, Apr 13th, May 25th due

to lack of across-the-board data for those days

Question 1: Relationship between Currency pair volume and variance

 Caveat: Aware of the concerns over the reliability of

volume data and interpretation and small window of data:

• Called data provider to verify meaning and accuracy; Lack of free

fx volume data to check…

 Hypothesis: Commodity volatility should be related to

respective “commodity currency” volumes as traders want to move to adjust portfolios for risk

 Lyons(1994), Admati and Pfleiderer(1988), Easley and

O‟Hara (1992): Event-uncertainty theory, hot-potato theory, Analysis of FX: volume begets volume

Question 1: Currency pair volume and Commodity realized variance

Question 1: Currency pair volume and Commodity realized variance

Question 1: Currency pair volume and Commodity realized variance

Question 1: Currency pair volume and Commodity realized variance

Question 1: Relationship between Currency pair volume and variance

 Can volume be a useful predictor of realized variance of

its respective commodity

 Hypothesis: Information about an impending change in

commodity volatility will cause traders to make adjustments in respective currency

 Regress lagged volume of commodity currencies on

realized variance of respective commodity

Question 1: Relationship between Currency pair volume and variance

Lag 1 Volume on RV of Gold

AUD CHF NZD ZAR GBP CAD Constant 0.0015 4.7143e- 004 9.8187e- 004 6.9084e- 004 5.8580e- 004 0.0014 Beta

• 1.1945e-

004

• 3.0394e-

005

• 7.9778e-

005

• 5.5753e-

005

• 3.9823e-

005

• 1.1294e-

004 F-Test 8.8839 0.3392 3.7435 1.9116 0.5388 6.3431 p-value 0.0035 0.5614 0.0554 0.1694 0.4643 0.0131 R- squared 0.0689 0.0028 0.0303 0.0157 0.0045 0.0502

Question 1: Relationship between Currency pair volume and variance

Lag 1 Volume on RV of Oil

CAD NOK GBP AUD Constant 0.0052 6.2762e-004 5.8344e-004 0.0054 Beta

• 4.1377e-

004

• 9.9115e-

006

• 5.9715e-

006

• 4.2852e-

004 F-Test 14.7557 0.0051 0.0020 20.2121 p-value 0.0002 0.9430 0.9648 0.0000 R-squared 0.1095 0.0000 0.0000 0.1442

Question 1: Relationship between Currency pair volume and variance

 Highest R-squared are for the AUD/USD, NZD/USD, CAD/USD  From a initial search on the Internet, these 3 pairs are the most

consistently noted as “currency commodities”



High R-squared in mismatched pair/commodity: Perhaps change in volatility in trade gives traders incentive to adjust other commodity pair to hedge risk

 However, in the case of a relationship, across-the-board negative

betas seem to support the hot-potato theory IF information about volatility changes are not well-known

 Possibility: Perform analysis with higher lag and regress oil and gold

• n all pairs and correlations between commodity currencies

Question 2: Relationship between Currency pair & commodity variance

 Question: Can volatility in a commodity be a good

predictor for volatility in respective „commodity currencies”?

 Employ the HAR-RV model

• Regress for RV (t+1) of a particular currency pair with its lagged

daily RV(t), weekly RV(t-5), and monthly RV(t-22)

• Add in HAR-RV regressors for gold
• Add in HAR-RV regressors for oil
• Compare!

 For this presentation, only AUD/USD and CAD/USD

are shown for time concerns

Question 2: Relationship between Currency pair & commodity variance

Question 2: Relationship between Currency pair & commodity variance

Question 2: Relationship between Currency pair & commodity variance

Regress for AUD/USD RV

* indicates significance at the 5% level

AUD AUD GOLD AUD OIL Constant 0.0000 0.0000 0.0000 Beta_d 0.4137* 0.4086* 0.0761* 0.3678* 0.0132 Beta_w

• 0.0537
• 0.0371
• 0.0329
• 0.0586
• 0.0041

Beta_m

• 0.0244
• 0.0560

0.0062

• 0.1651

0.0298 p-value of F-test 0.0003 0.0006 0.0007 R-squared 0.1754 0.2213 0.2186

Question 2: Relationship between Currency pair & commodity variance

Regress for AUD/USD RV

* indicates significance at the 5% level

AUD GOLD OIL Constant 0.0000 Beta_d 0.3785* 0.0708*

• 0.0002

Beta_w

• 0.0447

0.0049

• 0.0049

Beta_m

• 0.1695
• 0.0380

0.0295 p-value of F- test 0.0010 R-squared 0.2585

Question 2: Relationship between Currency pair & commodity variance

Regress for CAD/USD RV

* indicates significance at the 5% level

CAD CAD GOLD CAD OIL Constant 0.0000 0.0000 0.0000 Beta_d 0.1867 0.1871 0.0467* 0.1678 0.0071 Beta_w

• 0.1108
• 0.0508
• 0.0295
• 0.0938
• 0.0090

Beta_m 0.0769 0.0699

• 0.0224

0.0462 0.0060 p-value of F-test 0.1588 0.0268 0.4009 R-squared 0.0523 0.1394 0.0632

Question 2: Relationship between Currency pair & commodity variance

Regress for CAD/USD RV

* indicates significance at the 5% level

CAD GOLD OIL Constant 0.0000 Beta_d 0.1808 0.0449

• 0.0031

Beta_w

• 0.0458

0.0045

• 0.0240

Beta_m 0.0359

• 0.0317

0.0062 p-value of F- test 0.0927 R-squared 0.1478

Question 2: Relationship between Currency pair & commodity variance

 Only the lag 1 (daily) regressor is individually significant in

these regressions

• Daily Gold on AUD/USD and daily AUD/USD on AUD/USD
• Daily Gold on CAD/USD

 Significant regressors are all positive in these cases; however

immediate intuitive on the relationship is unclear

 Notice that CAD/USD regressors were not individually or

jointly significant when regressed on CAD/USD and had low r-squared => HAR-RV model may be inadequate due to small window of data or due to uninformative past movements in RV

Question 2: Relationship between Currency pair & commodity variance

 Run HAR-RV using higher sampling frequencies (10 min,

15 min) to calculate daily RV

 Run HAR-RV on the commodity RV and SPY RV and

look for any relationships

 Look for relationships between currency pairs using

HAR-RV

 Assess the viability of HAR-RV model with the short

time window and implications on interpretation outside

• f this window

Question 3: Currency pair & commodity co-jumps

 Do currency pairs and their respective commodities

jump together?

 Hypothesis: I expect to see more instances of co-jumps

between commodity currency and the commodity itself because I expect macroeconomic announcements that

• ur revelant to a currency pair to also be relevant to

the respective commodity

Question 3: Currency pair & commodity co-jumps

 Raw analysis: Run the max-adjusted bipower and max-

adjusted tripower BNS Jump tests and Median Jump test and search for common days declared as jump days between commodities and currencies at the 5%, 1%, and 0.1% significance levels

 Use the correlation statistic from Roeber (1993) to

express standardized jump correlation, where C is number of common jumps and J are the number of jumps for each respective currency pair

b a b a b a

J J C * /

, , 



Question 3: Currency pair & commodity co-jumps

Question 3: Currency pair & commodity co-jumps

CAD NOK AUD 5% Level 3 3 1 Co-Jump Days 09-Jan-2009 12-Jun-2009 17-Jun-2009 09-Jan-2009 14-Jan-2009 24-Jun-2009 04-Jun-2009 1% Level

• 1

Co-Jump Days

• 04-Jun-2009

0.1% Level

• 1

Co-Jump Days

• 04-Jun-2009

Roeber Coefficient (5%,1%,0.1%) 0.1309; - ; - 0.1414; - ; - 0.0485; 0.1890; 0.5774 Max-Adjusted Tri-Power Test OIL “CO-JUMPS”

Question 3: Currency pair & commodity co-jumps

AUD CHF NZD 5% Level 3 9 3 Co-Jump Days 16-Jan 08-May 22-May 14-Jan, 16-Jan, 05- Mar,25-Mar, 23- Apr, 08-May, 22- May, 16-Jun, 23- Jun 03-Mar 19-May 22-May 1% Level

• 3

1 Co-Jump Days

• 16-Jan, 23-Apr,

08-May 03-Mar 0.1% Level

• 2
• Co-Jump Days
• 23-Apr, 08-May
• Roeber Coef

.1328; - ; - 0.3051; 0.2224; 0.3381 0.1054; 0.0772 ; - Max-Adjusted Tri-Power Test GOLD “CO-JUMPS”

Question 3: Currency pair & commodity co-jumps

ZAR CAD 5% Level 4 3 Co-Jump Days 16-Jan-2009 16-Apr-2009 30-Apr-2009 21-May-2009 09-Jan-2009 12-Jun-2009 17-Jun-2009 1% Level 1

• Co-Jump Days

21-May-2009

• 0.1% Level
• Co-Jump Days
• Roeber Coef

0.2025; 0.1890; - 0.1309 ; - ; - Max-Adjusted Tri-Power Test GOLD “CO-JUMPS”

Question 3: Currency pair & commodity co-jumps

Question 3: Currency pair & commodity co-jumps

Question 3: Currency pair & commodity co-jumps

Question 3: Currency pair & commodity co-jumps

Question 3: Currency pair & commodity co-jumps

Question 3: Currency pair & commodity co-jumps

 CHF/USD is the only currency pair that has a common

jump to the 0.1% level with the “correct” commodity

 AUD/USD has a co-jump at the 0.1% level with gold  We see a couple of common jump across currency

pairs, but only at the 5% significance level

Question 3: Currency pair & commodity co-jumps

 Check and correct for bugs in code  Implement formalized co-jump tests  Use the Lee-Mykland test outlined in Lee-Mykland

(2008) to test for jumps in specific returns

• Employed the Lee-Mykland test correction suggested in Jansen &

Tauchen (2009)

 Use the BNS Co-Jump Test  Focus on one topic? Suggestions.