[PPT] - Can Oil Prices Forecast Exchange Rates? Domenico Ferraro, Ken Rogo PowerPoint Presentation

SLIDE 1

Can Oil Prices Forecast Exchange Rates?

Domenico Ferraro, Ken Rogo¤ and Barbara Rossi June 2011

Rossi () Oil Prices & Exchange Rates June 2011 1 / 42

SLIDE 2

Motivation

Can Oil Prices Forecast Exchange Rate Movements?

Rossi (Duke) Oil Prices & Exchange Rates June 2011 2 / 42

SLIDE 3

Motivation

Can Oil Prices Forecast Exchange Rate Movements? Focus on Canada for three reasons:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 2 / 42

SLIDE 4

Motivation

Can Oil Prices Forecast Exchange Rate Movements? Focus on Canada for three reasons:

1

crude oil represents a substantial component of Canada’s total exports

Rossi (Duke) Oil Prices & Exchange Rates June 2011 2 / 42

SLIDE 5

Motivation

Can Oil Prices Forecast Exchange Rate Movements? Focus on Canada for three reasons:

1

crude oil represents a substantial component of Canada’s total exports

2

Canada has a su¢ciently long history of market-based ‡oating exchange rate

Rossi (Duke) Oil Prices & Exchange Rates June 2011 2 / 42

SLIDE 6

Motivation

Can Oil Prices Forecast Exchange Rate Movements? Focus on Canada for three reasons:

1

crude oil represents a substantial component of Canada’s total exports

2

Canada has a su¢ciently long history of market-based ‡oating exchange rate

3

Canada is a small-open economy => crude oil price ‡uctuations serve as an observable and essentially exogenous terms-of-trade shock

Rossi (Duke) Oil Prices & Exchange Rates June 2011 2 / 42

SLIDE 7

Motivation

Can Oil Prices Forecast Exchange Rate Movements? Focus on Canada for three reasons:

1

crude oil represents a substantial component of Canada’s total exports

2

Canada has a su¢ciently long history of market-based ‡oating exchange rate

3

Canada is a small-open economy => crude oil price ‡uctuations serve as an observable and essentially exogenous terms-of-trade shock

... although we check robustness with other countries/commodity prices.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 2 / 42

SLIDE 8

Motivation

Can Oil Prices Forecast Exchange Rate Movements?

Rossi (Duke) Oil Prices & Exchange Rates June 2011 3 / 42

SLIDE 9

Motivation

Can Oil Prices Forecast Exchange Rate Movements? In DAILY data, YES!

Rossi (Duke) Oil Prices & Exchange Rates June 2011 3 / 42

SLIDE 10

Main Contributions I

We …nd little systematic out-of-sample forecasting relation between oil prices and the exchange rate at monthly and quarterly frequencies, but...

Rossi (Duke) Oil Prices & Exchange Rates June 2011 4 / 42

SLIDE 11

Main Contributions I

We …nd little systematic out-of-sample forecasting relation between oil prices and the exchange rate at monthly and quarterly frequencies, but... ...there is an out-of-sample forecast relationship in daily data:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 4 / 42

SLIDE 12

Main Contributions I

We …nd little systematic out-of-sample forecasting relation between oil prices and the exchange rate at monthly and quarterly frequencies, but... ...there is an out-of-sample forecast relationship in daily data:

contemporaneous, realized oil prices do predict daily nominal exchange rates between Canada and the U.S., and their predictive ability is strongly signi…cant

Rossi (Duke) Oil Prices & Exchange Rates June 2011 4 / 42

SLIDE 13

Main Contributions I

We …nd little systematic out-of-sample forecasting relation between oil prices and the exchange rate at monthly and quarterly frequencies, but... ...there is an out-of-sample forecast relationship in daily data:

contemporaneous, realized oil prices do predict daily nominal exchange rates between Canada and the U.S., and their predictive ability is strongly signi…cant the predictive ability of the lagged realized oil prices is more ephemeral, and allowing for time variation is crucial

Rossi (Duke) Oil Prices & Exchange Rates June 2011 4 / 42

SLIDE 14

Main Contributions I

We …nd little systematic out-of-sample forecasting relation between oil prices and the exchange rate at monthly and quarterly frequencies, but... ...there is an out-of-sample forecast relationship in daily data:

contemporaneous, realized oil prices do predict daily nominal exchange rates between Canada and the U.S., and their predictive ability is strongly signi…cant the predictive ability of the lagged realized oil prices is more ephemeral, and allowing for time variation is crucial

On the contrary, in-sample …t is stronger in monthly and quarterly data than in daily data

Rossi (Duke) Oil Prices & Exchange Rates June 2011 4 / 42

SLIDE 15

Main Contributions II

Similar results hold for other commodity prices/exchange rates:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 5 / 42

SLIDE 16

Main Contributions II

Similar results hold for other commodity prices/exchange rates:

in Norwegian Krone-U.S. dollar exchange rate and oil prices, we …nd signi…cant predictive ability of both contemporaneous and lagged oil prices

Rossi (Duke) Oil Prices & Exchange Rates June 2011 5 / 42

SLIDE 17

Main Contributions II

Similar results hold for other commodity prices/exchange rates:

in Norwegian Krone-U.S. dollar exchange rate and oil prices, we …nd signi…cant predictive ability of both contemporaneous and lagged oil prices for the South African Rand-U.S. dollar exchange rate and gold prices we also …nd signi…cance with both contemporaneous and lagged commodity prices

Rossi (Duke) Oil Prices & Exchange Rates June 2011 5 / 42

SLIDE 18

Main Contributions II

Similar results hold for other commodity prices/exchange rates:

in Norwegian Krone-U.S. dollar exchange rate and oil prices, we …nd signi…cant predictive ability of both contemporaneous and lagged oil prices for the South African Rand-U.S. dollar exchange rate and gold prices we also …nd signi…cance with both contemporaneous and lagged commodity prices for the Australian-U.S. dollar and oil prices and the Chilean Peso-U.S. dollar exchange rate and copper prices, we …nd strong and signi…cant predictive ability only with contemporaneous commodity prices as predictors.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 5 / 42

SLIDE 19

Literature Review I

Our results are related to literature on high frequency e¤ects

f macro news announcements: Andersen et al., 2003, Faust

et al., 2007, Kilian and Vega, 2008

Rossi (Duke) Oil Prices & Exchange Rates June 2011 6 / 42

SLIDE 20

Literature Review I

Our results are related to literature on high frequency e¤ects

f macro news announcements: Andersen et al., 2003, Faust

et al., 2007, Kilian and Vega, 2008

=> We identify an alternative macroeconomic determinant: commodity price changes

Rossi (Duke) Oil Prices & Exchange Rates June 2011 6 / 42

SLIDE 21

Literature Review I

Our results are related to literature on high frequency e¤ects

f macro news announcements: Andersen et al., 2003, Faust

et al., 2007, Kilian and Vega, 2008

=> We identify an alternative macroeconomic determinant: commodity price changes

Our results are also related to studies on the in-sample …t of commodity currencies and commodity prices at monthly/quarterly frequencies:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 6 / 42

SLIDE 22

Literature Review I

Our results are related to literature on high frequency e¤ects

f macro news announcements: Andersen et al., 2003, Faust

et al., 2007, Kilian and Vega, 2008

=> We identify an alternative macroeconomic determinant: commodity price changes

Our results are also related to studies on the in-sample …t of commodity currencies and commodity prices at monthly/quarterly frequencies:

Amano and Van Norden (1998a,b) and Chen and Rogo¤ (2003) for in-sample …t of real exchange rates/commodity indexes

Rossi (Duke) Oil Prices & Exchange Rates June 2011 6 / 42

SLIDE 23

Literature Review I

Our results are related to literature on high frequency e¤ects

f macro news announcements: Andersen et al., 2003, Faust

et al., 2007, Kilian and Vega, 2008

=> We identify an alternative macroeconomic determinant: commodity price changes

Our results are also related to studies on the in-sample …t of commodity currencies and commodity prices at monthly/quarterly frequencies:

Amano and Van Norden (1998a,b) and Chen and Rogo¤ (2003) for in-sample …t of real exchange rates/commodity indexes Issa, Lafrance and Murray (2008) and Cayen, Coletti, Lalonde and Maier (2010) for in-sample …t and instabilities / factor analysis

Rossi (Duke) Oil Prices & Exchange Rates June 2011 6 / 42

SLIDE 24

Literature Review I

Our results are related to literature on high frequency e¤ects

f macro news announcements: Andersen et al., 2003, Faust

et al., 2007, Kilian and Vega, 2008

=> We identify an alternative macroeconomic determinant: commodity price changes

Our results are also related to studies on the in-sample …t of commodity currencies and commodity prices at monthly/quarterly frequencies:

Amano and Van Norden (1998a,b) and Chen and Rogo¤ (2003) for in-sample …t of real exchange rates/commodity indexes Issa, Lafrance and Murray (2008) and Cayen, Coletti, Lalonde and Maier (2010) for in-sample …t and instabilities / factor analysis

=> Our results focus on out-of-sample forecasting, and document short-lived e¤ect identi…able only at high frequencies

Rossi (Duke) Oil Prices & Exchange Rates June 2011 6 / 42

SLIDE 25

Literature Review II

Our paper is related to literature on predicting nominal exchange rates using macroeconomic fundamentals:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 7 / 42

SLIDE 26

Literature Review II

Our paper is related to literature on predicting nominal exchange rates using macroeconomic fundamentals:

Meese and Rogo¤’s (1983a,b) puzzle

Rossi (Duke) Oil Prices & Exchange Rates June 2011 7 / 42

SLIDE 27

Literature Review II

Our paper is related to literature on predicting nominal exchange rates using macroeconomic fundamentals:

Meese and Rogo¤’s (1983a,b) puzzle Evidence of predictive ability at longer horizons: Mark, 1995; Chinn and Meese, 1995; Engel, Mark and West, 2007; Faust, Rogers and Wright (2003); Kilian and Taylor (2003).

Rossi (Duke) Oil Prices & Exchange Rates June 2011 7 / 42

SLIDE 28

Literature Review II

Our paper is related to literature on predicting nominal exchange rates using macroeconomic fundamentals:

Meese and Rogo¤’s (1983a,b) puzzle Evidence of predictive ability at longer horizons: Mark, 1995; Chinn and Meese, 1995; Engel, Mark and West, 2007; Faust, Rogers and Wright (2003); Kilian and Taylor (2003).

Our paper focuses instead on short-horizon predictive ability, for which the empirical evidence in favor of the economic models has been more controversial.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 7 / 42

SLIDE 29

Literature Review II

Our paper is related to literature on predicting nominal exchange rates using macroeconomic fundamentals:

Meese and Rogo¤’s (1983a,b) puzzle Evidence of predictive ability at longer horizons: Mark, 1995; Chinn and Meese, 1995; Engel, Mark and West, 2007; Faust, Rogers and Wright (2003); Kilian and Taylor (2003).

Our paper focuses instead on short-horizon predictive ability, for which the empirical evidence in favor of the economic models has been more controversial.

We focus on linear models but we check performance of nonlinear models: see Hamilton (2003), Kilian and Vigfusson (2011), etc. for nonlinear relationships between oil and output, and Alquist, Kilian and Vigfusson (2011) for forecasting oil prices.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 7 / 42

SLIDE 30

Roadmap

I. Can Realized Oil Prices Forecast Exchange Rates?
II. Can Lagged Oil Prices Forecast Exchange Rates?
III. Other Commodities/Exchange Rates
IV. Are Non-linearities Important?

Rossi (Duke) Oil Prices & Exchange Rates June 2011 8 / 42

SLIDE 31

Roadmap

I. Can Realized Oil Prices Forecast Exchange Rates?

YES! Why are we able to …nd predictive ability?

Choice of Fundamental? Frequency or Number of Observations? Is it stable? How about in-sample …t?

II. Can Lagged Oil Prices Forecast Exchange Rates?
III. Other Commodities/Exchange Rates
IV. Are Non-linearities Important?

Rossi (Duke) Oil Prices & Exchange Rates June 2011 9 / 42

SLIDE 32

Can Realized Oil Price Changes Forecast Exchange Rates?

The Regression:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 10 / 42

SLIDE 33

Can Realized Oil Price Changes Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt + ut, t = 1, ..., T,

Rossi (Duke) Oil Prices & Exchange Rates June 2011 10 / 42

SLIDE 34

Can Realized Oil Price Changes Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt + ut, t = 1, ..., T, ∆st = …rst di¤erence of the log of the Canadian/U.S. Dollar exchange rate (Barclays)

Rossi (Duke) Oil Prices & Exchange Rates June 2011 10 / 42

SLIDE 35

Can Realized Oil Price Changes Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt + ut, t = 1, ..., T, ∆st = …rst di¤erence of the log of the Canadian/U.S. Dollar exchange rate (Barclays) ∆pt = …rst di¤erence of the log of the oil price (West Texas Intermediate)

Rossi (Duke) Oil Prices & Exchange Rates June 2011 10 / 42

SLIDE 36

Can Realized Oil Price Changes Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt + ut, t = 1, ..., T, ∆st = …rst di¤erence of the log of the Canadian/U.S. Dollar exchange rate (Barclays) ∆pt = …rst di¤erence of the log of the oil price (West Texas Intermediate) ut = unforecastable error term

Rossi (Duke) Oil Prices & Exchange Rates June 2011 10 / 42

SLIDE 37

Can Realized Oil Price Changes Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt + ut, t = 1, ..., T, ∆st = …rst di¤erence of the log of the Canadian/U.S. Dollar exchange rate (Barclays) ∆pt = …rst di¤erence of the log of the oil price (West Texas Intermediate) ut = unforecastable error term We consider three frequencies: quarterly, monthly and daily

Rossi (Duke) Oil Prices & Exchange Rates June 2011 10 / 42

SLIDE 38

Can Realized Oil Price Changes Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt + ut, t = 1, ..., T, ∆st = …rst di¤erence of the log of the Canadian/U.S. Dollar exchange rate (Barclays) ∆pt = …rst di¤erence of the log of the oil price (West Texas Intermediate) ut = unforecastable error term We consider three frequencies: quarterly, monthly and daily 12/14/1984 to 11/05/2010, end of period.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 10 / 42

SLIDE 39

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 40

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T,

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 41

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T, Rolling "post out-of-sample" forecasts: see Meese-Rogo¤ (1983) and Andersen et al. (2003)

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 42

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T, Rolling "post out-of-sample" forecasts: see Meese-Rogo¤ (1983) and Andersen et al. (2003)

Choose an estimation window, R

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 43

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T, Rolling "post out-of-sample" forecasts: see Meese-Rogo¤ (1983) and Andersen et al. (2003)

Choose an estimation window, R Estimate parameter using last R observation

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 44

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T, Rolling "post out-of-sample" forecasts: see Meese-Rogo¤ (1983) and Andersen et al. (2003)

Choose an estimation window, R Estimate parameter using last R observation Get forecast using realized fundamentals, and forecast error

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 45

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T, Rolling "post out-of-sample" forecasts: see Meese-Rogo¤ (1983) and Andersen et al. (2003)

Choose an estimation window, R Estimate parameter using last R observation Get forecast using realized fundamentals, and forecast error Mimic forecaster in real time and roll through the data

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 46

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T, Rolling "post out-of-sample" forecasts: see Meese-Rogo¤ (1983) and Andersen et al. (2003)

Choose an estimation window, R Estimate parameter using last R observation Get forecast using realized fundamentals, and forecast error Mimic forecaster in real time and roll through the data

Predictive ability assessed by Diebold and Mariano’s (1995)

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 47

Can Realized Oil Price Changes Forecast Exchange Rates?

The Strategy:

∆st = α + β∆pt + ut, t = 1, ..., T, Rolling "post out-of-sample" forecasts: see Meese-Rogo¤ (1983) and Andersen et al. (2003)

Choose an estimation window, R Estimate parameter using last R observation Get forecast using realized fundamentals, and forecast error Mimic forecaster in real time and roll through the data

Predictive ability assessed by Diebold and Mariano’s (1995) Benchmarks are random walk without and with drift

Rossi (Duke) Oil Prices & Exchange Rates June 2011 11 / 42

SLIDE 48

Can Realized Oil Price Changes Forecast Exchange Rates in Daily Data? YES!

Rossi (Duke) Oil Prices & Exchange Rates June 2011 12 / 42

SLIDE 49

Can Realized Oil Price Changes Forecast Exchange Rates in Monthly/Quarterly Data? Barely...

Rossi (Duke) Oil Prices & Exchange Rates June 2011 13 / 42

SLIDE 50

Why Can We Find Predictive Ability? Is it the Fundamental?

Is it the choice of the fundamental? What if we use interest rates?

∆st = α + βit + ut, t = 1, ..., T,

Rossi (Duke) Oil Prices & Exchange Rates June 2011 14 / 42

SLIDE 51

Why Can We Find Predictive Ability? Is it the Fundamental?

Is it the choice of the fundamental? What if we use interest rates?

∆st = α + βit + ut, t = 1, ..., T, ∆it = logarithm of the interest rate di¤erential

Rossi (Duke) Oil Prices & Exchange Rates June 2011 14 / 42

SLIDE 52

Why Can We Find Predictive Ability? Is it the Fundamental?

Is it the choice of the fundamental? What if we use interest rates?

∆st = α + βit + ut, t = 1, ..., T, ∆it = logarithm of the interest rate di¤erential ut = unforecastable error term

Rossi (Duke) Oil Prices & Exchange Rates June 2011 14 / 42

SLIDE 53

Why Can We Find Predictive Ability? Is it the Fundamental?

Is it the choice of the fundamental? What if we use interest rates?

∆st = α + βit + ut, t = 1, ..., T, ∆it = logarithm of the interest rate di¤erential ut = unforecastable error term Canadian short-term interest rate is the daily overnight money market …nancing rate (Bank of Canada) and the U.S. short-term rate is the daily Federal funds e¤ective rate

Rossi (Duke) Oil Prices & Exchange Rates June 2011 14 / 42

SLIDE 54

Why Can We Find Predictive Ability? Is it the Fundamental?

Fundamental plays a big role: no predictive ability with interest rates!

Rossi (Duke) Oil Prices & Exchange Rates June 2011 15 / 42

SLIDE 55

Is the Predictive Ability Stable Over Time?

Predictive Ability mainly after 2004 in Daily data, very little or none in Monthly/Quarterly data...

Rossi (Duke) Oil Prices & Exchange Rates June 2011 16 / 42

SLIDE 56

Why Predictive Ability? Is it the Frequency or the Number of Observations?

Calculate the test in a way to make them comparable

Rossi (Duke) Oil Prices & Exchange Rates June 2011 17 / 42

SLIDE 57

Why Predictive Ability? Is it the Frequency or the Number of Observations?

Calculate the test in a way to make them comparable Select n. of observations for daily data = n. of observations for monthly and quarterly data.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 17 / 42

SLIDE 58

Why Predictive Ability? Is it the Frequency or the Number of Observations?

Calculate the test in a way to make them comparable Select n. of observations for daily data = n. of observations for monthly and quarterly data.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 17 / 42

SLIDE 59

Why Predictive Ability? Is it the Frequency or the Number of Observations?

Calculate the test in a way to make them comparable Select n. of observations for daily data = n. of observations for monthly and quarterly data. It is the frequency, not the number of observations!

Rossi (Duke) Oil Prices & Exchange Rates June 2011 17 / 42

SLIDE 60

Why Predictive Ability? Is it a Dollar E¤ect?

Both oil prices and the exchange rate are denominated in US Dollars

Rossi (Duke) Oil Prices & Exchange Rates June 2011 18 / 42

SLIDE 61

Why Predictive Ability? Is it a Dollar E¤ect?

Both oil prices and the exchange rate are denominated in US Dollars Consider oil prices and Canadian Dollar/British Pound exchange rate

Rossi (Duke) Oil Prices & Exchange Rates June 2011 18 / 42

SLIDE 62

Why Predictive Ability? Is it a Dollar E¤ect?

Both oil prices and the exchange rate are denominated in US Dollars Consider oil prices and Canadian Dollar/British Pound exchange rate

Rossi (Duke) Oil Prices & Exchange Rates June 2011 18 / 42

SLIDE 63

Why Predictive Ability? Is it a Dollar E¤ect?

Both oil prices and the exchange rate are denominated in US Dollars Consider oil prices and Canadian Dollar/British Pound exchange rate It is not a dollar e¤ect!

Rossi (Duke) Oil Prices & Exchange Rates June 2011 18 / 42

SLIDE 64

Robustness: Recursive Estimation

Results are robust to the use of a recursive window estimation procedure

Rossi (Duke) Oil Prices & Exchange Rates June 2011 19 / 42

SLIDE 65

Roadmap

I. Can Realized Oil Prices Forecast Exchange Rates?
II. Can Lagged Oil Prices Forecast Exchange Rates?
III. Other Commodities/Exchange Rates
IV. Are Non-linearities Important?

Rossi (Duke) Oil Prices & Exchange Rates June 2011 20 / 42

SLIDE 66

Can Lagged Oil Prices Forecast Exchange Rates?

The Regression:

Rossi (Duke) Oil Prices & Exchange Rates June 2011 21 / 42

SLIDE 67

Can Lagged Oil Prices Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt1 + ut, t = 1, ..., T,

Rossi (Duke) Oil Prices & Exchange Rates June 2011 21 / 42

SLIDE 68

Can Lagged Oil Prices Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt1 + ut, t = 1, ..., T, ∆st = …rst di¤erence of the logarithm of the Canadian/U.S. dollar exchange rate

Rossi (Duke) Oil Prices & Exchange Rates June 2011 21 / 42

SLIDE 69

Can Lagged Oil Prices Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt1 + ut, t = 1, ..., T, ∆st = …rst di¤erence of the logarithm of the Canadian/U.S. dollar exchange rate ∆pt = …rst di¤erence of the logarithm of the oil price

Rossi (Duke) Oil Prices & Exchange Rates June 2011 21 / 42

SLIDE 70

Can Lagged Oil Prices Forecast Exchange Rates?

The Regression:

∆st = α + β∆pt1 + ut, t = 1, ..., T, ∆st = …rst di¤erence of the logarithm of the Canadian/U.S. dollar exchange rate ∆pt = …rst di¤erence of the logarithm of the oil price ut = unforecastable error term

Rossi (Duke) Oil Prices & Exchange Rates June 2011 21 / 42

SLIDE 71

Can Lagged Oil Price Changes Forecast Exchange Rates in Daily Data?

Not on average, nor in monthly/quarterly data, but there might

Rossi (Duke) Oil Prices & Exchange Rates June 2011 22 / 42

SLIDE 72

Can Lagged Oil Price Changes Forecast Exchange Rates?

Yes, after taking into account instabilities in the relative forecasting performance

Rossi (Duke) Oil Prices & Exchange Rates June 2011 23 / 42

SLIDE 73

Can Lagged Interest Rate Di¤erentials Forecast Exchange Rates?

Never!

Rossi (Duke) Oil Prices & Exchange Rates June 2011 24 / 42

SLIDE 74

Roadmap

I. Can Realized Oil Prices Forecast Exchange Rates?
II. Can Lagged Oil Prices Forecast Exchange Rates?
III. Other Commodities/Exchange Rates
IV. Are Non-linearities Important?

Rossi (Duke) Oil Prices & Exchange Rates June 2011 25 / 42

SLIDE 75

Other Commodities: Norwegian Krone, contemp. price

Very strong predictive ability in daily data with contemp. prices

Rossi (Duke) Oil Prices & Exchange Rates June 2011 26 / 42

SLIDE 76

Other Commodities: Norwegian Krone, lagged price

Predictive ability with realized fundamentals robust, with lagged p sporadic

Rossi (Duke) Oil Prices & Exchange Rates June 2011 27 / 42

SLIDE 77

Other Commodities: S.A. Rand and Gold, contemp.

Very strong predictive ability in daily data with contemp. prices

Rossi (Duke) Oil Prices & Exchange Rates June 2011 28 / 42

SLIDE 78

Other Commodities: S.A. Rand and Gold, lagged p

Predictive ability with realized fundamentals robust, with lagged p sporadic

Rossi (Duke) Oil Prices & Exchange Rates June 2011 29 / 42

SLIDE 79

Other Commodities: Chilean Peso and Copper, contemp.

Very strong predictive ability in daily data with contemp. prices

Rossi (Duke) Oil Prices & Exchange Rates June 2011 30 / 42

SLIDE 80

Other Commodities: Chilean Peso and Copper, contemp.

Only predictive ability with realized fundamentals

Rossi (Duke) Oil Prices & Exchange Rates June 2011 31 / 42

SLIDE 81

Other Commodities: Australian $ and Oil, contemp.

Very strong predictive ability in daily data with contemp. prices

Rossi (Duke) Oil Prices & Exchange Rates June 2011 32 / 42

SLIDE 82

Other Commodities: Australian $ and Oil

Only predictive ability with realized fundamentals

Rossi (Duke) Oil Prices & Exchange Rates June 2011 33 / 42

SLIDE 83

Roadmap

I. Can Realized Oil Prices Forecast Exchange Rates?
II. Can Lagged Oil Prices Forecast Exchange Rates?
III. Other Commodities/Exchange Rates
IV. Are Non-linearities Important?

Rossi (Duke) Oil Prices & Exchange Rates June 2011 34 / 42

SLIDE 84

Are Non-linearities Important?

Consider a model with Asymmetries (Kilian and Vigfusson (2009), Alquist, Kilian and Vigfusson (2011))

Rossi (Duke) Oil Prices & Exchange Rates June 2011 35 / 42

SLIDE 85

Are Non-linearities Important?

Consider a model with Asymmetries (Kilian and Vigfusson (2009), Alquist, Kilian and Vigfusson (2011)) The exchange rate response is asymmetric in oil price increases and decreases: ∆st = α+ + β+∆pt + γ+∆p+

t + ut

(1) where ∆p+

t =

∆pt if ∆pt > 0 0 otherwise. .

Rossi (Duke) Oil Prices & Exchange Rates June 2011 35 / 42

SLIDE 86

Are Non-linearities Important?

Consider a model with Threshold e¤ects (Hamilton):

Rossi (Duke) Oil Prices & Exchange Rates June 2011 36 / 42

SLIDE 87

Are Non-linearities Important?

Consider a model with Threshold e¤ects (Hamilton): “large” changes in oil prices have additional predictive power for the nominal exchange rate: ∆st = αq + βq∆pt + γq∆pq

t + ut

where ∆pq

t =

∆pt if ∆pt > 80th quantile of ∆pt or < 20th quantile 0 otherwise.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 36 / 42

SLIDE 88

Non-linearities: Contemporaneous Price Model – Daily data

Some predictability in Threshold models but only for very large window sizes

Rossi (Duke) Oil Prices & Exchange Rates June 2011 37 / 42

SLIDE 89

Non-linearities: Contemp. Price Model, Monthly a& Quarterly data

Non-linear models are never better and sometimes signif. worse

Rossi (Duke) Oil Prices & Exchange Rates June 2011 38 / 42

SLIDE 90

Non-linearities: Lagged Price Model – Daily data

Non-linear models are never better and sometimes signif. worse

Rossi (Duke) Oil Prices & Exchange Rates June 2011 39 / 42

SLIDE 91

Non-linearities: Lagged Price Model – Monthly and Quarterly data

Non-linear models are never better and sometimes signif. worse

Rossi (Duke) Oil Prices & Exchange Rates June 2011 40 / 42

SLIDE 92

Conclusions

Our empirical results suggest that oil prices can predict the Canadian/U.S. dollar nominal exchange rate at daily frequency out-of-sample. However, the predictive ability is not evident at quarterly and monthly frequencies.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 41 / 42

SLIDE 93

Conclusions

Our empirical results suggest that oil prices can predict the Canadian/U.S. dollar nominal exchange rate at daily frequency out-of-sample. However, the predictive ability is not evident at quarterly and monthly frequencies.

When using contemporaneous realized daily oil prices, the predictive ability is robust to the choice of the in-sample window size and it does not depend on the sample period under consideration.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 41 / 42

SLIDE 94

Conclusions

Our empirical results suggest that oil prices can predict the Canadian/U.S. dollar nominal exchange rate at daily frequency out-of-sample. However, the predictive ability is not evident at quarterly and monthly frequencies.

When using contemporaneous realized daily oil prices, the predictive ability is robust to the choice of the in-sample window size and it does not depend on the sample period under consideration. When using lagged oil prices, the predictive ability is more ephemeral and only shows up in daily data after allowing the relative forecasting performance of the oil price model and the random walk to be time-varying.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 41 / 42

SLIDE 95

Conclusions

Our empirical results suggest that oil prices can predict the Canadian/U.S. dollar nominal exchange rate at daily frequency out-of-sample. However, the predictive ability is not evident at quarterly and monthly frequencies.

When using contemporaneous realized daily oil prices, the predictive ability is robust to the choice of the in-sample window size and it does not depend on the sample period under consideration. When using lagged oil prices, the predictive ability is more ephemeral and only shows up in daily data after allowing the relative forecasting performance of the oil price model and the random walk to be time-varying.

Both out-of-sample and in-sample analyses suggest that frequency of the data is important to detect the predictive ability of oil prices

Rossi (Duke) Oil Prices & Exchange Rates June 2011 41 / 42

SLIDE 96

Conclusions and Future Work

Non-linearities do not signi…cantly improve upon the simple linear oil price model.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 42 / 42

SLIDE 97

Conclusions and Future Work

Non-linearities do not signi…cantly improve upon the simple linear oil price model. Overall, reason why existing literature has been unable to …nd evidence of predictive power in oil prices is that they focused on low frequencies where the short-lived e¤ects

f oil price changes wash away

Rossi (Duke) Oil Prices & Exchange Rates June 2011 42 / 42

SLIDE 98

Conclusions and Future Work

Non-linearities do not signi…cantly improve upon the simple linear oil price model. Overall, reason why existing literature has been unable to …nd evidence of predictive power in oil prices is that they focused on low frequencies where the short-lived e¤ects

f oil price changes wash away

At the same time, our results also raise interesting questions.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 42 / 42

SLIDE 99

Conclusions and Future Work

Non-linearities do not signi…cantly improve upon the simple linear oil price model. Overall, reason why existing literature has been unable to …nd evidence of predictive power in oil prices is that they focused on low frequencies where the short-lived e¤ects

f oil price changes wash away

At the same time, our results also raise interesting questions. Does the Canadian/U.S. dollar exchange rate respond to demand or supply shocks to oil prices? See Kilian (2009). However, unfeasible at the daily frequency.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 42 / 42

SLIDE 100

Conclusions and Future Work

Non-linearities do not signi…cantly improve upon the simple linear oil price model. Overall, reason why existing literature has been unable to …nd evidence of predictive power in oil prices is that they focused on low frequencies where the short-lived e¤ects

f oil price changes wash away

At the same time, our results also raise interesting questions. Does the Canadian/U.S. dollar exchange rate respond to demand or supply shocks to oil prices? See Kilian (2009). However, unfeasible at the daily frequency. We leave these issues for future research.

Rossi (Duke) Oil Prices & Exchange Rates June 2011 42 / 42