Professional Documents
Culture Documents
Irm Ecn405 Researchpaper1
Irm Ecn405 Researchpaper1
Ian McGinnis
Spring 2016
ABSTRACT. This paper employs time-series regression and OLS estimates in order to analyze the
factors which determine the USD-GBP exchange rate from 1970 through present. Through
literature review and economic consideration, exchange rate was regressed on GDP ratio, money
supply ratio, unemployment rate ratio, and the previous years exchange rate. The results suggest
that all but unemployment rate ratio are significant variables in determining the current exchange
rate. The model was determined to be a significantly good fit, overall, and few issues were found
through testing. Overall, results suggest that past years exchange rates are modeled very good
with this model and future work building on this model could be significant.
I. Introduction
The relationship between the United States and Great Britain is an important one on many fronts.
The relationship, deemed The Special Relationship has been one of the most important foreign
relationships for both the U.S. and G.B. for the past century (Harris, 2006). While this affiliation
could be outlined in pages upon pages of literature, there is a single component of it that affects
millions of people each day from the big banks to individual consumers, the exchange rate.
Exchange rates are determined in the foreign exchange market (FOREX), a market
composed of every currency from every country with trading 24 hours a day from Monday-
Friday, similar to stocks. This market is by far the largest in the world with trading totaling
roughly $1.9 trillion a day (Levinson, 2006). Books could be written about this fascinating
market, but the interest of this paper lies in one simple point, what determines these exchange
rates?
The objective of this paper is to use macroeconomic variables and a time-series model in
order to test the existing literature of exchange rate theory. By creating a reliable model that
follows both sound economic theory and passes tests to be a statistically consistent model, an
analysis of the factors that determine the USD-GBP exchange rate will be provided. I believe this
is an important (and interesting) project because works with similar scopes may be instrumental
in order to provide policy recommendations. Also, if development of these models could become
more reliable, personal fortunes are able to be made in the FOREX market (but you have to get
there first!)
Empirical studies that have tested various exchange rates and their determining variables
have concluded with mixed results. This section will outline the economic literature that was
crucial to this paper. The following papers provided important economic theory background and
Obeng, Enu, Osei-Gyimah, and Opoku, using data from 1980-2012. The authors employed
multiple statistical methods such as ordinary least squares regression, graphing of a scatter plot,
and PPMC. The simple linear relationship between the Ghanaian Cedi and their GDP growth rate
was determined to be positive and strong. Also, the authors did tests for autocorrelation,
heteroskedasticity, and multicollinearity, where the former two were found to be absent. This
paper will provide guidance in terms of economic theory, due to the finding of the positive
relationship between GDP and exchange rate (Attah-Obeng, Enu, Osei-Gyimah, and Opoku
2013). Also, the various tests they performed will be similar to tests that will be run in this paper
Gregory P. Hoppers paper titled What determines the Exchange Rate: Economics
Factors or Market Sentiment? was useful in explaining the current monetary model which is
used to explain exchange rates. The model can be broken down to the simple explanation that
exchange rates are determined by the relative price levels of the two countries. What determines
these price levels is where the problem gets interesting. The monetary model uses the supply and
demand of money to explain the price levels, which implies that a faster growing money supply
in one country will depreciate that countries currency. This monetary model has been supported
in a paper by Frenkel (1978). This paper provides me with the economic backing that relative
money supplies are an important factor in the monetary model approach to determining exchange
rates, even if Hooper has some disagreements with the usefulness of this model (Hopper 1997).
Finally, crucial findings for the determination of the model that will be used in this paper
were explained by Bansal, Chereddy, Mehta, and Susanto in 2013. In an analysis of the Indian
Rupee exchange rate, using regression analysis, the authors tested many variables in order to find
the most significant variables in determining the exchange rate. The variables that were found to
be the best (in terms of significance) were unemployment rate ratio, inflation rate ratio, interest
rate ratio, government debt/GDP ratio, budget balance ratio, and government bond yield ratio.
(Bansal, Chereddy, Mehta, and Susanto, 2013). Therefore, this article was of great use in order to
A time-series model will be estimated for this analysis. A lagged variable is used when the impact
of the independent variable can be seen over multiple time periods (Studenmund, 2011, p. 405).
It should be noted that from here on out in this paper, any mention of GDP should be
assumed to be a calculation of GDP per capita. All GDPs are real (converted to a base year) and
to find GDP per capita the real GDP is divided by the population for that given year.
With the independent variables of GDP Ratio, Money Supply Ratio, and Unemployment
Rate Ratio, macroeconomics shows that there should be a lag between these indicators and their
effect on the economy. Therefore, each variable will be lagged by a single time period (1 year), in
order to account for this. Also, it is obvious to assume that the exchange rates from years past
would have an effect on the current interest rate, therefore a dynamic model will be used. The
following model, Equation III.1 will be the basis for the analysis:
Y t = 0 + 0 X t 1+ Y t1+ t (III.1)
The variables discussed above will be represented by Xs and the dependent variable,
exchange rate will be Y. The following (Equation III.2) is the model that will be estimated in the
final analysis and the null and alternative hypotheses (and a brief reasoning) for each independent
GDPRATIO t1
H 0: 0 0
HA: 0 < 0
This variable, the ratio of US/GB GDP, I would expect to have a negative coefficient. As
one country grows faster than the other, the currency for that country should appreciate, therefore,
as US_GSD/GB_GDP gets larger, the exchange rate of USD/GBP would go down, showing
MS RATIOt 1
H 0: 1 0
HA: 1 > 0
This variable is the ratio of the money supplies of the dollar and the pound. As one money supply
increases relative to the other, that currency should be depreciating because of the increased
inflation associated with money supply growth. Therefore, it is expected that 1 would be
positive.
UR RATIOt 1
H 0: 2 0
HA: 2 <0
This variable is the ratio of the rate of unemployment of the U.S. to that of the U.K. As this ratio
EX RATEt1
H 0: 0 0
HA: 0 < 0
Last years exchange rate should be positively correlated with the current rate, so it is expected
Exchange rate (EX_RATE) and Gross Domestic Product (GDP_RATIO) data was found on a site
called Measuring Worth. This website has monetary data for centuries back and for many
countries, it was instrumental for this paper. United States money supply data was found on the
St. Louis Fed website and UK money supply data was located on the Bank of England website
(MS_RATIO_LAG). U.S. unemployment rate data is on the website for the Bureau of Labor
Statistics and this data for the U.K. is on the Office of National Statistics site (UR_RATE_LAG).
Data for the four independent variables (with lags) and the dependent variable used in the final
USD/GBP. GDP Ratio has one less observation because the 2016 GDP has not yet been
calculated.
Using the data from Table 1 and equation III.3 this model was estimated and Table 2
G
MS_RATIO_LAG .023 .010 2.312 .026**
UR_RATE_LAG -.131 .117 -1.124 .268
EX_RATE_LAG .638 .118 5.425 .000**
R = .884 R2 = .781 F = 34.804
* - indicates significance at the 10% level
The R2 is .781 meaning 78.1% of the total variation in this model is explained by the
independent variables. This is a strong correlation and shows that the given variables do a good
job of explaining the dependent variable. Also, as noted by the asterisks in Table 2 in the
that this variable fit well due to chance and that it most likely represents the population rather
than just the sample. This is also true of EX_RATE_LAG, which is expected because of the strong
association between a variable and its lag. Also, GDP_RATIO_LAG has a significance level of .
065 which means that there is only a 6.5% chance that the observation reflects only the sample
and not the population. These numbers suggest that the null hypotheses for MS_RATIO_LAG,
EX_RATE_LAG, and GDP_RATIO_LAG can be rejected at the 10% level. The only null
The F-test is a test for overall goodness of fit. Using Table B-3 on p. 589 of Studenmund,
a critical F-statistic for 4 independent variables and 43 observations is 3.83 < 34.804. Therefore,
To test for multicollinearity, VIFs were calculated in SPSS. These are variance inflation
factors which test the amount a certain variable can be explained by all other independent
variables in the given equation (Studenmund, 2011, p. 259). Table 3 gives the VIF values for each
Table 3-VIFs
Variable VIF
GDP_RATIO_LAG 3.599
MS_RATIO_AG 6.518
UR_RATE_LAG 2.592
EX_RATE_LAG 2.765
A VIF > 5 is considered to show severe multicollinearity. This would suggest that the
variable MS_RATIO_LAG has multicollinearity. However, due to the strong significance of the
variable, the strong economic theory that supports it, and the little redundancy that is noticed, it
has been decided that the variable will stay in the equation. Every other variable has a good VIF,
constant variance of the error term. In order to test for heteroskedasticity, the Park test will be
conducted with a proportionality factor of time. This proportionality factor is a good choice
because what needs to be analyzed is if as the time-series model goes on, the variance in the
residuals has a changing variance or not. Therefore, the natural log of the squared residuals will
be regressed with the year variable. The regression results are noted in Table 4:
E
LN(YEAR) -112.867 52.747 -2.140 .038
R2=.098
At a 5% level, the coefficient of this variable is significant. Therefore, we can reject the
null hypothesis of homoskedasticity. This does not prove that there is heteroskedasticity in our
model, however it does make a pretty strong case for it. This could possibly mean our estimates
are not BLUE, but it cannot be sure. A graph shows somewhat constant variance when residuals
Graph 1 provides a case for homoskedasticity due to the constant variance appearance of
Serial correlation occurs when the error terms from different time periods are correlated.
In order to test for serial correlation in a time-series model, a Lagrange Multiplier test will be
used, which uses residuals as the dependent variable and independent variables from the original
equation and a lagged term of the residuals as the independent variables. Then, using OLS and t-
tests, testing the null hypothesis that the coefficient of the lagged residuals term will reveal serial
correlation. With LM=NR2, LM=42(.256) =10.752. From Table B-8 of Studenmund, the critical
value of this test is 11.07 (5 def., 5%), so we can fail to reject the null hypothesis that the
coefficient of the lagged residuals is 0, so we can say there is a good chance there is no serial
When time-series variables are nonstationary, one of the properties of stationarity is not
met. With this test, it was concluded that we fail to reject the null hypothesis of nonstationarity
(with t=1-.294 and significance =.203). Therefore, a cointegration test was conducted, which is a
Dickey-Fuller test on the residuals of the original model. With t=-4.395 and significance = .000,
we can reject the null hypothesis of unit root in the residuals, and conclude that there is
cointegration in our model and the OLS estimates are not spurious.
V. Conclusions
This analysis deemed given variables as significant under initial estimation. Also, tests such as
VIFs and F-tests show low levels of multicollinearity and high goodness of fit. However, there
were a few problems that could be fixed with more time, expertise, and resources in the future.
One major problem found in the final model estimation was nonstationarity. This could
be expected due to the nature of macroeconomic variables. By inspecting the data, it can be
Dickey-Fuller test on the residuals, cointegration was decided to be present in the model. Another
small problem was the possibility of heteroskedasticity in this model. This could be a
specification issue that is unnoticed or could be fixed with HC standard errors, hopefully making
Future work could build on this model. Suggestions such as fixing the nonstationarity,
more observations, and more independent variables may be able to provide an estimation that is
statistically significant and passes a multitude of criteria. Also, a further analysis could be
expansions. This could help determine the exchange rate depending on the state of the economy.
Gregory P. Hopper hypothesized that the exchange rate may be determined by market
sentiment, and it could be a self-fulfilling prophecy. If this is the case, then market expectations
and consumer expectations will be the determining factors of the exchange rate, which is much
The policy implications of exchange rate determination could be huge for smoothing
monetary and fiscal policies. This could eliminate (or decrease) economic recessions in the future
Bansal, Rishav, Chereddy, Manusha, Mehta, Kairavi, and Susanto, Hendry. What factors
Clegg, Richard. "Time Series: Unemployment Rate." Unemployment Rate - Office for National
Statistics. Office for National Statistics, 20 Apr. 2016. Web. 03 May 2016.
"Databases, Tables & Calculators by Subject." Bureau of Labor Statistics Data. Bureau of Labor
Enu, Prudence Attah-Obeng Patrick, and F. Osei-Gyimah CDK Opoku. "An Econometric
Analysis of the Relationship between Gdp Growth Rate and Exchange Rate in Ghana."
Harris, Robin. Beyond Friendship: The Future of Anglo-American Relations. Washington, DC:
Hopper, Gregory P. What Determines the Exchange Rate: Economic Factors or Market
Johnston, Louis and Williamson, Samuel H., "What Was the U.S. GDP Then?" MeasuringWorth,
URL: http://www.measuringworth.org/usgdp/
Levinson, Marc. Guide to Financial Markets. Princeton, NJ: Bloomberg, 2006. Print.
"M2 Money Stock." St. Louis - FRED. 21 Apr. 2016. Web. 03 May 2016.
Officer, Lawrence H., "Dollar-Pound Exchange Rate From 1791," MeasuringWorth, 2016
URL: http://www.measuringworth.com/exchangepound/
Studenmund, A. H. Using Econometrics: A Practical Guide. Boston, MA: Addison Wesley, 2011.
Williamson, Samuel H., "What Was the U.K. GDP Then?" MeasuringWorth, 2016
URL: http://www.measuringworth.com/ukgdp/