1 Exchange Rate Predictability: Fundamentals versus Technical Analysis Ibrahim Jamali Ehab Yamani Associate Professor of Finance Visiting Assistant Professor of Finance American University in Beirut Jackson State University [email protected][email protected]+1 (817) 673-6883 January 17, 2018 Abstract This paper compares the predictive ability of macroeconomic variables with that of technical indicators in generating out-of-sample forecasts for exchange rate returns. We use four measures for the macroeconomic variables based on the standard theories of exchange rate determination: uncovered interested rate parity, purchasing power parity, monetary fundamentals, and Taylor rule. We also use three popular trend following technical trading strategies in foreign exchange markets, namely, simple moving averages (MA) indicator, momentum (MOM) oscillator, and relative strength index (RSI). JEL classification: G14 G15 F31 Keywords: Exchange Rate Predictability; Forecasting; Fundamentals; Technical Trading.
29
Embed
Exchange Rate Predictability: Fundamentals versus ...fmaconferences.org/SanDiego/Papers/Jamali Yamani.pdf2 Exchange Rate Predictability: Fundamentals versus Technical Analysis Abstract
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Exchange Rate Predictability: Fundamentals versus Technical Analysis
Ibrahim Jamali Ehab Yamani
Associate Professor of Finance Visiting Assistant Professor of Finance
American University in Beirut Jackson State University
Predicting exchange rate movements is undoubtedly a daunting task. Several anomalies, which
characterize the state of international finance, exacerbate the difficulty in predicting exchange
rates. First, the literature establishes the existence of a βdisconnectβ between macroeconomic (and
monetary) fundamentals and exchange rate movements (Bacchetta and van Wincoop, 2006; Sarno
and Taylor, 2002; Evans and Lyons, 2002). The βexchange rate disconnectβ puzzle implies that
exploiting the informational content of fundamentals does not yield forecast improvements vis-Γ -
vis the random walk which Meese and Rogoff (1983) show is a stringent benchmark for assessing
the out-of-sample predictability in exchange rate changes.1
Second, the existing literature provides ample evidence against Uncovered Interest Parity
(UIP). The absence of empirical support for UIP, according to which exchange rate changes should
be equal to the interest rate differential between two countries, is closely connected to the βforward
premium puzzleβ which is a another widely researched anomaly in international finance. Under
1 A more positive assessment of the predictive power of fundamentals is provided in Engel, Mark and West (2007)
and Li, Tsiakas and Wang (2015). Engel, Mark and West (2007) argue that the random walk is an unnecessarily
stringent benchmark against which to compare the predictive ability of fundamentals.
3
risk neutrality and rational expectations, one of the implications of UIP is that the forward rate is
an unbiased predictor of the future spot rate (Li, Tsiakas and Wang, 2015). Nonetheless, the
considerable evidence on the βforward premium puzzleβ for the currencies of developed economies
implies that forward rates are biased predictors of the future spot rate.2
Perhaps the best assessment of the state of the exchange rate predictability literature is given
in Della Corte and Tsiakas (2012). The authors argue that, since Meese and Rogoff (1983), the
literature has come βfull circleβ from finding no predictability, to uncovering predictability at long
horizons (Mark, 1995) and then back to failing to find evidence of predictability in currency
exchange rates (Cheung, Chinn and Pascual, 2005). After coming full circle, more recent
contributions to the literature provide compelling evidence of predictability in exchange rate
movements at short horizons (Molodtsova and Papell, 2009; Li, Tsiakas and Wang, 2015;
Anatolyev, Gospodinov, Jamali and Liu, 2017).
The presence of short-horizon predictability in exchange rate movements is consistent with the
widespread popularity of technical analysis among currency traders. At its core, technical trading
attempts to discern and exploit trends in asset prices. There have been several studies that examined
the profitability of technical trading strategies, such as Gencay (1999), LeBaron (1999), Lee et al.
(2001), Neely and Weller (2013), Raza et al. (2014), Katusiime et al. (2015), and Zarrabi et al.
(2017). While traders have long used technical trading rules in the foreign exchange market,
academic research provides somewhat mixed evidence regarding the efficacy and profitability of
technical analysis. Neely, Weller and Dittmar (1997) and Neely and Weller (2001) employ the
genetic programming algorithm to identify technical trading rules which generate economically
significant out-of-sample profitability in the foreign exchange market. Both studies report
2 For comprehensive reviews of the forward premium anomaly literature, see Engel (1996, 2015).
4
supportive evidence of technical trading rulesβ ability to generate out-of-sample profits.3 GenΓ§ay
(1999) provides evidence that simple technical trading rules outperform the random walk out-of-
sample. In contrast, Neely and Weller (2003) find that accounting for transaction costs erodes the
profitability of technical trading rules when high-frequency data on four currencies are employed.4
In a thorough review of the literature, Park and Irwin (2007) synthesize the conclusions of recent
studies as being supportive of the profitability of technical analysis in foreign exchange and equity
markets.5 The profitability of technical trading rules is not surprising in light of the recent
contribution by Levine and Pedersen (2016) who show that the moving average crossovers, which
are a popular technical indicator, as well as other filters, are capable of detecting time series
momentum.
This paper examines the predictive power of fundamentals, technical indicators as well as high
frequency measures of risk and commodity prices in predicting the currency rate movements of
developing countries. Such an exploration contributes to the literature along several lines. First,
the existing literature examines the predictive power of fundamentals for the developed countriesβ
currencies. However, to the best of our knowledge, no such exploration is systematically
undertaken for developed countries currencies. In fact, the literatureβs findings concerning the
predictive power of fundamentals need not generalize to developing countriesβ currencies. For
3 Sweeney (1986) provides evidence of the effectiveness of technical analysis in foreign exchange market. Neely and
Weller (2001) find that exploiting information on the Federal Reserveβs intervention in the currency market enhances
the profitability of the trading rules identified via the genetic programming algorithm. In a related paper, Neely (2000)
finds that the profitability of trading rules cannot be ascribed solely to central bank intervention. Rather, the author
provides evidence that technical trading rule are profitable before the central bank intervention. 4 The conclusions from the strand of research examining linear and non-linear implementations of technical analysis
in equity markets (Sweeney, 1988; Gençay 1996, 1998; Gençay and Stengos, 1998; Neely, Rapach, Tu and Zhou,
2014) lean towards uncovering out-of-sample profitability. However, a number of studies (see, for example, Ready,
2002; Bessembinder and Chang, 1998) starting with Sullivan, Timmermann and White (2000) argue that researchers
should be mindful of data measurement and, more importantly, data snooping biases before drawing such a conclusion. 5 However, the authors themselves do not find favorable results from applying technical trading rules to U.S. futures
(Park and Irwin, 2010).
5
instance, while the forward premium anomaly is a staple of the developing countriesβ currencies,
existing research (Bansal and Dahlquist, 2000; Frankel and Poonawala, 2010) suggests that the
forward premium puzzle is much less pronounced for the currencies of developing economies.
This, in turn, implies that while the forward rate may still be a biased predictor of future exchange
rate movements of emerging markets, it will indicate the correct directional change in the exchange
rate movements (Frankel and Poonawala, 2010).6
Second, this paper is the first to examine the predictive ability of technical indicators for
emerging market economies. The lower turnover in emerging market currencies (Bank of
International Settlements, 2016) and the lesser competition among traders might imply that
technical trading indicators generate profitable signals. This latter view is echoed in Frankel and
Poonawala (2010) who assert that βEmerging market currencies probably have more easily-
identified trends of depreciation than currencies of advanced countriesβ.
The remainder of the paper is organized as follows. Data and variables are set forth in section
2. Section 3 examines the in-sample analysis of exchange rate returns using both fundamental and
We collect the WMR/Reuters spot and one-month forward exchange rates for a cross-section
of twenty-three developing and developed economies against the United States Dollar (USD).
Thirteen countries in our sample are classified by the World Bank and the International Monetary
Fund as developing whereas ten are developed countries. More specifically, our cross-section
6 It is interesting to note, in this context, that Bansal and Dahlquist (2000) relate the attenuation of the forward premium
puzzle for emerging market economies to macroeconomic fundamentals such as per capita GNP, average inflation
rates and inflation volatility. This implies that fundamentals might possess predictive power for the exchange rate
changes of developing countries.
6
comprises the following developing country currencies: the Mexican New Peso (MXN), Hong
Kong Dollar (HKD), Indian Rupee (INR), Indonesian Rupiah (IDR), Philippine Peso (PHP),
Kuwaiti Dinar (KWD), New Taiwan Dollar (TWD), Saudi Riyal (SAR), Singapore Dollar (SGD),
Thai Baht (THB), Czech Koruna (CZK), Hungarian Forint (HUF) and South African Rand
(ZAR).7 In order to benchmark our results against those of influential studies in the literature (Della
Corte and Tsiakas, 2012; Li, Tsiakas and Wang, 2015; Burnside et al., 2011a; Lustig et al., 2011;
Daniel et al., 2017; Bekaert and Panayotov, 2016), we also provide results for the ten most liquid
currencies of developed countries in the world. The G10 currencies we include in our cross-section
are: British pound (GBP), Canadian dollar (CAD), Swiss franc (CHF), Euro (EUR), Japanese yen
(JPY), Australian dollar (AUD), New Zealand dollar (NZD), Swedish krona (SEK) and Norwegian
krone (NOK).
Our data spans the period from December 1996 to June 2017. Our starting date is dictated by
the availability of one-month forward rate data for the developing currencies while our sample is
contained to end in June 2017 given that Gross Domestic Product (GDP) (see section 2.2) data are
available with a time lag. The monthly spot and one-month forward quotes are sample from daily
data as the last observation of the month. The returns on currency i in month t is given by: βπ ππ‘ =
ln(πππ‘) β ln(πππ‘β1) = π ππ‘ β π ππ‘β1 for π = 1,2, β¦ ,23 where itS denotes the exchange rate expressed
in terms of US dollar price of a unit of the foreign currency. The return on currency i is the
dependent variable in our predictive models.
7 Hong Kong, Thailand, and Saudi Arabia have pegged their currencies to the USD during part of our sample period.
However, we elect to keep them in our cross-section, as in Verdelhan (2017) and Lustig, Roussanov, and Verdelhan
(2011), because their forward prices are not inconsistent with covered interest rate parity.
7
2.2. Macroeconomic data
We obtain macroeconomic data for the twenty-three countries comprising our cross-section
from Datastream. More specifically, estimation and prediction from the models with fundamentals
requires data on GDP, inflation and the money supply of each of the countries. We collect data on
the seasonally adjust GDP and non-seasonally adjusted M1 as a measure of the money supply and
Consumer Price Index (CPI) of each of the countries.8 As noted in the online appendix, the source
of the macroeconomic data are the central bank and national statistical agencies for each of the
countries. In specific, the money supply data are collected by the central bank while the source of
the GDP and CPI data are the national statistical agencies of each of the countries.
Following Della Corte and Tsiakas (2012), we deseasonalize M1 by implementing the
procedure of Gomez and Maravall (2000). We also use Gomez and Maravall (2000)βs approach to
seasonally adjust the CPI for each of the countries. Given that GDP is only available at the
quarterly frequency (for all the countries), we linearly interpolate the GDP series from to obtain
data at the monthly frequency using the Chow and Lin (1971) procedure.9 GDP and M1 figures
are expressed in USD using the spot exchange rate against the USD. The CPI data for New Zealand
and Australia are available only at the quarterly frequency so we linearly interpolate the CPI for
the latter two countries using the Chow and Lin (1971) procedure.
We should highlight some important data gaps for the developing economies. Data on the CPI
are not available for India and Kuwait and Saudi Arabia while we could not obtain GDP data for
8 The Datastream mnemonic for each of the series can be found in the online appendix. 9 We closely follow Table 1 of Della Corte and Tsiakas (2012) when collecting the M1 data for the developed
countries. However, we opt not to reply on the industrial production indexes as they do. Our choice is driven by the
fact that the industrial production data are not available for many of the developing countries. Given that we would
like to compare the predictive performance of the different models for developing and developed countries on an equal
footing, we instead use GDP numbers for the developed economies and interpolate these. The same considerations
drive us to use the national statistical agenciesβ CPI indexes for all the countries instead of relying on the OECD CPI
data used in Della Corte and Tsiakas (2012). The OECD data are available only for the developed economies.
8
India. The GDP data for Indonesia are available only starting in 2011:Q2 while those of the
Philippines and Turkey are available only starting 1998:Q1. The M1 data for Thailand are available
only starting November 2015 while those of Turkey are available starting December 2005. In light
of these data constraints, we are unable to estimate and predict from some of the models with
fundamentals for these currencies.
3. Econometric Methodology
This section describes the modelling approach that we follow. Throughout our analysis, the
returns on currency i in month t is the dependent variable in our predictive regression models, and
given by the change in the log of spot exchange rate (βπ ππ‘+1 = ππ(πππ‘+1) β ππ(πππ‘)) for π =
1,2,3, β¦ . ,23, where πππ‘ denotes the nominal spot exchange rate expressed in terms of US dollar
price of a unit of the foreign currency. We initially present the predictive regression models we
use to empirically examine the predictive power of fundamentals and technical indicators in
predicting the currency rate movements, βπ ππ‘+1, and then we describe the statistical procedures we
use to evaluate our predictive regression models against the benchmark random walk (RW) model.
3.1. The Predictive Power of Fundamentals
Since the seminal contribution of Meese and Rogoff (1983), the RW model constitutes the
benchmark against which the statistical accuracy of exchange rate forecasting models is assessed.
Like Della Corte and Tsiakas (2012), most of our predictive models are cast within the general
framework of a predictive regression. The simple predictive regression is given by:
Under UIP, the null hypothesis 1,0:0 H is not rejected.
Li, Tsiakas and Wang (2015) note that the implications of rejecting UIP are twofold. The first
is that the forward rate is a biased predictor of the future spot rate. The sizeable literature on the
forward premium anomaly for developed currencies provides empirical evidence, which confirms
that the forward rate is a biased predictor of the future spot rate. While abundant empirical evidence
against UIP exists for the currencies of developed currencies, existing studies (Bansal and
Dahlquist, 2000; Frankel and Poonawala, 2010) indicate that the anomaly is much less pronounced
for the currencies of developing economies. This potentially makes the forward rate a useful
predictor of the future spot rate. The second implication of rejecting UIP is that the exchange rate
change is not equal to the interest rate differential.
3.1.3. Purchasing Power Parity
A Purchasing Power Parity (PPP) exchange rate guarantees that a unit of the currency has the
same purchasing power in two economies (Sarno and Taylor, 2002).12 As Mark (2001) notes, the
10 CIP is typically tested using the regression ππ‘ β π π‘ = πΌ + π½(ππ‘ β ππ‘
β) + ππ‘. If CIP holds, the null hypothesis π»0: πΌ =0, π½ = 1 should not be rejected. For studies providing empirical support for at high frequencies prior to the financial
crisis see, for example, Akram, Rime and Sarno (2008) and Fong, Valente, Fung (2010). Recent contributions to the
literature provide evidence of short-lived deviations from CIP during and after the financial crisis (Baba and Packer,
2009; Du, Tepper, and Verdelhan, 2016; Borio, McCauley, McGuire, Sushko, 2016; Mancini Griffoli and Ranaldo,
2011). 11 Studies which use the Fama (1984) approach include Froot and Thaler (1990), Baillie and Bollerslev (1989,2000),
Bansal and Dahlquist (2000), Frankel and Poonawala (2010) and Ahmad, Rhee and Wong (2012). 12 A more elaborate and accurate statement of a PPP exchange rate is one which βwould equate the two relevant
national price levels if expressed in a common currencyβ (Sarno and Taylor, 2002).
11
commodity-arbitrage view of PPP in Samuelson (1964) requires that the Law of One Price (LOOP)
hold for tradeable goods. The PPP hypothesis can be tested using the regression:
πβ1π=0 ) for each currency pair.14 An upward (downward) trend is usually
13 Yet another potential explanation of the feeble link between fundamentals and exchange rate changes is
nonlinearities. Sarno, Valente and Leon (2006) provide evidence of nonlinearities in UIP. 14 The 1-month/9-month MA strategy is very commonly used by currency traders and by many academic scholars
(e.g., Gencay, 1999). Further, LeBaron (1999) shows that trading rule profitability is not overly sensitive to the actual
length of the moving average.
13
identified when the spot rate is greater (less) than the moving average. Buy and sell signals are
3.3. Statistical Evaluation of Predictive Regressions
15 We calculate first the monthly relative strength π π measured as the ratio of total average gains to total average
losses (π π = π΄π£. πΊππππ π΄π£. πΏππ π ππ β ) for each currency pair. Average gains (losses) are calculated by totaling all
gains (losses) from the past 14 months and dividing by 14, where monthly gain (loss) is determined if the spot rate in
the current month is higher (lower) than the previous monthβs spot rate. The RS is then converted to an index value
that ranges between 0 and 100, using the following equation: π ππΌ = 100 β [100/(1 + π π].
14
We first run out-of-sample (OOS) monthly forecasts by estimating our predictive regression
models using fundamental variables (βοΏ½ΜοΏ½ππ‘+1 = οΏ½ΜοΏ½ + οΏ½ΜοΏ½π₯ππ‘) as well as technical indicators(βοΏ½ΜοΏ½ππ‘+1 =
οΏ½ΜοΏ½ + οΏ½ΜοΏ½π§ππ‘), where οΏ½ΜοΏ½ and οΏ½ΜοΏ½ are the OLS estimates computed from regressing currency returns on a
constant and one of our predictive regressors. We obtain the OOS monthly forecasts using rolling
regressions by restimating the model parameters every time we increase the beginning and ending
dates by a new monthly observation, using a fixed window equal to 140 observations. More
specifically, our first OOS one month ahead forecast is for September 2008, using an IS period
from January 1997 to August 2008, and our last OOS forecast is for August 2017, using an IS
period from August 2006 to July 2017. This exercise produces 108 OOS forecasts.16
We then evaluate the out-of-sample predictive power of the above empirical exchange rate
models by comparing the forecasts given by the fundamental-based regressors (in equation 1) and
the technical indicators (in equation 6) against the benchmark random walk model (in equation 9).
The comparative performance of the models is assessed using two statistics commonly used in the
literature: the out-of-sample R-square of Campbell and Thompson (2008), and the mean squared
forecast error (MSFE) adjusted statistic of Clark and West (2007). First, the Campbell and
Thompson (2008) out-of-sample R-square (π πππ 2 ) measures the proportional reduction in MSFE
for the one-month ahead conditional forecasts (βοΏ½ΜοΏ½π‘+1|π‘) using both fundamental and technical
indicators (i.e., where π½ β 0), relative to the one-month ahead unconditional forecast
(βοΏ½Μ οΏ½π‘+1|π‘) using the random walk model (i.e., where π½ = 0). Mathematically, the π πππ 2 is given by
π πππ 2 = 1 β
πππΉπΈ(βοΏ½ΜοΏ½π‘+1|π‘)
πππΉπΈ(βοΏ½Μ οΏ½π‘+1|π‘) (9)
16 Our IS sample period is comparable to Li, Tsiakas, and Wang (2015) who use 11 year time horizon as IS period.
The motivation for choosing August 2008 as the ending date of our IS period is that a number of previous studies
(e.g., Li, Tsiakas, and Wang, 2015; Buncic and Piras, 2016) find that there is a change in the predictability in the pre
Lehman Brothers collapse period compared to the Lehman collapse period which started in September 2008.
15
A positive (negative) π πππ 2 thus indicates that the predictive regression forecast model outperform
(underperform) the benchmark RW model. Second, the Clark and West statistic tests the null
hypothesis that the π πππ 2 is less than or equal to zero, so that MSFE of the RW model is less than
or equal MSFE of the alternative model.
4. Empirical Results
A natural starting point for our analysis is a comparison of exchange rate changes across
developed and emerging countries. Table 1 presents descriptive statistics for the study variables:
Lee, C.I., Pan, M, Liu, Y.A., 2001. On Market Efficiency of Asian Foreign Exchange rates:
Evidence from a joint variance ratio test and technical trading rules. Journal of International
Financial markets, Institutions and Money 11, 199-214.
18
Li, J., Tsiakas, I., Wang, W., 2015. Predicting Exchange Rates Out of Sample: Can Economic
Fundamentals Beat the Random Walk? Journal of Financial Econometrics 13 (2), 293-341.
Lustig, H., Roussanov, N., Verdelhan, A., 2011. Common risk factors in currency markets. Review
of Financial Studies 24 (11), 3731-3777.
Mark, N. C. (1995). Exchange rates and fundamentals: Evidence on long-horizon predictability.
The American Economic Review, 201-218.
Meese, R. A., & Rogoff, K. (1983). Empirical exchange rate models of the seventies: Do they fit
out of sample?. Journal of international economics, 14(1-2), 3-24.
Neely, C. J. (2000). The Temporal Pattern of Trading Rule Returns and Central Bank Intervention:
Intervention Does Not Generate Technical. Federal Reserve Bank of St. Louis Working Paper
Series, (2000-018)
Molodtsova, T., & Papell, D. H. (2009). Out-of-sample exchange rate predictability with Taylor
rule fundamentals. Journal of International Economics, 77(2), 167-180.
Neely, C.J., Rapach, D.E., Tu, J., Zhou, G., 2014. Forecasting the Equity Risk Premium: The Role
of Technical Indicators. Management Science 60(7):1772-1791
Neely, C. J., & Weller, P. A. (2001). Technical analysis and central bank intervention. Journal of
International Money and Finance, 20(7), 949-970.
Neely, C.J., Weller, P.A., 2013. Lessons from the evolution of foreign exchange trading strategies.
Journal of Banking and Finance 37, 3783β3798.
Neely, C. J., & Weller, P. A. (2003). Intraday technical trading in the foreign exchange market.
Journal of International Money and Finance, 22(2), 223-237.
Neely, C., Weller, P., & Dittmar, R. (1997). Is technical analysis in the foreign exchange market
profitable? A genetic programming approach. Journal of financial and Quantitative Analysis,
32(4), 405-426.
Park, C. H., & Irwin, S. H. (2007). What do we know about the profitability of technical analysis?.
Journal of Economic Surveys, 21(4), 786-826.
Raza, A., Marshall, B.R., Visaltanachoti, N., 2014. Is there momentum or reversal in weekly
currency returns? Journal of International Money and Finance 45, 38β60.
Samuleson, P. A. (1964). Theoretical notes on trade problems, Review of Economics and Statistics
46, 145-154.
Sarno, L., & Sojli, E. (2009). The feeble link between exchange rates and fundamentals: Can we
blame the discount factor?. Journal of Money, Credit and Banking, 41(2β3), 437-442.
Sarno, L., & Taylor, M. P. (2002). The Economics of Exchange Rates. Cambridge University Press.
Sarno, L., & Valente, G. (2009). Exchange rates and fundamentals: Footloose or evolving
relationship?. Journal of the European Economic Association, 7(4), 786-830.
Sarno, L., Valente, G., & Leon, H. (2006). Nonlinearity in deviations from uncovered interest
parity: an explanation of the forward bias puzzle. Review of Finance, 10(3), 443-482.
Sweeney, R. J. (1986). Beating the foreign exchange market. The Journal of Finance, 41(1), 163-
182.
Taylor, A. M., & Taylor, M. P. (2004). The purchasing power parity debate. The Journal of
Economic Perspectives, 18(4), 135-158.
Verdelhan, A., 2017. The share of systematic variation in bilateral exchange rates. Journal of
Finance, forthcoming.
Zarrabi, N., Snaith, S., Coakley, J., 2017. FX technical trading rules can be profitable sometimes!
International Review of Financial Analysis 49, 113β127.
19
Table 1: Descriptive Statistics
This table reports summary statistics for the log spot exchange rate changes (i. e., πππ‘+1 = log(St+1) β log(St)), where ππ‘ is the spot exchange rate of the
foreign currency against the USD, and 4 macroeconomic variables for the full sample period spanning from December 1996 to July 2017. We use a sample
of 13 developing countries: Mexican New Peso (MXN) (from Latin America); Hong Kong Dollar (HKD), Indian Rupee (INR), Indonesian Rupiah (IDR),
Philippine Peso (PHP), Kuwaiti Dinar (KWD), New Taiwan Dollar (TWD), Saudi Riyal (SAR), Singapore Dollar (SGD), Thai Baht (THB) (from Asia); Czech
Table 2: Out-of-Sample Forecasting Results using Rolling Regressions β Developed Countries
The table reports the out-of-sample π πππ 2 , McCracken (2007) MSE-F, and change in RMSE-statistics for the predictive regression models for the developed market
currencies, against the null of a random walk (RW). Panel A shows the results using fundamental indicators given by,
βοΏ½ΜοΏ½ππ‘+1 = οΏ½ΜοΏ½ + οΏ½ΜοΏ½π₯ππ‘ , βοΏ½ΜοΏ½ππ‘+1 is one-step ahead forecast of the log spot exchange rate changes; π₯ππ‘ is one of the 3 macroeconomic variables (based on UIP, PPP, and MF). Panel B presents
the results using technical trading indicators given by
where π§ππ‘ is one of the 3 technical trading signals (based on MA, MOM, and RSI). We obtain the OOS monthly forecasts using rolling regressions using a fixed
window equal to 140 observations. This exercise produces 108 OOS forecasts, so that our first OOS forecast is for September 2008, using IS period from January
1997 to August 2008, and our last OOS forecast is for August 2017, using IS period from August 2006 to July 2017. The last row in each panel shows the estimates
for the pooled sample that includes all the sample currencies. *, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively.
Table 3: Out-of-Sample Forecasting Results using Rolling Regressions β Developing Countries
The table reports the out-of-sample π πππ 2 , McCracken (2007) MSE-F, and change in RMSE-statistics for the predictive regression models for the developing market
currencies, against the null of a random walk (RW). Panel A shows the results using fundamental indicators given by,
βοΏ½ΜοΏ½ππ‘+1 = οΏ½ΜοΏ½ + οΏ½ΜοΏ½π₯ππ‘ , βοΏ½ΜοΏ½ππ‘+1 is one-step ahead forecast of the log spot exchange rate changes; π₯ππ‘ is one of the 3 macroeconomic variables (based on UIP, PPP, and MF). Panel B presents
the results using technical trading indicators given by
where π§ππ‘ is one of the 3 technical trading signals (based on MA, MOM, and RSI). We obtain the OOS monthly forecasts using rolling regressions using a fixed
window equal to 140 observations. This exercise produces 108 OOS forecasts, so that our first OOS forecast is for September 2008, using IS period from January
1997 to August 2008, and our last OOS forecast is for August 2017, using IS period from August 2006 to July 2017. The last row in each panel shows the estimates
for the pooled sample that includes all the sample currencies. *, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively. NA indicates not
applicable due to data availability during the sample period.
Panel A: Fundamental Indicators
UIP PPP MF
π πππ 2 MSE-F RMSE π πππ
2 MSE-F RMSE π πππ 2 MSE-F RMSE
Philippine -0.028 -2.978 -0.023 -0.051 -5.182 -0.042 0.006 0.647 0.005 Chez Rep. 0.003 0.355 0.006 -0.027 -2.832 -0.052 -0.027 -2.799 -0.051 Hong Kong -0.001 -0.136 -7.61E-05 1.23E-04 0.013 7.29E-06 -0.005 -0.612 -3.42E-04 Indonesia -0.010 -1.138 -0.016 -0.089 -8.736 -0.134 NA NA NA
India -0.011 -1.254 -0.016 NA NA NA NA NA NA Kuwait -0.081 -7.990 -0.033 NA NA NA NA NA NA
Hungary -0.004 -0.457 -0.010 -0.036 -3.709 -0.086 -0.023 -2.412 -0.055 Mexico -0.007 -0.835 -0.014 -0.016 -1.721 -0.029 -0.037 -3.842 -0.067 Saudi -0.317 -25.553 -0.007 NA NA NA NA NA NA
Taiwan 0.005 0.571 0.004 -0.014 -1.511 -0.010 -0.009 -1.036 -0.007 Thailand -0.034 -3.529 -0.027 0.003 0.409 0.003 NA NA NA Turkey 0.021 2.338 0.040 0.021 1.017 0.017 NA NA NA
Table 4: Out-of-Sample Forecasting Results using Recursive Regressions β Developed Countries
The table reports the out-of-sample π πππ 2 , McCracken (2007) MSE-F, and change in RMSE-statistics for the predictive regression models for the developed market
currencies, against the null of a random walk (RW). Panel A shows the results using fundamental indicators given by,
βοΏ½ΜοΏ½ππ‘+1 = οΏ½ΜοΏ½ + οΏ½ΜοΏ½π₯ππ‘ , βοΏ½ΜοΏ½ππ‘+1 is one-step ahead forecast of the log spot exchange rate changes; π₯ππ‘ is one of the 3 macroeconomic variables (based on UIP, PPP, and MF). Panel B presents
the results using technical trading indicators given by
where π§ππ‘ is one of the 3 technical trading signals (based on MA, MOM, and RSI). We obtain the OOS monthly forecasts using rolling regressions using a fixed
window equal to 140 observations. This exercise produces 108 OOS forecasts, so that our first OOS forecast is for September 2008, using IS period from January
1997 to August 2008, and our last OOS forecast is for August 2017, using IS period from August 2006 to July 2017. The last row in each panel shows the estimates
for the pooled sample that includes all the sample currencies. *, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively.
Panel A: Fundamental Indicators
UIP PPP MF
π πππ 2 MSE-F RMSE π πππ
2 MSE-F RMSE π πππ 2 MSE-F RMSE
Australia Canada
Euro Japan
New Zealand Norway Sweden
Switzerland UK
25
Table 4: continued
Panel B: Technical Indicators
MA MOM RSI
π πππ 2 MSE-F RMSE π πππ
2 MSE-F RMSE π πππ 2 MSE-F RMSE
Australia Canada
Euro Japan
New Zealand Norway Sweden
Switzerland UK
26
Table 5: Out-of-Sample Forecasting Results using Recursive Regressions β Developing Countries
The table reports the out-of-sample π πππ 2 , McCracken (2007) MSE-F, and change in RMSE-statistics for the predictive regression models for the developing market
currencies, against the null of a random walk (RW). Panel A shows the results using fundamental indicators given by,
βοΏ½ΜοΏ½ππ‘+1 = οΏ½ΜοΏ½ + οΏ½ΜοΏ½π₯ππ‘ , βοΏ½ΜοΏ½ππ‘+1 is one-step ahead forecast of the log spot exchange rate changes; π₯ππ‘ is one of the 3 macroeconomic variables (based on UIP, PPP, and MF). Panel B presents
the results using technical trading indicators given by
where π§ππ‘ is one of the 3 technical trading signals (based on MA, MOM, and RSI). We obtain the OOS monthly forecasts using rolling regressions using a fixed
window equal to 140 observations. This exercise produces 108 OOS forecasts, so that our first OOS forecast is for September 2008, using IS period from January
1997 to August 2008, and our last OOS forecast is for August 2017, using IS period from August 2006 to July 2017. The last row in each panel shows the estimates
for the pooled sample that includes all the sample currencies. *, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively. NA indicates not
applicable due to data availability during the sample period.
Panel A: Fundamental Indicators
UIP PPP MF
π πππ 2 MSE-F RMSE π πππ
2 MSE-F RMSE π πππ 2 MSE-F RMSE
Philippine Chez Rep. Hong Kong Indonesia
India Kuwait
Hungary Mexico Saudi
Singapore South Africa
Taiwan Thailand Turkey
27
Table 5: continued
Panel B: Technical Indicators
MA MOM RSI
π πππ 2 MSE-F RMSE π πππ
2 MSE-F RMSE π πππ 2 MSE-F RMSE
Philippine
Chez Rep.
Hong Kong
Indonesia
India
Kuwait
Hungary
Mexico
Saudi
Singapore
South Africa
Taiwan
Thailand
Turkey
28
Table 6: In-Sample Forecasting Results (January 1997 β June 2017)
Under the null hypothesis of no predictability using technical indicators, the table reports the least squares estimates for the following two bivariate predictive
where βπ ππ‘+1 is the log spot exchange rate changes (i. e., βπ ππ‘+1 = ln(St+1) β ln(St)); π₯ππ‘ is one of the 3 macroeconomic variables (based on UIP, PPP, and MF);
and π§ππ‘ is one of the 3 technical trading signals (based on MA, MOM, and RSI). If currency returns are predictable from fundamental or technical regressors, the
slope coefficient estimate should be insignificantly different from zero. The R2 -statistics are computed for the full estimation period spanning from January 1997
to June 2017 (246 observations). Panel A shows the results for developed market currencies, and Panel B presents the results for emerging market currencies. The
last row in each panel shows the estimates for the pooled sample that includes all the sample currencies. *, ** and *** indicate significance at the 10%, 5% and