Economic Computation and Economic Cybernetics Studies and Research, Issue 4/2019; Vol.

53
________________________________________________________________________________

Professor Maricruz OLAZABAL-LUGO, PhD

Universidad Autónoma de Occidente, México
E-mail: maricruz.olazabal@udo.mx
Professor Ernesto LEON-CASTRO, PhD
Universidad de la Salle Bajio, México
E-mail: eleon@delasalle.edu.mx
Professor Luis Fernando ESPINOZA-AUDELO, M.Sc.
Universidad Autónoma de Occidente, México
Professor Jose Maria MERIGÓ, PhD
Universidad de Chile, Chile
Professor Anna María GIL LAFUENTE, PhD
Universidad de Barcelona, España

FORGOTTEN EFFECTS AND HEAVY MOVING AVERAGES IN

EXCHANGE RATE FORECASTING

Abstract. The paper presented different exchange rate forecasting models

based on fundamental economics using different aggregation information
operators such as heavy moving average, forgotten effects and expertons. The
heavy ordered weighted moving average weighted average (HOWMAWA) operator
is introduced. This new operator includes the weighted average in the typical
heavy ordered weighted moving average (HOWMA) operator, considering a
degree of importance for each concept that includes the operator. The use of
expertons and forgotten effects methodology represents the information of experts
in the field, with which hidden variables or second-degree relations were obtained.
Once these items were detected, they were included in the econometric models, and
the forecast of exchange rate of USD/MXN was performed using time series and
HOWMA and HOWMAWA operators. The results show that the inclusion of the
forgotten effects and heavy moving average operators improves our results and
reduces the forecast error.
Keywords: Experton, Forgotten effects, HOWMAWA operator,
econometric forecasting.
JEL Classification: D70, E17, D81

79

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________
1. Introduction
Among the most volatilizes products in open economics is the exchange rate. That
is why, it is necessary to generate new models that can include that characteristic
and that will help to obtain an acceptable forecast error (Majhiet al., 2009).
Among the weakness of the forecasting models are that they usually try to
simplify the reality (Krugman, 2000).Thus, although some of the variables are
included in the models, there are other that are excluded and that can have an
important effect in the results, because of that it is important to identify them.
A methodology that can be used in order to identify the forgotten variables is the
forgotten effects methodology developed by Kaufmann & Gil-Aluja (1988). This
method use different matrixes in order to obtain relationship between different
variables and in this sense obtain a degree of forgetfulness (Vizuete-Lucianoet al.,
2013).
In order to obtain the information that is needed to formulate the matrixes the
experton technique (Kaufmann, 1988) is used. This method uses a cumulative
distribution function and a linguistic expression based on a decimal scale from 0 to
1. By doing this, it is possible to consolidate a group of experts opinion.
The technique that was used in order to aggregate the knowledge and
expectations of the financial markets of the experts was the heavy ordered
weighted moving average (HOWMA) operator. This operator is an extension of the
traditional ordered weighted average (OWA) operator developed by Yager (1988).
Among the advantages is that it is possible to include a heavy weighted vector with
a moving average series.. In doing so, we can incorporate the knowledge and
expectations of financial market experts into the typical moving average, allowing
us to under- or overestimate the results according to the available information
(Yager, 2008).
The HOWMA operator was extended by aggregation a weighted average in the
same formulation and in this sense we can provide different degree of important to
the weighted moving average or the weighted average. This new formulation is
called heavy ordered weighted moving average weighted average (HOWMAWA)
operator. The main properties and cases have been developed.
This new operator has been used in order to forecast USD/MXN exchange rate
for 2015 by the use of econometric models based on economic fundamentals. In
order to generate the models information from 1994-2014 was used; then in order
to generate the values of the different variables in the model two different
techniques were used: time series and HOWMAWA operator. After that, in order
to obtain forgotten variables that will be added to the econometric models the
forgotten effects methodology and the experton technique were used.
The paper is organized as follows. In Section 2, the main formulations used in
the paper are presented. Section 3presents some generalizations and specific cases
of the heavy moving averages operators. In Section 4, the forgotten variables are
detected based on forgotten effects methodology and expertons technique. Section

80

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

5 presents an analysis of the different methods and techniques used to calculate

USD/MXN exchange rate. Section 6 summarizes the conclusions of the paper.
2. Preliminaries
This section presents the main formulations that are used in the paper. It presents
expertons technique, forgotten effects methodology and HOWMA operator.
2.1Expertons
The expertons (Kaufmann & Gil-Aluja, 1993; Kaufmann, 1988) suggests that, to
obtain realistic data from a phenomenon that is not directly measurable, an
aggregated set of valuations that are given by experts is useful. Kaufmann (1988)
indicates that an experton is an extension of a probabilistic set (Hirota, 1981)in
which each 𝛼 − 𝑐𝑢𝑡 is replaced by a range of probabilities collected by expert
opinion (Merigóet al. 2014).
Assume that E is a referential set, finite or not; r is a certain number of experts who
are asked to give their opinion about each element of E, which is represented as
follows:
𝑗
∀𝑎 ∈: [𝛼∗ (𝑥), 𝛼𝑗∗ (𝑥)] ⊂ [0,1],
where ⊂ is a set of inclusion and j is the jth expert.
A static that concerns each 𝑥 ∈ 𝐸 is the lower and upper bound, for which a
𝑗
cumulative complementary law 𝐹∗ (𝑎, 𝑥) is established for 𝛼∗ (𝑥) and 𝐹 ∗ (𝑎, 𝑥) is
established for 𝛼𝑗∗ (𝑥). Thus,
∀𝑥 ∈ 𝐸, ∀𝑥 ∈ [0,1]: 𝐴̃(𝑥) = [𝐹∗ (𝛼, 𝑥), 𝐹 ∗ (𝛼, 𝑥)],
where ~ is the nature of the concept.
The referential set E is the following experton:
∀𝑥 ∈ 𝐸, ∀𝛼 ∈ 𝐸[0,1]: [𝐹∗ (𝛼, 𝑥), 𝐹 ∗ (𝛼, 𝑥)] = 1.
The empty experton is then given by
1, 𝛼 = 0
∀𝑥 ∈ 𝐸: [𝐹∗ (𝛼, 𝑥), 𝐹 ∗ (𝛼, 𝑥)] = {
0, 𝛼 ≠ 0.
2.2 Forgotten effects methodology
The forgotten effects methodology introduced by Kaufmann and Gil-Aluja(1988)
is supported on the assumption of the existence of the following two sets:
A = {ai ⁄i = 1,2, … , n},
B = {bj ⁄j = 1,2, … , m}.
It is hypothesized that the incidence of 𝑎𝑖 on 𝑏𝑗 prevails if the value of the
membership function characteristic of (𝑎𝑖 , 𝑏𝑗 ) is valued in the range [0,1], i.e.,
∀(𝑎𝑖 , 𝑏𝑗 ) ⟹ 𝜇(𝑎𝑖 , 𝑏𝑗 ) ∈ [0,1].
The set of the pair of evaluated elements is known as the direct impact matrix,
which shows the cause-effect relationship in different degrees caused by the
corresponding assembly A (causes) and the set B (effects).

81

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________

b1 b2 b3 … bm
a1 𝜇𝑎1 𝑏1 𝜇𝑎1 𝑏2 𝜇𝑎1 𝑏3 … 𝜇𝑎1 𝑏𝑚

a2 𝜇𝑎2 𝑏1 𝜇𝑎2 𝑏2 𝜇𝑎2 𝑏3 … 𝜇𝑎2 𝑏𝑚

M= a3 𝜇𝑎3 𝑏1 𝜇𝑎3 𝑏2 𝜇𝑎3 𝑏3 … 𝜇𝑎3 𝑏𝑚

…

… … … … …

an 𝜇𝑎𝑛 𝑏1 𝜇𝑎𝑛 𝑏2 𝜇𝑎𝑛 𝑏3 … 𝜇𝑎𝑛 𝑏𝑚

Now, assume that there is a third set of elements, called C, which is expressed as
follows:
𝐶𝑘
𝐶 = { = 1,2, … , 𝑘}.
𝑘
This new set of elements represents the effects of the set B, having an incidence
matrix as follows:
c1 c2 c3 … cm
b1 𝜇𝑏1 𝑐1 𝜇𝑏1 𝑐2 𝜇𝑏1 𝑐3 … 𝜇𝑏1 𝑐𝑚

b2 𝜇𝑏2 𝑐1 𝜇𝑏2 𝑐2 𝜇𝑏2 𝑐3 … 𝜇𝑏2 𝑐𝑚

N= b3 𝜇𝑏3 𝑐1 𝜇𝑏3 𝑐2 𝜇𝑏3 𝑐3 … 𝜇𝑏3 𝑐𝑚

…

… … … … …

bn 𝜇𝑏𝑛 𝑐1 𝜇𝑏𝑛 𝑐2 𝜇𝑏𝑛 𝑐3 … 𝜇𝑏𝑛 𝑐𝑚

Note that we have two incidence matrices that share the set Bas a common
element. Thus, the relationship of the three sets can be expressed as follows:
𝑀 ⊂ 𝐴x𝐵 and 𝑁 ⊂ 𝐵x𝐶.
With this information, we obtain the forgotten effects in A and C, using the set B
as the base. To do so, the max-min operator is used (expressed by symbol ∘),
generating a new incidence matrix as follows:
𝑀 ∘ 𝑁 = 𝑃,
𝑃 ⊂ 𝐴x𝐶.
This new relationship is formulated as
∀(𝑎𝑖 , 𝑐𝑧 ∈ 𝐴x𝐶),
𝜇(𝑎𝑖 , 𝑐𝑧 )𝑀 ∘ 𝑁 = ∀𝑏𝑗 (𝜇𝑀(𝑎𝑖 , 𝑏𝑗 )˄𝜇𝑁(𝑏𝑗 , 𝑐𝑧 ).
The resulting incidence matrix from performing the operation max-min is as
follows:

82

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

c1 c2 c3 … cm
a1 𝜇𝑎1 𝑐1 𝜇𝑎1 𝑐2 𝜇𝑎1 𝑐3 … 𝜇𝑎1 𝑐𝑚

a2 𝜇𝑎2 𝑐1 𝜇𝑎2 𝑐2 𝜇𝑎2 𝑐3 … 𝜇𝑎2 𝑐𝑚

P= a3 𝜇𝑎3 𝑐1 𝜇𝑎3 𝑐2 𝜇𝑎3 𝑐3 … 𝜇𝑎3 𝑐𝑚

…

… … … … …

an 𝜇𝑎𝑛 𝑐1 𝜇𝑎𝑛 𝑐2 𝜇𝑎𝑛 𝑐3 … 𝜇𝑎𝑛 𝑐𝑚

The P matrix defines the causal relationships between elements of sets A and C
in the intensity or degree that leads to the consideration of those belonging to B.
To assign values to the matrices, Kaufmann and Gil-Aluja(1988) propose a
decimal scale that is composed of 11 values from 0 to 1[0,0.1,0.2,0.3, … ,1]. This
scale facilitates their adaptation and treatment since people are used to thinking and
working in decimal form.
2.3Heavy moving average operators
The main advantage of the use of a heavy weighting vector and moving average in
exchange rate forecasting is that we can include in the same formulation historical
data and expectations of the financial market through the experts. The HOWMA
operator can be defined as follows (León et al., 2016; León et al., 2018a).
Definition 1. A HOWMA operator is defined as a given sequence {𝑎𝑖 }𝑁 𝑖=1 , where
one obtains a new sequence {𝑠𝑖 }𝑁−𝑛+1
𝑖=1 , which is multiplied by a heavy weighting
vector, such that
𝑚+𝑡

𝐻𝑂𝑊𝑀𝐴(𝑠𝑖 ) = ∑ 𝑤𝑗 𝑏𝑗 , (1)
𝑗=1+𝑡
where 𝑏𝑗 is the jth largest element of the collection a1 , a2 , … , an , and W is an
associated weighting vector of dimension m with 𝑊: 1 ≤ ∑𝑚+𝑡 𝑖=1+𝑡 𝑤𝑖 ≤ 𝑛 and 𝑤𝑖 ∈
[0,1]. Observe that, here, we can also expand the weighting vector from −∞ to ∞.
Thus, the weighting vector w becomes −∞ ≤ ∑𝑛𝑗=1 𝑤𝑗 ≤ ∞.
It is important to note that the characteristics of the HOWMA operator are: a)
monotonic because, if 𝑎𝑖 ≥ 𝑑𝑖 , for all i, then 𝐻𝑂𝑊𝑀𝐴(𝑎1 , … , 𝑎𝑛 ) ≥
𝐻𝑂𝑊𝑀𝐴(𝑑1 , … , 𝑑𝑛 ),b) commutative because any permutation of the arguments
has the same evaluation. Also, the HOWMA operator is unbounded that because
the weighted vector can range from 1 to ∞ or even from −∞ to ∞
Finally, the beta value explained by Yager (2002) applies to the HOWMA
operator. In this sense, the beta value normalizes the vector W and is defined as
𝛽(𝑊) = (|𝑊| − 1)/(𝑛 − 1). Since |𝑊| ∈ [1, 𝑛], 𝛽 ∈ [0,1], which is why, if 𝛽 =
1, then we obtain the total operator and, if 𝛽 = 0, then we obtain the typical
moving average.

83

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________

An additional extension that can be developed using as a base the HOWMA

operator is when we include a weighted average, this can helps us to include two
different aggregation operators (weighted moving average and weighted average)
where it is possible to give a different degree of importance. The new formulation
is called the heavy ordered weighted moving average weighted average
(HOWMAWA) operator. The formulation is as follows.
Definition 2. An HOWMAWA operator of dimension m is a mapping
𝐻𝑂𝑊𝑀𝐴𝑊𝐴: 𝑅 𝑚 → 𝑅 that has an associated weighting vector W of dimension m
with 𝑊: 1 ≤ ∑𝑚+𝑡
𝑖=1+𝑡 𝑤𝑖 ≤ 𝑛 and 𝑤𝑗 ∈ [0,1], such that
𝑚+𝑡

𝐻𝑂𝑊𝑀𝐴𝑊𝐴 (𝑎1+𝑡 , … , 𝑎𝑚+𝑡 ) = ∑ 𝑣̂𝑗 𝑏𝑗 , (2)

𝑗=1+𝑡
where 𝑏𝑗 is the jth largest argument of 𝑎𝑖 , each argument 𝑎𝑖 has an associated
weight (WA) 𝑣𝑖 with 𝑣𝑖 ∶ 1 ≤ ∑𝑚+𝑡 𝑖=1+𝑡 𝑤𝑖 ≤ 𝑛 and 𝑣𝑖 ∈ [0,1], 𝑣
̂𝑗 = 𝜃𝑤𝑗 + (1 − 𝜃)𝑣𝑗
with 𝜃 ∈ [0,1], 𝑣𝑗 is the weight 𝑣𝑖 ordered according to 𝑏𝑗 , that is, according to the
jth largest of 𝑎𝑖 , m is the total number of arguments considered from the entire
sample, and t indicates the movement made in the average from the initial analysis.
The HOWMAWA operator can provide a wide range of cases by using generalized
and quasi-arithmetic means (Merigó&Yager, 2013). The Quasi-HOWMAWA
operator is presented because include the generalized mean as a specific case.
Definition 3. A Quasi-HOWMAWA operator of dimension m is a mapping
𝐻𝑂𝑊𝑀𝐴𝑊𝐴: 𝑅 𝑚 → 𝑅 that has an associated weighting vector W of dimension m
with 𝑊: 1 ≤ ∑𝑚+𝑡𝑖=1+𝑡 𝑤𝑖 ≤ 𝑛 and 𝑤𝑗 ∈ [0,1], such that:
𝑚+𝑡
−1
𝑄𝑢𝑎𝑠𝑖 − 𝐻𝑂𝑊𝑀𝐴𝑊𝐴(𝑎1+𝑡 , … , 𝑎𝑚+𝑡 ) = 𝑔 ∑ 𝑣̂𝑗 𝑔(𝑏𝑗 ), (3)
𝑗=1+𝑡
where 𝑏𝑗 is the jth largest argument of the 𝑎𝑖 , each argument 𝑎𝑖 has an associated
weight (WA) 𝑣𝑖 with 𝑣𝑖 ∶ 1 ≤ ∑𝑚+𝑡 𝑖=1+𝑡 𝑤𝑖 ≤ 𝑛 and 𝑣𝑖 ∈ [0,1], 𝑣̂𝑗 = 𝛽𝑤𝑗 + (1 −
𝛽)𝑣𝑗 with 𝛽 ∈ [0,1], 𝑣𝑗 is the weight 𝑣𝑖 ordered according to 𝑏𝑗 , that is, according
to the jth largest of 𝑎𝑖 , m is the total number of arguments considered from the
entire sample, t indicates the movement made in the average from the initial
analysis, and 𝑔(𝑏) is a strictly continuous monotonic function. If more specific
cases want to be analyzed, most of them are explained in Merigó&Yager (2013).
4. Detecting exchange rate forgotten variables
In order to detect forgotten variables that can be included in the econometric
models, the opinion of five experts in the exchange rate market is used. This group
includes three people who are financial advisors in the forex market and another
two who are academics who previously worked in this field. It is important to note
that the experts that are used can be increase or decrease according to the problem
and situation. Also, it is relevant to add that there is no specific optimal number

84

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

because it depends on the problem and the results may vary if we different experts
or number of them. With the information provided, we use the experton technique,
and the results are presented in Tables1-3. Note that the selection of the variables
that are used as cause or effects are the ones that compose the PPP, IRP and BoP
models. Also other macroeconomic variables that have been important in exchange
rate such as growth expectations, monetary politics, country risk, global market
and oil price are included (Tsai, 1994; Bergholt &Lujala, 2012; Zhang et al, 2012;
Reboredo, 2012; Blanchard & Adler, 2015; Ghosh et al., 2015).
Table 1. Effect-Effect
𝑃𝐶𝐼𝑈𝑆𝐴 𝑃𝐶𝐼𝑀𝐸𝑋 𝑟𝑈𝑆𝐴 𝑟𝑀𝐸𝑋 TB FDI FPI R
𝑃𝐶𝐼𝑈𝑆𝐴 1 0.4 0.6 0.3 0.3 0.3 0.3 0.3
𝑃𝐶𝐼𝑀𝐸𝑋 0.4 1 0.6 0.6 0.4 0.4 0.4 0.4
𝑟𝑈𝑆𝐴 0.8 0.3 1 0.6 0.2 0.4 0.5 0.3
𝑟𝑀𝐸𝑋 0.4 0.7 0.7 1 0.4 0.5 0.5 0.4
TB 0.3 0.5 0.3 0.4 1 0.6 0.6 0.5
FDI 0.2 0.4 0.5 0.4 0.5 1 0.6 0.5
FPI 0.3 0.3 0.4 0.5 0.5 0.6 1 0.4
R 0.3 0.4 0.4 0.5 0.4 0.6 0.6 1

Table 2. Cause-Cause
Growth Monetary Country Oil price Global
expectations politics risk markets
Growth
expectations 1 0.8 0.5 0.7 0.7
Monetary
politics 0.8 1 0.7 0.6 0.6
Country risk 0.7 0.7 1 0.5 0.7
Oil price 0.7 0.5 0.4 1 0.8
Global markets 0.5 0.4 0.3 0.7 1

Table 3. Cause-Effect
𝑃𝐶𝐼𝐸𝑈𝐴 𝑃𝐶𝐼𝑀𝐸𝑋 𝑟𝐸𝑈𝐴 𝑟𝑀𝐸𝑋 TB FDI FPI R
Growth
0.4 0.6 0.6 0.7 0.8 0.7 0.7 0.5
expectations
Monetary
0.7 0.8 0.7 0.8 0.7 0.6 0.6 0.5
politics
Country
0.3 0.6 0.5 0.8 0.5 0.7 0.7 0.5
risk
Oil price 0.5 0.3 0.3 0.3 0.6 0.4 0.5 0.5
Global
0.6 0.4 0.6 0.4 0.4 0.6 0.6 0.4
markets

where 𝑃𝐶𝐼𝑈𝑆𝐴 = the consumer price index inthe USA, 𝑃𝐶𝐼𝑀𝐸𝑋 = the consumer price
index in Mexico, 𝑟𝑈𝑆𝐴 = the interest rate in the USA, 𝑟𝑀𝐸𝑋 = the interest rate in

85

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________
Mexico, TB = the trade balance in Mexico, FDI= foreign direct investment in
Mexico, FPI= foreign portfolio investment in Mexico, andR= international
reserves in Mexico.
With the information from Tables 1-3, the forgotten effects methodology
developed by Kauffman and Gil-Aluja (1988) is used to identify the forgotten
variables. The results are presented in Table 4.
Table 4. Forgotten effects in determining the forward exchange rate
Cause Effect M 𝑴∘𝑵 M* = Q=
𝑴∘𝑵∘𝑩 M*-M
Growth expectations 𝑃𝐶𝐼𝑈𝑆𝐴 0.4 0.7 0.7 0.3
Growth expectations 𝑃𝐶𝐼𝑀𝐸𝑋 0.6 0.8 0.8 0.2
Growth expectations 𝑟𝑈𝑆𝐴 0.6 0.7 0.7 0.1
Growth expectations 𝑟𝑀𝐸𝑋 0.7 0.8 0.8 0.1
Growth expectations TB 0.8 0.8 0.8 0
Growth expectations FDI 0.7 0.7 0.7 0
Growth expectations FPI 0.7 0.7 0.7 0
Growth expectations R 0.5 0.5 0.6 0.1
Monetary politics 𝑃𝐶𝐼𝑈𝑆𝐴 0.7 0.7 0.7 0
Monetary politics 𝑃𝐶𝐼𝑀𝐸𝑋 0.8 0.8 0.8 0
Monetary politics 𝑟𝑈𝑆𝐴 0.7 0.7 0.7 0
Monetary politics 𝑟𝑀𝐸𝑋 0.8 0.8 0.8 0
Monetary politics TB 0.7 0.8 0.8 0.1
Monetary politics FDI 0.6 0.7 0.7 0.1
Monetary politics FPI 0.6 0.7 0.7 0.1
Monetary politics R 0.5 0.5 0.6 0.1
Country risk 𝑃𝐶𝐼𝑈𝑆𝐴 0.3 0.7 0.7 0.4
Country risk 𝑃𝐶𝐼𝑀𝐸𝑋 0.6 0.7 0.7 0.1
Country risk 𝑟𝑈𝑆𝐴 0.5 0.7 0.7 0.2
Country risk 𝑟𝑀𝐸𝑋 0.8 0.8 0.8 0
Country risk TB 0.5 0.7 0.7 0.2
Country risk FDI 0.7 0.7 0.7 0
Country risk FPI 0.7 0.7 0.7 0
Country risk R 0.5 0.5 0.6 0.1
Oil price 𝑃𝐶𝐼𝑈𝑆𝐴 0.5 0.6 0.6 0.1
Oil price 𝑃𝐶𝐼𝑀𝐸𝑋 0.3 0.6 0.6 0.3
Oil price 𝑟𝑈𝑆𝐴 0.3 0.6 0.6 0.3
Oil price 𝑟𝑀𝐸𝑋 0.3 0.7 0.7 0.4
Oil price TB 0.6 0.7 0.7 0.1
Oil price FDI 0.4 0.7 0.7 0.3
Oil price FPI 0.5 07 0.7 0.2
Oil price R 0.5 0.5 0.6 0.1
Global markets 𝑃𝐶𝐼𝑈𝑆𝐴 0.6 0.6 0.6 0
Global markets 𝑃𝐶𝐼𝑀𝐸𝑋 0.4 0.5 0.6 0.2
Global markets 𝑟𝑈𝑆𝐴 0.6 0.6 0.6 0
Global markets 𝑟𝑀𝐸𝑋 0.4 0.5 0.6 0.2
Global markets TB 0.4 0.6 0.6 0.2

86

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

Global markets FDI 0.6 0.6 0.6 0

Global markets FPI 0.5 0.5 0.6 0.1
Global markets R 0.4 0.5 0.6 0.2

As can be seen in Table 4, oil price and country risk are the indirect effects with
higher sum (1.8 and 1.0, respectively). Also, it is important to note that the indirect
effects with 0.4 values are the oil price in relation with interest rate (though growth
expectations) and country risk with the consumer price index in the USA (through
monetary policy) (See Figure 1 and 2).
Note: To develop these relationships, we use FuzzyLog Software. It is freely
available at http://www.fuzzyeconomics.com/jaimegil.html.

Figure 1. Element interposed between oil prices and the interest rate in Mexico.

Figure 2. Element interposed between country risk and the consumer price index in
the USA.

6. Forecast of the USD/MXN exchange rate for 2015

To forecast the USD/MXN exchange rate for 2015, we use information from the
1994-2014 period because 1994 was when Mexico first implemented the floating

87

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________
exchange rate regime. Additionally, we use three different econometric models
based on Purchase Power Parity (PPP), Interest Rate Parity (IRP) and Balance of
Payment (BoP) theories. The models are presented in Table 5.

Table 5. Econometric models used to forecast USD/MXN 2015

Traditional With forgotten effects
PPP model 𝑡𝑐𝐹 𝑡𝑐𝐹
= 0.107 + 0.914𝑡𝑐−1 = −0.016 + 0.913𝑡𝑐−1
+ 0.0235𝑣 − 0.0195𝑝𝑐𝑖𝑈𝑆𝐴 + 0.0228𝑣 + 0.0457𝑝𝑐𝑖𝑈𝑆𝐴
+ 0.0408𝑝𝑐𝑖𝑀𝐸𝑋 + 0.0354𝑝𝑐𝑖𝑀𝐸𝑋
−0.0098𝑚𝑚𝑒
R-square = 99.0%
R-square(adjust) = 99.0% R-square = 99.0%
R-square(adjust) = 99.0%
IRP model 𝑡𝑐𝐹 = 0.0908 + 0.960𝑡𝑐−1 𝑡𝑐𝐹
+ 0.0255𝑣 = 0.0850 + 0.958𝑡𝑐−1
− 0.00019𝑟𝑈𝑆𝐴 + 0.0226𝑣 − 0.00009𝑟𝑈𝑆𝐴
− 0.00228𝑟𝑀𝐸𝑋 − 0.00032𝑟𝑀𝐸𝑋
+0.00401𝑚𝑚𝑒
R-square = 99.0%
R-square(adjust) = 98.9% R-square = 99.0%
R-square(adjust) = 98.9%
BoP model 𝑇𝐶𝐹 𝑇𝐶𝐹 = 0.178 + 0.979𝑇𝐶−1 +
= 0.175 + 0.978𝑇𝐶−1 + 4.59𝑉 4.60𝑉 − 0.000042𝑇𝐵 −
− 0.000041𝑇𝐵 0.000022𝐹𝑃𝐼 −
− 0.000023𝐹𝑃𝐼 0.000015𝐹𝐷𝐼 + 0.00284𝑅 −
− 0.000016𝐹𝐷𝐼 + 0.00226𝑅 0.00112𝑀𝑀𝐸

R-square = 98.8% R-square = 98.8%

R-square(adjust)= 98.7% R-square(adjust) = 98.7%

where 𝑡𝑐𝐹 is the forward exchange rate, 𝑡𝑐−1 is the exchange rate with a lag, v is
volatility, 𝑝𝑐𝑖𝑈𝑆𝐴 is the consumer price index in the USA, 𝑝𝑐𝑖𝑀𝐸𝑋 is the consumer
price index in Mexico, 𝑟𝑈𝑆𝐴 is the interest rate in USA, 𝑟𝑀𝐸𝑋 is the interest rate in
Mexico, and mme is the Mexican crude oil price; note that all of these variables are
expressed in logarithmic form. Additionally, 𝑇𝐶𝐹 is the forward exchange rate,
𝑇𝐶−1 is the exchange rate with a lag, V is volatility, TB is the trade balance in
Mexico, FDI is foreign direct investment in Mexico, FPI is foreign portfolio
investment in Mexico, R is international reserves in Mexico, and MMEis the
Mexican crude oil price.
In the case of the HOWMA and HOWMAWA operators, a sequence of 𝑛 = 6, a
weighting vector 𝑊 = (0.05,0.15,0.15,0.25,0.40)and a weighting vector of the

88

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

weighted average 𝑉 = (0.05,0.05,0.10,0.20,0.20,0.40) were used based on the

information obtained from the financial market experts. Additionally, degrees of
importance of 70% for the OWA and 30% for the weighted average were
considered.
The results of using the different methods are shown in Tables 6-8.
Note that, with the information above, not only can we make decisions based on
one result but we now have a range of different scenarios to analyze and improve
our vision of the future. It is important to note that, with the HOWMA and
HOWMAWA operators, it is easier to generate new scenarios based on changes in
the weighting vector and decision makers’ expectations of the future, which is
something that is not possible using time series.
Also it is important to advertise that this methodology is based in the
information that is provided by the decision maker or the group that will make the
decision, that is why the weighting vector is very influenced by them. This can be
considered an advantage because as we explained before we can visualize
scenarios that cannot be taken into account if we only use time series but also it is
relevant to note that maybe this is one of the weakness of the methods because the
weight vector can vary from person to person, that is why is important to include
techniques that help us to unify different opinions like the experton, this in order to
generate a more accurate weight vector.

89

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________
Table 6. USD/MXN forecast using PPP model and different operators

Spot
exchange Time HOW HOWMA
Time HOWM HOWM
Time rate Error series Error Error MA Error Error WA and Error
Series A AWA
USD/MX and FE and FE FE
N
01-15 14.6808 15.6757 0.9949 15.8839 1.2031 14.4197 -0.2611 14.5507 -0.1301 14.2532 -0.4276 14.3696 -0.3112
02-15 14.9230 15.6272 0.7042 15.8420 0.9190 14.4971 -0.4259 14.6505 -0.2725 14.5154 -0.4076 14.6504 -0.2726
03-15 15.2136 15.2111 -0.0025 15.4178 0.2042 14.8056 -0.4080 14.9846 -0.2290 14.7428 -0.4708 14.8963 -0.3173
04-15 15.2208 15.2666 0.0458 15.4674 0.2466 15.1007 -0.1201 15.3050 0.0842 14.9880 -0.2328 15.1604 -0.0604
05-15 15.2475 15.4326 0.1851 15.6321 0.3846 15.4455 0.1980 15.6786 0.4311 15.1598 -0.0877 15.3517 0.1042
06-15 15.4692 15.7162 0.2470 15.9168 0.4476 15.7565 0.2873 16.0171 0.5479 15.4292 -0.0400 15.6407 0.1715
07-15 15.9225 15.8889 -0.0336 16.0907 0.1682 16.0647 0.1422 16.3512 0.4287 15.6150 -0.3075 15.8422 -0.0803
08-15 16.5032 15.6731 -0.8301 15.8782 -0.6250 16.3817 -0.1215 16.6976 0.1944 15.8376 -0.6656 16.0843 -0.4189
09-15 16.8519 15.6922 -1.1597 15.8910 -0.9609 16.7113 -0.1406 17.0577 0.2058 16.0610 -0.7909 16.3276 -0.5243
10-15 16.5813 15.8567 -0.7246 16.0594 -0.5219 17.0465 0.4652 17.4243 0.8429 16.5040 -0.0773 16.7944 0.2131
11-15 16.6325 15.7631 -0.8694 15.9705 -0.6620 17.3888 0.7563 17.7990 1.1666 16.6033 -0.0291 16.9123 0.2798
12-15 17.0365 16.0779 -0.9586 16.3045 -0.7320 17.7368 0.7003 18.1806 1.1442 16.8365 -0.2000 17.1668 0.1303

Average 15.8569 15.6568 -0.2001 15.8629 0.0060 15.9462 0.0893 16.2248 0.3679 15.5455 -0.3114 15.7664 -0.0905

90

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

Table 7. USD/MXN forecast using IRP model and different operators

Spot
exchang Time HOWM HOWM
Time HOWM HOWM
Time e rate Error series Error Error A and Error Error AWA Error
Series A AWA
USD/M and FE FE and FE
XN
01-15 14.6808 15.6435 0.9627 15.6071 0.9263 14.4255 -0.2553 14.4156 -0.2652 14.2560 -0.4248 14.4160 -0.2648
02-15 14.9230 15.5716 0.6486 15.5329 0.6099 14.4952 -0.4278 14.4829 -0.4401 14.5220 -0.4010 14.4835 -0.4395
03-15 15.2136 15.1264 -0.0872 15.0923 -0.1213 14.8092 -0.4044 14.7938 -0.4198 14.7530 -0.4606 14.7947 -0.4189
04-15 15.2208 15.1832 -0.0376 15.1519 -0.0689 15.1098 -0.1110 15.0920 -0.1288 15.0034 -0.2174 15.0930 -0.1278
05-15 15.2475 15.3492 0.1017 15.3180 0.0705 15.4622 0.2147 15.4418 0.1943 15.1766 -0.0709 15.4431 0.1956
06-15 15.4692 15.6489 0.1797 15.6176 0.1484 15.7792 0.3100 15.7569 0.2877 15.4525 -0.0167 15.7584 0.2892
07-15 15.9225 15.8342 -0.0883 15.8022 -0.1203 16.0965 0.1740 16.0739 0.1514 15.6440 -0.2785 16.0761 0.1536
08-15 16.5032 15.6062 -0.8970 15.5731 -0.9301 16.4208 -0.0824 16.3962 -0.1070 15.8718 -0.6314 16.3988 -0.1044
09-15 16.8519 15.6222 -1.2297 15.5931 -1.2588 16.7585 -0.0934 16.7321 -0.1198 16.1005 -0.7514 16.7351 -0.1168
10-15 16.5813 15.7903 -0.7910 15.7599 -0.8214 17.1021 0.5208 17.0738 0.4925 16.5606 -0.0207 17.0773 0.4960
11-15 16.6325 15.6985 -0.9340 15.6662 -0.9663 17.4533 0.8208 17.4232 0.7907 16.6588 0.0263 17.4271 0.7947
12-15 17.0365 16.0393 -0.9972 15.9985 -1.0380 17.8103 0.7738 17.7783 0.7418 16.8979 -0.1386 17.7827 0.7462

Average 15.8569 15.5928 -0.2641 15.5594 -0.2975 15.9769 0.1200 15.9551 0.0982 15.5748 -0.2821 15.9572 0.1003

91

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________

Table 8. USD/MXN forecast using BoP model and different operators

Spot
exchang Time HOWM HOWM
Time HOWM HOWM
Time e rate Error series Error Error A and Error Error AWA Error
Series A AWA
USD/M and FE FE and FE
XN
01-15 14.6808 15.6639 0.9831 15.8697 1.1889 14.5893 -0.0915 14.8200 0.1392 14.5853 -0.0955 14.8147 0.1339
02-15 14.9230 15.7500 0.8270 15.9520 1.0290 14.6074 -0.3156 14.8369 -0.0861 14.6019 -0.3211 14.8296 -0.0934
03-15 15.2136 15.3460 0.1324 15.5446 0.3310 14.8923 -0.3213 15.1216 -0.0920 14.8853 -0.3283 15.1123 -0.1013
04-15 15.2208 15.4047 0.1839 15.6086 0.3878 15.1824 -0.0384 15.4122 0.1914 15.1893 -0.0315 15.4154 0.1946
05-15 15.2475 15.5288 0.2813 15.7326 0.4851 15.5398 0.2923 15.7696 0.5221 15.5351 0.2876 15.7610 0.5135
06-15 15.4692 15.7800 0.3108 15.9896 0.5204 15.8456 0.3764 16.0768 0.6076 15.8400 0.3708 16.0665 0.5973
07-15 15.9225 15.9280 0.0055 16.1437 0.2212 16.1832 0.2607 16.4181 0.4956 16.1768 0.2543 16.4065 0.4840
08-15 16.5032 15.7062 -0.7970 15.9175 -0.5857 16.5081 0.0049 16.7447 0.2415 16.5003 -0.0029 16.7311 0.2279
09-15 16.8519 15.7956 -1.0563 16.0088 -0.8431 16.8501 -0.0018 17.0888 0.2369 16.8410 -0.0109 17.0732 0.2213
10-15 16.5813 15.7810 -0.8003 16.0010 -0.5803 17.1984 0.6170 17.4392 0.8579 17.1884 0.6071 17.4220 0.8407
11-15 16.6325 15.7541 -0.8784 15.9720 -0.6605 17.5561 0.9236 17.7992 1.1667 17.5441 0.9116 17.7792 1.1467
12-15 17.0365 16.2798 -0.7567 16.4927 -0.5437 17.9184 0.8819 18.1640 1.1275 17.9049 0.8684 18.1416 1.1051

Average 15.8569 15.7265 -0.1304 15.9361 0.0792 16.0726 0.2157 16.3076 0.4507 16.0660 0.2091 16.2961 0.4392

92

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

To evaluate the errors obtained by the forecasts for each one of the models and
techniques, the results are analyzed using the average absolute percentage error
(MAPE), mean absolute deviation (MAD) and the mean square deviation (MSD),
the formulas of which are the following. The results are presented in Table 9-11.

∑(|𝑦𝑡 − 𝑦̂𝑡 |/𝑦𝑡

𝑀𝐴𝑃𝐸 = x100,
𝑛
∑|𝑦𝑡 − 𝑦̂𝑡 |
𝑀𝐴𝐷 = ,
𝑛
∑|𝑦𝑡 − 𝑦̂𝑡 |2
𝑀𝑆𝐷 = ,
𝑛

Table 9. Error analysis of the PPP models

Time Series HOWMA HOWMAWA
Time Series HOWMA HOWMAWA
and F.E. and F.E. and F.E.
MAD 0.5844 0.5896 0.3355 0.4731 0.3114 0.2403
MSD 0.4678 0.4443 0.1572 0.3568 0.1537 0.0761
MAPE 3.9807 4.0160 2.2856 3.2227 2.1212 1.6370
The ranking of the results for the PPP models are: HOWMAWA and FE
>HOWMAWA>HOWMA>HOWMA and F.E. > Time Series > Time Series and
F.E.

Table 10. Error analysis of the IRP models

Time Series HOWMA HOWMAWA
Time Series HOWMA HOWMAWA
and F.E. and F.E. and F.E.
MAD 0.5796 0.5900 0.3490 0.3449 0.2865 0.3456
MSD 0.5180 0.5355 0.1796 0.1703 0.1389 0.1716
MAPE 3.9477 4.0189 2.3775 2.3495 1.9517 2.3542
The ranking of the results for the IRP models are: HOWMAWA>HOWMA and
F.E. >HOWMAWA and F.E. >HOWMA> Time Series > Time Series and F.E.

Table 11. Error analysis of the BoP models

Time Series HOWMA HOWMAWA
Time Series HOWMA HOWMAWA
and F.E. and F.E. and F.E.
MAD 0.5844 0.6147 0.3438 0.4804 0.3408 0.4717
MSD 0.4678 0.4508 0.2099 0.3702 0.2050 0.3561
MAPE 3.9807 4.1873 2.3418 3.2721 2.3216 3.2127
The ranking of the results for the BoP models are:
HOWMAWA>HOWMA>HOWMAWA and F.E. >HOWMA and F.E. > Time
Series > Time Series and F.E.

93

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________
Also, it is important to note that the three best models taking into account all
models according to the error analysis are:
1. PPP model with HOWMAWA and F.E.
2. IRP model with HOWMAWA
3. PPP model with HOWMAWA
As we can see with the analysis using different operators and techniques we can
obtain different scenarios about the future of the USD/MXN. Note that the
HOWMAWA operator give in all the models the best solutions, considering or not
the forgotten effects. Also, an interesting finding is that the PPP model with
HOWMAWA and F.E. are the best results taking into account the error, this can
help us to identify that the aggregation operators can be a useful technique in order
to forecast. Finally, the forgotten effects also help to improve in some cases the
results but not always, this is an important finding because then there are some
other things that we are not taking into account in the models and finding them will
improve the results.

7. Conclusions
This paper introduced new hidden variables to three traditional models that are
based on economic fundamentals: the purchase power parity, the interest rate parity
and the balance of payment models. To introduce the forgotten items, we use
expertons and the forgotten effects methodology. Therefore, these techniques use
the information of a group of decision makers who are experts in the field; with the
use of matrices, we can obtain the second-degree effects and, with that information,
the forgotten effects. The objective is to integrate these new variables into the
econometric models and to use different methods to forecast the exchange rate.
Additionally, a new extension of the OWA operator, which is called the heavy
ordered weighted moving average weighted average (HOWMAWA) operator, was
introduced. This new operator considers a degree of importance for each concept
that is used in the HOWMA operator. We have analyzed this new operator, giving
its definition and studying its properties; in addition, some interesting specific
cases have been included.
This new operator and some others have been used to forecast the2015
USD/MXN exchange rate. As noted above, all of the traditional models improve
with the additional hidden variable detected through the forgotten effects
methodology and the HOWMA and HOWMAWA operators. Based on this
finding, we can conclude that the inclusion of uncertainty techniques and
knowledge based systems in traditional econometric models, in the case of
exchange rate forecasting, is an efficient tool to reduce the forecasting error. In
future research, we expect to develop a new extension of the OWA operator by
Bonferroni means (Blanco-Mesa et al., 2018; 2019), order induced variable (León
et al., 2018b) or distance measures (Merigó&Yager, 2013).

94

DOI: 10.24818/18423264/53.4.19.05
Forgotten Effects and Heavy Moving Averages in Exchange Rate Forecasting
__________________________________________________________________

REFERENCES

[1]Bergholt, D., & Lujala, P. (2012), Climate-related Natural Disasters,

Economic Growth, And Armed Civil Conflict. Journal of Peace Research, 49(1),
147-162;
[2]Blanchard, O. & Adler, G. (2015), Can Foreign Exchange Intervention
Stem Exchange Rate Pressures from Global Capital Flow Shocks?National
Bureau of Economic Research;
[3]Ghosh, A. R., Ostry, J. D. & Chamon, M. (2016), Two Targets, Two
Instruments: Monetary and Exchange Rate Policies in Emerging Market
Economies. Journal of International Money and Finance, 60, 172-196;
[4]Hirota, K. (1981), Concepts of Probabilistic Sets. Fuzzy sets and systems, 5,
31-46;
[5]Kaufmann, A. (1988), Theory of Expertons and Fuzzy Logic. Fuzzy Sets
and Systems, 28(3), 295-304;
[6]Kaufmann, A. & Gil-Aluja, J. (1988), Models for the Investigation of
Forgotten Effects (In Spanish). Santiago de Compostela: Milladoiro;
[7]Krugman, P. (2000), Thinking about the Liquidity Trap. Journal of the
Japanese and International Economics, 14, 221-237;
[8]Leon-Castro, E., Aviles-Ochoa, E., Gil-Lafuente, A. M. (2016),
Exchange Rate Usd/Mxn Forecast through Econometric Models, Time
Series and HOWMA Operators. Economic Computation & Economic
Cybernetics Studies & Research, ASE Publishing; 50(4), 135-150;
[9]León-Castro, E., Avilés-Ochoa, E., Merigó, J. M. & Gil-Lafuente, A. M.
(2018a), Heavy Moving Averages and Their Application in Econometric
Forecasting. Cybernetics and Systems, 49(1), 26-43;
[10]León-Castro, E., Avilés-Ochoa, E. &Merigó, J. M. (2018b), Induced
Heavy Moving Averages. International Journal of Intelligent Systems, 33(9),
1823-1839;
[11]Majhi, R., Panda, G. & Sahoo, G. (2009), Efficient Prediction of
Exchange Rates with Low Complexity Artificial Neural Network Models.
Expert Systems with Applications, 36(1), 181-189;
[12]Merigó, J. M. & Yager, R. R. (2013), Generalized Moving Averages,
Distance Measures and OWA Operators. International Journal of
Uncertainty, Fuzziness and Knowledge-Based Systems, 21(4), 533-559;
[13]Merigó, J. M., Casanovas, M. & Yang, J. B. (2014), Group Decision
Making with Expertons and Uncertain Generalized Probabilistic Weighted
Aggregation Operators. European Journal of Operational Research, 235, 215-
224;
[14]Reboredo, J. C. (2012), Modelling Oil Price and Exchange Rate Co-
Movements. Journal of Policy Modeling, 34(3), 419-440.
[15]Tsai, P. (1994), Determinants of Foreign Direct Investment and Its Impact
on Economic Growth. Journal of Economic Development, 19(1), 137-163;

95

DOI: 10.24818/18423264/53.4.19.05
Maricruz Olazabal-Lugo , Ernesto Leon-Castro, Luis F. Espinoza-Audelo,
Jose M. Merigó, Anna María Gil Lafuente
__________________________________________________________________
[16]Vizuete-Luciano, E., Gil-Lafuente, A. M., Garcia-Gonzalez, A. & Boria-
Reverter, S. (2013), Forgotten Effects of Corporate Social and Environmental
Responsibility: A Case Study of Catalonian Economy. Kybernetes, 42(5), 736-
753;
[17]Yager, R. R. (2002), Heavy OWA Operators. Fuzzy Optimization and
Decision Making,1(4), 379-397;
[18]Yager, R. R. (1988), On Ordered Weighted Averaging Aggregation
Operators in Multicriteria Decisionmaking. IEEE Transactions on Systems,
Man and Cybernetics, 18(1), 183-190;
[19]Yager, R. R. (2008), Time Series Smoothing and OWA Aggregation. IEEE
Transactions on Fuzzy Systems, 16(4), 994-1007;
[20]Zhang, J., Wang, L. & Wang, S. (2012), Financial Development and
Economic Growth: Recent Evidence from China. Journal of Comparative
Economics, 40(3), 393-412;
[21]Blanco-Mesa, F., León-Castro, E. &Merigó, J. M. (2018), Bonferroni
Induced Heavy Operators in ERM Decision-Making: A Case On Large
Companies in Colombia. Applied Soft Computing, 72, 371-391;
[22]Blanco-Mesa, F., León-Castro, E. &Merigó, J. M. (2019), Bonferroni
Means with Induced Ordered Weighted Average Operators. International Journal
of Intelligent Systems, 34(1), 3-23.

96

DOI: 10.24818/18423264/53.4.19.05