1. Introduction
The global financial industry is quietly changing
under the catalysis of artificial intelligence (AI) [
1
]. AI
represents a clear opportunity to advance the
transformation of the finance industry by providing users with
greater value and increasing firms’ revenues [
2
]. The
goal of AI is to invent a machine which can sense,
remember, learn, and recognize like a real human being
[
3
]. The deep integration of AI technology and finance
is the inevitable result of deepening development and
Exploring Innovation in these fields [
1
]. These
innovations have the potential to directly influence both
the production and the characteristics of a wide range
of products and services, with important implications
for productivity, employment, and competition [
4
]. AI
also improve work eficiency at the business and
create a whole process of intelligent finance [
1
].
Applications of AI systems are generally viewed as positive
for economic growth and productivity [
2
]. Deep
learning is a recently-developed field belonging to
Artificial Intelligence [
3
]. It attempts to learn hierarchical
representations from raw data and is capable of
learning simple concepts first and then successfully
building up more complex concepts by merging the simpler
ones [
5, 6, 7
]. Companies such as Google, Facebook,
IBM, Microsoft and others use this algorithm for
developing next-generation intelligent applications [
8
].
In finances there are two major problems:1) to predict
future returns (i.e., stock prices, currencies, indices,
product demand); or 2) to make categorical
classification (i.e. credit scoring (“good, “bad”), bankruptcy
(“True”, “False”)). While the issues in finances remain
almost the same over the last several decades, novel
methods, and growing amount of data are changing
the field, especially Machine Learning and Artificial
Intelligence techniques [
9
]. Furthermore, exploitation
of additional data sources allows to achieve better
results, e.g. satellite images can be used for predicting
economic activity, voice information provides
information about emotions, textual information, extracted
from news and comments gives sentiments of writers
and audience, etc [
10
]. However, extraction of useful
knowledge out of such data heap is not trivial, it
requires considerable efort [
11, 12
]. Portfolio
management tasks have more challenges, because there are
two main issues with portfolio formation: (1)
selection of assets with highest revenue, and (2)
determination other value composition of assets in the portfolio
to achieve the goal of maximal potential returns with
minimal risk [
13
]. Therefore, this paper is divided in
two parts: 1) diferent deep learning architectures are
discussed; 2) application of the aforementioned
methods in finances is discussed.
2. Literature review
unit (image, video, time series). This layer produces
huge amount of features that makes overfitting
probThe term "artificial intelligence" is applied when a ma- lems and expensive computation [
8
]. Pooling leayers
chine mimics "cognitive" functions that humans asso- reducses this problem by aggregating multiple feature
ciate with other human minds, such as "learning" and values into a single value. Max-pooling is mostly used
"problem solving” [
14
]. In other words, it tries to mimic pooling operation, in Keras instead of this operation
the human brain, which is capable of processing the could be used Average-pooling, Global-max-pooling
complex input data, learning diferent knowledges in- or Global-average-pooling operations [20]. Rectified
tellectually and fast, and solving diferent kinds of com- linear unit (ReLU) is an activation function meant to
plicated tasks well [
3
]. zero out negative values, whereas a sigmoid “squashes”
AI has been part of human thoughts and slowly evolv- arbitrary values into the interval [
0, 1
] producing
someing in academic research labs [
14
]. Machine learning thing that can be interpreted as a probability [19].
is the subset of AI. Machine learning is the study of These three operations are repeated over tens or
huncomputer algorithms that can be improved automat- dreds of layers, with each layer learning to detect
difically through experience [
1
]. Machine learning al- ferent features [16]. The classification phase consists
gorithms overcome following strictly static program from two layers dropout and fully connected. Dropout
instructions by making data-driven predictions or de- consists of randomly dropping out (setting to zero) a
cisions, through building a model from sample inputs number of output features of the layer during
train[
14
]. In machine learning, artificial neural networks ing [19].
are a family of models that mimic the structural ele- The fully connected layer that produces a vector of
gance of the neural system and learn patterns inherent K dimensions where K is the number of classes that
in observations [15], see Fig. 1. The term “deep” refers the network will be able to predict. This vector
conto the number of layers in the network—the more lay- tains the probabilities for each class of any image being
ers, the deeper the network [16]. Traditional neural classified [16, 21]. The quality of model is evaluated
networks contain only 2 or 3 layers, while deep net- by the cost function in fully connected layer (sigmoid,
works can have hundreds [16]. Deep learning has been softmax or other).
explosively developed today. Compared with shallow
learning, deep learning reaches the state of arts in many 2.2. Deep Belief Network
researches [17].
In contrast to the shallow architectures like kernel The power of Deep Belief Network (DBN) (Fig. 3 and
machines which only contain a fixed feature layer (or Fig. 4) lies in their ability to reconstruct both the input
base function) and a weight-combination layer (usu- vector and the learning feature vectors, which is
imally linear), deep architectures refers to the multi-layer plemented using a layer-by-layer learning strategy [22].
network where each two adjacent layers are connected Each layer of a DBN consists of a Restricted
Boltzto each other in some way [
3
]. This introduces the un- mann Machines (RBM). RBMs follow the principle of
precedented flexibility to model even highly complex, the probability distribution to complete its learning
cynon-linear relationships between predictor and out- cle [23]. Each RBM is concluded from a visible layer (v)
come variables, a quality that has allowed deep neural and a hidden layer (h). Number of neurons is set up
networks to outperform models from traditional ma- in each layer. The neurons between diferent layers
chine learning in a variety of tasks [18]. Deep learn- are fully connected, and the neurons in the same layer
ing methods have only now become so powerful, due are not connected [23]. When an RBM has learned, its
to technical reasons of: computational power (hard- feature activations are used as the “data” for training
ware), availability of large datasets and optimization the next RBM in the DBNs [24]. RBMs is as an
unsualgorithms [18],[19]. pervised network which considers the visible layer to
the hidden layer as a subnetwork. Then, this hidden
2.1. Convolution Neural Network layer is considered as a visible layer to the next layer
and so on [24]. The higher-level features are learned
Convolution neural network (CNN) algorithm is sepa- from the previous layers and the higher-level features
rated into two main parts: feature detection and clas- are believed to be more complicated and better reflects
sification (see Fig. 2). the information contained in the input data’s
struc
Feature detection phase consist from convolution, tures [
3
]. DBN training is divided into two steps:
forpooling and rectified linear unit (ReLU) layers. Convo- ward pre-training process and reverse fine-tuning
prolutional filters activates certain features from data set cess [25]. During the pre-training phase, the RBMs are
2.5. Deep Q-Learning or Deep
Reinforcement Learning
Deep Q-Learning (DQL) or Deep reinforcement
learning (DRL) concept is replaceable in scientific literature
(6). In DQL is always used reinforcement learning
algorithm, or in DRL is often used Q-learning function,
because of it is dealing with high-dimensional state
space inputs [
30
], [
31
]. A reinforcement learning (RL)
process involves an agent learning from interactions
Figure 1: The connection of AI, ML and DL with its environment in discrete time steps in order to
update its mapping between the perceived state and a
probability of selecting possible actions (policy) [
32
].
trained one-by-one until the hidden layer of the last In other words, RL is commonly used to solve an
seRBM. During this phase, the parameter of each RBM quential decision making problem [
30
]. The RL
probcan be obtained [23]. The back-propagation network lem is normally formalized using the Markov decision
(BP) is set in the last hidden layer of the DBN [25]. BP process (MDP) and includes a set of states S, set of
acis applied to fine tune the parameter using the output tions A, transition function T as action distributions,
labels of the sample data [23]. reward function R and discount factor [
33
]. The
solution to the MDP is a policy : S → A and the
pol2.3. Deep Boltzmann Machine icy should maximize the expected discounted
cumulative reward [
30
]. Q-learning, as a typical
reinforceDeep Boltzmann Machine (DBM) have only one undi- ment learning approach, mimics human behaviors to
rected network [24]. DBM as DBN is comprised of a take actions to the environment, in order to obtain the
Restricted Boltzmann Machines (RBM). The main dif- maximum long-term rewards [
34
]. The DQL process
ference is related to the interaction among layers of can be viewed as iteratively optimizing network
paRBMs [25]. For the computation of the conditional rameters process according to gradient direction of the
probability of the hidden units h1, both the lower visi- loss function at each stage [
35
]. Therefore, the
inexble layer v and the upper hidden layer h2 are incorpo- act approximate gradient estimation with a large
varirated, that makes DBM diferentiated from DBN and ance can largely deteriorate the representation
perforalso more robust to noisy observation [15]. There are mance of deep Q network by driving the network
pano direct connections between the units in the same rameter deviated from the optimal setting, causing the
layers. DBM parameters of all layers can be optimized large variability of DQL performance [
35
]. The
advanjointly by following the approximate gradient of a vari- tages of deep Q-learning is good results and ease of use
ational lower-bound on the likelihood objective [26]. (code can be modified easy for diferent physical
prob
Diferent from the DBN, the DBM can incorporate lems) [
36
].
top-down feedback, which can better propagate
uncertainty and hence deal with ambiguous input more
robustly [27]. 2.6. The Extreme Learning Machine
2.4. Deep Neural Network
Due to the novelty of the concept Deep neural network
(DNN) (Fig. 5)in the scientific literature can be
identiifed for all the algorithms analyzed in this paper.
However, in recent years the concept of DNN has become
known as Artificial Neural Network (ANN) with
hidden layers [
9
] [
28
]. DNN typically is feedforward
network so it can be understood as the Multilayer
Perceptron (MLP or MP). MLP consists of an input layer,
several hidden layers and one output layer ant it’s widely
used for pattern classification, recognition and
prediction [
29
].
The extreme learning machine (ELM) is a single-hidden
layer feedforward network, proposed by Huang in 2012.
In the traditional feed-forward ANN, the training of
the network is iterative, while the process is
transformed into an analytical equation in the ELM [
37
].
In ELM the weights between input and hidden layer
are assigned randomly following a normal distribution
and the weights between hidden and output layers are
learnt in single step by a pseudo-inverse technique.
During the training, the hidden layer is not learned
but the weight matrix of output layer is obtained by
solving the optimization problem formulated by some
learning criteria and regularizations [
38
], as showed
in the theory the output weights solved from
regularized least squares problem [
39
]. Therefore, ELM ofers
benefits such as fast learning speed, ease of
implementation, and less human intervention when compared to
the standard neural networks [
40
]
2.7. Generative Adversarial Network
same time, and finally the output is almost the same as
the real data [44].
The general idea of Generative adversarial network
(GAN) is that it aims to train a generator to recon- 2.8. Recurrent Neural Learning
struct high-resolution images for fooling a
discriminator that is trained to distinguish generative images Recurrent Neural Learning (RNN) (Fig. 9) is diferent
from real ones [41] (Fig. 8). This idea involves two from the traditional feedforward neural networks,
becompeting neural network models: one of them takes cause have feedback connections, which can be
benoise as input and produces some samples (genera- tween hidden units or from the output to the hidden
tor) and the other model (discriminator) accepts both units [44, 45]. This connections address the temporal
the data outputted by the generator and the real data, relationship of inputs by maintaining internal states
meanwhile, separates their sources [42]. The Discrim- that have memory . An RNN is able to process the
inator trains itself to discriminate real data and gen- sequential inputs by having a recurrent hidden state
erated data better while the Generator trains itself to whose activation at each step depends on that of the
ift the real data distribution so as to fool Discrimina- previous step [
5, 46
]. In other words, RNN not only
tor [43]. These two neural networks are trained at the processes the current element in the sequence, but also
However, it has been observed that it is dificult to train the RNNs to deal with long-term sequential data, as the gradients tend to vanish [5].
2.9. Long short-term memory
rent information should be treated as input in order to
generate the current state [51], whilst the forget gate
determine which information to be forgotten from the
memory state [52]. Finally, the output gate filters the
information that can be actually treated as significant
and produces the output [52]. The “gate” structure is
implemented using the sigmoid function, which
denotes how much information can be allowed to pass.
For one hidden layer in LSTM, activation function is
used in forward propagation, and gradient is used in
backward propagation [
38
].
2.10. Gated recurrent unit
Gated Recurrent Unit (GRU) is aimed to solve the van
ishing gradient problem which comes with a standard
RNN [
53
]. GRU consists of two gates: update gate (zt)
and reset gate (rt). Update gate decides how much
the unit updates its activation, or content, and reset
gate allows forget the previously computed state [
54
].
GRU is a less complex compared with LSTM, it does
not possess any internal memory and output gate like
LSTM [49].
Methods
Articles were included from electronic libraries: Sci
ence Direct, IEEE, Scopus, ACM, Emerald,
SpringerLink, JSTOR, EBSCO and others.
Analyzed period started from 2017 till 2020. The
review was conducted in January 2020. Keywords “Deep
learning” and “Finance” were used for the article’s
selection. All methods presented in this review matches
a term “Deep learning”, wherefore individually search
by each method was not developed. The same with a
term “Finance”, which includes accounting, financial
markets, risks and etc. Therefore, this paper presents
a big picture of developing scientific articles in Deep
learning in Finance category. 33 papers were selected
and analyzed. The analyzed articles can be categorized
by the problematic of given task: to predict future
returns or two make classification of results. Sometimes,
for better results are used natural language processing
algorithms (Fig. 12 and Tab 1).
The classification algorithms in finance most often
have been applied for credit scoring, which divides
loans into “good” and “bad”. For this problem
solving author’s used DBN [
29
], modified LSTM [52] and
CNN [
55
] networks. The results cannot be compared
due to diferent classifier evaluation methods used and
data source diferences. In credit scoring topic is a big
problem with unbalanced data set, i.e. authors [
55
]
used data set, where credit worthy instances were 91,55
proc. and CNN accuracy rate was 91,64 proc. In the
bankruptcy and investment market structure was used
CNN network or in tax evasion DQL network.
Articles in financial field is interested to obtain knowledge
from words and used it as some indicators. Therefore,
is seen a trend to use natural language processing
techniques. The goal of natural language processing (NLP)
is to process text using the computational linguistics,
text analysis, machine learning, statistical and
linguistics knowledge in order to analyze and extract
significant information [
56
]. Researches in financial field
are using sentiment analysis for better stock price
prediction or bankruptcy classification. Sentiment
analysis is the essential task for NLP, which can be divided
into three categories: lexicon-based sentiment
analysis, machine learning-based sentiment analysis and
the hybrid approach [
56
].
Lexicon-based sentiment analysis was used only in
one article [
11
], due to the need of opinion lexicon in
this field. Machine learning-based sentiment
analysis uses in bag-of-words method [48], [
57
] and word
embeddings [
48, 58, 57, 10
] with CNN [
58, 57
], LSTM
[
48, 10
] methods.
In [
4
] research was used bag of words and word
embeddings methods for LSTM, results showed that
LSTM models can outperform all traditional machine
learning models based on the bag-of-words approach,
especially when further pre-train word embeddings
with transfer learning. The main financial article’s
focus is in future returns prediction, especially in stock
prices or stock indexes. The main reason is data source
availability for scientific research. In this field very
often, scientific researches combine diferent methods
together [
49, 59, 60
] or make some model
modification’s [
13, 50, 61, 62
] for better prediction results. Some
authors [
48, 63
] analyze several diferent deep learning
models results for the deeper future model
development, see Fig. 13.
Most popular methods are CNN and LSTM.
However, DBM and GAN method’s was not found any
adjustment in finance field.
In some papers data is not normalized, i.e.
cryptocurrency prices [51] or demand [18]. Therefore,
predictive accuracy measurements, such as RMSE, MPE
and others, can be comparable with diferent other
authors works or sometimes even in the same paper, i.e.
RMSE for Bitcoin is 2.75×103 or for Ripple 0.0499 [51].
4. Conclusions
learning machine, generative adversarial network,
recurrent neural learning, long short-term memory, gated
recurrent unit; and it’s applicability in finance field.
This review reveals that financial article’s:
1. mainly focus for the forecasting task than
classification;
2. starts using natural language processing
techniques, mostly sentiment analysis, for better
results prediction;
3. uses not ‘basic’ the deep learning methods, i.e.
they are often combined with several diferent
models or merged to voting classifier.
Furthermore, this analysis has shown the importance
of balanced data set and normalization of the data, which
is submitted to deep learning networks.
The main limitation of this work is representation
only a big picture of developing scientific articles in
Deep learning in Finance category. Therefore, in
future research is needed to extend search keyword’s in
electronic libraries, i.e. search by each method