Comparing the prediction accuracy of LSTM and arima models for time-series with permanent fluctuation
| Autor | Ghahreman Abdoli - Mohsen MehrAra - Mohammad Ebrahim Ardalani |
| Cargo | Professor at University of Tehran, Faculty of Economics - Professor at University of Tehran, Faculty of Economics - Ph.D. Candidate at University of Tehran at Alborz campus |
| Páginas | 314-339 |
Periódico do Núcleo de Estudos e Pesquisas sobre Gênero e Direito
Centro de Ciências Jurídicas - Universidade Federal da Paraíba
V. 9 - Nº 02 - Ano 2020
ISSN | 2179-7137 | http://periodicos.ufpb.br/ojs2/index.php/ged/index
314
COMPARING THE PREDICTION ACCURACY OF LSTM AND
ARIMA MODELS FOR TIME-SERIES WITH PERMANENT
FLUCTUATION
Ghahreman Abdoli1
Mohsen MehrAra 2
Mohammad Ebrahim Ardalani3
Abstract: In developing countries with
an unstable economic system, permanent
fluctuation in historical data is always a
concern. Recognizing dependency and
independency of variables are vague and
proceeding a reliable forecast model is
more complex than other countries.
Although linearization of nonlinear
multivariate economic time-series to
predict, may give a result, the nature of
data which shows irregularities in the
economic system, should be ignored.
New approaches of artificial neural
network (ANN) help to make a
prediction model with keeping data
attributes. In this paper, we used the
Tehran Stock Exchange (TSE) intraday
data in 10 years to forecast the next 2
months. Long Short-Term Memory
(LSTM) from ANN chooses and outputs
compared with the autoregressive
1 Professor at University of Tehran, Faculty of Economics, email address: abdoli@ut.ac.ir
2 Professor at University of Tehran, Faculty of Economics, email address: mmehrara@ut.ac.ir
3 Ph.D. Candidate at University of Tehran at Alborz campus, email address: ardalani@ut.ac.ir
integrated moving average (ARIMA)
model. The results show, although, in
long term prediction, the forecast
accuracy of both models reduce, LSTM
outperforms ARIMA, in terms of error of
accuracy, significantly.
Keywords: Prediction Model, LSTM,
ARIMA, Forecast Accuracy, Tehran
Stock Exchange.
Introduction
The Stock market prediction
itself is violent and high risk, due to
equivocal and unforeseeable nature of it,
as well as complicated financial
indicators. It is affected by noise and
many hidden factors. No one knows the
exact time of bearish or bullish the
market. The Forecast situation becomes
more difficult when the exchange market
Periódico do Núcleo de Estudos e Pesquisas sobre Gênero e Direito
Centro de Ciências Jurídicas - Universidade Federal da Paraíba
V. 9 - Nº 02 - Ano 2020
ISSN | 2179-7137 | http://periodicos.ufpb.br/ojs2/index.php/ged/index
315
reflects an economic system without
stability, regulation and money
discipline. Developing countries with
permanent fluctuation in economic data
are the first candidate. Developing
countries involve a variety of
fundamental concerns in economics such
as poverty, the relations between state
and market, the gap between rich and
poor, population growth, structural
change, international debt and finance
and in some cases permanent sanctions.
Somewhat because of this, economic
data and time-series generated in these
countries, volatile abruptly over time in
response to political currents, in such a
way that causes the major roots of
changes to be hidden. (Mukherjee et al.,
2013) In this situation, even if a research
group wants to make a prediction model,
in the process of converting related
qualitative indexes into quantitative
indexes, it is inevitable to keep all the
time-series nature. Data carries all the
economic, social and even political
events that have happened through the
time. Regarded to permanent
fluctuations and unrecognized
independent variables, we must make
data fit for any regression or modeling. It
means, differentiation comes to help and
subsequently we ignore the core of data.
We go just for making a statistical model
which most of the time does not reflect
the reality of the market. Thus; making a
robust forecasting model for a stock
market in developing countries given by
unmodified data is a challenging issue.
Linear regression is one of the
solutions for stock market prediction
however, the idea works better for short
terms. (Ariyo et al., 2014) Besides, as
mentioned above, the linearization of
nonlinear multivariate economic time-
series to regress, may not give the best
response.
Cutting-edge approaches in
data analysis allow gaining better
forecast models. With the emerge of
neural networks, researchers went for
assessing its accuracy of prediction
versus linear forecasting models such as
ARIMA. Results in most of the time
talked about the benefits of non-linear
solutions. (Lin and Yeh, 2009; Binner*
et al., 2005) However, in some research,
ARIMA (C.-C. Wang et al., 2011) and
sometimes hybrid models (Hansen and
Nelson, 2003) resulted better.
There is a wide spectrum of
applications for Deep-learning models in
which financial forecasting is one of
Para continuar a ler
PEÇA SUA AVALIAÇÃO