A LSTM-Method for Bitcoin Price Prediction: A Case Study Yahoo Finance Stock Market

Abstract

Bitcoin is a kind of Cryptocurrency and now is one of type of investment on the stock market. Stock markets are influenced by many risks of factor. And bitcoin is one kind of cryptocurrency that keep rising in recent few years, and sometimes sudden fall without knowing influence behind it on the stock market. Because it’s fluctuations, there’s a need and automation tool to predict bitcoin on the stock market. This research study learns how to create model prediction bitcoin stock market prediction using LSTM, LSTM (Long Short Term Memory) is another type of module provided for RNN later developed and popularized by many researchers, like RNN, the LSTM also consists of modules with recurrent consistency. The Method that we apply on this research, also technique and tools to predict Bitcoin on stock market yahoo finance can predict the result above $ 12600 USD for next days after prediction, in the last section we make conclusions and discuss future works.

Full text

A LSTM-Method for Bitcoin Price Prediction: A
Case Study Yahoo Finance Stock Market
1
st
Ferdiansyah
Department of Informatics,
Universitas Bina Darma
Palembang, Indonesia
ferdi@binadarma.ac.id
4
th
Deris Stiawan
Faculty of Computer Knowledge,
Universiti Sriwijaya,
Bukit Besar, Palembang, Indonesia
deris@unsri.ac.id
2
nd
Siti Hajar Othman
School of Computing, Faculty of
Engineering,
Universiti Teknologi Malaysia,
Johor Bahru, Johor, Malaysia
hajar@utm.my
5
th
Yoppy Sazaki
Faculty of Computer Knowledge,
Universiti Sriwijaya,
Bukit Besar, Palembang, Indonesia
yoppysazaki@gmail.com
3
rd
Raja Zahilah Raja Md Radzi
School of Computing, Faculty of
Engineering,
Universiti Teknologi Malaysia,
Johor Bahru, Johor, Malaysia
zahilah@utm.my
6
th
Usman Ependi
Department of Informatics,
Universitas Bina Darma
Palembang, Indonesia
u.ependi@binadarma.ac.id
Abstract—Bitcoin is a kind of Cryptocurrency and now is one of
type of investment on the stock market. Stock markets are
influenced by many risks of factor. And bitcoin is one kind of
cryptocurrency that keep rising in recent few years, and
sometimes sudden fall without knowing influence behind it on the
stock market. Because it’s fluctuations, there’s a need and
automation tool to predict bitcoin on the stock market. This
research study learns how to create model prediction bitcoin
stock market prediction using LSTM, LSTM (Long Short Term
Memory) is another type of module provided for RNN later
developed and popularized by many researchers, like RNN, the
LSTM also consists of modules with recurrent consistency. The
Method that we apply on this research, also technique and tools
to predict Bitcoin on stock market yahoo finance can predict the
result above $ 12600 USD for next days after prediction, in the
last section we make conclusions and discuss future works.
Index Terms—Cryptocurrency, Bitcoin prediction, Bitcoin
Stock Market Prediction, LSTM.
I. I
NTRODUCTION
Cryptocurrency has been around for several years and has
now become quite popular, widespread, and also surrounded
and there is a lot of controversy from innovative developments.
Cryptocurrencies are a digital currency where transactions
can be done by online transactions, unlike the common
currency, cryptocurrency is designed based on cryptography.
Bitcoin is one kind of Cryptocurrency no regulation from
any party and decentralized. The unique characteristic of
Bitcoin is daily price fluctuations and always change every
day. The value Bitcoin Exchange rate to (USD) is $ 12,354.73
USD on 28 June 2019 in yahoo finance stock market[1] and
sometimes keep rising and sudden fall on march the value is $
3900 USD.
The Stock markets are influenced by many uncertainties
factor such as political issue, the economic issue at impacted
to local or global levels. To interpretation key of success,
factor to providing accurate predictions is complicated work.
For the market, we can analyze with any techniques such as
technical indicator, price movements, and market technical
analysis.[2]
To solve the problem above, regarding the fluctuations
there’s a need automation tool for prediction to help investors
decide for bitcoin or other cryptocurrency market investment.
Nowadays the automation tools are usually used in common
stock market predictions, and we can do the same works and
strategy on this domain cryptocurrencies.
LSTM (Long Short Term Memory) is another type of
module provided for RNN. LSTM was created by Hochreiter
& Schmidhuber (1997)[3] and later developed and
popularized by many researchers. Like RNN, the LSTM) also
consists of modules with recurrent consistency.
This paper studies about bitcoin and stock market
predictions, method, technique, and tools from a big number
of resources paper, and other available sources.
II. L
ITERATURE
R
EVIEW
A. Cryptocurrency and Bitcoin
The history of cryptocurrency (Cryptographic currency)
begins in the 1980s started with David chaum, In his paper, he
proposed a novel of a cryptographic scheme to blind the
content of the message before it is signed so that the signer
cannot determine the content. These blind signatures can be
publicly verified just like a regular digital signature. Chaum
proposed digital cash approach in such a way that is
untraceable by another party.[4]
The rise of cryptocurrency started on B-money In 1998,
Wei Dai proposed b-money[5], an anonymous and distributed
electronic cash system, In that method, describes two protocols
based on network that cannot be traced, where senders and
receivers are identified only by digital such as their public
keys, and each message will be signed by its sender to receiver.
206
Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on March 01,2021 at 14:12:51 UTC from IEEE Xplore. Restrictions apply.

Bit Gold In 1998, Nick Szabo[6] propose models a new
digital currency, the models based on cryptographic system
puzzles, which after being solved, were sent to the Byzantine-
fault-tolerant public registry and assigned to the public key of
the solver.
Hashcash proposed by Adam Back, Haschash, a system
relied on a cryptographic hash function to derive a probabilistic
proof of computational work as authentification system Pow
(Proof of work)[7]
And the last is RPOW Hal finney propose currency system
based on a reusable proof of work (RPOW) in 2004[8].
Between 2008 and 2009, Bitcoin was made as to the first
decentralized cryptocurrency by Satoshi Nakamoto. Nakamoto
published the Bitcoin whitepaper in 2008[9], and after January
3
rd,
2009, the genesis block of the bitcoin protocol was created.
Nowadays it is most successful cryptocurrency in terms of
market capitalization, beside above 700 altcoins that circulated
in the world (eg. Litcoin, Etherum) based on Bitcoin have been
proposed since the launch of Bitcoin.
B. Bitcoin Price Predictions Methods
There are many people doing research about the prediction
of cryptocurrency. Greaves et al. [10] is a proposed technique
using Logistic Regression and SVM and analyzed using Graph
to predict bitcoin price. Huisu Jang et al. [11] they concern
about a study on modeling and prediction bitcoin with
Bayesian Neural Network and giving some knowledge about
bitcoin. Edwin sin et al. [12] provide topic Bitcoin price
prediction using Ensemble of Neural. Networks. Arief Radityo
et al.[2] proposed a prediction of bitcoin using Artificial Neural
Network Technique. They combine with market technical
indicators but the results are worse of performance and training
time
Papers using LSTM, Sean et al [13] They propose method
the price of Bitcoin using RNN and combine Using Recurrent
Neural Network and Long Short Term Memory and Ruchi
Mittal et al [14] is propose an Automated cryptocurrencies
prices prediction using machine learning technique based on
the historical trend (daily trend). Chih-Hung et al. [15] are
created new forecasting framework bitcoin price using LSTM,
they proposed with two various LSTM models (conventional
LSTM and LSTM with AR(2) model) with 208 records dataset,
compared with MSE, RMSE, MAE, and MAPE. Fei Qian et
al.[16] produced a common stock market prediction model
based on LSTM under Different factor that impacts the market,
in this research their selected three stock with similar trends.
The LSTM prediction model is performed well.
The researches above proposed various method to
prediction bitcoin. In this paper, we analyze and constructing a
model to predict bitcoin using LSTM
C. Bitcoin price predictions Technique
Technically Bitcoin stock market prediction it’s the same
with prediction technique on the common stock exchange but
in the other way, when you try another technique and strategy
like sentiment analysis maybe we can’t get different results
orthe strategy it does not work. Because beside many factors
that impact on stock exchange prediction, Bitcoin it’s
decentralized and not regulated by any party so it’s different
from common currency or the common stock market.
We can use the same algorithm to Bitcoin prediction using
Machine Learning (eq. SVM, Naïve Bayes, Regression)
[10][11] or any other Advance Machine learning technique to
improve the results like Deep Learning using Neural Network
(eq. ANN and RNN)[2]
On Prediction, we can predict Bitcoin using technique on
the specific subject that we wanted. Example, we want to
predict only by the signal or the price, or we can predict just for
current day or next day close value based on Long Short Term
Memory (LSTM),[14] historical price and other technique like
regime prediction to detect current day’s trend on market, to
help investor to make decision to investment.
To make more accurate and enrich the result we can
combine the prediction algorithm with another method or
technique on prediction Bitcoin
In this section, we describe some of the technique that
mentions previous research on some papers.
Market technical analysis is a method that studies price
movements by looking at historical price data that occur in the
market through media charts. By studying this historical data a
conclusion can be drawn for making investment decisions in
the market.
There are several reasons why we should use TA, the first,
not necessarily fundamental analysis can be applied in trading.
I do not say Fundamental is not important, but here that needs
to be underlined fundamental analysis will be far more
important if our position as an Investor where the investment
period can take more than one year.
Secondly, with the media chart, it will be seen how the
journey of a stock price where it is very helpful for traders to
analyze in anticipation of future price changes and see patterns
of patterns that occur in the price movements of a stock, so
traders do not trade in ' darkness or without clear direction.[17]
Another technique to interpretation trend is Technical
Indicator [18]is a series of data points obtained by applying
the formula to securities price of data. The combinations of
price data, such as close, low, high, low and open can be used
as data point certain period of time.
Time series Data Analysis in the terms of future price
predictions popular methods is using Autoregressive
integrated moving average (ARIMA) models are a popular
choice for forecasting over a short term condition, it works
when data exhibits consistent or stable pattern (constant) over
time with least possible outliers. But this does not work
always in the real-time scenario, where data fluctuated
drastically and it is highly volatile.[19]
Trading Strategy, this strategy is that we maintain the
position of +1 Bitcoin, 0 Bitcoin or −1 Bitcoin. [20]
III. METHODOLOGY
LSTM (Long Short Term Memory) is another type of
module provided for RNN. LSTM was created by Hochreiter
& Schmidhuber (1997)[3] and later developed and
popularized by many researchers. Like RNN, the LSTM
network (LSTM network) also consists of modules with
207
Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on March 01,2021 at 14:12:51 UTC from IEEE Xplore. Restrictions apply.

recurrent consistency.
LSTM is an updated version from RNN, the difference is
the connection between the hidden layers of RNN. The
explanation structure of RNN is shown in Figure 1. RNN &
LSTM have a similar structure, the other different is memory
cell of structure hidden layer. And the design of three special
gates effectively solve the gradient problems. The LSTM
memory structure of hidden layer shown in Figure 2. [16]
Figure 1. The Expanded Structure of RNN[21]
In Figure 1 explains the RNN has shortcomings, the
shortcomings can be seen in the input X
o
, X
1
has a very large
range of information X
t,,
X
t+1
so that when 
1+1
requires
information those that are relevant X
o,
X
1
to RNN are unable to
learn to link information because of old memory saved will be
increasingly useless as time goes by because it is overwritten or
replaced with new memory, this problem was discovered by
Bengio, et al. (1994)[22] .
Unlike the RNN, LSTM does not have the disadvantage is
that LSTM can manage the memory at each input by using
memory cells and gate units.
Figure 2. LSTM memory cell structure of the hidden
layer[16]
•
i
= the input gate, it’s means
information will be updated in the cell
•
t
= the forget gate, it’s means
information should be dropped from the cell.
•
t
= the output gate, it means how much
information is output
•
t
= the candidate value for the states of the
memory cell at time t.
•
t
= the state of the current memory cell at time t,
which calculated by the combination of
I and
t
t
and
t -1
through element-wise multiapplication
•
t
= it’s mean output value filtered by output gate
•
• = is denoted denotes sigmoid function with the
range 0 to 1, the function is used to put the value
between -1 and 1.
A. Propose Method
Figure 3. Prediction Method
Based on Figure 3 The process start from collect the data,
dataset is collected from yahoo finance stock market based on
the USD Exchange rate and collected from CCC -
CryptoCompare. Currency in USD, with 5 years period from
27-06-2014 until 27-06-2019, this is historical data prices, so in
this research used time-series data on this experimental, with
1829 number of datasets, on CSV format. Figures 3 show the
sample data of datasets.
Figures 4. Sample data
208
Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on March 01,2021 at 14:12:51 UTC from IEEE Xplore. Restrictions apply.

IV. EXPERIMENTAL
AND
RESULTS
Figures 4 below shown pre-processing result to load the
dataset into machine and algorithm, and then shown the last
day close price data of bitcoin before we train and test and
predict the results.
Figures 5. Close price History data bitcoin
For Step Split and training the data, we divided 4 years period
of data for training and 1 year for testing. Split at 1462 data in.
While splitting the data into train and validation, we cannot use
random splitting since that will destroy the time
component[23]. So we set the last year’s data into test and the
4 years’ data before that into train the dataset.
A. Scenario analysis results using RMSE
Before finalizing the results we try to Measure the results
using RMSE, RMSE is (the Root Mean Square Error). The
RMSE will always be larger or equal to the MAE. The
RMSE metric evaluates how well a model can predict a
continuous value. The RMSE units are the same units as
your data’s dependent variable/target (so if that’s dollars,
this is in dollars), which is useful for understanding
whether the size of the error is meaningful or not. The
smaller the RMSE, the better the model’s
performance.[24]
This is the formula RMSE =
i - i
)
2
[25]
Where N is the total number of observations,
i
is the actual
value; whereas,
i
is the predicted value. The main benefit
using RMSE is that it penalizes large errors. It also scales the
scores in the same units as the forecast values).
We combine the epoch and model dropout to search the best
results.The number of epoch we used: 10, 100, 1000, 200, 400,
800, 2000, and 5000. With the number of model dropout we
combined with 0, 0,1 and 0,5.
No Epoch Model Dropout RMSE Results
1 10 0 631.74963
2 100 0 455.98107
3 1000 0 825.37505
4 200 0 360.64511
5 400 0 354.18368
6 500 0 288.59866
7 800 0 292.78967
8 2000 0 477.91428
9 5000 0 474930575
10 500 0,1 602.140637
11 500 0,5 313.6623
Table 1. RMSE Results
Based on Table 1 The smaller results is using 500
epoch with model dropout 0. With RMSE results 288.59866. as
RMSE mention on above, The smaller the RMSE, the better
the model’s performance, after that we use to produce the
results.
Figures 6. Prediction result price of bitcoin
Based on figures 6 that result from prediction using LSTM
shown by the graph with epoch 500, model dropout 0, and,
Yellow Line is a result for the close prediction, the Blue and
Green line is from Data Training. The price result is above
$12600 for next days based on the model.
V. C
ONCLUSION AND
F
UTURE
W
ORKS
Our Proposed model has been succeeded to provide the
result prediction bitcoin from yahoo finance stock market. Our
model with time series techniques can build produce the
results and the results can predict the price for the next days
with split the data to train and test that we mention in the
article above. But the disadvantage is the result it’s not good
enough regarding the RMSE, maybe under the hundreds or
209
Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on March 01,2021 at 14:12:51 UTC from IEEE Xplore. Restrictions apply.

near close to 50 Score RMSE. Afterward, as we mentioned
before in the article, the stock market is influenced by many
uncertainty factors. The Stock markets are influenced by many
uncertainties factor such as political issue, the economic issue
at impacted to local or global levels. So prediction price
bitcoin using LSTM can’t good enough to make the decision
to invest in bitcoin, it is another side for taking the decisions.
Future research will focus on, modified LSTM layers,
adding dropout and modified number of epochs, and using
different instability dataset to test how good the prediction
results or try to use sentiment analysis combined with LSTM
method to see the impact of the uncertainty in value bitcoin.
A
CKNOWLEDGMENT
This work was supported by the Department of
Informatics, Universitas Bina Darma. Jalan A. Yani No. 3
Palembang, Indonesia.
R
EFERENCES
[1] B. U. (BTC-USD) and C.-C. C. in USD, “Yahoo
Finance,” 2019. [Online]. Available:
https://finance.yahoo.com/quote/BTC-USD?p=BTC-
USD. [Accessed: 14-Mar-2019].
[2] A. Radityo, Q. Munajat, and I. Budi, “Prediction of
Bitcoin exchange rate to American dollar using
artificial neural network methods,” in Advanced
Computer Science and Information Systems
(ICACSIS), 2017 International Conference on, 2017,
pp. 433–438.
[3] S. Hochreiter and J. Schmidhuber, “Long short-term
memory,” Neural Comput., vol. 9, no. 8, pp. 1735–
1780, 1997.
[4] A. Judmayer, N. Stifter, K. Krombholz, and E.
Weippl, “Blocks and Chains: Introduction to Bitcoin,
Cryptocurrencies, and Their Consensus Mechanisms,”
Synth. Lect. Inf. Secur. Privacy, Trust, 2017.
[5] W. Dai, “B-money proposal,” White Pap., 1998.
[6] N. Szabo, “Bit gold,” Website/Blog, 2008.
[7] A. Back, “Hashcash.” 1997.
[8] Hal Finney, “RPOW - Reusable Proofs of Work,”
Agosto 15, 2004. 2004.
[9] S. Nakamoto and others, “Bitcoin: A peer-to-peer
electronic cash system,” 2008.
[10] A. Greaves and B. Au, “Using the bitcoin transaction
graph to predict the price of bitcoin,” No Data, 2015.
[11] H. Jang and J. Lee, “An Empirical Study on Modeling
and Prediction of Bitcoin Prices With Bayesian Neural
Networks Based on Blockchain Information,” IEEE
ACCESS, vol. 6, pp. 5427–5437, 2018.
[12] E. Sin and L. Wang, “Bitcoin Price Prediction Using
Ensembles of Neural Networks,” in 2017 13TH
INTERNATIONAL CONFERENCE ON NATURAL
COMPUTATION, FUZZY SYSTEMS AND
KNOWLEDGE DISCOVERY (ICNC-FSKD), 2017,
pp. 666–671.
[13] S. McNally, J. Roche, and S. Caton, “Predicting the
price of Bitcoin using Machine Learning,” in Parallel,
Distributed and Network-based Processing (PDP),
2018 26th Euromicro International Conference on,
2018, pp. 339–343.
[14] R. Mittal, S. Arora, and M. P. S. Bhatia,
“AUTOMATED CRYPTOCURRENCIES PRICES
PREDICTION USING MACHINE LEARNING,”
2018.
[15] C.-H. Wu, C.-C. Lu, Y.-F. Ma, and R.-S. Lu, “A New
Forecasting Framework for Bitcoin Price with
LSTM,” in 2018 IEEE International Conference on
Data Mining Workshops (ICDMW), 2018, pp. 168–
175.
[16] F. Qian and X. Chen, “Stock Prediction Based on
LSTM under Different Stability,” in 2019 IEEE 4th
International Conference on Cloud Computing and
Big Data Analysis (ICCCBDA), 2019, pp. 483–486.
[17] J. J. Murphy, “Technical Analysis Of The Financial
Markets,” Pennsylvania Dental Journal. 1999.
[18] E. Kristensen, S. Østergaard, M. A. Krogh, and C.
Enevoldsen, “Technical Indicators of Financial
Performance in the Dairy Herd,” J. Dairy Sci., 2008.
[19] C. Scheier and W. Tschacher, “Appropriate algorithms
for nonlinear time series analysis in psychology,” in
Nonlinear dynamics in human behavior, World
Scientific, 1996, pp. 27–43.
[20] D. Shah and K. Zhang, “Bayesian regression and
Bitcoin,” in 2014 52nd Annual Allerton Conference on
Communication, Control, and Computing (Allerton),
2014, pp. 409–414.
[21] M. W. P. Aldi, J. Jondri, and A. Aditsania, “Analisis
Dan Implementasi Long Short Term Memory Neural
Network Untuk Prediksi Harga Bitcoin,”
eProceedings Eng., vol. 5, no. 2, 2018.
[22] Y. Bengio, P. Simard, and P. Frasconi, “Learning
Long-Term Dependencies with Gradient Descent is
Difficult,” IEEE Trans. Neural Networks, 1994.
[23] AISHWARYA SINGH, “Stock Prices Prediction
Using Machine Learning and Deep Learning
Techniques (with Python codes),” OCTOBER 25,
2018, 2018. [Online]. Available:
https://www.analyticsvidhya.com/blog/2018/10/predic
ting-stock-price-machine-learningnd-deep-learning-
techniques-python/. [Accessed: 20-Jun-2019].
[24] Squark, “ROOT MEAN SQUARE ERROR OR
RMSE.” [Online]. Available:
https://squarkai.com/root-mean-square-error-or-
rmse/#.XWVCLegzZPY. [Accessed: 28-Aug-2019].
[25] J. Brownlee, “Time series prediction with lstm
recurrent neural networks in python with keras,”
Available Mach. com, p. 18, 2016.
210
Authorized licensed use limited to: UNIVERSITY TEKNOLOGI MALAYSIA. Downloaded on March 01,2021 at 14:12:51 UTC from IEEE Xplore. Restrictions apply.