You are on page 1of 21

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/347948296

Effectivness of Monte Carlo Simulation and ARIMA Model in Predicting Stock


Prices

Conference Paper · December 2017

CITATIONS READS

0 2,101

3 authors:

Almira Arnaut-Berilo Azra Zaimovic


University of Sarajevo University of Sarajevo
23 PUBLICATIONS 112 CITATIONS 41 PUBLICATIONS 141 CITATIONS

SEE PROFILE SEE PROFILE

Nedzmija Turbo-Merdan

3 PUBLICATIONS 0 CITATIONS

SEE PROFILE

All content following this page was uploaded by Azra Zaimovic on 22 May 2021.

The user has requested enhancement of the downloaded file.


ISSN 2490-2616

University of Tuzla
Faculty of Economics

Internacional Scientific Conference


"Economy of Integration"

CONFERENCE
PROCEEDINGS
Tuzla, December 7th-9th, 2017

e os
EKONOMSKI FAKULTET PODGORICA
5th International Scientific Conference
“Economy of Integration”

ICEI 2017
“The Role of Economic Thought in Modern Environment”

CONFERENCE PROCEEDINGS
Editors:
Emira Kozarević
Jasmina Okičić

University of Tuzla
Faculty of Economics
7th-9th December, 2017
Tuzla, Bosnia and Herzegovina
Almira Arnaut-Berilo, PhD
Sarajevo School of Economics and Business, University of Sarajevo, BiH
E-mail: almira.arnaut@efsa.unsa.ba

Azra Zaimović, PhD


Sarajevo School of Economics and Business, University of Sarajevo, BiH
E-mail: azra.zaimovic@efsa.unsa.ba

Nedžmija Turbo-Merdan, MA
E-mail: nedzmija_t@hotmail.com

EFFECTIVENESS OF MONTE CARLO SIMULATION


AND ARIMA MODEL IN PREDICTING STOCK PRICES

EFEKTIVNOST MONTE CARLO SIMULACIJE I ARIMA


MODELA U PREDVIĐANJU CIJENA DIONICA

Abstract

The efficient market hypothesis states that stock prices are a reflection of information
and rational expectations, and all new information about the issuer are almost
immediately reflected in the current stock price. Stock prices should not be accurately
predicted by looking at the historical prices. Stock prices could be described by
statistical process called “random walk”, i.e. single day’s deviations from the central
value are random and unpredictable.
The aim of this research is to test the effectiveness of Monte Carlo simulation model and
Autoregressive Integrated Moving Average model (ARIMA) in predicting stock prices.
Selected methods use different input data in predicting stock prices, including different
time horizons; ARIMA model uses time series data to predict future prices, while Monte
Carlo simulation uses random walk component. We compare daily predictions of stock
prices estimated by both methods for five companies from Dow Jones Industrial Average
(DJIA) index. Predictions have been generated for one month period, from 10th March
2016 until 10th April 2016. For the purpose of evaluation of prediction models we use
Mean Absolute Error (MAE), Mean Absolute Percent Error (MAPE), Mean Square
Error (MSE) and Root Mean Square Error (RMSE) methods.
We came to the conclusion that there is a slight advantage in accuracy of predictions
of ARIMA model for shorter periods (5 days), while Monte Carlo simulation has
undoubted advantage for longer periods (21 days). Also, we noticed linkages between

946
single stock’s predictions for periods of 5, 10 and 15 days for both models. Our results
show that in shorter periods stock prices have elements of predictability, what questions
the level of efficiency of the world leading stock exchange. In longer periods (21 days)
results indicate market efficiency, i.e. model with random walk component (Monte Carlo
simulation) has advantage compared to the model based on historical prices (ARIMA).

Keywords: efficient market hypothesis, ARIMA model, Monte Carlo simulation,


random walk

JEL: G11, G32

Sažetak

Hipotezom o efikasnosti tržišta tvrdi se da su cijene dionica odraz informacija i


racionalnih očekivanja, te da se sve nove informacije o emitentu gotovo odmah
bivaju uključene u trenutnu cijenu dionice. Cijene dionica se ne bi trebale moći
precizno predvidjeti na bazi historijskih cijena. Drugim riječima, cijene dionica se
mogu opisati putem statističkog procesa nazvanog “slučajni hod”, tj. pojedinačna
dnevna odstupanja od srednje vrijednosti su slučajna i nepredvidiva.
Cilj ovog istraživanja je testiranje efektivnosti Monte Carlo simulacije i ARIMA modela
u predviđanju cijena dionica. Odabrane metode koriste različite ulazne podatke
u predviđanju cijena dionica, uključujući različite vremenske intervale; ARIMA
model koristi vremensku seriju podataka u predviđanju budućih cijena, dok Monte
Carlo simulacija koristi element slučajnog hoda. Mi poredimo dnevna predviđanja
cijena dionica procijenjena na bazi obje metode za pet dionica iz indeksa Dow Jones
Industrial Average (DJIA). Predviđanja su generirana za jednomjesečni period, od 10.
marta 2016. do 10. aprila 2016. godine. U svrhu procjene koji model predviđanja daje
bolje rezultate koristimo metode Mean Absolute Error (MAE), Mean Absolute Percent
Error (MAPE), Mean Square Error (MSE) i Root Mean Square Error (RMSE).
Došli smo do zaključka da je blaga prednost u tačnosti predviđanja na strani ARIMA
modela za kraće periode (5 dana), dok Monte Carlo simulacija ima nedvojbenu
prednost kada se radi o dužim periodima predviđanja (21 dan). Također, uočili smo
povezanost između predviđanja za pojedinačne dionice za periode od 5, 10 i 15
dana kod oba modela. Naši rezultati pokazuju da cijene dionica imaju elemente
predvidivosti u kraćim periodima, što dovodi u pitanje nivo efikasnosti vodeće berze
u svijetu. U dužim periodima (21 dan) rezultati upućuju na tržišnu efikasnost, tj.
model s komponentom slučajnog hoda (Monte Carlo simulacija) ima prednost u
odnosu na model baziran na historijskim cijenama (ARIMA).
Ključne riječi: hipoteza o efikasnosti tržišta, ARIMA model, Monte Carlo simulacija,
slučajni hod
JEL: G11, G32

947
1. Introduction

Interest in the topic of predicting the movements of stock prices experienced


its expansion in 1930’s, when world’s stock exchanges started its faster
development. Since then, this issue is studied intensively and as a result a
large number of different models that attempt to predict the future path of
movement of stock prices have been developed. Some of these models are
neural networks, fuzzy inference systems and machine learning techniques,
various ARIMA models, Monte Carlo simulation and GARCH model.

The aim of this research is to test the effectiveness of Monte Carlo (MC)
simulation model and autoregressive integrated moving average model
(ARIMA) in predicting stock prices. Selected methods use different input
data in predicting stock prices, including different time horizons; ARIMA
model uses time series data to predict future prices, while Monte Carlo
simulation uses random walk component. Monte Carlo simulation model
is being used for valuation of options, while due to high volatility it is not
widely accepted for stock prices predictions. The aim of our research is to
test whether the MC model can be competitive to one of widely used models
for short-term forecasting of stock prices, namely ARIMA model. The basic
idea of this research is to compare different models; the sample selection, we
think, doesn’t diminish the importance of the research.

We carried out analysis of comparison of the predicting power of two models


on five companies from Dow Jones Industrial Average (DJIA) listed on the New
York Stock Exchange: Pfizer, Coca-Cola, General Electric, Exxon and 3M.

For an initial insight into the dynamics of stock prices, daily prices for all five
companies are used in the period from March 2006 till March 2016. In the next
step, we forecasted stock prices by both models for one month period from
10th March till 10thApril 2016, i.e. 21 active trading days on NYSE. Individual
forecasted prices are compared with actual prices, and then forecasted prices
by both models are compared to each other.

For comparison were used statistical tools for evaluating the accuracy of
forecasting models (mean absolute error, mean absolute percentage error and
root mean square error).Vertical analysis is done for deriving conclusions
regarding the forecasting power of two models for time horizons of 5, 10, 15
and 21 days. The paper is divided into five parts. Section 2 Literature review

948
provides an overview of theoretical background of the research on efficient
markets and forecasting models, and shows the results on previously conducted
researches. Section 3 explains the methodology and data used for the analysis.
Section 4 gives the results of ARIMA and MC forecasting process. Finally,
brief summary and concluding remarks are given in Section 5.

2. Literature review

Efficient market hypothesis (EMH) states that asset’s prices fully reflect all
available information in the market, so it is not possible to beat the market,
since prices change only due to new information or changes in investors’
required rates of return (Malkiel & Fama, 1970). An implication of EMH is
that stocks are always in equilibrium, i.e. stocks trade at their fair value. It
is impossible to gain excessive returns or to outperform the market through
fundamental or technical analysis and market timing. Above average returns
are connected with investments in riskier securities.

Random walk theory is stock market theory that stats that stock price changes
are random, i.e. next price changes are random from previous price. According
to this theory stock price changes have the same distribution and they are
independent from each other, meaning that past movements of stock prices
or direction of stock prices cannot be used to predict future prices. The idea
is that information flow unhindered and all new information are immediately
embedded in stock price, so tomorrow’s stock price changes will reflect only
tomorrow’s information, independent from today’s price change. The idea is
also referred to as the weak form of efficient market hypothesis. Dupernex
(2007) on the other hand states that random walk theory does not imply that
stock market is efficient with rational investors, as efficient market theory stats.

Dai and Zhang (2013) find US stock market semi-strong efficient, meaning
that all publicly available information is included in the current price. They
used machine learning theory to confirm the semi-strong efficiency; prediction
of 3M stock next day price had low accuracy of 50%, while the accuracy
increased to 79% when authors tried to predict long-term trend.

Xu and Berkely (2012) used information from Yahoo Finance and Google
Trend to simplify the process of evaluation of online news. The author
analyses Apple Inc. and combines the conventional time series analysis
technique (ARMA - autoregressive moving average) with the information

949
from the Internet to predict weekly changes in stock price. Important news
related to a selected stock over a five-year period are recorded and the weekly
Google trend index values on this stock are used to provide a measure of
the magnitude of these events. The result of this experiment has shown
significant correlation between the changes in weekly stock prices and the
values of important news computed from the Google trend website. Author
concludes that algorithm proposed in this study can potentially outperform
the conventional time series analysis in stock price forecasting.

Autoregressive integrated moving average (ARIMA) is generalization of an


autoregressive moving average (ARMA) model fitted to time series data for their
better understanding or forecasting future values (Box & Jenkins, 1970). This
model is one of the best methods in financial forecasting, and different studies
have shown its’ efficient ability of short-term forecasting (Ariyo, Adewumi & Ayo,
2014). ARIMA model is more robust and more efficient in time series forecasting
even then popular Artificial Neural Networks technic (Kyungjoo, Sehwan & John,
2007; Merh, Saxena & Pardasani, 2010; Sterba & Hilovska, 2010).

Mondal, Shit and Goswami (2014) conducted a test of ARIMA model on


sample of 56 stocks from seven sectors listed at the National Stock Exchange
in India. The accuracy of predictions was above 85% for all sectors, with the
best accuracy in consumer goods sector.

Monte Carlo (MC) simulation is a quantitative technic developed from John


von Neumann during the World War II (Eckhardt, 1987). The basic idea is
using randomness to solve problems that might be deterministic in principle.
Monte Carlo simulation finds its application in much wider areas then financial
forecasting. Some examples we can see in papers from Tadeu et al. (2012),
Whiteside (2008), Hoesli, Jani and Bender (2005) as well as Borkar, Cevher
and McClellan (2007).

Boyle (1976) shows that MC simulation is useful for option pricing when
underlying stock returns are generated as join of continues as well as sudden
processes, and that it can be used for numerical forecasting of European stock
call options that pay dividends. Jabbour and Liu (2005) find that the accuracy
of MC simulation is increased by larger number of attempts in simulations.

Clewlow and Strickland (1998) and Hull (2012) state that MC is computer
inadequate because of the high generated variances. Landauskas (2011) in his

950
paper compares standard MC simulation with Markov chain MC simulation
(MCMC). After 300 executed trajectories, average stock price after 50 trades
was very similar in both methods ($ 18.08 and $ 18.10). Author also states
that larger number of intervals used in model development leads to higher
accuracy, but also demands more time to get result.

3. Methodology and data

ARIMA model as one of the most prominent models for financial forecasting
can be described as follows:
Yt = φ0 + φ1Yt-1 + φ2Yt-2 + ... + φpYt-p + εt ‒ θ1εt-1 ‒ θ2εt-2 ‒ θqεt-q, (1)
Where symbols represent:
■■ Yt– forecasted stock price in period t (day, month, year or similar);
■■ Yt-p– real stock price in period t‒p ;
■■ φp and θq – coefficients;
■■ p and q – order of autoregressive component and moving average,
respectively.

The main task of time series analysis is discovering of model that will
appropriately describe the process that is being generated by time series.
Box-Jenkins approach to model selection is one of the most often used
methods for time-series analysis. This approach is characterized by four
stages: identification stage, estimation stage, diagnostic checking stage and
forecasting stage (Bahovec & Erjavec, 2009). The application of the ARIMA
methodology for time series analysis is due to Box and Jenkins (1970).

The general statistical methodology of ARIMA model can be presented as


follows:
Step 1 - The model is identified for the observed data.
Step 2 - The model parameters are estimated.
Step 3 - If the hypotheses of the model are validated, go to Step 4, otherwise
go to Step 1 to refine the model.
Step 4 - The model is ready for forecasting.

The main idea of MC simulation is to perform an experiment by simulation


with probabilistic components of system, whereby random numbers are being
generated in order to assign values to the components of system (Render,
Stair & Hanna, 2012).

951
According to Crnjac-Milić and Masle (2013) MC simulation was found
useful when stock returns have been generated from continuous and discrete
returns. MC simulation model is being derived from Markov process, Wiener
process and Brownian motion. If we analyze discrete time period, model can
be presented as follows:

ΔS =μSΔt + σSϵ√Δt (2)

Where symbols represent


• ΔS – change in stock price;
• μ – stock return;
• Δt – time in which stock price is being observed (day, month, year or similar);
• S – initial stock price or price from the last observed period;
• σ – standard deviation and
• ϵ – random component.

In equation (2), μΔt is expected return, while σϵ√Δt is a stochastic component


of return (Hull, 2012). Random component ϵ simulates Markov process,
meaning that selected samples of ϵ should be independent from each other.

For the comparison of the forecasts obtained through the Monte Carlo
simulation and ARIMA model, price movements for next five companies
are analyzed: Pfizer, Coca-Cola, General Electric, Exxon and 3M. These
companies are part of Dow Jones Industrial Average (DJIA) index and they
are also quoted on the New York Stock Exchange (NYSE). The historical
prices of all five companies are analyzed in the period from March 2006 to
March 2016. The aim is to test how the results of the ARIMA model and
Monte Carlo simulation are changed if the actual stock prices of companies
vary in different ways, regardless of the industry to which the company
belongs. For each model, a set of steps is implemented by using Eviews 9
SV and MS Excel 2010. The prices are forecasted at the daily level for the
period from 10 March to 10 April 2016, or 21 days of active trading. The
observation period is divided into four parts: 5, 10, 15 and 21 trading days.
First, the individual forecasted prices for both models are compared with the
actual ones. Then, forecasted prices by both models are compared with each
other. To help with the comparison, statistics are used to evaluate the accuracy
of prognostic models: mean absolute error (MAE), mean absolute percentage
error (MAPE) and root of a mean square error (RMSE).

952
4. Results

The subsections below describe the processes of ARIMA model development


for Pfizer Inc.

4.1. ARIMA model for Pfizer Inc.

Pfizer stock data used in this study cover the period from March 2006 to
March 2016 having a total number of 2507 observations. Figure 1 shows the
original pattern of the series and gives a general overview whether the time
series is stationary or not. From the graph below we can see that the time
series have random walk pattern.

Figure 1. Movements of daily closing prices of Pfizer Inc. Source

Source: Authors’ research (Eviews 9 SV)

By visual analysis we can conclude that this is a non-stationary series that


follows a random walk pattern. The analysis of the correlogram provides
additional confirmation that this is a non-stationary series. After the first
differentiation of the series the graph and the correlogram indicate stationarity.
In order to obtain a formal confirmation that it is a stationary series after
the first differentiation the Augmented Dickey-Fuller (ADF) unit root test is
obtained (Figure 2).

953
Figure 2. ADF unit root test for D(PFIZERCL)

Source: Authors’ research (Eviews, 9 SV)

In the analyzed case, the ADF test value is -37.84941, which is far less than
the critical value of the test (-3.43; -2.86; -2.56), which leads to a strong
rejection of the zero hypothesis that D(PFIZERCL) has a unit root, and
accepting an alternative hypothesis that D(PFIZERCL) has no unit root with
the usual level of significance.

The parameter I in the ARIMA model measures the integrations of the model
(I between AR and MA) and in the case of the observed model it is 1, which
means that the series is the first-order differential. Table 1 shows the different
parameters of the autoregressive (p) and moving average (q) among the
several ARIMA model experimented upon. Comparison is made so that the
optimal model is one which has: (1) the highest value of the coefficient of
determination, (2) the lowest value of the Schwarz information criterion and
(3) the lowest standard error of regression. Among the models of integration
of order , the coefficient of determination is the highest for the model (1, 1, 2).
The standard error of regression is the lowest also for the model (1, 1, 2), and
the Schwarz information criterion is approximate to the parameters of other
models of the order of integration, so this model is taken as optimal.

Table 1. Statistical results of different ARIMA parameters for Pfizer stock

Coefficient of Schwarz Standard error of


ARIMA
determination criterion regression
(1,0,0) 0.997180 0.542848 0.315767
(1,0,1) 0.997183 0.545029 0.315682
(1,1,1) 0.003735 0.537663 0.315311
(1,1,0) 0.000843 0.537434 0.315706

954
(0,1,1) 0.000948 0.537329 0.315689
(2,1,0) 0.002827 0.535448 0.315392
(0,1,2) 0.003088 0.535187 0.315351
(2,1,1) 0.003723 0.537672 0.315313
(1,1,2) 0.003979 0.537416 0.315273
Source: Authors’ research

The parameter estimation is done (Figure 3) using the method of Maximum


Likelihood. The coefficient AR (1) is at the limit of statistical significance at
5% and amounts -0.029924. The coefficient MA (2) is statistically significant
and amounts -0.058546. SIGMASQ is a mark for the estimated variation of
innovation or “white noise”.

Figure 3. ARIMA (1,1,2) estimation output with D(PFIZERCL)

Source: Authors’ research (Eviews, 9 SV)

Based on the Durbin-Watson test results we conclude that there is no


autocorrelation of residuals. We also perform the autocorrelation test of
residuals’ deviations and since there are no significant spikes of ACFs and
PACFs, it means that the residuals of the selected ARIMA model are white
noise, and no other significant patterns are left in the time series. Therefore,
there is no need to consider any AR (p) and MA (q) further.

The model (1, 1, 2) can be written as follows:

955

PFIZERt = ‒ 0,029924t-1 ‒ 0,058546 εt-2 + εt. (3)

The subsections below describe the processes of Monte Carlo simulation
development for Pfizer Inc.

4.2. Monte Carlo simulations for forecasting stock prices of Pfizer Inc.

The starting MC model for calculating the input data is presented and
explained earlier (2):

ΔS =μSΔt + σSϵ√Δt

Data on expected annual growth and standard deviation are determined based
on historical prices. We considered a period of 30 years (from 1985 to 2015)
with an average annual return for that period of 16.80% (μ = 0.16794) and a
standard deviation of 33.66% (σ = 0.33665).

The price change is observed at the daily level as the results of the simulation
are compared with real prices within one month period of time. For this model,
it is assumed that Pfizer stocks are traded actively for 254 days a year, so one
year is divided by the number of days in order to get the required parameter.
Thus, Δt = 1/254 or 0.00394, while the second root from Δt is 0.06275.

The real price (S) on 10th March was $ 30.5, so the model can be written as:

ΔS = 0.16794 · S · 0.00394 + 0.33665 · S · ϵ · 0.06275 (4)

According to the Geometric Brownian motion on which the MC simulation


is based, the model thus presented means that the price of a stock follows
a random walk. This is consistent with a weak form of the efficient market
hypothesis: past price information has already been incorporated and
subsequent price movements are conditionally independent from past price
movements.

After 1000 simulations of the daily prices, the expected value was calculated
and the result obtained is the estimated daily stock price based on the MC
model. The final result is visible in Table 2, in a comparative view with the
results of the ARIMA model predictions.

956
Table 2. Prices obtained through the forecast of two models in relation
to the real prices of the stocks of Pfizer Inc.

Sample period Actual values Predicted values ARIMA Predicted values MC


1 30.10 29.40483 30.55675
2 29.54 29.50121 30.56519
3 29.04 29.41259 30.55577
4 29.34 29.49407 30.51950
5 29.45 29.41915 30.49679
6 30.07 29.48804 30.50409
7 30.38 29.42470 30.52727
8 30.19 29.48294 30.46389
9 30.08 29.42939 30.49685
10 29.78 29.47863 30.48413
11 30.05 29.43336 30.55115
12 30.07 29.47498 30.52850
13 29.64 29.43671 30.48524
14 30.04 29.47190 30.53086
15 30.72 29.43954 30.71015
16 31.36 29.46929 30.77961
17 32.93 29.44194 30.76318
18 32.76 29.46709 30.79568
19 32.5 29.44397 30.79217
20 31.89 29.46523 30.93653
21 31.96 29.44568 31.01251
Source: Authors’ research

We compare the ex post one week ahead forecasts of stock prices using three
different evaluation statistics to ensure that our inferences regarding the
relative efficiency of the forecasting models are not driven by the particular
criterion used in these comparisons. The statistics are the Root Mean Squared
Error (RMSE), the Mean Absolute Error (MAE) and Mean Absolute Percent
Error (MAPE).

957
Table 3. Comparative ARIMA and MC model statistics
for different periods of time for Pfizer Inc.

Time period
Model
5 days 10 days 15 days 21 day
The mean absolute error (MAE)
ARIMA model 0.25829 0.44877 0.51675 1.16276
MC model 1.04480 0.72002 0.63372 0.848865
Mean absolute percent error (MAPE)
ARIMA model 0.87074 1.49505 1.71510 3.67968
MC model 3.55643 2.43679 2.13855 2.74831
The root mean squared error (RMSE)
ARIMA model 0.36007 0.53977 0.61381 1.60025
MC model 1.09947 0.83659 0.749033 1.024226
Advantage ARIMA ARIMA ARIMA MC
Source: Authors’ research

In Table 3 we see that the statistics for the ARIMA model for periods of 5, 10
and 15 days are much lower than the same indicators for the MC simulation.
This means that the ARIMA model gives results closer to the real ones. If the
total period of 21 days is observed, MC simulation gives more precise results.

4.3. Comparative ARIMA model and Monte Carlo simulation model statistics

Similarly, the corresponding ARIMA models and MC models were selected


for the remaining 4 stocks and the results of the comparisons are given in
Tables 4 to 7.

Table 4. Comparative ARIMA and MC model statistics for


different periods of time for Coca Cola

Coca Cola

ARIMA model COCACOLAt = 29.94265 + 0.999573 COCACOLAt-1 + 0.998660εt-1 + εt.
MC model ΔS = 0.15943 · S · 0.00394 + 0.23938 · S · ϵ · 0.06275
Time horizon
5 days 10 days 15 days 21 days
The mean absolute error (MAE)
ARIMA 0.868828 1.004986 1.338585 1.596852
MC 0.699725 0.784929 1.056698 1.219699

958
Mean absolute percent error (MAPE)
ARIMA 0.019147 0.022095 0.029117 0.034532
MC 0.015418 0.017257 0.022981 0.026378
The root mean squared error (RMSE)
ARIMA 0.901527 1.029633 1.444867 1.714948
MC 0.734412 0.807169 1.148668 1.310382
Advantage MC MC MC MC
Source: Authors’ research

Table 5. Comparative ARIMA and MC model statistics for different periods of time for
General Electric

General Electric

ARIMA model GEt = 27.23002 + 0.998697 GE t-1 − 0.231098εt-1 + εt.
MC model ΔS = 0.14525 · S · 0.00394 + 0.25661 · S · ϵ · 0.06275
Time horizon
5 days 10 days 15 days 21 days
The mean absolute error (MAE)
ARIMA 0.353942 0.686105 1.020895 0.980553
MC 0.387337 0.64592 0.9259 0.814732
Mean absolute percent error (MAPE)
ARIMA 0.011556 0.022173 0.032544 0.031376
MC 0.012662 0.020895 0.029543 0.026057
The root mean squared error (RMSE)
ARIMA 0.455237 0.790311 1.175119 1.103603
MC 0.467556 0.720884 1.044524 0.935381
Advantage ARIMA MC MC MC
Source: Authors’ research

Table 6. Comparative ARIMA and MC model statistics for


different periods of time for Exxon

Exxon

ARIMA model EXXONt = 79.03694 + 0.992641 EXXON t-1 − 0.152387εt-1 + εt.
MC model ΔS = 0.13012 · S · 0.00394 + 0.15147 · S · ϵ · 0.06275
Time horizon
5 days 10 days 15 days 21 days

959
The mean absolute error (MAE)
ARIMA 0.51880 0.95413 1.11439 0.86722
MC 0.53117 0.92719 0.94768 0.79938
Mean absolute percent error (MAPE)
ARIMA 0.62197 1.13829 1.32680 1.30333
MC 0.63548 1.10566 1.12809 0.95426
The root mean squared error (RMSE)
ARIMA 0.71518 1.12048 1.28306 0.98519
MC 0.78263 1.10219 1.11170 0.99400
Advantage ARIMA ARIMA ARIMA MC
Source: Authors’ research

Table 7. Comparative ARIMA and MC model statistics for


different periods of time for 3M

3M

ARIMA model 3Mt = 109.4349 + 0.999529 3M t-1 − 0.062219εt-1 + εt.
MC model ΔS = 0.13764 · S · 0.00394 + 0.19119 · S · ϵ · 0.06275
Time horizon
5 days 10 days 15 days 21 days
The mean absolute error (MAE)
ARIMA 2.71791 3.88420 4.83041 5.47981
MC 2.25786 3.11321 3.81109 4.06980
Mean absolute percent error (MAPE)
ARIMA 1.67007 2.36747 2.92442 3.30713
MC 1.38759 1.89775 2.30753 2.45718
The root mean squared error (RMSE)
ARIMA 2.81926 4.09706 5.16102 5.78750
MC 2.32453 3.27595 4.06431 4.26339
Advantage MC MC MC MC
Source: Authors’ research

Based on selected criterions we find that ARIMA model has advantage in 3


out of 5 stocks in sample in shortest time horizon of 5 days. It is interesting
that advantage of ARIMA model disappears with longer time horizons, and in
the end MC model has advantage over ARIMA model in all five cases for the
longest time horizon of 21 days.

960
4.4. Discussion of results

In order to get an insight of effectiveness of Monte Carlo simulation model


and ARIMA model in predicting stock prices, we forecasted stock prices of
five stocks from DJIA in next 5, 10, 15 and 21 days from 10th March 2016
using the mentioned models. The main idea was to test how selected methods
that use different input data in predicting stock prices, including different
time horizons are effective in predicting stock prices. While ARIMA model
uses time series data to predict future prices, Monte Carlo simulation uses
random walk component in predicting prices. The table 8 summarizes the
advantages of applied models in predicting stock prices.

Table 8. Advantages of applied models for different time horizons

TIME HORIZON
Company
5 DAYS 10 DAYS 15 DAYS 21 DAYS
Pfizer Inc. ARIMA ARIMA ARIMA MC
Coca-Cola MC MC MC MC
General Electric ARIMA MC MC MC
Exxon ARIMA ARIMA ARIMA MC
3M MC MC MC MC
Source: Authors’ research

Based on the results of the predicting power of both models, we can conclude
that ARIMA model has same advantage in predicting stock prices in shorter
periods of time (5 days), while there is an undoubted advantage on side of
MC model for longer periods of time (21 days). Also, we noticed linkages
between single stocks predictions in a way that there is only one change in
advantage from one method to another, and that advantage of MC method is
increasing with the lengthening of the time.

The persistence in quality of predictions of one or other method leads us to the


possible generalization of results. It is especially interesting that MC can be
seen as method of better quality in predicting stock prices with time lag of 21
days. MC simulation model is formed to describe random walk pattern, and
weak form market efficiency assume that future securities’ prices are random
and not influenced by past events. Is it possible that assumption about day’s
stock prices reflect all the data of past prices is being satisfied with some time
lag? We assume that market time lag appears due to information imperfection

961
or some other process of market information absorption, after what they are
being included in the market price.

5. Conclusion

This paper analyzes the effectiveness of Monte Carlo simulations and ARIMA
model in predicting stock prices. Our results show that in shorter period stock
prices have elements of predictability according to historical price, what questions
the level of efficiency of the world leading stock exchange. In longer periods (21
days) results indicate weak-form market efficiency, i.e. model with random walk
component (Monte Carlo simulation) has advantage compared to the model based
on historical prices (ARIMA). A combination of these findings could provide
useful information to investors when to buy or sell stocks and Monte Carlo
simulation could be used to determine which stock to sell short. We consider
that this research can be extend in two ways: (1) for possible generalization of
obtained conclusions about advantages of one or the other method in predicting
stock prices for which we suggest the expansion of sample and (2) further analysis
of time lag needed to absorb new information in market and its’ inclusion in the
stock price as according to MC model which is based on random walk principle.

References

1) Ariyo, A. A., Adewumi, A. O. & Ayo, C. K. (2014) Stock price prediction using
the ARIMA model. In Al-Dabass, D., Orsoni, A., Cant, R., Yunus, J., Ibrahim,
Z. & Saad, I. (Eds.), 2014 UKSim-AMSS 16th International Conference on
Computer Modelling and Simulation (pp. 106-112). Cambridge: IEE.
2) Bahovec, V. & Erjavec, N. (2009) Uvod u ekonometrijsku analizu. Zagreb: Element.
3) Borkar, M., Cevher, V. & McClellan, J. H. (2007) Decentralized state initialization
with delay compensation for multi-modal sensor networks. The Journal of VLSI Signal
Processing Systems for Signal, Image, and Video Technology. 48(1-2), pp. 109-125.
4) Box, G. E. P. & Jenkins, G. M. (1970). Times series Analysis Forecasting and
Control. San Francisco: Holden-Day.
5) Boyle, P. P. (1976) Options: A Monte Carlo Approach. Journal of Financial
Economics. 4, pp. 323-338.
6) Clewlow, L. & Strickland, C. (1998) Pricing interest rate exotics by Monte Carlo
simulation. In Dupire B. (Ed.), Monte Carlo: Methodologies and Applications
for Pricing and Risk Management. London: Risk Books.
7) Crnjac-Milić, D. & Masle, D. (2013) Mogućnost primjene Monte Carlo metode
na primjeru agroekonomskog problema prilikom donošenja odluka u uvjetima
rizika. Ekonomski vjesnik. XXVI(1), pp. 309-313.

962
8) Dai, Y. & Zhang, Y. (2013) Machine learning in stock price trend forecasting.
Available at: http://cs229.stanford.edu/proj2013/DaiZhang-MachineLearningIn
StockPriceTrendForecasting.pdf (Accessed: September 5, 2017).
9) Dupernex, S. (2007) Why might share prices follow a random walk?. Student
Economic Review. 21, pp. 167-179
10) Eckhardt, R. (1987) Stan Ulam, John von Neumann and the Monte Carlo
method. Los Alamos Science. 15, pp. 131-136.
11) Hoesli, M., Jani, E. & Bender, A. (2005) Monte Carlo Simulations for Real
Estate Valuation. Research Paper No. 148. Available at: https://ssrn.com/
abstract=770766 (Accessed: September 25, 2017).
12) Hull, J. (2012) Options, futures and other derivatives. Boston: Pearson Education.
13) Jabbour, G. M. & Liu, Y. K. (2005) Option Pricing and Monte Carlo Simulations.
Journal of Business & Economics Research. 3(9), pp. 1-6.
14) Kyungjoo, L., Sehwan, Y. & John, J. J. (2007) Neural network model vs.
SARIMA model in forecasting Korean stock price index. Issues in Information
Systems. 8(2), pp. 372-378.
15) Landauskas, M. (2011) Modelling Of Stock Prices By The Markov Chain Monte
Carlo Method. Intellectual Economics. 5(10), pp. 244–256.
16) Malkiel, B. G. & Fama, E. F. (1970) Efficient capital markets: A review of theory
and empirical work. The Journal of Finance. 25(2), pp. 383-417.
17) Merh, N., Saxena, V. P. & Pardasani, K. R. (2010) A comparison between hybrid
approaches of ANN and ARIMA for Indian stock trend forecasting. Business
Intelligence Journal. 3(2), pp. 23-43.
18) Mondal, P., Shit, L. & Goswami, S. (2014) Study of effectiveness of time
series modeling (ARIMA) in forecasting stock prices. International Journal of
Computer Science, Engineering and Applications. 4(2), pp. 13-29.
19) Render, B., Stair, R. M. & Hanna, M. E. (2012) Quantitative Analysis for
Management. New York: Pearson.
20) Sterba, J. & Hilovska, K. (2010). The implementation of hybrid ARIMA
neural network prediction model for aggregate water consumption prediction.
Aplimat—Journal of Applied Mathematics. 3(3), pp. 123-131.
21) Tadeu, H. F. B., Silva, J. T. M., Malafaia, G. C. & de Azevedo, D. B. (2012)
Brazilian airport infrastructure: Analysis using the Monte Carlo simulation and
multiple regressions. African Journal of Business Management. 6(15), pp. 52-68.
22) Whiteside, J. D. (2008) A Practical Application of Monte Carlo Simulation in
Forecasting. AACE International transactions EST.04. Available at: http://www.
icoste.org/AACE2008%20Papers/Toronto_est04.pdf (Accessed: September 25,
2017).
23) Xu, S. Y. & Berkely, C. U. (2014) Stock price forecasting using information from
Yahoo Finance and Google trend. Available at: https://www.econ.berkeley.edu/
sites/default/files/Selene%20Yue%20Xu.pdf (Accessed: September 5, 2017).

963
View publication stats

You might also like