See discussions, stats, and author profiles for this publication at: [Link]

net/publication/320886731

An automatic stock trading system using Particle Swarm Optimization

Conference Paper · June 2017

DOI: 10.1109/ECTICon.2017.8096283

CITATIONS READS

4 506

4 authors, including:

Chukiat Worasucheep Sirapop Nuannimnoi

King Mongkut's University of Technology Thonburi Academia Sinica
20 PUBLICATIONS   106 CITATIONS    5 PUBLICATIONS   5 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Automatic financial trading algorithms View project

All content following this page was uploaded by Sirapop Nuannimnoi on 11 July 2018.

The user has requested enhancement of the downloaded file.

The 14th International Conference ECTI-CON 2017, Phuket, Thailand, 27-30 June 2017

An Automatic Stock Trading System using

Particle Swarm Optimization
Chukiat Worasucheep, Sirapop Nuannimnoi, Ratchanon Khamvichit, Papon Attagonwantana
Applied Computer Science Program, Department of Mathematics
Faculty of Science, King Mongkut’s University of Technology Thonburi, Thailand

Abstract—This paper proposes a trading strategy based on a

learning method to combine a set of technical trading signals. The
II. LITERATURE REVIEW
learning employs a modified Particle Swarm Optimization to Over the past twenty years, many academic research have
optimize the weights of signals. The set of weighted signals is then been conducted to develop or improve technical trading systems.
used to determine trading decisions, i.e. to buy, to sell or to hold. A Many of them apply computational intelligence techniques to
trading simulation is conducted using historical daily stock prices identify optimal trading rules [3]. A genetic programming is
of twenty stocks from NYSE and SET markets. The performance used by Allen and Karjalainen [4] for trading daily S&P500
is evaluated using the return on investment with the testing subset prices from 1928 to 1995, however without a consistent beat on
of such data. The results are compared with buy-and-hold strategy the buy-and-hold (B&H) strategy. Núñez-Letamendia [5]
and the signal follow strategy of each individual signal. demonstrates the robustness of genetic algorithms-tuned trading
Keywords : Particle Swarm Optimization, Trading strategy, strategy on 25 stocks from the Madrid Stock Exchange. Kwon
Technical indicators and Moon [6] presents a hybrid neurogenetic system with a
context based ensemble method and tested on 36 stocks in NYSE
I. INTRODUCTION
and NASDAQ, showing on average a notable improvement over
Algorithmic trading has gained a very high interest in the the B&H strategy. Massimiliano [3] proposes a multi-objective
domain of computational finance for several decades. Generally, genetic algorithm, boosting and statistical learning methods
it is very difficult to for an investor to make trading decisions together for trading strategy of S&P 500 index. Genetic network
without financial advices [1]. Recently, in order to support them, programming with sarsa learning is employed for efficient stock
many techniques have been proposed and used to either predict trading decision-making using technical indicators and
prices or discover trading rules based on technical analysis candlestick charts [7]. A multi-objective particle swarm
approach. A number of technical indicators are used for optimization is proposed for end-of-day historical stock trading
generating trading signals (buy, sell or hold) [2]. Unfortunately [8]. The system optimizes the weights of several technical
they often create inconsistent signals for a particular stock. Many indicators over two objective functions, i.e. percent profit and
computational algorithms have been developed to apply these Sharpe ratio. More recent research survey can be reached from
technical indicators for automatic trading of stocks or other Yong [Link]. [9] in 2015 which provided a systematic review on
financial data, most of them employ genetic programming, the state-of-the-art application of evolutionary computation
genetic algorithm or similar algorithms [3]-[9]. techniques for rule discovery in stock algorithmic trading.
To resolve the trading signal inconsistencies, this paper
proposes a new trading strategy built from signals generated III. THE PROPOSED ALGORITHM
from technical indicators. The proposed algorithm weights a set This section briefly explains the proposed trading algorithm,
of signals, each of which can be to buy, to sell, or to hold, based namely PSOTrader. The required inputs are daily closing prices,
on an evolutionary learning from historical data. The weighted open prices, volumes, high prices and low prices of the stock in
signal set then will be used as trading strategy. The algorithm consideration. The steps of the algorithm are described below.
requires daily closing, opening, highest and lowest prices and 1. Nine chosen technical trading signals are calculated from
volumes of a stock to be traded as inputs. These inputs are used the inputs for each historical data record.
to calculate widely used technical indicators and generate 2. The vectors of trading signals and closing prices are fed into
trading signals. Then a modified Particle Swarm Optimization the optimization process, in which our modified PSO
(PSO) algorithm [10] is used to optimize the weights of optimizes the weights for all signals.
indicators for an optimal trading strategy. PSO is a well-known
swarm intelligence technique that is based on the social behavior 3. The weighted signals are summed up to determine a trading
of a group of computing agents [11]. Finally, the rates of return decision for each day.
from the proposed trading strategy are compared with the ones 4. The return of the investment, as a percentage of final
from the buy-and-hold (B&H) strategy which is widely used by portfolio value per initial portfolio value, is then calculated
most passive mutual funds, and from each indicator’s signal- for performance comparison.
follow trading strategy. A. Trading Signals from Technical Indicators
The remaining of this paper are as follows. Section II reviews There are many technical indicators developed and widely
some recent trading systems based on technical analysis. Section used, but they often provide inconsistent trading signals. In this
III describes the proposed algorithm. Section IV explains the work, the indicators below are considered for the creation of
experimental setups as well as results and discussion. At last, trading signal vectors [2], which the algorithm uses to calculate
Section V concludes this paper with some possible future works. the weights for each indicator’s signal.

978-1-5386-0449-6/17/$31.00 ©2017 IEEE

505
The 14th International Conference ECTI-CON 2017, Phuket, Thailand, 27-30 June 2017

1. Simple Moving Average (SMA). Here, we use 100 days [12]. In that version of PSO, the cognitive coefficient ( ) and
moving average. When the closing price moves above the the social coefficient ( ) were set to vary linearly from the
100-day SMA, a buy signal is generated. A sell signal is beginning to the end. Parameter inertia weight factor also
given when price moves below the SMA line. decreases linearly with time.
2. Two moving averages are commonly used for “crossover” C. Trading Decision
buy or sell indications. Three pairs of (20, 50), (50, 100),
and (20, 100) are chosen to be our crossovers. The buy During the simulation process, the from equation
signal happens when the shorter line moves above the (1) is used to determine whether to buy, to sell or to hold the
longer line, and the sell signal occurs when vice versa. particular stock on that day with the conditions below.
3. Moving Average Convergence Divergence (MACD). The − If is higher than , then buy.
sell signal occurs when the MACD signal line moves − If is lower than − , then sell.
above the zero line, while the buy signal occurs when the − If is in the range of (− , ), hold the position.
line moves below the zero line.
Parameter is the decision threshold value for making buy/sell
4. Relative Strength Index (RSI) value ranges in 0 to 100. decision and ∈ (-0.1, 1.0). If approaches 1.0, the algorithm
Movements above 70 are considered overbought and suggests buying. In contrast, if approaches -1.0, the algorithm
provides sell signal, while an oversold condition giving a suggests selling.
buy signal would be a move below 30.
5. Stochastic Oscillator (STO) help identify trend velocity Later in the end of each simulation process, the return of
and movement. Sell signal occurs when the %D line moves investment calculated from return = (final portfolio value -
above value 80, while the buy signal occurs when the %D initial portfolio value) / initial portfolio value is used as PSO’s
moves below value 20. objective function value. That is the goal is to find the optimal
set of weights for the particular stock that gives the highest return
6. Commodities Channel Index (CCI) identifies new trend or on investment during the considered period.
detect risk condition. CCI over +100 considers a sell signal
while CCI below -100 considers a buy signal. IV. EXPERIMENTATION
7. William Percent Range (%R) identifies the reversal point. To verify the performance of the proposed algorithm, ten
Readings over 80 or under 20 identify overbought (sell highly active stocks from different industries in NYSE, and ten
signal) and oversold (buy signal), respectively. stocks from The Stock Exchange of Thailand (SET) are used in
B. Optimization with PSO an experiment. Historical data of daily stock prices from 2015 to
2016 (about 505 days or records) are downloaded from
Particle Swarm Optimization (PSO) is a swarm intelligence [Link] for used in the simulation. These 20 stocks
technique developed for solving an optimization problem of real have different trends within the tested period; some are sideway,
numbers [10]. A swarm consists of a set of particles, each of while others are downward or upward.
which represents a possible solution. Each particle consists of a
position vector xi = (xi1, xi2,..., xiD) and a velocity vector vi = (vi1, A. Parameters
vi2,..., viD) where D is dimension or the number of decision In this experiment, parameters of PSO are set as widely used.
variables to be optimized. PSO algorithm searches for the and are set to be changed through time from 2.5 to 0.5, and
optimal solution in an iterative manner. In a common practice, from 0.5 to 2.5, respectively [12]. The inertia weight, , is
in each iteration the velocity of each particle is updated by using decreased linearly from 0.9 to 0.4 for more exploitation ability
the latest velocity, the cognitive information of the particle and at the beginning, and then more exploration ability at the end.
the social information from the whole swarm. Then the updated We set the number of particles at 20 and 25 to investigate the
velocity is used for changing the position of such particle. This difference, and the maximum number of allowable objective
iterative updates move the swarm to the optimal solution of the function calls at 150,000 per run as stopping criterion.
problem. Algorithmic details of PSO can be obtained from
literature [10][11][12]. B. Performance Comparison
For this problem, the vector of weights for trading signals is After the algorithm learns from the training dataset, the set
represented in the PSO as the position vector of each particle. In of optimized weights are used for trading with the testing dataset.
this work, we use PSO to search for optimal values of the Then the return on investment during the testing period is
weights from the decision equation shown below. calculated as described in subsection III.C. The obtained return
on investment is compared to those from the simple buy-and-
=( + + ⋯ + )/ ∑ …. (1) hold strategy (B&H), and from the signal-follow strategy for
each individual indicator’s. The signal-follow strategy means
where trading signals , , …, are generated from each that trading decision (buy, sell or hold) is determined by the only
indicator as described in subsection A. , , …, are indicator in consideration. Here in this experiment, we compare
weights for each corresponding signal and n is the number of with the following most common trading signals. SMA100,
trading signals in use. These weights are to be optimized by a EMA20/50, EMA20/100, EMA50/100, MACD, RSI, STO, CCI
modified PSO and thus they are the decision variables of the and Williams %R. The buy and sell signals are determined as
optimization problem. Each determines the trading discusses in subsection III.A.
decision for day , and its meaning is described in the next
subsection. The PSO used in this paper was modified from a The experiment also studies the effects of some parameters
version with time-varying acceleration coefficients proposed in in the proposed PSOTrader algorithm. Swarm size of PSO

506
The 14th International Conference ECTI-CON 2017, Phuket, Thailand, 27-30 June 2017

generally has some effects on the optimization result and is V. CONCLUSION AND FUTURE WORKS
normally set between 20 and 30. So we tests the swarm size (n) Many technical trading signals are not very reliable and
of both 20 and 25. From subsection III.C, parameter (decision sometimes even conflicting. This paper proposes a trading
threshold) affects the frequency of trading. More frequent strategy that learns to follow an optimally-weighted set of
trading may suffer higher cost of trading commission fee. In this trading signals instead of following only one signal. The optimal
experiment, we study 3 different values of : 0.1, 0.3 and 0.5. weights are obtained from a learning using a modified PSO
The initial investment is set to 1,000,000 baht for simulating algorithm. Then, the decision to either buy, sell or hold the
each stock. The trading commission fee is set at 0.2% of every current volume of stocks is determined using those weighted
trading amount of both buying and selling orders. For each signals. The experimental results demonstrate that the proposed
buying order, a quarter amount of the whole money is used to algorithm clearly outperforms both buy-and-hold strategy and
buy the stock. For each selling order, a quarter volume of stocks all individual-signal following strategies in terms of return on
(if bought using the whole amount) is sold. This follows the investment.
basic concept of gradual investment, not buying the stock in just
one time using all the money. Possible directions for future works include adaptive setting
of the buying and selling proportion for each trading, further
For each stock, proportion of training data subset and testing
study of other technical signals for a more profitable trading
data subset is 90:10, meaning that the first 90% of dataset is used
decision, and improved optimization algorithms for optimizing
for modeling the PSO trading equation (1), which is then used
the weights. Risk on investment such as the Sharpe ratio or the
for testing the performance with the remaining 10% dataset. For
maximum drawdown, defined as the maximum loss from a peak
other trading strategies, the trading signals are calculated for the
to a trough of a portfolio before a new peak is attained, should
whole dataset, but the trading is simulated only with the 10%
also be considered in the experiment. Also different trading
testing subset, for a fair comparison.
period length (other than 2 years) should be tested.
C. Results and Discussions
REFERENCES
Table I and Table II report returns on investments for stocks
[1] J.M. Berutich, F. Lopez, F. Luna, and D. Quintana. “Robust Technical
in NYSE and SET respectively, from trading by using each Trading Strategies Using GP for Algorithmic Portfolio Selection,” Expert
strategy. The average returns (for all stocks) are summarized in Systems with Applications, vol. 46, 2016, pp. 307-315.
Table III for the sake of clarity. Excluding our proposed [2] S. Leigh. Essential Technical Analysis: Tools and Techniques to Spot
algorithm, SMA100 signal provides the lowest returns on Market Trends. New York, NY: Wiley, 2002.
average while Williams %R provides the highest returns on [3] K. Massimiliano, “Investment using evolutionary learning methods and
average. Thus also shown in Table I and Table II are the numbers technical rules,” European Journal of Operational Research, vol. 207,
of buying and selling executed by using SMA100 and Williams December 2010, pp. 1717-1727.
%R signals as well as the proposed PSOTrader algorithm, for [4] F. Allen, R. Karjalainen, “Using genetic algorithms to find technical
investigation. trading rules,” Journal of Financial Economics, vol. 51, pp. 245–271.
[5] L. Núñez-Letamendia, “Fitting the control parameters of a genetic
As seen in the tables, trading signals are not reliable to every algorithm: an application to technical trading systems design,” European
stock. In some stocks, some trading signals work well, whereas Journal of Operational Research, vol. 179, pp. 847–868.
in other stocks, those trading signals underperform. The [6] Y.K. Kwon, B.R. Moon, “A hybrid neurogenetic approach for stock
performance of most trading signals for NYSE stocks and for forecasting,” IEEE Transactions on Neural Networks, vol. 18, 2007, pp.
SET stocks are mixed. 851–864.
From Table III, on average, B&H strategy performs badly (- [7] Y. Chen, et al., “A genetic network programming with learning approach
0.95%), but SMA100 performs the worst (-1.01%) for this set of for enhanced stock trading model,” Expert Syst. Appl., vol. 36, 2009,
pp.12537–12546.
tested stocks. The best performer is our PSOTrader with swarm
[8] A. C. Briza, Prospero C. Naval Jr., “Stock trading system based on the
size = 25 and = 0.1, yielding a profit of 3.81%. A simulation multi-objective particle swarm optimization of technical indicators on
with swarm size = 25 works better than 20 for every value. end-of-day market data,” Applied Soft Computing, vol. 11, January 2011,
The lower value, the more frequently the algorithm trades pp. 1191-1201.
(buys and sells) as seen in Table I. This takes more chance of [9] H. Yong, K. Liu, X. Zhang, L. Su, E.w.t. Ngai, and M. Liu. "Application
trading that is enough to earn a higher profit, as seen in Table III. of Evolutionary Computation for Rule Discovery in Stock Algorithmic
Trading: A Literature Review." Applied Soft Computing, vol.36, 2015,
However a high frequency of trading does not guarantee a pp. 534-51.
high profitability. Look at the cases of SMA100 and Williams [10] J. Kennedy and R. Eberhart, “Particle swarm optimization,” In IEEE
%R. SMA100 makes more frequent trading than Williams %R International Conference on Proceedings Neural Networks, Nov-Dec.
does, but SMA100 gives less returns. The reason should be that 1995, vol. 4, pp. 1942–1948.
both employ only one technical signal which is not reliable. In [11] Y. del Valle, G.K. Venayagamoorthy, S., Mohagheghi, J.-C. Hernandez,
contrast, the trading decision in our PSOTrader depends upon a R.G. Harley, “Particle Swarm Optimization: Basic Concepts, Variants and
Applications in Power Systems,” IEEE Trans. on Evol. Comp, vol. 12, pp.
combined set of technical signals rather than only one signal. 171 – 195, April 2008.
The proposed strategy successfully learns the optimal weights [12] A. Ratnaweera, S.K. Halgamuge, and H.C. Watson. “Self-Organizing
for each component signal. The obtained weighted signal thus Hierarchical Particle Swarm Optimizer With Time-Varying Acceleration
provides a more efficient trading strategy and results in the Coefficients,” IEEE Transactions on Evolutionary Computation, vol. 8,
highest profit on average. 2004, pp. 240-255.

507
The 14th International Conference ECTI-CON 2017, Phuket, Thailand, 27-30 June 2017

TABLE I THE RETURNS ON INVESTMENTS FOR THE NYSE STOCKS

Strategy AAPL AMZN CVS GM GOOGL HSBC KO PG TSLA YHOO
B&H -1.51% -8.66% -9.05% 8.97% -4.52% 5.83% -1.80% -2.10% 4.56% -9.88%
SMA100 -0.90% -0.58% 0.13% -3.16% -0.09% 1.80% -1.25% -0.33% 0.76% -1.98%
#Buy/#Sell 9/9 12/11 13/13 14/14 11/11 9/8 14/13 16/15 12/12 12/12
EMA20/50 0.61% -0.19% -2.83% 1.03% 0.10% 1.26% 0.26% -0.89% 1.14% 0.91%
EMA20/100 0.36% -0.13% -3.10% 1.44% 0.41% 1.22% 0.24% 0.42% -1.38% -0.40%
EMA50/100 0.31% -0.79% -2.62% 1.40% -0.03% 1.02% 0.27% -0.95% 1.30% -0.56%
Signal follow MACD 0.20% -2.60% 2.26% -0.10% 0.12% -1.72% 0.59% -0.24% 2.41% -1.45%
RSI 3.36% 0.00% 0.00% 0.00% 1.58% 3.70% 0.00% 0.00% 0.00% -5.03%
STO -0.31% -3.64% -3.76% 4.23% 3.85% 1.67% 0.43% -0.78% 4.16% -2.77%
CCI 0.00% -0.90% 0.39% 1.38% 0.00% 0.00% 0.21% -0.04% 3.34% 0.12%
Williams %R 0.30% 0.36% -3.14% 5.70% 2.81% 2.28% 2.21% 2.20% 10.59% 0.76%
#Buy/#Sell 4/4 5/2 4/4 5/3 5/4 5/2 4/3 5/3 5/5 7/3
PSOTrader0.1 Training 2.68% 8.35% 0.60% 7.47% 3.43% 0.06% 2.85% 0.26% 7.51% 1.40%
n = 20 Testing 8.35% -0.68% 1.46% 16.23% 4.77% 4.96% 2.46% -0.66% 0.59% -4.72%
#Buy/#Sell 14/14 11/8 11/11 11/11 8/8 11/5 14/13 23/17 14/14 16/11
PSOTrader0.1 Training 2.13% 18.25% 0.31% 8.30% 3.18% -2.47% 2.23% -2.75% 6.96% 0.15%
n = 25 Testing 10.18% -0.40% -2.07% 3.66% 3.90% 3.70% 3.89% 0.06% 19.38% -6.62%
#Buy/#Sell 14/14 14/14 12/12 11/10 11/9 9/3 12/11 9/8 12/12 13/4
PSOTrader0.3 Training 0.51% 16.33% 0.13% 11.04% 2.92% 0.05% 0.52% 1.51% 4.99% 0.69%
n = 20 Testing 2.24% -0.63% -1.54% -0.44% 0.75% 1.22% 0.50% 0.69% 12.72% 0.57%
#Buy/#Sell 5/2 3/1 4/1 1/1 5/5 2/1 6/3 2/0 6/4 2/1
PSOTrader0.3 Training 1.67% 0.23% 0.00% 10.06% 2.40% 0.99% -0.61% -0.05% 6.07% 1.39%
n = 25 Testing 1.50% -0.73% -2.50% -0.52% 1.65% 2.87% 0.16% 1.79% 13.36% -0.25%
#Buy/#Sell 3/3 3/3 5/4 2/2 3/3 2/2 4/3 10/6 6/4 3/1
PSOTrader0.5 Training 0.00% 0.94% 1.21% 7.35% 0.00% 1.16% -0.54% 1.51% 0.49% 0.00%
n = 20 Testing 0.00% 0.06% -1.38% 2.26% 0.00% 0.00% -0.01% 0.25% 7.21% 0.00%
#Buy/#Sell 0/0 3/3 1/0 3/1 0/0 0/0 3/2 1/1 4/2 0/0
PSOTrader0.5 Training 0.00% 0.00% 0.00% 3.54% 0.00% -0.57% -0.61% 1.51% 4.98% 0.00%
n = 25 Testing 0.00% -2.61% 0.00% 0.00% 0.00% 0.24% -0.03% 0.25% 9.57% 0.00%
#Buy/#Sell 0/0 1/0 0/0 0/0 0/0 1/1 2/0 1/1 3/1 0/0

TABLE II THE RETURNS ON INVESTMENTS FOR THE SET STOCKS

Strategy ADVANC AOT BANPU BEM CPALL KBANK KTB PTT SCC TRUE
B&H -6.45% 2.96% -0.40% -3.01% 1.22% 0.17% -0.40% 5.27% -2.75% 2.47%
SMA100 -2.02% -0.57% -6.13% 0.00% -2.04% -0.51% 0.37% 0.18% -1.62% -2.28%
#Buy/#Sell 12/12 11/10 14/13 16/15 13/12 9/9 12/12 12/12 11/11 15/15
EMA20/50 -1.80% -1.51% -2.43% -0.35% -0.84% -0.37% 0.53% 0.01% -0.85% 0.28%
EMA20/100 0.00% -0.60% 0.00% -0.74% 0.00% 0.44% 0.62% 1.10% 0.27% -0.23%
EMA50/100 -1.93% -0.09% 1.10% -0.21% -1.64% -0.76% 0.29% -0.09% -1.91% 0.73%
Signal follow MACD -1.42% 0.88% -2.93% -1.72% 0.79% -0.29% -2.12% -1.28% -0.94% -2.34%
RSI -2.00% 5.98% 2.00% 0.00% 0.00% 4.83% 2.49% 7.77% 0.69% 1.28%
STO 1.31% 8.29% 2.27% -0.65% 1.81% 0.32% 2.69% 2.33% 3.95% 0.22%
CCI -0.38% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 1.62% 2.63% 0.00%
Williams %R -2.31% 6.13% 3.05% 1.56% 3.79% 2.00% 2.61% 2.70% 2.88% 1.28%
#Buy/#Sell 4/2 7/5 3/1 5/1 5/3 4/4 5/4 3/3 5/2 3/0
PSOTrader0.1 Training 1.44% -0.31% 9.91% 10.57% -1.37% 1.58% 0.18% -3.05% 1.81% -0.43%
n = 20 Testing 1.05% -2.11% -0.76% -0.70% 4.83% 7.96% 5.69% 6.95% 6.22% 1.91%
#Buy/#Sell 16/5 8/6 17/9 11/1 21/15 21/15 19/13 18/18 14/8 12/7
PSOTrader0.1 Training 1.53% -2.32% 9.91% 7.55% 1.49% 3.48% 0.44% -2.07% 1.30% -3.53%
n = 25 Testing -0.81% 6.58% -0.76% -0.98% 3.88% 7.85% 6.31% 8.88% 7.58% 1.88%
#Buy/#Sell 9/3 12/10 17/9 13/5 16/12 15/10 22/15 17/13 16/8 9/3
PSOTrader0.3 Training 4.47% -2.36% 2.13% 9.62% 2.57% 3.56% 1.62% 1.48% 0.00% 2.35%
n = 20 Testing 3.70% 3.13% 2.20% -0.57% 1.56% 2.30% 3.45% 0.69% 2.63% 2.63%
#Buy/#Sell 7/1 5/2 4/1 10/0 3/1 6/4 7/6 3/3 3/1 6/2
PSOTrader0.3 Training 2.54% -3.85% 0.00% 7.86% 2.10% 4.12% 0.88% 1.06% 1.53% 1.37%
n = 25 Testing 1.42% 10.61% 2.08% 2.31% -0.55% 4.96% 3.73% 2.25% 3.32% 2.25%
#Buy/#Sell 5/1 14/10 2/0 7/1 3/2 6/2 11/8 6/5 12/7 5/0
PSOTrader0.5 Training 1.16% 0.00% 0.00% 0.00% 3.07% 0.49% 0.00% 0.92% 0.00% 0.00%
n = 20 Testing 1.86% 0.00% 0.00% 0.00% 0.15% 0.00% 0.93% 1.98% 0.00% 0.00%
#Buy/#Sell 3/0 0/0 0/0 0/0 1/0 0/0 1/1 4/3 0/0 0/0
PSOTrader0.5 Training 1.48% 0.00% 0.00% -1.28% 4.09% 0.58% 3.49% 1.05% 0.00% 1.81%
n = 25 Testing 0.00% 0.00% 0.00% 0.00% -0.05% 3.01% 4.71% 0.00% 0.66% 1.28%
#Buy/#Sell 0/0 0/0 0/0 0/0 2/1 3/0 7/4 0/0 1/0 3/0

TABLE III THE AVERAGE RETURNS

NYSE SET NYSE SET
Strategy Average Strategy Average
Average Average Average Average
B&H -0.95% -1.82% -0.09% PSOTrader0.1 Training 2.75% 3.46% 2.03%
SMA100 -1.01% -0.56% -1.46% n = 20 Testing 3.19% 3.28% 3.10%
EMA20/50 -0.30% 0.14% -0.73% PSOTrader0.1 Training 2.70% 3.63% 1.78%
EMA20/100 0.00% -0.09% 0.08% n = 25 Testing 3.81% 3.57% 4.04%
EMA50/100 -0.26% -0.06% -0.45% PSOTrader0.3 Training 3.21% 3.87% 2.54%
Signal follow MACD -0.59% -0.05% -1.14% n = 20 Testing 1.89% 1.61% 2.17%
RSI 1.33% 0.36% 2.30% PSOTrader0.3 Training 1.99% 2.22% 1.76%
Stoch 1.28% 0.31% 2.25% n = 25 Testing 2.49% 1.73% 3.24%
CCI 0.42% 0.45% 0.39% PSOTrader0.5 Training 0.89% 1.21% 0.56%
Wiiliams %R 2.39% 2.41% 2.37% n = 20 Testing 0.67% 0.84% 0.49%
PSOTrader0.5 Training 1.00% 0.88% 1.12%
n = 25 Testing 0.85% 0.74% 0.96%

508

View publication stats